WIDE BANDGAP SEMICONDUCTORS
This page intentionally left blank
WIDE BANDGAP SEMICONDUCTORS Growth, Processing and A...
90 downloads
1146 Views
24MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
WIDE BANDGAP SEMICONDUCTORS
This page intentionally left blank
WIDE BANDGAP SEMICONDUCTORS Growth, Processing and Applications
Edited by
Stephen J. Pearton University of Florida Gainesville, Florida
NOYES PUBLICATIONS Park Ridge, New Jersey, U.S.A. WILLIAM ANDREW PUBLISHING, LLC Norwich, New York, U.S.A.
Copyright 9 2000 by Noyes Publications No part of this book may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording or by any information storage and retrieval system, without permission in writing from the Publisher. Library of Congress Catalog Card Number: ISBN: 0-8155-1439-5 Printed in the United States Published in the United States of America by Noyes Publications / William Andrew Publishing, LLC Norwich, New York, U.S.A. 1098765432
1
Library of Congress Cataloging-in-Publication Data Pearton, S. J. Processing of wide bandgap semiconductors / by Stephen J. Pearton. p. cm. Includes bibliographical references. ISBN 0-8155-1439-5 1. Semiconductors--Design and construction. 2. Wide gap semiconductors. Compound semiconductors. I. Title. TK7871.85.P395 621.3815'2--dc21
2000 00-027325 CIP
MATERIALS SCIENCE AND PROCESS TECHNOLOGY SERIES
Editors Rointan F. Bunshah, University of California, Los Angeles (Series Editor) Gary E. McGuire, Microelectronics Center of North Carolina (Series Editor) Stephen M. Rossnagel, IBM Thomas J. Watson Research Center (Consulting Editor)
Electronic Materials and Process Technology CHARACTERIZATION OF SEMICONDUCTOR MATERIALS, Volume 1: edited by Gary E. McGuire CHEMICAL VAPOR DEPOSITION FOR MICROELECTRONICS: by Arthur Sherman CHEMICALVAPOR DEPOSITION OF TUNGSTEN ANDTUNGSTEN SlLICIDES: by John E. J. Schmitz CHEMISTRY OF SUPERCONDUCTOR MATERIALS: edited by Terrell A. Vanderah CONTACTS TO SEMICONDUCTORS: edited by Leonard J. Brillson DIAMOND CHEMICAL VAPOR DEPOSITION: by Huimin Liu and David S. Dandy DIAMOND FILMS AND COATINGS: edited by Robert F. Davis DIFFUSlON PHENOMENA IN THIN FILMSAND MICROELECTRONIC MATERIALS: edited by Devendra Gupta and Paul S. Ho ELECTROCHEMISTRY OF SEMICONDUCTORS AND ELECTRONICS: edited by John McHardy and Frank Ludwig ELECTRODEPOSlTION: by Jack W. Dini HANDBOOK OF CARBON, GRAPHITE, DIAMONDS AND FULLERENES: by Hugh O. Pierson HANDBOOK OF CHEMICAL VAPOR DEPOSITION, Second Edition: by Hugh O. Pierson HANDBOOK OF COMPOUND SEMICONDUCTORS: edited by Paul H. Holloway and Gary E. McGuire HANDBOOK OF CONTAMINATION CONTROL IN MICROELECTRONICS:edited by Donald L. Tolliver HANDBOOK OF DEPOSITION TECHNOLOGIES FOR FILMS AND COATINGS, Second Edition: edited by Rointan F. Bunshah HANDBOOK OF ION BEAM PROCESSING TECHNOLOGY: edited by Jerome J. Cuomo, Stephen M. Rossnagel, and Harold R. Kaufman HANDBOOK OF MAGNETO-OPTICAL DATA RECORDING: edited by Terry McDaniel and Randall H. Victora HANDBOOK OF MULTILEVEL METALLIZATION FOR INTEGRATED CIRCUITS: edited by Syd R. Wilson, Clarence J. Tracy, and John L. Freeman, Jr. HANDBOOK OF PLASMA PROCESSING TECHNOLOGY: edited by Stephen M. Rossnagel, Jerome J. Cuomo, and William D. Westwood HANDBOOK OF POLYMER COATINGS FOR ELECTRONICS, 2nd Edition: by James Licari and Laura A. Hughes HANDBOOK OF REFRACTORY CARBIDES AND NITRIDES: by Hugh O. Pierson HANDBOOK OF SEMICONDUCTOR SILICON TECHNOLOGY: edited by William C. O'Mara, Robert B. Herring, and Lee P. Hunt
vi
Series
HANDBOOKOF SEMICONDUCTORWAFER CLEANINGTECHNOLOGY:edited byWemer Kem HANDBOOK OF SPUTTER DEPOSITION TECHNOLOGY: by Kiyotaka Wasa and Shigeru Hayakawa HANDBOOK OF THIN FILM DEPOSITION PROCESSESAND TECHNIQUES: edited by Klaus K. Schuegraf HANDBOOK OF VACUUM ARC SCIENCE AND TECHNOLOGY: edited by Raymond L. Boxman, Philip J. Martin, and David M. Sanders HANDBOOK OF VLSl MICROLITHOGRAPHY: edited by William B. Glendinning and John N. Helbert HIGH DENSITY PLASMA SOURCES: edited by Oleg A. Popov HYBRID MICROCIRCUITTECHNOLOGY HANDBOOK, Second Edition: by James J. Licari and Leonard R. Enlow IONIZED-CLUSTER BEAM DEPOSITION AND EPITAXY: by Toshinori Takagi MOLECULAR BEAM EPITAXY: edited by Robin F. C. Farrow SEMICONDUCTOR MATERIALS AND PROCESS TECHNOLOGY HANDBOOK: edited by Gary E. McGuire ULTRA-FINE PARTICLES: edited by Chikara Hayashi, R. Ueda and A. Tasaki
Ceramic and Other Materials--Processing and Technology ADVANCED CERAMIC PROCESSINGAND TECHNOLOGY,Volume 1:edited by Jon G. P. Binner CEMENTED TUNGSTEN CARBIDES: by Gopal S. Upadhyaya CERAMIC CUTTING TOOLS: edited by E. Dow Whitney CERAMIC FILMS AND COATINGS: edited by John B. Wachtman and Richard A. Haber CORROSION OF GLASS, CERAMICS AND CERAMIC SUPERCONDUCTORS: edited by David E. Clark and Bruce K. Zoitos FIBER REINFORCED CERAMIC COMPOSITES: edited by K. S. Mazdiyasni FRICTION AND WEAR TRANSITIONS OF MATERIALS: by Peter J. Blau HANDBOOK OF CERAMIC GRINDING AND POLISHING: edited by loan D. Mavinescu, Hans K. Tonshoff, and Ichiro Inasaki HANDBOOK OF INDUSTRIAL REFRACTORIES TECHNOLOGY: by Stephen C. Carniglia and Gordon L. Barna SHOCK WAVES FOR INDUSTRIAL APPLICATIONS: edited by Lawrence E. Murr SOL-GEL TECHNOLOGY FOR THIN FILMS, FIBERS, PREFORMS, ELECTRONICS AND SPECIALTY SHAPES: edited by Lisa C. Klein SOL-GEL SILICA: by Larry L. Hench SPECIAL MELTING AND PROCESSING TECHNOLOGIES: edited by G. K. Bhat SUPERCRITICAL FLUID CLEANING: edited by John McHardy and Samuel P. Sawan
Other Related Titles HANDBOOK OF PHYSICAL VAPOR DEPOSITION (PVD) PROCESSING: by Donald M. Mattox
Preface
There has been a resurgence of interest in wide bandgap semiconductors in recent times, for two important classes of applications, namely blue/green light emitters and high power/high temperature electronics. For the first set of applications, ZnSe and GaN are leading the race, due to their direct bandgaps which give them a huge advantage over the indirect gap SiC. Nichia Chemical Industries of Japan, led by the efforts of Shuji Nakamura, announced its first blue GaN light-emitting diode product in 1993, followed by quantum well blue and green devices in 1995/1996. Currently, more than 10 million blue LEDs are sold per month. Attention has turned to work on high brightness (15 lumens per watt) white LEDs using a blue LED to excite a yttrium aluminum garnet phosphor. These may have applications in whitelighting situations, with the advantage ofmuch longer lifetimes than incandescent bulbs. Nichia has also demonstrated continuous-wave blue-violet laser diodes with room-temperature lifetimes acceptable for commercial applications. The active layer in these devices is InGaN with A1GaN cladding layers. The LEDs are useful for full-color displays, whereas the main commercial laser application is high density optical storage on CD-ROMs. A notable feature of current GaN light-emitter technology is the fact that all devices are currently grown heteroepitaxially on A1203, SiC, or magnesium aluminate substrates. The resulting high defect density (109-10 ~~cm 2) does not appear to affect the light output of LEDs, but can
io
VII
viii
Preface
cause problems in laser diodes through the migration of the p-contact metal along dislocations which may short-out the p-n junction. For this reason, there have appeared some novel lateral overgrowth techniques on SiO 2 patterned GaN templates that produce defect-flee regions above the masked areas. The active regions of the laser diodes are then processed in these areas, with the result that their lifetimes under high current operation are much longer than in devices grown in a blanket (non-patterned) fashion. While II-VI based laser diodes were the first to be demonstrated, their lifetimes are currently in the several hundred hours range and are limited by the ease of defect generation and migration in these soft materials. A typical laser diode structure consists of ZnSe/ZnMgSSe/ZnSSe/ZnCdSe/ ZnMgSSe/ZnSe layers grown on GaAs substrates by Molecular Beam Epitaxy. In this materials system, Metal Organic Chemical Vapor Deposition lags somewhat, due to poorer control over precursor purity and in-situ thickness control. This is in contrast to the GaN system, where MOCVD seems to have an advantage over MBE for photonic devices because of higher quality material due to the higher growth temperature. For the second major class of applications, high power electronics, SiC is by far the most mature, with diamond and GaN as other candidates. Diamond actually has the most appropriate material parameters, but problems with producing large single crystals and lack ofn-type dopability have retarded its progress. GaN has the advantage of the availability of heterostructures and excellent transport properties, but has relatively poor thermal conductivity. SiC has excellent thermal conductivity, demonstrated breakdown voltages of several kV and more well-developed substrates and device processing techniques. To date, the highest rfpower (850 W per mm at 850 MHz CW) was demonstrated by a 4H-SiC MESFET, and the highest total power (450 W pulsed at 600 MHz) produced by a SiC static induction transistor. A SiC power module containing four SITs has demonstrated a 1 kW capability at 600 MHz. The basic driving force is the requirement for electronics in adtomobiles, aircraft, and ships that can function directly on engines to lower the weight and cost of control functions. As an example, it is estimated that approximately 800 pounds could be eliminated on an F16 fighter jet if current mechanical, hydraulic, and pneumatic systems were replaced with advanced power electronics. Si-based electronics is limited to -- 100~ for reliability reasons, requiring active cooling systems. NASA and other agencies have needs for advanced electronics capable of operation at 600~ for temperatures X2T_.g
.
.
.
t
t
.
ff Plasma". Ha mode: ~4NI 110 W~ 7-10 -7 torr
~1
1st pos. series
300
400
500
600
I
700
800
900
IIl~
1000
1100
wavelength [nm]
Figure 12. Emission spectra of radio-frequency nitrogen plasma in the high brightness mode (upper part) and the low brightness mode (lowerpart) (from Ref. 56).
20
Wide Bandgap Semiconductors
Although no comparable emission spectrum of the DC plasma is available, the fact that this plasma source is operated at much lower power than an RF source (15 W compared to 150-300 W for an RF source) makes it reasonable to assume that the DC plasma does not contain any atomic nitrogen, and the much lower doping levels are a result of doping with excited nitrogen molecules. Two properties of RF plasma doped layers give additional insight in the processes occurring during compensation. First, p-ZnSe doped with RF plasma is rather unstable against annealing, and even anneals below growth temperature lead to a considerable increase in resistance,[ 57] approaching that of layers doped with DC plasma. Thus, it seems that the highly conductive "RF" state can be converted into the resistive "DC" state by a diffusion process. Second, the free carrier concentration in RF plasma doped layers decreases drastically if the atomic nitrogen flux is increased. This behaviour can be seen in Fig. 13, where the p-doping levels obtained by different groups[56][58] (solid symbols) are plotted as a function of the total nitrogen content incorporated in the layer. The variation of the atomic nitrogen flux was obtained by an increase of the RF plasma power used. For low nitrogen incorporation, the nitrogen was nearly completely electrically activated. When the total amount of nitrogen was increased, the free hole concentration first saturates and then drops abruptly for very high nitrogen content. In analogy to annealing, an excess of atomic nitrogen was another possibility to induce a transition from a highly conductive to a resistive state. All these experimental findings can be understood in a model that assumes the existence of two nitrogen complexes which contain one or two nitrogen atoms, respectively.[ 59] Many microscopic models of such centers have been proposed in the literature.[56116~-[62]We assume that the second one is energetically more stable and is formed automatically when nitrogen molecules are the doping species. However, its formation is kinetically limited as soon as atomic nitrogen is offered, since the nitrogen atoms first have to find each other by diffusion. It is now plausible that annealing has such a drastic effect on conductivity, and also helps to interpret the data shown in Fig. 13. As soon as too much atomic nitrogen is offered, the probability for the formation of nitrogen pairs increases. To quantify this latter point, the data in Fig. 13 have been modelled by a Monte Carlo simulation of the nitrogen pair formation process via diffusion (open symbols in Fig. 13, the solid line is a guide for the eye). This simulation assumes a nitrogen sublattice with fixed lattice points, and an average lattice constant determined by the absolute nitrogen content. One incoming nitrogen atom is then moved randomly for a fixed number of
Doping Limits and Bandgap Engineering
21
steps, and the probability of meeting another nitrogen atom is determined by averaging over 100 diffusing atoms at each given composition. This probability is then translated into a free carrier concentration by the assumption that for each nitrogen pair, two acceptors are lost, and a double donor is formed. The number of diffusion steps is calibrated by the condition that an absolute nitrogen concentration of 1018 cm "3 should correspond to a free hole concentration of 3.1017 cm "3 (an experimental finding of much of the research,[421156][58])and is then kept constant for all other concentrations. Under these assumptions, the experimental features, and especially the abrupt drop at high nitrogen content, can be exactly reproduced.
1018 . "J
}
-!
~1017. tO
i,
~
.
~1016O
r,.) ~1015
Q
Kurtzetal.
II
Ohkawa et al.
121
Monte Carlo
1014 .
t
1016
1017
1018
1019
Nitrogen Concentration ( cm -3 ) Figure 13. Free hole concentration in ZnSe doped with RF nitrogen plasma as a function of the total nitrogen concentration (solid symbols, Refs. 56 and 58) together with results of a Monte Carlo simulation (open symbols). The solid line is a guide for the eye.
22
Wide Bandgap Semiconductors
These ideas can be taken together with the Fermi level pinning model to explain the different behaviour for doping with RF and DC plasma, as shown in Fig. 11. As long as atomic nitrogen in a moderate concentration is offered, the compensating defect is the complex involving one nitrogen atom, which pins the Fermi level at a value of 120 meV above the ZnSe valence band edge. This pinning is responsible for the decreasing doping levels in ZnMgSSe claddings. If activated nitrogen molecules are offered, the energetically more favorable complex involving two nitrogen atoms becomes dominant and pins the Fermi level 580 meV above the ZnSe valence band edge. As in the case of n-doping, it is again of interest to compare these pinning positions with those occurring in III-V compounds. From the limited obtainable p-doping in InP with its relatively low valence band edge, one can estimate a pinning position about 500 meV below the GaAs valence band edge in these materials.[ 34] Under the assumption of 1.1 eV for GaAs/ZnSe[ 52] (as in the case of n-doping), this would correspond to a position about 600 meV above the ZnSe valence band edge. This is much higher than for RF plasma doping in II-VI compounds. From this, it may be concluded that the compensation mechanisms are very different in nature between III-V's and RF plasma doped II-VI's. This is not too astonishing if one considers the very special compensation situation which is encountered in nitrogen doping of II-VI compounds, as it has been described above. However, the value for the pinning level is close to what is observed for DC plasma doped II-VI compounds, suggesting that the compensation mechanism occurring in this case is closer to what is observed in III-V' s. As an important consequence, the compensation picture given here implies that the complex formed during RF plasma doping is metastable and can be converted into the stable configuration formed during DC doping by an annealing process. While this metastability does not seem to be a significant problem for the conductivity of the p-layers in a laser structure during operation, it has important consequences on the formation of Ohmic contacts to ZnSe, a point to which we will return in Ch. 5.
3.5
Competing Doping Limitation Mechanisms
Obviously, there exist upper doping limits which are determined exclusively by the positions of the band edges. However, this should not be misinterpreted in a way that Fermi level pinning is the only limiting
Doping Limits and Bandgap Engineering
23
mechanism. One obvious example of a competing mechanism is limited solubility of the dopant. This is the reason for the saturation observed in the n-doping level of CdMgTe:I visible in Fig. 8, as well as an analogous saturation in the p-doping level of ZnSeTe seen in Fig. 12. A similar mechanism is the formation of stable competing phases of the dopant with one of the constituents of the host. An example of this is the strong decrease in free hole concentration if Mg is added to ZnTe, which cannot be due to Fermi level pinning, since both ZnTe and MgTe have very high valence band edges. This behaviour has been explained with the formation energy of magnesium nitride, which is much higher than that of MgTe[ 63] and favors the formation of the nitride instead of the substitutional incorporation. Such a mechanism may also be present in selenides, where it has been shown that an addition of Mg to ZnSSe led to a slight decrease in the p-doping level[ 64] (although this may also be att~buted to an increase in the S sticking coefficient due to the presence of Mg and the resulting lowering of the valence bond edge). An analogous explanation has been given also for the relatively poor free hole concentrations in nitrogen doped CdTe.[ 63] The size of the dopant with respect to the host lattice may also be of importance in some cases. The high n-doping of CdMgTe plotted in Fig. 9, for example, is only obtained with the large halogen Iodine, which fits relatively well to the large CdMgTe lattice. If the halogens bromine or chlorine are used, the resulting doping levels are much lower, the lowest values resulting for the small chlorine ion.[47] Such effects may well be due to the formation of DX-like lattice deformations,[231which are more likely for small dopants. In a real crystal, the pinning mechanism competes with all these other limitations. In this context, the main value of the pinning model is that it allows a prediction of the maximum obtainable doping level if all competing mechanisms can be overcome.
4.0
D O P I N G AND BAND S T R U C T U R E E N G I N E E R I N G
4.1
Surface Segregation
It is evident from Fig. 11 that the key factor for the improvement of p-doping levels is to raise the position of the valence band edge. Since, in an MBE process, the incorporation of a dopant occurs at the surface of the growing crystal, the formation of compensating centers should rather be
24
Wide Bandgap Semiconductors
determined by the conditions at the surface than by those in the bulk. Consequently, it could be sufficient to increase the valence band edge at the surface in order to obtain a low degree of compensation. One possible way to do this is to take advantage of surface segregation. Segregation usually occurs if atoms of very different size are offered simultaneously, so that it becomes energetically favorable for the larger atom to occupy surface sites where more space is available than in the bulk. In the case of ZnSeTe, this effect is so strong that nearly no tellurium is incorporated during MBE growth. [65] In analogy, strong segregation can be expected if sulfur, as a very small atom, is offered together with the much larger tellurium. In the ease of ZnSTe (or ZnMgSTe) this combination offers, in addition, the possibility to be lattice matched to GaAs. In Fig. 14, the free hole concentration of MBE grown ZnSTe layers determined by CV profiling is plotted versus the bulk Te content of the layers (solid symbols).[ 66] The layers were grown using elementary Zn, S, and Te sources. The obtained carrier concentration is nearly independent of the bulk Te content, whereas the Fermi level pinning model would predict a very strong dependence on Te content (solid line, with a large error in Te content, since the dependence of the valence band on the Te content can only be estimated). This picture changes if one does not plot the bulk Te content, which is determined by x-ray diffraction, but the surface Te content, measured by x-ray photoemission spectroscopy for some selected layers (open symbols, the arrows connect identical samples). This surface Te content is always much higher than the bulk content and is above 20% even for layers with a bulk Te content as low as 5%. It must be noted that this strong segregation effect does not occur if a ZnS source is used instead of the elementary S source. As a consequence, the surface data points now lie close to the calculation, except for high Te content, where they are considerably lower. Since, in this range, the carrier concentration is already close to the 1019 em -3 range, this is probably an effect of limited solubility, in analogy to the ease of ZnSe/ZnTe superlattiees (see Fig. 11). Thus, surface segregation and the shift of the surface valence band edge associated to it allow an increase of the doping level. Unfortunately, there is only a limited potential for applications of this effect due to two negative by-products of the segregation. First, it causes a vertical gradient in composition which can be observed as a strong broadening of x-ray reflexes. Second, the large potential fluctuations associated with these inhomogeneities seem to have a tendency to trap carriers, so that transport finally occurs only via hopping.[ 66] Nevertheless, the conductivity of material with a carrier concentration close to 1019 em -3 is high enough to be useful for p-contacts, as will be shown in Ch. 5.
Doping Limits and Bandgap Engineering
25
10 20
10 = o
19
l
1018
l
l
i
.p,,~
t-d
10
17
o
~
Calculation I"! Volume Te Content (x-ray) I Surface Te Content (XPS)
10 z6
o
10~5 0
10
20 30 Te Content in ZnSTe
40
Figure 14. Free hole concentration of RF-plasma doped ZnSTe as a function of the bulk Te content (solid symbols) and the surface Te content (open symbols). The solid line is a calculation based on the Fermi level pinning model.
4.2
Superlattices
A second attractive possibility to influence the valence band position is the use of superlattices instead of mixed crystals. In that case, the positions of the first conduction and valence minibands that determine the superlattice energy gap can be shifted up and down by simply changing the superlattice period. Since the effective hole masses are much larger than those of electrons, confinement effects are much more pronounced in the conduction than in the valence band, a situation that is in favor of a higher valence band edge as compared to a mixed crystal with a comparable energy gap. That this really affects the doping properties is demonstrated in Fig. 15, where the free hole concentration of ZnSe/ZnTe superlattices doped with DC nitrogen plasma is plotted versus the superlattice period.[ 3~ The average Te content is 15% for all samples. An increase of the period from 2.4 to 5 nm leads to a strong increase of the hole concentration from 2.1016cm -3 to 3.1019cm"3. The solid line is the
Wide Bandgap Semiconductors
26
calculated carrier concentration according to our model for miniband positions determined from a Kronig Penney calculation of ideally abrupt superlattices, with heavy hole masses of 0.78 for ZnSe and 0.6 for ZnTe. In fact, this simple model reproduces the large increase in carrier concentration, except for very small periods, where, due to interdiffusion, the dopability approaches that of a mixed crystal of the same composition.
10 20
r
'~ 10
ZnTe Layer Thickness (ML) 1,5 2,0 2,5
1,0
19
1
3,0
Calculation Data
o 18 "'~ 10 I
o 017 =O 1 r,.)
l--
--
I
/
I
Mixed
O
~
10 ~6 2
3 4 5 ZnSe/ZnTe SL Period ( nm )
6
Figure 15. Free hole concentration ofZnSe/ZnTe superlattices with an average Te content of 15% as a function of the superlattice period. The solid line is the result of a model r the dashed line indicates the doping level obtained for a ZnSeTe mixed crystal with 15% Te.
Figure 16 gives the results of an analogous calculation for two examples of RF plasma doped superlattices, namely (MgS)n/(ZnSe)n and (Zn0.sMg0.sS)3J(ZnTe)n, two combinations that are reasonably well lattice matched to GaAs. The former has the advantage that it uses the same components as the quaternary ZnMgSSe, the latter contains ZnTe with its very high valence band edge (which in turn leads to a high heavy hole superlattice miniband). For comparison, the calculation and the data[ 16]for ZnMgSSe mixed crystals are also given in Fig. 16. The dotted part of the curve for (Zno.sMg0.sS)an/(ZnTe)n represents the region where the ZnTe
Doping Limits and Bandgap Engineering
27
thickness is lower than 2 ML, and the calculation is expected to be inaccurate due to interdiffusion.
I
g"" 10 Is
"
o
!
!
!
'
.... ZnMgSfZnTe . . . . . . MgS/ZnSe ZnMgSSe 9 ZnMgSSedata
\
o 9~
o 017 "~1
O
9
"
9 %~
% 9
"%
o
9 %
o 1016
9
9 %
9
9,
O O
% 9
1015
,
9 %
%
9149
.. 9
218
'
I
'
I
3.'0 3.2 3.4 Energy Gap ( eV )
'
I
3.6
Figure 16. Calculatedfreehole concentrationfor (MgS)n/(ZnSe)nand(Zno.sMgo.sS)3n/(ZnTe)n superlattices as a function of their energy gap. For comparison,calculated and experimental values for ZnMgSSe mixed crystals (Ref. 16) are also shown.
For MgS/ZnSe an increase of the hole concentration by a factor of five to ten compared to ZnMgSSe is expected. For cladding layers with an energy gap of 2.95 eV, as they are currently used in blue-green lasers, this would represent an increase of the hole concentration from 1 to 5.1017cm -3, which could help improve device performance. The predicted behavior of the Zn0.sMg0.sS/ZnTe superlattices is still more promising. Even combinations with an energy gap up to 3.2 eV are predicted to exhibit hole concentrations above 1017cm -3. For such a cladding material, the use of a ZnSe quantum well would no longer be unrealistic. Nevertheless, it has to be considered that the improvement in vertical transport, which is decisive for a laser device, will be somewhat smaller than the predicted increase in dopability. The carriers will have to tunnel through the barriers introduced by the superlattice structure. In addition, the conduction band edge of such a cladding layer would be
28
Wide Bandgap Semiconductors
already so high that the n-dopability is expected to be very bad. As a consequence, for a deep blue room temperature laser, we propose to use an asymmetric device design with a ZnMgSSe cladding on the n-side, and a Zn0.sMg0.sS/ZnTe cladding on the p-side. The schematic energy band diagram of such a proposed device is shown in Fig. 17. Since the dopability is directly connected to the band edge positions, the dopability scale given at the fight y-axis is directly related to the energy scale. As an additional feature, the use of ZnSTe, which should also allow high p-doping levels, is proposed as a lattice matched p-contact. Such asymmetric devices have already been realized, in principle, for the example ofZnMgSe/ZnSeTe LEDs. [67][68] n
...:. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
=
..........................
1014
........................................................................................................................... ..............................................................................................................................
10, r~
oo
oo
o
~2 >
c5
=1
0
N
c5
N
~
~._
N
i iiiii iiiiiiiiiiiiiii;iiiiiiiiiiiiiiiii'~ ".............................................................................................................................. 1 0 1 4
p
Figure 17. Schematic energy band diagram under fiat band conditions for a proposed asymmetric blue laser diode with a ZnMgS/ZnTe superlattice cladding and a ZnSe active zone.
Interesting new possibilities may arise from the use of the highly pdopable compound BeTe, which is nearly lattice matched to GaAs.[ 3] Superlattices between BeTe and ZnSe have already been produced in high quality. However, these superlattices are of the staggered type with an overlap of the two energy gaps of only about 1.9 eV.[ 3] They cannot be used as
Doping Limits and Bandgap Engineering
29
cladding layers in a laser. For this purpose, BeTe should be combined with a material with a much larger energy gap than ZnSe. The best candidate from a confinement and lattice match point of view is cubic MgS.[ 2] In fact, miniband calculations for such superlattiees reveal that they would be potential candidates for highly p-doped cladding layers with energy gaps in excess of 3.2 eV. However, the combination BeTe/MgS is somewhat exotic and may be difficult to realize. A more convenient altemative could be a ZnBeSe/ZnBeTe superlattice. The addition of Be to ZnSe reduces the lattice constant, while the addition of Zn to BeTe enlarges it, so that a combination of a selenide and a telluride with constant Zn to Be ratio, and equal thicknesses, is approximately stain balanced on GaAs. A Be content of about 50% is estimated to lead to approximately zero conduction band offset between the two materials, whereas the valence band of ZnBeSe is much lower than that of ZnBeTe. According to the miniband calculation, short period superlattices of this type can have energy gaps as large as 3.15 eV, if they can be fabricated without significant interdiffusion. At the same time, the valence band of these superlattices is more than 300 meV above that of ZnSe. The obtainable p-doping levels should consequently be limited only by solubility, and no longer by Fermi level pinning. Since it can expected that such a highly doped superlattice may be directly contacted with a metal, the valence band grading in this structure is not at the contact, but between the p-cladding and the p-waveguide.
5.0
OHMIC CONTACT TO
p-ZnSe
Soon after the introduction of nitrogen doping, it became evident that it was difficult to obtain good ohmic contacts to p-ZnSe. The reason for this was that the valence band of ZnSe is very low compared to the work function of even noble metals as Au or Pt. Consequently, a large Schottky barrier was formed between ZnSe and these metals. Additionally, the obtained doping levels in p-ZnSe remained limited to about 1018 cm "3, so that no effective tunneling through that barrier can lower the high contact resistance. Both problems are much less severe for ZnTe, which exhibits a higher valence band edge and much higher p-doping levels. As a result, several groups developed p-contacts to ZnSe that involve ZnTe. In order to reduce the problem of the large valence band offset between ZnTe and ZnSe, which acts as an additional barrier for holes, two
30
Wide Bandgap Semiconductors
strategies were followed. Fan et al. introduced a ZnSe/ZnTe superlattice with a quasi continuously varied Te content between the ZnSe and the ZnTe;[ 7] whereas Ishibashi et al. designed a series of ZnTe quantum wells in ZnSe which should allow sequential resonant tunneling of holes from the ZnTe through the quantum wells to the ZnSe.[ 81The decisive element for a good contact is believed to be an optimized grading in the first case and a precise design of the resonant tunneling structure in the second case. Finally, both approaches led to the development of ohmic contacts and allowed a reduction of the laser threshold voltage at room temperature to values well below 5 V[ 13] so that the p-contact problem for blue ZnSelasers seemed to be solved. New results, however, show that the contact properties are mainly governed by a diffusion process. This is demonstrated in Fig. 18, where IV characteristics of several ZnSe diodes are compared.[ 69] The contact used is a multi-quantum well structure similar to the one published in Ref. 8, followed by a ZnTe cap layer of 20 nm thickness. Based on the observation ofTaike et al., doping of the entire contact structure resulted in nonohmic contacts.[ 7~ The ZnTe quantum wells in the structure and the first 15 nm ofthe ZnTe cap were not doped. Only the topmost 5 nm of the ZnTe were heavily doped with nitrogen under the same plasma conditions as used for the p-ZnSe. This highly doped ZnTe layer was then contacted with gold which was evaporated, without breaking the vacuum, in a metallization chamber coupled to the MBE system. The only difference between the diodes was the substrate temperature at which the contact was grown; the temperature for the growth of the diode itself was 280~ for all samples. The solid curves in Fig. 18 represent diodes where the temperature for contact growth was set to 250, 230, and 200~ respectively. It was evident that a reduction of the growth temperature leads to a drastic improvement of the IV curve. This was a strong indication that the contact resistance was dominantly determined by a diffusion process and not, as commonly believed, by the exact design of the quantum wells in the grading or the resonant tunneling structure. A further reduction of the growth temperature to 160~ resulted in the deposition of polycrystalline ZnTe and a deterioration of the contact (see thin solid line in Fig. 18). The importance of diffusion is additionally underlined by the IV characteristics shown in Fig. 19. In this sample series, the growth temperature for all contacts was 200~ Only the thickness of the ZnTe cap was reduced from 10 to 5 nm until finally the cap was completely omitted. In
Doping Limits and Bandgap Engineering
31
order to guarantee a highly doped layer below the gold in these very thin contacts, in contrast to the samples shown in Fig. 19, the entire contact was doped with nitrogen. It can also be seen that the thickness of the ZnTe strongly affects the IV characteristics. For the contact where the ZnTe cap was completely omitted, the voltage which must be applied to obtain a current density of 600 A/cm 2 (a typical threshold current density for a ZnSe-based laser) was as low as 4 V, which was even lower than the best value reported for a similar contact structure.[ 13] This result can be also understood in terms of diffusion, if one assumes that the ZnTe cap acts either as a source or a sink for the diffusing species. A reduction of the ZnTe thickness limits the diffusion process by either reducing the amount of the species available for diffusion or the reservoir to which the diffusing species can migrate.
10,
,
'
" '
"'
'
'
~ZnTe
'!
'
I'
200 ~ ,
'
'
I'
160~
'
'
I'
230~
'
250~
8
"
n-ZnSe
4
Ii I ' : , 1
2
I! I ,I/j
0 I
-15
,
I
-10
.
I
-5
,
I
0
,
I
5
,
/I
i1
I
l
10
l
15
,,'1 i
I
20
i
I
25
.
l
30
Voltage ( V )
Figure 18. Current-density vs voltage for four ZnSe diodes with a ZnSe/ZnTe multiquantum well contact at the p-side. The structure of all diodes is identical, but the substrate temperature during the growth of the multi-quantum well contact was varied between 160 and 250~ The dotted lines represent results of a simulation.
32
Wide Bandgap Semiconductors
10 ,~
8
4
0 -2
0
2
4 6 Voltage ( V )
8
10
12
Figure 19. Current-density vs voltage for four ZnSe diodes with a ZnSe/ZnTe multiquantum well contact at the p-side. All contacts were grown at a substrate temperature of
200~ but the thickness of the overlyingZnTe cap layerwas varied between 0 and 10 nm. An additional experiment demonstrates that the diffusion induced deterioration of the contacts does not occur in the graded multi-quantum well itself, but rather in the underlying p-ZnSe. A diode was grown by MBE in which the p-ZnSe was replaced by p-ZnSTe, a mixed crystal which can by grown lattice matched on ZnSe for a Te concentration of 35% and doped p-type with nitrogen plasma.[661 This layer was followed by a ZnSTe stepgraded layer, consisting of five layers of 4 nm thickness for which the Te content was adjusted to 45%, 55%, 65%, 75%, and 85% by reducing the sulphur flux from the valved cracker sulphur source. This step-graded layer replaced the ZnSe/ZnTe multi-quantum wells used in the previous samples. The structure was terminated with a 20 nm ZnTe cap. The sample structure is shown as an insert at the left side of Fig. 20. Since high quality ZnSTe can only be grown at high substrate temperatures, the entire diode, as well as the contact, was grown at 280~ The IV characteristics of this sample are shown as a solid line in Fig. 20 Although it was grown at a high substrate temperature and with a relatively thick ZnTe cap, the IV curve is similar to the one of the best ZnSe diode. This diode was produced at a much lower substrate temperature and without a ZnTe cap to minimize
Doping Limits and Bandgap Engineering
33
diffusion. In the n-ZnSe/p-ZnSTe diode, the diffusion process observed in the pure ZnSe diodes was obviously absent or at least greatly reduced. To make sure that this improvement is not due to the replacement of the ZnSe/ ZnTe multi-quantum well by the ZnSTe grading layer, a second, nearly identical, sample was grown. As the only change, an additional p-ZnSe layer was introduced between the n-ZnSe and the p-ZnSTe. The sample structure is depicted as the right insert in Fig. 20. In contrast to the previous sample, the IV characteristics of this sample (dashed line in Fig. 20) was much worse (although still much better than for a pure ZnSe diode with a contact grown at such a high temperature), which means that interdiffusion is again important, in this case. From this we conclude that interdiffusion causes a deterioration of the p-ZnSe. On the other hand, if one succeeds in suppressing interdiffusion, the detailed design of the grading itself seems to be of minor importance, as can be seen from the fact that comparably good contacts can be achieved for gradings as different as a ZnSe/ZnTe multi-quantum well or a step-graded ZnSTe layer.
10 -
.
I
8
6 !
~
~[
I
n-ZnSe _ n-~~.~~~
/
I [ [
[.... p-ZnTe -1
/
iznsTe gradin~ i p,ZnS0.65Teo.351
; ,.
2 L)
0 ,
-2
I
0
,
I
2
,
I
,
I
4 6 Voltage ( V )
,
I
8
,
I
10
,
12
Figure 20. Current-density vs voltage for two diodes containing p-ZnSTe layers. The detailed structures are given in the inserts. The only difference between the two diodes is that the p-ZnSe layer was omitted in one case.
34
Wide Bandgap Semiconductors
The microscopic nature of the interdiffusion process that affects the p-ZnSe cannot be deduced directly from these experiments. However, additional insight can be gained by analyzing the IV characteristics in more detail. The low current for the contacts grown at 230 and 250~ can only be understood as a result of a tunneling process. In fact, a reasonable description of the IV curves can be obtained with a model for tunneling through a triangular barrier introduced by a highly resistive ZnSe layer of a given thickness. Although this model does not give a perfect fit of the observed IV curves, it describes at least qualitatively the observed shift of the experimental IV curves to higher voltages with increasing contact growth temperature (Fig. 18). Best fits are shown as dashed lines in Fig. 18. Although they are not unambiguous in the sense that both a variation of the barrier height and the free carrier density have a similar influence, one can nevertheless see clear trends. For a fixed barrier height of 0.5 eV (which is lower than that for gold on ZnSe,[ 71] but high enough to prevent significant thermionic emission), the thickness of the resistive layer used for the fit decreases from 145 nm for a substrate temperature of 250 ~ to 42 nm for a temperature of 200 ~ A change of the barrier height results in slight variations of these values, but the relative thickness change of the resistive layer remains constant. In Fig. 21, the thickness of the resistive layer obtained in this way is plotted as a function of temperature. These values are compared to a calculation giving the relative change of the diffusion length as a function of the temperature. For diffusion from a finite as well as for an infinite source, the diffusion length is proportional to the square root of the diffusion constant times the diffusion time [72] (which is approximately constant for all structures presented here). The diffusion constant D itself depends on temperature according to an Arrhenius law of the form: Eq. (1)
D = exp(-E a/kT)
where E a is the activation energy of the diffusion process and k the Boltzmann constant.[ 72] The relative change of the diffusion length obtained in this manner is plotted as solid lines in Fig. 21 with E a as a parameter. The comparison to the experimental values shows that the observed behavior can be understood for activation energies on the order Of 1 eV. This value is too low for diffusion via regular lattice sites, and corresponds rather to values observed for diffusion via interstitial sites or along dislocations.[ 72]
Dop&g Limits and Bandgap Engineering
35
160 140
1,o
~ 120
0,8
100 N
r~ ~,,J~
0,6
>
80
9 v,,,~
~
60
~
40
~
20
t~
~
0
*~,,I
o
0,4
o,2
0
190
200
210
220
230
240
250
0,0
Contact Growth Temperature ( ~ )
Figure 21. Thickness of a resistive layer, as obtained from a fit to the IV curves shown in Fig. 18, as a function of the temperature at which the contact was grown. The solid lines are calculated relative changes of the diffusion length with the activation energy E a as a parameter.
Based on these observations, our suggestion is that nitrogen diffuses along dislocations from the ZnTe cap and the grading into the p-ZnSe. The observation of such diffusion by secondary ion mass spectrometry (SIMS) has indeed been reported in the literature[ 7~ and could be driven by the different solubility of nitrogen in ZnSe and ZnTe.[ 2]] The free hole concentration in p-ZnSe would then be drastically reduced by the formation of nitrogen-nitrogen pairs (see See. 3.4 and Fig. 13) so that the diffusing nitrogen could lead to overcompensation in the p-ZnSe. These results can shed light on a few additional features of ZnSe/ ZnTe contacts. First, these contacts, and especially the resonant tunneling variant, seem not to be very reproducible. As an example, the device with a record lifetime of 101 hours reported by Ishibashi et al. needed a threshold voltage of 11 V[ 1~ (a value much larger than the lowest voltage of 4.7 V reported by the same group for a diode with a nominally identical
36
Wide Bandgap Semiconductors
contact).[ 13] This irreproducibility was still stronger when the results of different groups were compared. Several workers did not obtain ohmic contacts at all, although the same contact design and equivalent equipment was used.[ 7~ If the properties of the contact are determined by tunneling through a barrier created by nitrogen diffusion, even a slight variation of this barrier will have a strong influence on the IV characteristics, since the properties of the barrier appear exponentially in the tunneling probability. Thus, minor changes in growth conditions or composition of the plasma can lead to strong effects in IV characteristics. Second, aging experiments on laser diodes show that the threshold voltage tends to increase during operation.[ 73] This indicates that a contact degradation takes place during heavy duty operation. In the picture given here, this is an indication of nitrogen diffusion during device operation. Following the argument given above, even minor diffusion effects may be seen drastically in the IV curve. Although this diffusion during operation did not seem to be a factor that limited device lifetime, as long as the density of extended defects was high, it could become important for the structurally more perfect devices that are now available. Based on these observations, one can give some ideas how pcontacts to ZnSe can be further improved. A reduction of the dislocation density in the contact may help to suppress diffusion or at least to increase the activation energy of the diffusion process. In fact, the IV curve of the diode with a lattice mached ZnSTe spacer between the p-ZnSe and the graded contact (which is shown at the right side of Fig. 20) is better than that of a pure ZnSe diode grown at the same conditions. This indicates that the lattice matched spacer reduces interdiffusion. In that sense, the use of nearly lattice matched BeTe instead of ZnTe in contacts[ 3] may be an advantage. Completely avoiding nitrogen as a dopant in the contact region is still more promising. A first step towards this direction has been done by the use of p-Ge as a contact material.[ TM] An extension of this concept to graded ZnSe/Ge structures might even allow the realization of a lattice matched, nitrogen free contact, although significant difficulties for the epitaxy of ZnSe on Ge may be encountered in the practical realization of such a structure.
Doping Limits and Bandgap Engineering 6.0
37
CONCLUSIONS
Most experimental results on the maximum doping levels obtained in wide gap II-VI materials can be explained by a model that assumes a pinning of the Fermi level at a fixed position with respect to the vacuum level. In the case of n-type doping, this position is about 130 meV above the ZnSe conduction band edge for nearly all II-VI compounds. For pdoping, two different values (namely 120 meV and 580 meV above the ZnSe valence band edge) are obtained for doping with radio frequency and DC nitrogen plasma, respectively. This discrepancy can be understood by assuming the formation of complexes related to nitrogen-nitrogen pairs as the most stable compensating center, an assumption which also is able to describe the abrupt drop in the free hole concentration of RF plasma doped ZnSe if too much nitrogen is offered. In agreement with the model, an upward shift of the valence band edge through surface segregation, or the use of carefully designed superlattices, has the effect of increasing the obtained p-doping level. This fact can be used to gain more flexibility in the design of optoelectronic devices emitting in the blue spectral region. Finally, the formation of ohmic contacts to p-ZnSe is a process governed by diffusion. The results presented can be understood under the assumption that nitrogen diffuses along dislocations from the ZnTe cap into the underlying ZnSe. The excess of nitrogen leads to a highly resistive zone that must be overcome by a tunneling process and thus increases the operation voltages of ZnSe based diodes. This phenomenon may be reduced by the use of lattice matched contacts or contacts that do not involve nitrogen.
REFERENCES 1. Gunshor, R. L., Kobayashi, M., Kolodziejski, L. A., Otsuka, N., and Nurmikko, A. V., J. Crystal Growth, 99:390 (1990) 2. Okuyama, H., Nakano, K., Miyajima, T., and Akimoto, K., Jpn. J. Appl. Physics, 30:L1620 (1991) 3. Landwehr,G., and Waag, A., Proc. Int. Syrup. on Blue Lasers and LEDs, p. 17, Chiba (1996)
38
Wide Bandgap Semiconductors
4. Park, R. M., Troffer, M. B., Rouleau, C. M., DePuydt, J. M., and Haase, M. A., Appl. Phys. Lett. 57:2127 (1990) 5. Ohkawa, K., Karasawa, T., and Mitsuyu, T., J. Crystal Growth, 111:797 (1991) 6. Haase, M., Qui, J., DePuydt, J., and Cheng, H., Appl. Phys. Lett., 59" 1272 (1991) 7. Fan, Y., Han, J., He, L., Saraie, J., Gunshor, R. L., Hua, G. C., and Otsuka, N., Appl. Phys. Lett., 61:3160 (1992) 8. Ishibashi, A., and Mori, Y., J. Crystal Growth, 138:677 (1994) 9. Okuyama, H., Miyajima, T., Morinaga, Y., Hiei, F., Ozawa, M., and Akimoto, K., Electron. Lett., 28:1798 (1992) 10. Taniguchi, S., Hino, T., Itoh, S., Nakano, K., Nakayama, N., Ishibashi, A., and Ikeda, M., Electronics Letters, 32:552 (1996) 11. Albert, D., Olszowi, B., Spahn, W., Niimberger, J., Kom, M., Hock, V., Ehinger, M., Faschinger, W., Landwehr, G., J. Crystal Growth, 184/ 185:571(1998) 12. Kawasumi, T., Okuyama, H., Nakayama, N., Ishibashi, A., and Mori, Y., Electron. Lett., 31:1667 (1995) 13. Itoh, S., Nakayama, N., Matsumoto, S., Nagai, M., Nakano, K., Ozawa, M., Okuyama, H., Tomiya, S., Ohata, T., Ikeda, M., Ishibashi, A., and Mori, Y., Jpn. J. Appl. Phys, 33:L938 (1994) 14. Hua, G. C., Otsuka, N., Grillo, D. C., Fan, F., Han, J., Ringle, M. D., Gunshot, R. L., Hovinen, M., and Nurmikko, A. V., Appl. Phys. Lett., 65:1331 (1994) 15. Nurmikko, A. V., Jeon, H., Gunshor, R. L., and Han, J., J. Crystal Growth, 159:644 (1996) 16. Okuyama, H., Kishita, Y., Miyajima, T., and Ishibashi, A., Appl. Phys. Lett., 64:904 (1994) 17. Ishibashi, A., IEEE J. on Selected Topics in Quantum Electronics, 1:741 (1995) 18. Laks, D. B., Van de Walle, C. G., Neumark, G. F., and Pantelides, S. T., Phys. Rev. Lett., 66:648 (1991) 19. Van de Valle, C. G., Laks, D. B., Neumark, G. F., and Pantelides, S. T., Phys. Rev. B, 47:9425 (1993) 20. Van de Valle, C. G., Laks, D. B., Neumark, G. F., and Pantelides, S. T., Appl. Phys. Lett., 63:1375 (1993) 21. Fan. Y., Han. J., Gunshor. R. L., Brandt, M. S., Walker, J., Johnson, N. M., and Nurmikko, A. V., Appl. Phys. Lett., 65:1001 (1994) 22. Prior, K., Materials Science Forum, 182-184:11 (Trans Tech Publications), (1995)
Doping Limits and Bandgap Engineering
39
23. Chadi, D. J., Phys. Rev. Lett., 72:534 (1994) 24. Burkey, B. C., Khosla, R. P., Fischer, J. R., and Losee, D. L., J. Appl. Phys., 47:1095(1976) 25. Kachaturyan, K., Kaminska, M., Waber, E. R., Becla, P., and Street, R. A., Phys. Rev. B, 40:6304 (1989) 26. Terry, I., Penney, T., Molnar, S., Rigotty, J. M., and Becla, P., Solid State Commun., 84:235 (1992) 27. Han, J., Ringle, M. D., Fan, Y., Gunshor, R. L., and Nurmikko, A. V., AppL Phys. Lett., 65:3230 (1994) 28. Tao, I. W., Jurkowic, M., and Wang, W. I., Appl. Phys. Lett., 64:1848 (1994) 29. Ogawa, H., Irfan, G., Nakayama, H., Nishio, M., and Yoshida, A., Jpn. J. Appl. Phys., 33:L980 (1994) 30. Faschinger, W., Ferreira S., and Sitter, H., ,4ppl. Phys. Lett., 64:2682 (1994) 31. Ferreira, S., Sitter, H., and Faschinger, W., Appl. Phys. Lett., 66:1518 (1995) 32. Garcia, A., and Northrup, J., Phys. Rev. Lett., 74:1131 (1995) 33. Kurtz, E., PhD thesis, Univ. Wiirzburg, p. 32 (1996) 34. Walukiewicz, W., Materials Science Forum, 143-147:519 (1994) 35. Walukiewicz, W., Appl. Phys. Lett., 54:2094 (1989) 36. Walukiewicz, W.,J. Vac. Sci. Technol. B, 6:1257 (1988) 37. Ren, S. Y., Dow, J. D., and Shen, J., Phys. Rev. B, 38"10677 (1988) 38. Dow, J. D., Hong, R. D., Klemm, S., Ren, S. Y., Tsai, M. H., Sankey, O. F., and Kasowski, R. V., Phys. Rev. B., 43:4396 (1991) 39. Faschinger, W., Ferreira, S., Sitter, H., Krump, R., and Brunthaler, G., Materials Science Forum, 182-184:29, Trans Tech Publications, Switzerland (1995) 40. Marfaing, Y.,J. Vac. ScL Technol. B, 10:1444 (1992) 41. Marfaing, Y,J. Crystal Growth, 138, 305 (1994) 42. Qiu, J, DePuydt, J. M., Cheng, H., and Haase, M. A., ,4ppl. Phys. Lett., 59:2992 (1991) 43. Hauksson, S., Simpson, J., Wang, S. Y., Prior, K. A., and Cavenett, B. C., Appl. Phys. Lett., 61:2208 (1992) 44. Ferreira, S., Sitter, H., Faschinger, W., Krump, R., and Brunthaler, G., J. Crystal Growth, 146:418 (1994) 45. Rajakarunanayake, Y., Miles, R. H., Yu, G. Y., and McGill, T. C., Phys.Rev. B, 37:10212 (1988) 46. Langer, J. M., and Heinrich, H., Phys. Rev. Lett., 55:1414 (1985)
40
Wide Bandgap Semiconductors
47. Fischer, F., Waag, A., Bilger, G., Litz, T., SchoU, S., Schmitt, M., and Landwehr, G., J. Crystal Growth, 141:93 (1994) 48. Parbrook, P. J., and O'Donnel, K. P., Optical Properties of Wide Bandgap II-VI Superlattices, in: II-VI Semiconductor Compounds, (M. Jain, ed.), World Scientific, Singapore (1993) 49. Faschinger, W., Ferreira, S., Sitter, H., Krump, R., and Brunthaler, G., Materials Science Forum, 182-184:407, Trans Tech Publications, Switzerland (1995) 50. Tokumitsu, E., Jpn. J. Appl. Phys., 29:L698 (1990) 51. Nicolini, R., Vanzetti, L., Mula, G., Bratina, G., Sorba, L., Franciosi, A., Peressi, M., Baroni, S., Resta, R., Baldereschi, A., Angelo, J., and Geberich, W., Phys. Rev. Lett., 72:294 (1994) 52. Kowalczyc, S. P., Kraut, A., Waldrop, J., and Grant, R., J. Vac. Sci. Technol., 21:482 (1982) 53. Yoshida, S., Misawa, S., and Gonda, S., J. Appl. Phys., 53:6844, (1982) 54. Nakamura, S., Harada, Y., and Seno, M.,Appl. Phys. Lett., 58:2021 (1991) 55. Martin, G., Botchkarev, A., Rockett, A., and Morkoc, H., Appl. Phys. Lett., 68:2541 (1996) 56. Kurtz, E., Albert, D., Kraus, J., Hommel, D., and Landwehr, G., Proc. Int. Symp. on Blue Lasers and LEDs, Chiba, p. 429 (1996) 57. Einfeldt, S., Heinke, H., Behringer, M., Becker, C., Kurtz, E., Hommel, D., and Landwehr, G., J. Crystal Growth, 134:471 (1994) 58. Ohkawa, K., Tsujimura, A., Hayashi, S., Yoshii, Y., and Mitsuyu, T., Extended Abstracts of the Int. Conf. on Solid State Devices and Materials, Tsukuba, p. 330 (1992) 59. Faschinger, W., J. Crystal Growth, 197:557 (1999) 60. Kimura, K., Miwa, S., Yasuda, T., Kuo, L., Wang, T., Jin, C., Tanaka, K., and Yao, T., Proc. Int. Symp. on Blue Lasers and LEDs, Chiba, p. 167 (1996) 61. Hauksson, I, Simpson, J., Wang, S. Y., Prior, K. A., and Cavenett, B. C., Appl. Phys. Lett., 61:2208 (1991) 62. Zhu, Z., Brownlie, G., Horsburgh, G., Thompson, P. J., Wang, S. Y., Prior, K. A., and Cavenett, B. C.,J. Crystal Growth, 159:248 (1996) 63. Baron, T., Saminadayar, K., and Magnea, N., Appl. Phys. Lett., 67:2972 (1995) 64. Jobst, B., Strauf, S., B~iume, P., Kurtz, E., Schenk, H., Gutowski, J., Hommel, D., and Landwehr, G., Proc. Int. Symp. on Blue Lasers and LEDs, Chiba, p. 409 (1996)
Doping Limits and Bandgap Engineering
41
65. Turco-Sandroff, F. S., Nahory, R. E., Brasil, M. J. S. P., Martin, R. J., Besermann, R., Farrow, L. A., Worlock, J. M., and Weaver, A. L., jr. Crystal Growth, 111:762 (1991) 66. Kom, M., Niimberger, J., Faschinger, W., Spahn, W., Ehinger, M., and Landwehr, G., Proc.of the 23rd Int. Conf. on the Physics in Semiconductors, (M. Scheffler and R. Zimmermann, eds.), World Scientific, Singapore, p. 1023 (1996) 67. Faschinger, W., Krump, R., Brunthaler, G., Ferreira, S., and Sitter, H., Appl. Phys. Lett., 65:3215 (1994) 68. Krump, R., Brunthaler, G., Faschinger, W., Ferreira, S., and Sitter, H., Mater Sci. Forum, 182-184:349 (1995) 69. Niimberger, J., Faschinger, W., Schmitt, R., Kom, M., Ehinger, M., and Landwehr, G., AppL Phys. Lett., 70:1281 (1997) 70. Taike, A., Momose, M., Kawata, M., Gotoh, J., and Mochizuki, K., J. Crystal Growth, 159:714 (1996) 71. Suemune, I., Appl. Phys. Lett., 63:2612 (1993) 72. Schumicki, G., and Seegbrecht, P., Prozefltechnologie, Springer, p. 262ff (1991) 73. Ishibashi, A., IEEE J. Quantum Electronics, 1:741 (1995) 74. Ohki, A., Ohno, T., and Matsuoka, T., Proc. Int. Symp. on Blue Lasers and LEDs, Chiba, p. 252 (1996)
2 Epitaxial Growth of I I - V I C o m p o u n d s by MOVPE Wolfgang Gebhardt and Berthold Hahn
1.0
INTRODUCTION
This article reviews the present state of metal organic vapor phase epitaxy (MOVPE) of wide gap II-VI semiconductors, a class of materials, which is typically represented by ZnSe and related ternary and quaternary compounds. Epitaxial growth of large gap II-VI compounds has attracted a renewed interest, in the last years, since the successful operation of blue diodes and injection lasers based on ZnSe related material. Although MOVPE of II-VI compounds has not yet reached the same maturity as MBE growth, important progress has been made in the last years. Furthermore, we are convinced that there is no intrinsic problem connected with MOVPE which might hamper a successful production of II-VI devices. Both methods of epitaxy, MBE and MOVPE, still suffer from the lack of appropriate II-VI substrates, hence, (001)-GaAs wafers are widely used. The purity of MOVPE precursor compounds was, with exception of the hydrides, a severe problem in the past, but has now been solved for most of the precursor molecules given in Table 1.
42
Epitaxial Growth of ll-Vl Compounds by MOVPE
43
Table 1. Thermodynamic Data MOVPE Precursor Frequently Used for Growth and Doping of II-VI-Compounds[ 1]
Compound
t-(C4Hlo)OH
Shortnotation in MOVPE TBOH
82.2 (MP 25.5~
Vapor Pressure at 20~ (Pa) 5600(25~
-60.7
H2S t-(C4HIo)SH
Boiling Point (~ at Normal Pressure
TBSH
63
16 646
-41.5
H2Se
2926
(C2Hs)2Se
DESe
t-(C4Hio)2Se
DTBSe
122
580*
i-(C3Hs)ETe
DIPTe
49 at 18.12 Pa
282
(C2H5)2Zn
DEZn
118
1395
(CH3)2Zn-(C2H5)3N
DMZn-TEN
95
1733
(CH3)2Zn
DMZn
46
16 277 at 0~
(CH3-CsH4)2Mg
(M-CP)2Mg
C2H5I
El
72.3
13300
n-(C4Hlo)C1
n-BCI
78
11 000
(C2Ha)3Ga
TEGa
143
490
t-(C4HIo)PH2
TBP
54
216.9
t-(C4Hlo)AsH2
TBAs
66
19710
t-(C4HIo)NH2
TBN
46
33 701
21 at 40~
*This vapor pressure was derived from consumption measurements in the authors' group.
By far, most of the results presented in this article were obtained with a horizontal laminar flow reactor used in many commercial growth facilities. In our laboratory, an Aixtron MOVPE system AIX200 is used. These kind of reactor cells are operated between atmospheric and low pressure (1000 hPa-0.1 hPa). Their hydrodynamic properties, with
44
Wide Bandgap Semiconductors
respect to II-VI epitaxy, have been investigated[ 2] and reviewed by W. Kuhn. [3] There is a well known inhomogeneity of temperature and growth rate in cells along the reactor axis. The growth rate (GR) has a maximum at the gas inlet, which has to be taken into account when results of different groups are to be compared. The inhomogeneity in the used reactor is, at atmospheric pressure, about 10% over a 2" wafer and can be greatly reduced either by low pressure MOVPE or by installation of a rotating susceptor. In a typical MOVPE system three growth regimes may be distinguished (see Fig. 1): 9 A kinetic limited region at low temperatures where the rate of the surface reactions limits the GR 9A flat region which is defined by diffusion or transport
limited growth 9 A high temperature region in which surface desorption and thermal activation of further chemical reactions suppress growth increasingly Growth optimization should be performed (varying mass flow and temperature) within the diffusion limited regime where the GR is practically independent of temperature.
I
(9
I
I
transport
I I
(b) i
2 t--
kinetics (a)
'!' I
I I
0 o) o) o L_
..,.,.
I
I
I i
i
I i
ill.,
,,.,
1/T 0
Figure 1. Three regions of MOVPE growth.
i.
||
Epitaxial Growth o f l I - V l Compounds by M O V P E
2.0
BINARY COMPOUNDS
2.1
ZnTe
45
This substance grows easily on (001)-GaAs using DIPTe and DEZn, or DMZn-TEN as metal organics. The last mentioned adduct, first introduced by Jones, has been proved to be a very useful Zn-compound.[ 4] It is less volatile than DMZn, prevents prereactions in the gas phase, and leads to layers with a smooth surface and less deep center luminescence. The growth in our laboratory was usually performed with DIPTe and DEZn, or DIPTe and DMZn-TEN, at a substrate temperature of Ts = 340~ and a carrier gas flow of 5 stdl/min at atmospheric pressure. Typical input precursor flow rates are found in Fig. 2 which presents a typical plot of GR versus inverse temperature. The kinetic region in Fig. 2 fits a straight line which yields an activation energy (E~) and a prefactor of the surface reactions E r = 102 + 12 kJ/mol and A r = 6.108 pm/h for DIPTe + DEZn and E~ = 96 • 20 kJ/mol and Ar = 2" 108 ~tm/h for DIPTe + DMZn-TEN.
Growth Temperature [~ 400 '
I
380 '
360
I
'
:340
I
'
320
I
'
300
I
'
I
'
DIPTe+DEZn DIPTe+DMZn-TEN
s"A %
_
l.,-=..e C"
-1 ~IL%
s
%
s
E "--i
O---
~3'
t J
=__.J
s $
,,N=,
s
9
$ s s
r
~
J
J
o L_
0
[DEZn] =17.5 pmol/min [DMZn-TEN]=23.0 pmol/min [DI PTe]=34.1 pmol/min 0.1
0,1 ,
I
1.50
,
I
1.55
.,
l
1.60
,
i
1.65
~
i
1.70
,
.I
1.75
1000/TG[1/K] Figure 2. Growth rate of ZnTe versus 1/T for two different Zn precursors.P]
46
Wide Bandgap Semiconductors
Kuhn et al. investigated the co-pyrolysis DIPTe and DEZn in an isothermal quartz tube under H2-flow (see Fig. 3). They found the respective activation energies of 71 and 74 kJ/mol. The molar ratio of DIPTe and DEZn in this experiment ensured that the walls of the tube were homogeneously covered by ZnTe. As in pyrolysis of the separate metalorganics propene, propane, 2,3-dimethylbutane from DIPTe and ethene, ethane and n-butane from DEZn appear. Additionally, 2-methylbutane has been detected in significant concentrations which is diminished when He is used as carrier gas. Alkenes as reaction products give evidence for fl-elimination. Eq. (1)
(C2Hs)2Zn -~ C2H4 + C2H 6 + Zn
and Eq. (2)
(C3H7)2Te ~ C3H 6 + C3H s + Te
The production gain ofalkenes is reduced when He is the carrier gas, indicating that the carder H 2 also participates. Eq. (3)
(C2H5)2Zn + H 2 --~ C2H 6 + C2H5ZnH
and Eq. (4)
C2HsZnH ~ C2H 6 + Zn ..........................
r ...........................
I ......................
v ..........................
I .................
'
. . . . . . .
I ......................
100
' ......................
! .......................
..
:DPT:
80 a===
09 f-
.~ f-
60-
, . J
03
:~
40-
o
20-
e-
200
'
'
350
400
Temperature[~ Figure 3. Copyrolysis of DEZn and DIPTe.[S]
Epitaxial Growth of ll- VI Compounds by MO VPE
47
There is also a reaction of radicals which yields butane and substituted butanes. When DMZn is used, t-substitution is not possible. Pyrolysis experiments show a clear evidence for a contribution of H2-carrier gas.[ 6] Eq. (5)
(CH3)2Zn --~ CH 3. + - Z n CH 3
Eq. (6)
CH 3" + H 2 --~ CH 4 + H.
The additional effect of (C2H5)3N in the adduct compound DMZnTEN will be discussed with the growth of ZnSe. Before layer quality is discussed, we will consider the substrate preparation which was applied by us and by W. Kuhn in nearly all growth experiments of ZnTe. The polished surface of the GaAs wafers (not epiready) is degreased in propanol, then etched for 60 sec in HzSO4:HzO2:H20 in the proportion 4:1:1 at 40~ Immediately after this wet etching step, the substrate is rinsed in 18 Mr2 water. After drying by spinning, an in-situ deoxidation is followed by an annealing procedure of 15 min at 350~ in streaming Hz-gas. This process was later changed into a 15 min annealing step at 500~ ZnTe layers grown on GaAs suffer from a large lattice mismatch: Eq. (7)
f = as - at a t
which is -7.5% between 300~ and 400~ Here a s is the lattice constant of the substrate and a t that of the layer material under relaxed conditions. After the growth of a few monolayers, the strain is already largely relaxed by the formation of Lomer- and 60~ (see Fig. 4). This dislocation array yields a strain relaxation: Eq. (8)
6=
b.e(llO)
where the scalar product b'e(110) of the Burgers vector, and the unit vector along (110) is just the component ofb in this direction and d is the averaged distance between two misfit dislocations. The remaining unrelaxed strain at the interface ~1is given by the difference: Eq. (9)
~]1= 6 - f
48
Wide Bandgap Semiconductors
which is about -0.5%. When layers of various thicknesses are investigated (see Fig. 4), one finds a saturation of the relaxation at a thickness of about 0.8 ~tm. A very small residual compressive strain of about -0.05 % remains, even in thick layers, but is completely compensated for and even converted into tensile strain when cooled down to liquid Helium temperature for measurements. The dislocation density at the interface evaluated from high resolution transmission electron microscopy (HRTEM) images is about 6.10 ~2 cm -2, whereas, in a 3 ~tm thick layer, only about l0 s cm -2 can be determined by the broadening of x-ray rocking curve.
................... ' ............
I ...........
' ............
I ............
r .......
I . . . . .
' ........
I .........
'
.......
I
.......
r .............. [ ................
~ .............
e=,=~
c
0.0
o=,..
f.3
ffl &
II1 ---
9
/.D
9
-0.1
11.
-0.2 o
9 ion channeling DIPTe+DMZn-TEN 9 reflectance DIPTe+DEZn reflectance DIPTe+DMZn-TEN
. . . . . . . . . . . . . . . j . . . . . . . . . . . . . . . . 1 . . . . . . . . . . . . . . . . . . , . . . . . . . . . . . . . . . . . ! .................... ,. . . . . . . . . . . . . . . .
0.5
1.0
I .............
1.5
J .............
1 . . . . . . . . . . . . . . . . . , ................... !
2.0
.................. ~ .................... i . . . . . . . . . . . . .
2.5
~-. . . . . . . . . . . . . . .
3.0
Thickness [IJm] Figure 4. Parallel surface strain of ZnTe layersgrown on GaAs(100) as a function of layer thickness.[7H9]
Our laboratory has made some attempts to study the growth of cubic ZnTe on GaAs(111 A) 2 ~ off. The growth under standard conditions (DIPTe + DMZn-TEN, Ts = 340~ yielded layers which were nearly defect free. HRTEM images show a dark interface zone which extends about 10 nm into the ZnTe layer (see Fig. 5). The misfit dislocations have Burgers vectors parallel to the (111)-planes. Since these planes are glide planes of the zincblende structure, the dislocations are confined within these planes parallel to the interface. The FWHM of the x-ray (111)-reflection was 82 arc sec and, thus, close to the theoretical value of 72 arc sec for ZnTe-layers of 200 nm thickness.
Epitaxial Growth o f l l - V l Compounds by M O V P E
" .- -~,~~.,~- .
49
: ,~ ~ ,'.vr. . '~ : ~.
e. ~!
FWHM 24aw.sec GaAs( 14.1!
8Znre 2 (111) ~f cl~c
i O0 nm
4~
Omegaldegl
Figure 5. Cross section view of the ZnTe/GaAs (111) interface region together with the rocking curve of the (111 ) reflex.[ !~
Doping of ZnTe has been attempted with a variety of elements, however only p-doping has been achieved. The dopants used in our laboratory were: TBAs, TBP, TMBi [(CH3)3Bi ], EI, TEGa, EDMIn [(C2Hs)(CH3)2In ]. A successful doping up to p ~ 1018 cm -3 was obtained only with TBAs and TBP. Typical results from Hall measurements are given for three samples in Table 2 (see also Ref. 11).
Table 2. Acceptor Binding Energy AEA, Hole Concentration NA and the Compensation ND/NA in Three Samples Doped with TBP and TBAs
ZnTe:As- 12
Sample No.
ZnTe:P- 125
ZnTe:P- 124
AEA [meV]
46
35
58
NA [cm -s]
1.6.1017
7.8.1017
4.1.1016
N D [cm "3]
3.1.1016
1.6.1017
1.9.1016
No/NA
0.2
0.2
0.45
50
Wide Bandgap Semiconductors
The electron mobilities (~t) thus obtained are quite satisfactory. An example of ~t(T) is given together with a theoretical curve in Fig. 6.
10,8
1000
10"
10'e
750
E 10" ~'
~" "E
t
tO
'!ii I1'
[P]/Ere]: [P].~./Fe~
L
o
250
~ 10'2
10" 101o
5
10 15 20 1000/T [l/K]
25
00 1 ~50' 100....150' " " 200 ' ....250' ....300 Temperature [K]
Figure 6. Carrier concentrationand Hall mobilityof ZnTe:P.
2.2
ZnSe
The growth of ZnSe is usually performed with the same Zn sources as are commonly used for ZnTe. The usual Se sources are H2Se, DMSe, and DTBSe.[ 12] H2Se has the advantage of being readily available in relatively high purity. Disadvantages include its high toxicity and its tendency to prereact in the gas phase with the Zn-eompounds. Therefore, many laboratories have replaced H2Se by less hazardous compounds. Some have already been listed in Table 1. Heuken et al.[ 131have published a comparative study of ZnSe growth rates using various substituted selenium compounds (see Fig. 7), from which DTBSe allows by far the lowest growth temperatures (320~ < Ts). Further advantages of DTBSe are an exceptional broad region of diffusion limited growth (320~ < Ts < 360~ no detectable carbon incorporation, and no prereactions. High quality
Epitaxial Growth o f l l - V l Compounds by MOVPE
51
layers are thus grown at a relatively low selenium to zinc ratio (Se/Zn _=_ 2, compared to 18 necessary when H2Se is used).
550
500
450
400
350
300
o.1 f
......&
I t
--e-- DE_zn*DIPSe --"-- DEZn+DASe -+- DMZn+DESe
1.2
DEZn+DTBSe
1.3
_
1.4
1.5
1.6
1.7
1.8
1/TG [1 O00/K] Figure 7. ZnSe growth rates with DMZn and various Se-precursors.[ l 1][13]-[16]
Substrate Preparation. In the authors' laboratory, the standard procedure to grow ZnSe layers is the following: 9 The polished surface of the GaAs wafers (not epiready) is degreased in boiling propanol then rinsed in 18 MC~ water 9 A 60 see. etch at r.t. with H 2 S O 4 : H 2 0 in the proportion 4:1 9 Immediately after this wet etching step, the substrate is again rinsed in 18 M ~ water and dried by spinning 9 Lastly, an in-situ, 20 min desoxidation in flowing hydrogen at 500~ A slightly different procedure is applied to epi-ready GaAs substrates which, after wet etching, are only shortly in-situ annealed at 360~ G r o w t h Conditions with DTBSe. DMZn-TEN and DTBSe are used at Ts = 340~ with Hz-earrier gas flow of 5000 scem/min, Zn = 29 ~tmol/min, Se = 87 ~tmol/min, Se/Zn = 2-3. The pyrolysis of DMZn-TEN has been investigated in a clean growth reactor by recording the mass
52
Wide Bandgap Semiconductors
spectra between room temperature and 400~ The difference spectra reveal, as decomposition products, mainly ammonia, ethylene, and methane. The latter stems from DMZn in H 2, whereas NH 3 and C2H4 are products of the TEN pyrolysis. The amine is always adsorbed at the Se- or Sell-determined surface, a conclusion which was reached by in-situ experiments with a reflection difference spectrometer.[ 25] The complete decomposition of the amine on the surface leaves a Se-Zn-bond and releases NH 3, C2H4, C3H6, and C3Hs. Propene and iso-propane stem from the decomposition of DTBSe. Thus, the amine inhibits a quick reaction of Zn and Se on the surface. A model of the possible steps of this reaction has been suggested in Ref. 17. It is important to notice that H2-carrier gas does not participate in the thermal decomposition of DTBSe which disintegrates by ~-elimination. Therefore, H 2 may be safely replaced by an inert carrier gas like N 2 or He.[is] PL spectra may be used as a sensitive tool for the detection of point defects in the layers. As shown in Fig. 8, there is no deep luminescence in layers grown with DTBSe under the above mentioned conditions. Free and donor bound excitons dominate the near band gap luminescence. The donors are presumably chlorine atoms, impurities which are likely to originate in DTBSe. Although the purity of DTBSe has continuously improved, there is still much to be done to reduce background donor concentration. The density of line defects, such as dislocations and stacking faults, is another problem and is the principal source of degradation in II-VI LEDs and laser diodes. In order to reduce the number of ingrown dislocations, a 2-dimensional growth mode should be achieved as is commonly used in MBE. This is difficult to control in MOVPE, since all the powerful surface analytical methods, which need ultra high vacuum, cannot be used. In spite of this difficulty, MOVPE-growth was controlled by RHEED. For this purpose, growth was interrupted and the sample was cooled down to room temperature in flowing nitrogen. The sample was quickly transferred to a MBE-chamber. The RHEED pattern of samples grown under standard conditions (see above) showed a 3D-characteristic and [ 110]-oriented facets. When growth is continued in the MBE-chamber, a 2D-growth mode reconstitutes after a few monolayers. This procedure can be repeated several times with equal results. The results were independent of substrate preparation which was always the same. In Succeeding work, we were able to show that TEN in the adduct compound played a crucial role in the promotion of a 3D-growth mode during MOVPE. When DMZn was used, a 2D-growth mode could be achieved, although at relatively low temperatures.
Epitaxial Growth o f l I - V I Compounds by MOVPE
2.8
Energy [eV] 2.4
2.6
I
'
' I
'
""
2.2
I
"
'I
'
'
'
hh
.
,-.,
53
9
PL ZnSe 2K
"~
12
I,o
Z t"t
'
I
I
"
4380'
I
4 00
,
'
I
,
,
5000 Wavelength [A]
4400
.... 4 J l 2 0 -
i
"
'
.... 4~1.40 .... 4J160
.
.
.
.
.
5500
Figure 8. PL spectra of a 200 nm ZnSe layer at 2 K.
Growth Conditions with Other Se-Precursors. A comparative study of growth rates using DASe, DIPSe, DESe, and DTBSe together with DEZn at various VI/II ratios, has been carried out by Heuken et al.[ 13] In Fig. 7, the results are plotted versus 1/T. Data of another study using DEZn + DMSe and DEZN + MSeH are taken from Refs. 13, 16, and 15. In Refs. 14-16, there was broad experience with DMSe and DESe at growth temperatures above 400~ including photoassisted growth as low as 350~ This growth, however, needs H 2 as reactand (see below). In practice, it sometimes may be useful to choose an appropriate precursor in order to cover a convenient temperature range for growth. However, before planning the respective experiments, one should always ask if the chosen precursors arc available in the required purity. Photoassisted Growth. The lowest growth temperatures under standard conditions using alkyl selenides have been achieved, as shown above, with DMZn + DTBSe or TBSeH.[ 19l If necessary, the growth temperature Ts can be further reduced when the sample is irradiated with
54
Wide Bandgap Semiconductors
above bandgap light. In our laboratory, we have used a 150 W xenon lamp, which gives an intensity up to 45 mW/cm 2 at the sample surface. It was found that the diffusion limited region extended now to about 300~ when DMZn-TEN was used. In general, photoassisted growth facilitates doping but can also lead to an increase in the impurity background. Respective results have been published for DMSe and DESe by Fujita et al.[ 2~ and for DTBSe by Hahn et al.[211(See Fig. 9.) In general, the region of diffusion limited growth is shifted to lower temperatures and the reaction limited growth is governed by a lower activation energy.[ 171In the case of DTBSe + DMZn or DMZn-TEN, this yields, at 300~ a growth enhancement factor of 2.1 or 3.4 respectively. Fujita and Fujita summarized the known experimental facts and proposed the following model. The above band gap photons produce electron hole pairs with a quantum yield up to 10%. The holes accumulate at the ZnSe surface due to band bending.[ 221This effect changes the redox potential at the surface. Surprisingly, adsorption and decomposition of the Zn-precursor play important roles and determine the rate limiting step in photoassisted MOVPE (PA-MOVPE). The evidence was found by mass spectrometry [23] and surface sensitive experiments like RDS [24][25] and was also supported by ab-initio calculations.[ 26] When DMZn or DEZn molecules reach the positively charged surface, they bend and act as electron donors which react with the holes. Before a simple reaction scheme is formulated, the role of hydrogen should be emphasized. A Sedetermined surface grown by MBE shows reconstruction since the dangling bonds form dimers. However, if the surface is grown by MOVPE, all dangling bonds are occupied by hydrogen. Therefore, the reaction scheme may be written: Eq. (10)
(Se)surf-H + CH 3 Zn CH 3 ~ (Se)surf-Zn- CH 3 + CH 4 and
Eq. (11)
(Se)surf-Zn- CH 3 + CH 3 Se CH 3 --+ SeZn-(Se)surf CH 3 + C2H6
Here, (Se)sur f symbolizes a Se-atom at a normal lattice site but in surface position bound to two Zn-atoms in the lower lying layer. It is an essential aspect of this model, supported by all experimental results, that PA-MOVPE is a photocatalytic surface reaction. Furthermore, atomic hydrogen plays an important role, it covers the surface, is highly mobile, and reactive.
Epitaxial Growth of lI-VI Compounds by 510VPE
500
450
,
I
,-.,1
"
,,
I
,,
To [*C] 400
"
I
350
"
300
!
0 .......... 0 ~ ~ 0 - - - - - - - - - - - - - -
55
"
0"~
'
t.-
i
E "1 ,,._...=
L_ r
0.1"
Growth enhancement by illumination
--o-- DMZn+DESe ~ --..-- DMZn-TEN+DTBSe ~ I
1.3
'"
"
I
'
"
1.4
I
9
"
li~
1.5
1.6
"
"
'1
'
'
1.7
'"
'
!
1.8
1000/TG [l/K]
Figure 9. Growth enhancementunder irradiation with above band gap light; open symbols depict samples grown in photoassisted mode.[2~l[151
n-Doping. A variety of group-Ill-cations and group-VII-anions have been used, more or less successfully, as n-dopants of ZnSe.[ 27] The highest donor concentrations, well in the region of degeneracy, were obtained with C1 and I. Deep donor centers are not formed before very high concentrations (C1 and I > 1019 cm -3) are reached.[ 28] In Table 3, precursor molecules, doping results, and references are summarized.
Table 3. N-Type Doping of ZnSe with Group III and Group VII Elements
Dopant Atom
Precursor
I
C2H5I
Cl
n-C4HI0
Results
29, 30 n > 101Scm-3
27,31 27,32
AI Ga
Ref. #
(CH3)3Ga
n --__10Is cm"3, but forms deep centers
33, 34
56
Wide Bandgap Semiconductors
p-Doping. The p-doping of ZnSe and ZnS have a common problem believed to be due to the tendency in these compounds to have the Fermi level pinned in the upper region of the energy gap. This leads to donor states even with most of the group I and group V elements.[35][ 36] Only Li and N were found to be effective dopants yielding p > 1017 cm -3. However, lithium has been rejected because of its fast diffusion which occurs at moderate temperatures. This knowledge has led to numerous doping procedures in order to incorporate N-atoms into the lattice by MOVPE. They are summarized as follows" 9 Plasma doping with low pressure MOVPE[ 13] 9 MOVPE and PA-MOVPE with ammonia and substituted ammonia (organic amines)[37][38] 9 MOVPE with metal organic compounds with a direct nitrogen bond[ 39] 9 MOVPE and PA-MOVPE with azides or hydrazines[ 21][32][401141] The authors of this article have investigated a number of amines and found that, only those which have the structure R.NH 2, where R is an alkane or alkene radical, are effective p-dopants. The problem with most of these compounds, especially NH3, is the strength of the C-N bond which needs relatively high temperatures to dissociate. However, in order to achieve a sufficiently high doping concentration and to avoid interdiffusion in heterostructures, the growth temperature should be low. This is feasible when t-C4H10NH 2 (short notation TBN) is used. TBN has a weak C-N bond and already dissociates above 380~ The growth temperature can be further reduced by illumination. A standard procedure was described by Fujita et a1.[381[42] who used PA-MOVPE (1 > 300nm) with DEZn (6 ~mol/liter) + DMSe (12 ~mol/liter) at 200 Torr and 350~ The flow rate of TBN was varied between 6 and 12 I~mol/liter. A 30 minute post growth annealing in the reactor in flowing N 2 between 500 and 600~ was necessary to activate the nitrogen acceptors. Subsequent C-V-measurements revealed a hole concentration of up to 3.1017 cm -3. An extremely high purity of all precursors, especially with respect to halogen contamination, is an essential prerequisite of successful p-doping. The work cited above led to the suggestion that nitrogen is only incorporated if NH2-radicals reach the surface. On the other hand, hydrogen plays an important role and inactivates nitrogen acceptors. Therefore, it may be possible that at least one hydrogen atom stays either in the neighborhood of, or forms a bond with, the N-atom which occupies a Se-site.
Epitaxial Growth o f ll- VI Compounds by MO VPE
57
Therefore, compounds have been proposed which have a direct metal-nitrogen bond or which lack a NH-bond. Several compounds with direct metalnitrogen bonds have been synthesized and tested. However, no p-type conductivity has been found. [39][43] A member of the second group of compounds is ethylazide (C2HsN3) which easily decomposes into N 2 and N'. Although Nincorporation was achieved, [44]--[46] the layers showed high resistivity. The same is true for the application of substituted hydrazines, for which di-methyl- and phenyl-hydrazine (PhHz) have been used. The latter has also been applied by the present authors. Together with DTBSe + DMZn-TEN, very small molar concentrations of PhHz are sufficient to obtain samples in which nitrogen can be detected by SIMS. Under illumination with above band gap light, the optimum growth temperature is 300~ The N2-plasma treatment is a method borrowed from MBE where it has been quite successfully applied and leads to acceptor concentrations up to 4.1017 cm -3. In MOVPE growth and doping experiments, nitrogen incorporation has again been shown. However, p-conduction was not yet demonstrated.[ 13] For completeness, it should be mentioned, that p-ZnSe has been obtained by solid state diffusion using Li3N.[47]
2.3
ZnS
In early growth experiments, ZnS [48]-[50] (and ZnSSe) was grown with DMZn + H2S (+ H2Se) mostly between 3 50 and 400~ and often at a low pressure to avoid prereactions. Briot et al.[51] used DMZn-TEN + H2S and grew ZnS at 300~ with a reactor pressure of 40 Torr. The VI/II ratio was 5. The excellent optical properties of these layers were demonstrated by reflectivity and PL measurements.[ 52] Fujita et al.[ 53] replaced H2S by methylmercaptane (MSH) and grew ZnS between 400 and 500~ (see Fig. 10). In recent years, tertiary-butyl-mercaptane (TBSH) has become available.[ 54] Optimum growth conditions were found with DMZn-TEN +TBSH at a substrate temperature of Ts = 350~ and a reactor pressure ofP = 300 hPa.[ 55] Absorption measurements in the excitonic region of the material thus obtained have recently been published.[ 56] Low growth temperatures have also been reached with DMZn-TEN + DTBS. [57] (See Fig. 10.) Photoassisted growth allows lower growth temperatures as was explained for ZnSe. Fujita described the growth of ZnS from DMZn + DES under illumination with a xenon or Hg-lamp at temperatures between 100 and 150~ TMHowever, fragments of precursor molecules were incorporated leading to poor morphology of the ZnS-layers. The situation improved
58
Wide Ba n d g a p Semiconductors
when a ArF-Laser was additionally installed, leading to photodissociation of the MO-molecules and fairly good layers. In most practical applications, however, growth mastery should be attempted with as few external parameters as possible and the use ofDMZn-TEN +TBSH at Ts = 350~ may be preferred.
Ta[~ 700
600
500 450
400
350
300
250
t"
E ,,,n i.-...._a
0
-,--'
1
r
o 13 DMZn:TEN+TBSH
(.9
.
0.1
1.0
A
DMZn+DES
O
DMZn+MSH
[]
DEZn+DTBS
9
9 DMZn-TEN+DTBS .
I
1.1
.
,
J
.
1.2
,
I.
I
1.3
1.4
.
.
..
I
I
1.5
1.6
,
I
1.7
,
!
=
1.8
.I
1.9
=
2.0
1000/Ta[K] Figure 10. Growth rates of ZnS obtained with various precursors.[S3l[s4ltSTltSS]
Early doping experiments have been successfully performed with TMA1, HCI,[ 59] and EI.[ 6~ In the last mentioned reference, DMZn and HzS were used for growth which allowed growth temperatures as low as 260~ where the carrier concentration reached a maximum of about n = 2.1018 cm -3. In our laboratory, n-type ZnS was grown with DMZn-TEN + TBSH and n-BC1 as dopant at 330~ Degenerate carrier densities up to n < 5.10 TMcm -3 were easily reached.[ 611 Electroluminescence was another important field for industrial use of ZnS and especially of ZnS:Mn. For a description of the deposition of ZnS:Mn-layers by MOVPE, see Ref. 62.
Epitaxial Growth of ll- Vl Compounds by MO VPE 2.4
59
ZnO
ZnO has proven to be an excellent material for windows, antireflection coating, piezoelectric effects, and even for substrates on which to grow GaN. For this latter purpose, one needs single crystalline epitaxial layers grown on (0001)A1203 .[63] Conducting windows and antirefleetion coatings for solar cells [64]are deposited as polyerystalline layers on glass, CIS, [65] CdTe, and Si [66] from the gas phase where the crystallites are usually oriented with the e-axis normal to the substrate surface. Because of its large energy gap, the layers are transparent in visible light (~. > 380 nm). The transmission cutoff in the near IR is given by the plasma frequency and depends on donor concentration as well as on mobility (i.e., the carrier lifetime).[ 67] MOVPE of ZnO is usually performed with DEZn or DMZn and an oxidizing agent. Sometimes, the adduet DMZn:THF is used[ 68] where tetrahydrofurane (THF) may itself act as an oxidizer.[ 65] The substrate temperature ranges between 100 and 500~ Direct use ofO 2 or H20 in air, or a carrier gas such as N 2 or At', leads to prereaetions which may be partly avoided at reduced pressure. In the ease of CO 2, NO 2, and N20, however, the growth rate is rather small. Improvements were achieved by photoassisted MOVPE [67][69][70] and by using alcohols like t-BOH.[ 711 A comparative study of growth with H20 , CzH5OH , and t-BOH revealed superior properties oft-BOH, yielding defect free and highly oriented ZnO-films (see Fig. 11). [72] Kaufmann, et al. [71] investigated the precursor combinations tBOH, DMZn:THF, and DEZn:THF. They determined the activation energy in the kinetic region to be about 40 kJ/mol. There seems to be a diffusion limited region above 350~ for DMZn:THF where the growth rate reaches 0.45 ~tm/h. In all growth experiments, grain size grows with film thickness and is smaller, but of the same order of magnitude as the thickness. Films with good uniformity have been obtained. Analysis of the surface by REM shows erystallites with tetrapodlike morphology.[ TMSome solid Zn-preeursors have been considered, namely zinc-acetate, Zn(CH3COO)2174]; basic zinc-acetate, Zn4(CH3COO)6175]; and zinc-2-ethylhexanoate, Zn(C2HsCsHllCOO)2 .[76] These compounds, which are very cheap compared to DEZn, sublime at moderate temperatures and are used with a transport gas [74][76] or under vacuum conditions. [75] The oxygen is delivered by the acid radical. At 210~ a deposition rate of 5.4 ~tm/h was observed.[ TM]When this rate is plotted versus 1/T, an activation energy of 21 kJ/mole is obtained.
Wide Bandgap Semiconductors
60
T=[*C] 400 350
300
250
200
150
e-
E "-, 2 t_ .C:
o (.9
I
/"
_._ H,O
- - = - - t-C4H10OH
1
- - , - - CH3OH --v .... C=HsOH
[DEZn]=l.2sccm ,
9
I
1.6
,
I
1.8
,
I
2.0
,
I
,
2.2
I
I
1000/'1"=[1/K] Figure
11. The growth rates of ZnO films for varios oxidizing agents of DEZn.[723
In order to deposit stoichiometric material, it is advantageous to have independent control of the oxygen carrier. In some cases, a high ratio O/Zn 15 is used. Most as grown films are n-conducting (see Ref. 70). Oxygen vacancies probably act as donors. Intentional n-doping has been achieved with chlorine, when POC13 was used,[ 64] or with borane (B2H6) diluted with H2,[67] or even more efficiently with (CH3)3A1. Film resistivities as low as 0.008 W cm have been reported.[ 77] 2.5
CdSe
CdSe has been grown between 300 and 500~ on (111) GaAs in hexagonal, and on (001) GaAs in cubic, structure using the adduct DMCd.(C4HsS)2 and H2Se.[7s][791The VI/II ratio was about 3. The hexagonal layers showed better optical properties. The decomposition of DMCd starts at relatively low temperatures leading to prereactions especially with H2Se. This tendency is slightly inhibited when the thiophene adduct DMCd.(CiHsS)2 is used.[ 8~
Epitaxial Growth of ll-Vl Compounds by MOVPE
61
Parbrook et al. have grown cubic CdSe and Znl.xCdxSe (see also below) on (001) GaAs at 300~ under normal pressure using DMCd and DMSe. Limited diffusion growth is possible between 380 and 500~ with a VI/II ratio of 2.1 and a DMCd flow of 18 ~tmol/min. Although there is no evidence for hexagonal overgrowth, the layers have a high density of dislocations (Lomer and 60~ due to the 7% misfit between layer and substrate. The present authors used DTBSe + DMCd to obtain cubic CdSe layers on (001) GaAs at an optimum temperature near 360~ The VI/II ratio was 2. The layers were much less perfect than those of ZnTe on (001) GaAs which had about the same lattice misfit. This was probably due to the metastability of the cubic phase at growth temperatures. This conclusion was also drawn by Parbrook et al. since MBE samples showed the same width of x-ray rocking curves. No improvement of layer quality was observed when the adduct DMCd.(C4HsS)2 was used instead of DMCd. 2.6
CdS
Halsall et al.[81] found that CdS grows purely hexagonally only on the (111)A face of GaAs. A mixture of hexagonal and cubic phases was found when CdS was deposited on (001), (110), and (111)B faces. The precursor molecules were DMCd and HES, and the optimum growth temperature was 350~ Some growth improvement has been reported with DMCd-(C4H8S)2.[79] CdS on GaAs(111)A shows arrays of misfit dislocations confined near the interface. The relaxed epitaxial layers have mainly threading dislocations, but are of sufficiently high quality to allow a subsequent growth of wurtzite structure CdS/CdSe superlattiees.[821[831 MOVPE growth of purely cubic CdS has not yet been reported. However, mixed crystals of cubic ZnxCdl.xS were successfully grown on (001) GaAs at 400~ using DEZn, DMCd, and TBSH with x as low as 0.4 [84] (see also Sect. 3.6).
3.0
T E R N A R Y AND Q U A T E R N A R Y C O M P O U N D S
Ternary and quaternary solid solutions of II-VI compounds have been prepared and investigated in order to vary gap energies and lattice constants. The MOVPE technique makes the composition control especially easy, since the concentration of the constituents at the growing
62
Wide Bandgap Semiconductors
surface is regulated by the input partial pressure of the precursor compounds. Even a continuous grading of the energy bands can readily be achieved by MOVPE. If a ternary system mixed in the cation or anion sublattiee (A1.xBxC o r ACl.yDy) is t o be grown, it is necessary to allow an adequate overlap of the regions of optimal growth of the binaries AC and BC, or AC and AD. In general, the composition depends nonlinearly on the input partial pressures of the components. A quantitative description is given below.
3.1
Thermodynamics of Binary Compounds Under the conditions of thermodynamic equilibrium the binary reaction:
Eq. (12)
A + C 4-~ AC
is controlled by the reaction constant Kil_viwhieh is the ratio of the reaction rates defined by: Eq. (13)
Kii_w : = k ~ / k ~
The growth is completely determined by the chemical potential difference (A#) between the gas phase in the reactor and the solid interface. If we neglect the effect of a boundary layer, which may be correct in the ease of diffusion controlled growth, then we may write: Eq. (14)
A/t = R T ln(aK)
R is the universal gas constant and a the activity which replaces the concentration in non ideal solutions. Optimal growth conditions are usually found when the VI/II ratio R vI_II , given by the partial pressure in the gas phase, exceeds 2: Eq. (15)
R VI-lI P~176 =
>2
In this case, there is a depletion of cations at the semiconductor surface and the partial pressure ratio at the interface can be given by: Eq. (16)
P vl /P ll = (P~ v~)ZKll_ Vz/ a iz_vl
Note the quadratic dependence of the anionic partial pressure.
Epitaxial Growth of lI-Vl Compounds by MOVPE 3.2
63
Thermodynamics of Ternary Compounds
Given the ratio P v i / P m one would like to know the composition x of the growing layer. This is possible if the solid-vapor distribution function (SVDF) is known. Expressions for the SVDFs are given in Ref. 85. These authors use a model for regular solutions in which the interaction energy (f2) between neighboring atoms and the influence of the lattice distortion is taken into account.j86][87]The chemical activity is then given by: Eq. (17)
a i = x exp{(1-x)2f2/RT}
with the interaction constant if2 and
Tc = ~/R
given in Table 4.
Table 4. Interaction Constant (f2) and Critical Temperature (T~) for Three Ternary Systems
Compound
f2 (J/mol)
L (K)
Zn(SeTe)
14350
863
Zn(SSe)
6670
401
11300
680
(CdZn)Se
The equilibrium constant has to be determined from the well known relation: Eq. (18)
Ki=
exp{- AGi~
where the Gibb's free energy is given by: Eq. (19)
AGi 0 = Ani 0 + TAS i
using ASi extrapolated from 298K to the growth temperature T with:
64
Wide Bandgap Semiconductors T
Eq. (20)
AS = r! C--eTdT
with Cp(T)= a + bT + c7 2 The relevant values are given in Table 5. Table 5. Thermodynamic Data to Calculate the Free Energy of Formation
AS0298K J/mol
a
kJ/molk ZnS
-205
57.7
12.6
1.24
ZnSe
-160
70.3
11.99
1.38
ZnTe
-120
77.8
2.61
CdSe
-145
83.4
11.95
A n ~ 298K
-1.36
1.5
A solid-vapor distribution function (SVDF) can be derived which forms a unique relation between the ratio of the initial partial pressure F = p~176 +pOs) and the composition~. - -. [85]The following two equations for the SVDF are obtained when cationic mixed systems AxBl.xC are considered: Eq. (21)
F= x (1 - fu) +fn(1 + KAca.4c/KsAasA) "l
Eq. (22)
RvI_I](1-fvi) = 1 - ~ i
with the VI/II ratio: Eq. (23)
Rgl_ll= p~176
+ pOB)
The quantitiesfu andfv1are defined as the ratios of vapor pressure at the crystal surface Pk to input partial pressure pk~ of the component k:
Epitaxial Growth of ll-Vl Compounds by MOVPE
65
and ~
and describe the incorporated fraction of constituents. The SVDF for anionic mixed systems ACI_xDx obey similar equations: Eq. (24)
F = x (1 -fzI) +fvz( 1 + [K,4caAc/KBAaB,4]2) "1
Eq. (25)
Rv1-1il( 1 -~I) = 1-frz
Under the conditions of diffusion limited growth and choosing R vi_iz>2, the cationic component determines the growth rate and fn 10 -3 f2-cm2) will limit the maximum frequency and cause excessive Joule heating and failure of the devices.J7][8] Ideally, an ohmic contact could be formed by selecting metal work functions depending on the type of semiconductors. Practically, formation of ohmic contacts is a complicated and challenging issue due to mismatch of levels at the interface, or due to a phenomenon called Fermi level pinning which involves surface states. A critical parameter in the formation of metal/semiconductors (M/S) contacts is the condition of the surface and particularly whether or not the Fermi energy level, E F, is pinned. Comprehensive surface science studies oflII-V surface and interfaces, for several decades, have shown that Fermi level pinning affects the formation of both Schottky and ohmic contacts, as will be reviewed below. It has been generally demonstrated that pinning of E F is consistent with the characteristic of the bonding in the semiconductor; with covalently bonded semiconductors most often exhibiting pinned surfaces. On the other hand, ionically bonded semiconductors normally exhibit unpinned surfaces. The type of bonding also correlates well with the electronegativity difference between the elements in a compound semiconductor.[ 9] For the semiconductors to be discussed in this chapter, the character of the bonding varies considerably. The III-V semiconductors tend to exhibit more covalent bonding with lower electronegativity differences (Ax) than do the II-VI semiconductors. Therefore, we would expect the former to exhibit pinned surfaces more than the latter. For example, GaAs bonding is only 32% ionic, A~ between Ga and As is only ~0.4, and E F is pinned in nearly all conditions. For InP, bonding is 44% ionic with A;~ ~0.6 and E F is partially pinned. In the AlxlnyGal_x_yN system, GaN bonding is 70% ionic, A)r is ~1.4 and E F does not exhibit pinning. In the II-VI system, ZnSe bonding is 63% ionic, A~ is ~0.8 and this semiconductor also does not exhibit E F pinning. The consequence of these systematic trends will become evident when we discuss the formation of ohmic contacts to these different semiconductors. Covalent bonding and low A)r for the III-V compound semiconductors leads to Fermi level pinning. Therefore, practical ohmic contacts to GaAs and InP devices are predominantly produced by controlling the barrier width rather than the barrier height. The barrier width is controlled
82
Wide Bandgap Semiconductors
by incorporating dopants into the interface region of semiconductors. Annealing of ex situ (i.e., post-growth) metallizations is the practical method to cause the incorporation of dopant through metallurgical reactions. When the interface regions of semiconductors are heavily doped by the metallurgical reactions, current transport across the metal/semiconductor contact is dominated by Field Emission (FE) or Thermionic Field Emission (TFE). In this respect, our primary interest is to understand the details of metallizations used for ohmic contacts. Generally, ohmic contacts to compound semiconductors are formed by a lift-off process using multiple (two or more) element metallization, such as AuGeNi (developed in 1967 for n-type GaAs).[ l~ This same metallization is often used for n-InP. While it has long been used and often studied, AuGeNi metallization exhibits problems which include irregular morphology,[11][ 12]poor thermal stability, [13][14] and interdiffusion.[ 15] Tremendous efforts have been made not only to solve the practical problems of the AuGeNi system, but also to understand the mechanism of AuGeNi contacts.[16]-[ TM]Non-alloyed and Au-free contact schemes have been demonstrated for n-GaAs and InP. [19]'[21] Not withstanding all of the studies, the technology of ohmic contacts is much more advanced than the basis for the understanding which is largely empirical. A more fundamental understanding is essential because electrical contacts are critical to both proper performance of a device as well as its lifetime. A great deal of effort has been focused recently on better understanding of both Schottky and ohmic contacts. Brillson has edited a general review of these subjects.[ 22] In the present review, the focus will be only on ohmic contacts to GaAs, InP, GaN, and ZnSe rather than their ternary or quaternary variants. The purpose of this review is to clarify the current understanding of the effects of interracial reactions occurring during formation of ohmic contacts, rather than to compile various recipes for their formation. As will be shown, new information on the metallurgical reactions taking place at the M/S interface will allow us to better understand when to expect ohmic behavior.
I.I
Energy Barrier and Current Transport Mechanisms
An enormous difference in the electrical properties between metal and semiconductor usually results in the formation of a rectifying barrier at the metal/semiconductor interface. In the case of GaAs and InP, rectifying barriers are formed during processing creating an ohmic contact because of Fermi level pinning.[ 231 Therefore, the interface between the metal and
Ohmic Contacts to II-VI and III-V Compounds
83
semiconductor always has a barrier to transport of charge. To make a good ohmic contact, this barrier must be negated. Ideally, either rectifying (Sehottky) or ohmic contacts may form as a result of charge transfer between metal and semiconductor to align the Fermi level across the interface when a metal is brought into electrical contact with a semiconductor. Figure 1 illustrates the idealized formation of a contact. Figure 1a shows the energy band diagram of a metal and a n-semiconductor before contact (unless otherwise specified, all discussion refers to n-type semiconductors), where ~m and ~ are the work functions for the metal and the semiconductor, EF~ and EF~ are the equilibrium Fermi energy levels of electrons in the metal and the n-type semiconductor, and ~ is electron affinity of the semiconductor. The values E c and E v are the energies of the bottom of the conduction and top of the valence bands in the semiconductor. No surface states exist within bandgap, thus E~ and E v are flat up to the surface before contact. Due to the potential difference (~m--f~Is) when the metal and semiconductor are joined, electrons in the conduction band of the semiconductor flow into the metal to occupy lower energy states. This results in a negative charge on the metal and a positive charge on the semiconductor. Since there are a large number of nearly-flee electrons in the metal, the charge distributes uniformly. However, the limited charge in the semiconductor causes the development of an electric field which bends the bands near the surface out to a distance over which majority carriers are depleted (by donation to the metal). The band alignment and bending shown in Fig. l b is observed. The length of the charge depleted region in the semiconductor (W) is called the depletion length (Fig. lb). In GaAs, W ranges from 20 nm to 1000 nm depending on the doping concentration at room temperature.[Z41 The resulting band bending due to the space charge region in the semiconductor is described by the spatial variation of electrostatic potential, V(x): Eq. (1)
V(x) = (1/q).[Evb-Ev(x) ] = (1/q).[Ec6-Ec(x)]
where the subscript b refers to bulk (i.e., x = 89 and subscripts V and C refer to the valence and conduction bands. With the assumptions of semiinfinite boundary conditions and no interface states, the potential as a function of x can be described by Poisson's equation: Eq. (2)
d2 V(x)/dx 2 = -p(x)/~es
where es is the dielectric constant of the semiconductor and p ( x ) is the spatial distribution of charge. Provided that there are no free carriers in the
84
Wide B a n d g a p S e m i c o n d u c t o r s
depletion region of n-type semiconductor and complete ionization, i.e., p(x) = N o, (No = donor concentration) solving Poisson's equation yields:
Eq. (3)
V(x) = VS + Em(x - x2/2 W)
where E m is the electric field at x = 0 given by (qNoW)/6 s, W is the depletion length given by (2SSVbi/qND) 1/2, and Vbi is the built-in potential defined by Ecb-Ecs(= Om-Os)" Here, Vs is the surface band bending at x = 0. Equation (3) shows that the band edge is a parabolic function of x from x = 0 to x = IT"(Fig. l b). The effect of image charge on the band bending was not considered.
Vacuum level
-
IZI i(i) ~,Vacuum s level
_ _
~ - _ ~_ _ Efs
Ec
Eft-Ev n-Semiconductor
Metal
(a) Before contact
I Efm
Os
b l _~~_ ' . .
_E_c- Ets
,.F x -0
~x
(b) At equilibrium Figure 1. Formationof energybarrierand band bendingin electricaland thermal equilibrium.
Ohmic Contacts to II-VI and III-V Compounds
85
The energy barrier, ~0b, is defined as the energy distance between the Fermi level and the edge of the majority-carrier band (Ec). The rPb of an ideal M/S junction is shown (Fig. 1 b) to be equal to" Eq. (4)
9
=
+
= a
,.-Zs
where Zs is the electron affinity of semiconductor as defined in Fig. 1a. Eq. (4) is the Schottky-Mott rule or the Schottky Barrier Height (SBH). The first term, (q),n-q)~), represents the built-in potential and the second term (q)~-Z), is the potential difference between the conduction band edge and the Fermi level in the semiconductor. There are three transport mechanisms for electrons (major carriers) to cross such an energy barrier.[251[261Thermionic Emission (TE) transport occurs by thermal activation of electrons over the energy barrier from the conduction band of the semiconductor into the metal (Fig. 2a). TE is normally dominant when the semiconductor is low doped up to 1016cm-3 at room temperature. The second current transport mechanism is Field Emission (FE), which is a quantum mechanical effect whereby electrons penetrate through the M/S energy barrier at the Fermi level under the influence of an electric field (Fig. 2b). For FE to be predominant, high doping above 1019 cm 3 at room temperature is necessary. Thermionic Field Emission (TFE) is a combination of the TE and FE mechanisms (Fig. 2c) in which the carriers are thermally activated above Fermi level to a region where they can tunnel efficiently through a narrower portion of the barrier. By assuming that only electrons traversing from the semiconductor to the metal with sufficient energies to overcome the energy barrier can contribute to transport, the current density from the semiconductor to the metal is given by: Eq. (5)
Js-~m =.,4*"T2"exp(-q cl)b/k T).exp( q Va/kT) = JrH.exp(-q cI)a/kT)'exp( q Va/kT)
where Va is the applied bias, Tis the absolute temperature, and gYbis the M/ S barrier height. A* is known as Richardson's constant and given by (4rutm*k2/h 3) where m* is the effective mass of electrons in the semiconductor, k is Boltzmann's constant, and h is Planck's constant. For free electrons, A ~ is 120 m/cm2/K 2. Jrtt (known as the thermionic current coefficient) and defined by: Eq. (6)
J'm = A*.T2
- 120(m*/m)T2[A/em 2] - 10.8(m*/m)(T/300) 2 [MA/cm 2]
86
Wide Bandgap Semiconductors
where m is the mass of a free electron. For free electrons at room temperature (T = 300K), Ja'rt is 10.8 Million Amperes (MA)/cm 2. By considering din__,,s, the net current across the M/S contact is given by: Eq. (7)
gaet = gm~s + Jm.s = JTa.exp(-q ~b/kT)'exp(q Va/kT)-JTH'exp(-q~b/kT) = JTH'exp(-q ~b/kT).[exp(q Va/kT ) - 1] - J
.[exp(q Z a / k r ) - 1]
where .Is is saturation eurrent and defined by: Eq. (8)
Js = A*'T2"exp(-q~b/kT) = 120(m*/m)'T2exp(-q~b/kT) [A/cm2]
Metal
n-Semiconductor
r ~x'~ e"
Ef
E-c- Ef
(a) Thermionic Emission
Ef . . . . .
EcEf
(b) Field Emission
Ef
C
E-c- Ef
(c) Thermionic-Field Emission Figure 2. Current transport mechanisms across metal/semiconductor interfaces.
Ohmic Contacts to II-VI and III-V Compounds
87
For a large forward bias V~ >> kT/q, Eq. (7) is reduced to "/net = Js.exp(q V~/kT). Equation 7 determines the current-voltage (I-V) characteristics of M/S junctions in low doped semiconductors, which is the most frequently used method to determine the barrier height. Saturation current (J~) is obtained by extrapolating./net at zero Va from a plot ofln(J,,et ) versus forward bias Va. Then the barrier height ~ob can be determined from Eq. (8). At a large reverse bias voltage, V~ 2.7"kTeV
Ohmic Contacts to II-VI and III-V Compounds
89
where m r = m * / m , e~ = 6s/so and N D is the carrier concentration (cm-3). For n-type GaAs at 300K, m r --0.067, 6r -13, and Eq. (16) predicts ohmic behavior for N D > 1.2 x 1019 em-3.[s] The figure of merit for the formation of an ohmic contact is the specific contact resistance defined by Eq. (11). It is the contact resistance of an unit area for current flow perpendicular to the contact. Thus it has units of s z. Cox and Straek developed a bulk technique to measure contact resistance using different size circular dots defined on an epilayer.[ 341 This method is simple, but is only applicable to bulk structures (e.g., an epilayer on a conducting substrate), and its practical limit is ~ 10-5 ~-cm 2. The Transmission Line Method (TLM) is most frequently used to measure contact resistance in FET structures where current is confined in a highdoped conducting channel on a semi-insulating substrate.E3Sl Typically, the accuracy of TLM measurement is of the order of 10.6 ~-cm 2. For resistances of 10 -7 f~-cm 2, Kelvin Cross Bridge Resistor measurements are used, which require a relatively complicated geometry of the contact pads.j361 There are many other methods and variations to measure specific contact resistance.[371-[391 A particalar method should be selected depending on the type of application and required accuracy. In addition to the performance of ohmic contacts, the lifetime of ohmic contacts is often estimated by monitoring the variation of specific contact resistance over a period of time often versus temperature (to be discussed in Sec. 2.3 below). It should be noted that contact resistance measurements represent averaged electrical property of M/S contacts. This is because the dimensions of contact pads (dimension d in Fig. 3) typically range from-~ 10 (smallest) up to ~1000 gm, while the microstructure of the M/S contact is only homogeneous on the order of~l gm. In Fig. 3, current flow is not exactly perpendicular to the contact. The morphology of the contact is not ideal planar as depicted due to complicated metallurgical reactions, as discussed below.
1.2
Surface States and Fermi Level Pinning
In the development of Eq. (4), it is assumed that the M/S interface is free of structural defects and foreign atoms and is abrupt on the atomic scale (i.e., it is structurally ideal). In this case, the ideal barrier height (Oh) is determined by the work function of metal versus the electron affinity of the semiconductor. Experimentally measured barrier heights for many
90
Wide Bandgap Semiconductors
semiconductors, especially GaAs, show a very weak dependence on the metal work function.J4~ 42]
Metallized contact pad M/S~contact ~
~ / , ' /~
~
Semi-insulaing substrate conduction channel Figure 3. A schematicof Transmission Line Method (TLM)for measurementof specific contact resistance.
This deviation from ideal Schottky behavior was first attributed to the existence of surface states (intrinsic or extrinsic) by Bardeen in 1947.[4~ When surface states are present within the bandgap of semiconductors, they are occupied by electrons prior to contact formation with metals. Occupation is controlled by the Fermi level which is constant throughout the crystal. Consequently, these surface states pin the Fermi level resulting in band bending as shown in Fig. 4. As a result, the barrier height (Oh) is not given by Eq. (4) but must be modified by the details of Ess, the energy levels of the surface states. In Fig. 4, the barrier height can be described as Ob = Eg- Ess.[4~ As shown, the barrier height is nearly independent of the metal work function due to the Fermi level pinning. Fermi level pinning may be described by introducing extra charge (Qi), trapped in the surface states between E v to Ess, as shown in Fig. 4. Now charge neutrality is determined by: Eq. (17)
Qm + Qi + Qs = 0
Ohmic Contacts to II-VI and III-V Compounds
91
as contrasted to Qm + Qs = 0 in an ideal M/S interface. The potential variation (A), in Fig. 4 is due to Oi trapped at the surface states. The surface states may be a source of leakage current in M/S contacts and result in deviation from the ideal current-voltage characteristics predicted by Eq. (7), particularly in the low-voltage region.
Metal
n-Semi con ductor
Vacuum level A s
Om,T Erm _ __*_ _v_
Vacuum level ..
s Ec . . . . . Em
-I ss
~ I ' ,
Ev
Figure 4. Energy band diagram for a Schottky barrier formed by a metal and a semiconductor with surface states. Essis the surface state energy levels within the bandgap. LO
ev! / ,,I2: 0.6 ...... " 0.4
....
-
.-
."
_ "
,lCBM
LU Z ILl
0.2
o I," , - C s /
Go ....."
.Sb."Au,,0,y
.IVBM
,,.,"
,/
ev 1.2
In P (110 )
CBM
o A
O
0.8 0.4
--
A
Eg
l
VBM OVERLAYER PRODUCING PINNING ( S u b - M o n o l a y e r ) .
Figure 6. Fermi level pinningposition for n- or p-type (a) GaAs,(b) GaSb, or (c) lnP versus the overlayer adatom (from Ref. 49).
94
Wide Bandgap Semiconductors
Metal Induced Gap States. The Metal Induced Gap (MIG) model has been developed as an extension of the UDM.[5~ In the MIG model, surface states are postulated to be an intrinsic property of the semiconductor, and they exist over a continuum between the Valence Band Maximum (VBM) and the Fermi level. This is in direct contradiction to the discrete levels assumed in the UDM. The MIG states (surface states) are formed in the semiconductor at the initial interface, due to intimate contact with electrons from the metals. The tails of the metallic wave functions decay into the semiconductor and result in the interface states within the bandgap. Heine first pointed out that tails of the metal wavefunctions are derived from the Virtual Gap States (ViGS) of the complex band structure of the semiconduetor.[S~ In the MIG model, the character of the surface states changes across the bandgap from more acceptor-like, close to the Conduction Band Minimum (CBM), to more donor-like, close to the VBM. The energy at which the contributions from both bands are equal is called the branch point, and is located at mid-bandgap when the effective mass of electrons and holes are of equal value.[ 51] The Charge-Neutrality Level (CNL) is defined as the energy where the Fermi level coincides with the branch point. This determines the Schottky barrier heights. According to the MIG model combined with electronegativity concept, charge transfer occurs across the interface depending on the electronegativity difference between metals and semiconductors, and determines the final position of the Fermi level within the bandgap. For example, when the electronegativity of a metal is the same as the semiconductor, the final Fermi level pinning is located at the branch point of the semiconductor and there is no charge transfer. The final pinning position of the Fermi level should be above or below the branch point of the semiconductor when the metal exhibits a smaller or a larger eletronegativity than the semiconductor. With the Fermi level above or below the CNL, the net charge in the wavefunction tails of the metals has a negative and a positive sign, respectively, as a result of charge transfer between the metal and semiconductor. The Schottky barrier height of the M/S contact would be lowered with the Fermi level above the CNL, and increased with the Fermi level below the CNL, for n-type. The precise position of the Fermi level within the bandgap and corresponding barrier height depends on the occupation of the continuum of the MIG states, and the trend of occupation can be predicted from the electronegativites of metals and semiconductors:[ sll[s2] Eq. (18)
Ohmic Contacts to I I - V I and I I I - V Compounds
95
where tY~cnI is the CNL, Sx is a slope parameter defined as ~b/c m, and ~m and Zs are the eleetronegativities of metal and semiconductor, respectively. The slope parameter S x, (this is the same as S in Fig. 5) was related with the optical dielectric constant, Co, of the semiconductor, while (Y~cnl w a s theoretically predicted. Using Eq. (18), t~b -~ 0.7 eV above the VBM is predicted for Au/GaAs contact. In a recent refinement of the MIG model,[ 52] it was shown from experimental data[ 53] that the position of Fermi level pinning was a function of metal layer coverage. The Fermi level approached its final position as the surface was saturated from submonolayer coverage (isolated adatoms) to a continuous metallic film. The dependence of the Fermi level upon metal coverage (e.g., Fig. 7b), which can't be explained by the UDM, was explained in the MIG model by using the idea of metal-induced surface states at submonolayer coverage and by the continuum of metal-induced gap states at several monolayer coverage. At high coverage, only one Fermi level was determined, regardless of n- or p-type semiconductor. Chemical Reaction Models. Chemical reactivity between semiconductors and metals have been correlated to explain the deviation from the ideal Sehottky barrier; it was rationalized that chemical reactions would affect interfaeial electrical properties. The heat of phase formation was found to exhibit a linear relationship with the barrier heights (~b) between transition metals and Si.[ 54] Later, Freeouf proposed that ~b was determined by ~Siliciae (work function of silieides).[ 55J Even though this idea was developed to describe metal/Si contacts, it was later applied to various compound semiconductors and developed as the Effective Work Function (EWF) Model.E56] In the EWF model, ~b is given by: Eq. (19)
~b = ~eff-- Z instead of Ob =
(I)metal-
Z
where Z is the electron affinity of the semiconductor. r mainly due to the work function of the anion, ~Anion" The EWF model suggests that the Fermi level at the surface (or interface) is related to the work functions of microelusters of one or more interface phases resulting from either oxygen contamination (oxidation) or metal-semiconductor reactions with the metallization. Another chemical argument proposed is that chemical reactions at the interface on a microscopic scale modify the ideal Sehottky barrier via local charge transfer and creation of extrinsic interface states as a result of the interfaeial reaetion.[571[58] In this argument, discrete levels of defects, native or extrinsic, were used to account for the deviation from the ideal Sehottky behavior with chemical reactivity.
96
Wide Bandgap Semiconductors
Amphoteric Defect Model. Walukiewicz has attributed Fermi level pinning to amphoteric native defects in semiconductors.[59]-[ 61]This model relates the Fermi level pinning to the thermodynamic properties of the entire system of defects. A remarkable similarity was found between the semiconductor Fermi level (EFs) at the M/S interface and the Fermi level (EFI) in heavily irradiated III-V compound and column IV semiconductors (see Fig. 7). As shown in Fig. 7, the Fermi levels for n and p-type semiconductors merged into one level as the irradiating electron dose increased to greater than about 1017cm-2.[6~ Also note, there is a similarity between the convergence level in Fig. 7 and the pinned level shown in Fig. 6.
-/• 1.4-
>
I
C
t.u
%
1.0-
(a) _
9n - T y p e o p-Type ~ Semi-insul.
\eN
0.8~
_
" ~ - ' Z ~ . - - e._e_ ~ - e - & ,
0.6-
/
. . -
/
~_E 0 . 4 u_
I
GaAs
9 12 >, O}
1
1
N
0
0.2[ 1017
1 16
0
L 1018
1019
10 20
Electron Dose (cm -2) _A
>(1)
rw
1 29
0.8
~c-
~S
~9 0.4 LL
(b) -
,~
/.~
0 LAo V
~'-~-
..o- - z-o~-t== ~ , Ti/GaAs "'~
,,n-Type o p-Type , ,
1022
10 -1
1
10
102
Metal Coverage (~) Figure 7. Comparison of the Fermi level behavior (a) in electron irradiated GaAs and (b) for a submonolayer coverage of Ti (from Ref. 60).
Ohmic Contacts to 11-111and III-V Compounds
97
Table 1 lists the range of Fermi levei pinning positions deduced from the Schottky barrier heights for M/S contacts (Eps) and the Fermi level stabilization energy in heavily irradiated III-V compounds and column IV elemental semiconductors (EFt), plus the CNL predicted from the MIGS model (EcNL). Note that the Epi correlates better with the interface state pinning levels (E~s) than do the CNL values.
Table 1. Fermi Level Stabilization Energy in Irradiated Semiconductors (EFI) and at Metal-Semiconductor Interfaces (EFs)
Si InP GaAs
EEl (eV)
EFS (eV)
EClVL(eV)
0.4 1.0 0.5-0.7
0.3-0.4 0.8-1.1 0.5-0.7
0.36 0.76 0.5
Eclvl"represents Charge-NeutralityLevel from MIG model. All energies are with respect to the valence-band edges (from Ref. 60 and references therein).
According to this model, there is a Fermi-level stabilization energy (EFI) in covalent or weakly ionic semiconductors such as GaAs and InP, which is independent of the type of doping and the doping level. Therefore, this property is regarded as an intrinsic property of the semiconductors. As a consequence of this intrinsic property, native defects such as vacancies or substitutional dopants exhibit an amphoteric character depending on their energy level relative to the Fermi-level stabilization energy. For example, a Ga vacancy is a stable aceeptor in n-type GaAs, but it transforms to a donor complex (Asca + VAs) in p-type materials. This behavior results from a large electronic contribution to the total defect formation energy. Specifically, the formation energy of a Ga vacancy is lowered from --4eV to --0.2eV as the Fermi level varies from the VBM to the CBM under As-rich condition.[ 62] The defect formation energy varies until the Fermi level reaches the stabilized position (EF/), after which continued introduction of electrically active species does not affect the stabilized Fermi level. In the case of GaAs, the stabilized Fermi-level is located between E v +0.5 eV to E v +0.7 as can be seen in Table 1. In the amphoteric native defects model,
98
Wide Bandgap Semiconductors
the behavior of native defects is responsible for the Fermi level pinning, which appears to be similar to the UDM model. However, the behavior of defects is controlled by the stabilized Fermi level EFI, which is an intrinsic property of semiconductors similar to the MIGS model. Summary. For many years, all the models mentioned above were examined and compared. Nonetheless, the main idea of whether intrinsic or extrinsic effects play a primary role in Fermi level pinning still remains controversial.j63]-[ 66] It is now generally accepted that a single theory cannot explain Fermi level pinning. In the case of GaAs, it is readily and clearly seen that there is a range of-~0.3 eV in Fermi-level pinning (0.5-0.8 eV above the VBM) by reviewing the extensive experimental data collected to date. This variation implies that several mechanisms are simultaneously playing roles and even interacting with each other in the determination of SBH.
2.0
O H M I C C O N T A C T S T O GaAs
GaAs is the most widely used III-V semiconductor and as a result, ohmic contacts to it have been extensively studied. A few techniques have been developed to prepare an ohmic contact with no or little heat treatment after growth ofGaAs epitaxial layers. However, these in situ approaches to ohmic contact formation have not proven to be very practical. Therefore, the majority of this review will focus on contact schemes using deposition of metallization after the growth process has been completed and the sample removed from the MBE or MOCVD systems and stored in laboratory ambient for significant lengths of time. These techniques invariably required heating after deposition of the metallization, and they are called ex situ contact schemes.
2.1
In Situ Contact Scheme
In situ ohmic contact schemes do not utilize heat treatment to produce complicated alloying or metallurgical reactions for incorporation of doping elements into the surface of the GaAs substrate. Instead, very heavy doping is accomplished in situ during the growth of GaAs, or heterojunctions are formed to lower the barrier height between the contact metals and GaAs. In principle, heterojunctions may be either gradual or abrupt junctions, and both have been used for GaAs contacts. A gradual
Ohmic Contacts to II- VI and III- V Compounds
99
junction is one in which two bulk crystals are joined by a continuously varying composition (e.g., GaxInl_xAs/GaAs where x can vary from 0 to 1). An abrupt junction is a sharply defined interface between two homogeneous semiconductors (e.g., Ge/GaAs). For either doping or heterojunction formation, in situ contacts are nonalloyed because they are formed without post-deposition heat treatment. Nonalloyed contacts were developed to reduce some drawbacks of conventional metallization schemes, such as poor morphology, poor reproducibility, deep interdiffusion (deep junction), and poor thermal stability. They succeeded in minimizing some of these effects, but the penalty in product throughput was too great to be practical. Molecular beam epitaxy is frequently used to grow heavily doped n+-GaAs where n + can be as high as the mid-1019 cm -3 (i.e., above the critical carrier concentration calculated from Eq. 16 as being required to ensure TFE or FE transport across the M/S interface). The carrier concentrations produced by MBE are often as much as an order of magnitude above the values achieved in bulk crystal growth. MBE can produce such high carrier concentration because it is a non-equilibrium growth technique. Dopant incorporation may be controlled by surface kinetics rather than thermodynamic equilibrium conditions. Tin dopant has been studied because it is known to be less amphoteric as compared to Si or Ge. A free electron concentration as high as 6 • 1019 em -3 was achieved with Sn, resulting in specific contact resistance as low as 2 x 10 -6 ~_cm 2 simply by depositing metals without heat treatment.[ 67] Even though it tends to be amphoteric, Kirchner et al. reported--1 x 102~cm -3 Si doping which yielded contact resistances of ~1.3 x 10-6 ~_cm 2 with in situ metallization.[ 6s] As reported above, in situ formation of heterojunctions have been studied to form ohmic contacts. For example, Fig. 8 (a) and (b) show the band alignment between metal/GaAs and metal/n+-Ge/n+-GaAs, respectively. Because of Fermi level pinning, the Cb for metal/GaAs is 0.7-0.8 eV above the VBM as discussed already. The localized potential barriers of --0.45 eV at the metal/Ge interface and 0.06 eV at the Ge/GaAs heterojunction are much more favorable for ohmic behavior at room temperature .[691[7~ The dominant resistance will occur at the metal/Ge contact, but the Ge layer can easily be doped up to--102o cm -3, resulting in specific contact resistances of--10-6-10 -7 ~-cm 2 with smooth interfacial morphology.[69][ 71] The Ge/GaAs heterojunction is lattice matched (within 0.5%), and has compatible crystal structure. The thermal expansion coefficient of Ge (6.6 x 106/~ matches well with that of GaAs (6.0 x 10-6/~
100
Wide Bandgap Semiconductors
Metal
(a)
n+-GaAs
% Ec Ef
Ev
J Ec . . . .
Ef
(b)
I
Metal
!
,n§
,
n§
n-GaAs
Figure 8. Energy band diagram for metal contacts to GaAs showing the band alignment and bending for (a) a degenerately doped n-GaAs, and (b) the same GaAs with epitaxed Ge at the interface with a metal (from Ref. 7 l).
Another heterojunction scheme is to terminate the GaAs surface with Gal_xInxAS, in which the Fermi level is pinned in or near the conduction band as shown in Fig. 9.[72] Because there is a conduction band discontinuity between InAs and GaAs as shown in Fig. 9(e), a non-abrupt heterojunetion is necessary as shown in Fig. 9(d). In this contact scheme, tunneling is not required and low resistance contacts can be made for a wide range of doping without the need of alloying to form n + surface layers. InAs and Gal_xInxAs layers grown in situ by MBE produced contact resistances o f - 1 0 "7 f2-em2.[17][731However, this contact scheme is not generally practical because of MBE production throughput and
Ohmic Contacts to II-VI and III-V Compounds
101
inability to complete processing the device structure requires "dry" techniques in the MBE UHV environment.
Ec o)~////////////~
'
~,~M ! ETAL2//"
~
'
EF
n- GoAs
Ec b) ~
..
~..-~---~
. . .
EF
Ev ~,,~.__
n-InAs I.
c)-: MET.,;..,
.
.
.
.
.
'
. ,.~/,:.
Ec EF
s
ABRUPT NTERFAC :
:...,;;.::~;, I/,'.,
",9 , , . / / 3 " / . ,
NON ABRUPT INTERFACE
r
\
I
'
...........
.............; - ~ ~ ~ , - ~ Go i_x Inx As
Figure 9. Band diagram showing band b~nding and alignment for a metal conlact to (a) nGaAs, (b) n-InAs, (c) n-GaAs with a thin epitaxially abrupt layer ofn-InAs, and (d) n+-InAs layer on a graded n+-InxGal.xAS layer on n-GaAs (from Ref. 72).
102
Wide Bandgap Semiconductors
Fischer et al. [741175] have reported that sulfur passivation of the GaAs surface can at least partially remove surface pinning and result in ohmic contact formation. In their studies, solutions of P2Ss/Sx/NH4S were used to remove the native oxide on the GaAs surface and replace it with S bonding to the Ga and perhaps As. Samples of Si doped GaAs were then placed in a vacuum, S desorbed by heating to about 500~ and films of Au deposited in situ. Upon removal from the vacuum and without heat treatment, the contacts showed a linear I-V dependence demonstrating formation of an ohmic contact. Surface passivation with S, plus the H20/light treatment of GaAs surfaces reported by Woodall et al., [76]are the only known methods to at least partially unpin the GaAs surface for contact formation.
2.2
Ex Situ Contact Schemes Ex situ contact schemes are those using single or multilayer metal
thin films deposited in a separate chamber from that used in epilayer growth. Thermal annealing is generally required to form the ohmic contact. Single layer metallizations will be reviewed first, followed by a review of bilayer and multilayer metallizations. Single Metal/GaAs Contacts. While single metals on GaAs do not yield technologically feasible ohmic contacts, a review is worthwhile to learn about their interfacial reactions. Special attention has been given to several frequently used metals such as (Au, Ni, Pd, or Ge)/GaAs. It will become obvious that interfacial reactions between metals and GaAs play a critical role in the ex situ formation of ohmic contacts. It will also become obvious that equilibrium thermodynamics can be used to predict GaAs interfacial reactions, although non-equilibrium intermediate phases may also be encountered which may affect the contact properties. A u / G a A s Metallization. Au is frequently used as a contact metal for GaAs and other semiconductors because of its ease of deposition and etching, high conductivity which results in lower resistance interconnects, high ductility for bonding, and lack of oxide formation which results in good bonding and high reliability. Au begins to react with As depleted or near stoichiometric GaAs at ~250~ with rapid reaction at ---400~ resulting in large changes from the as deposited Schottky barrier heights.[771 Typically, Au/GaAs diodes exhibit a barrier height of--0.9 eV as deposited and low barrier Schottky or ohmic behavior after heat treatment.[ 781-[8~ The interfacial reaction between Au and GaAs upon annealing begins with the formation of a Au-Ga-As solid solution with very low As concentrations.[ 811
Ohmic Contacts to II-VI and III-V Compounds
103
The solubility of As in Au-Ga-As is so low that the reaction between Au and GaAs is commonly written: Eq. (20)
Au + GaAs ~ Au-Ga + As x (gas)
which represents the formation ofAu-rieh Au-Ga solid solutions with As x sublimation in the case of an open system (i.e., with a possible loss of Asx). In the ease of a closed system (i.e., no loss of As x possible), As forms precipitates because of the very low solubility of As in Au. [81]-[83]A sharp evaporation peak for As x has been observed by mass spectroscopy, suggesting rapid onset of a metallurgical reaction such as Au-Ga formation. Surface morphological changes have sometimes been interpreted to indicate formation of a liquid phase of Au-Ga alloys at the temperature ranges.[78][821184][85] However, in other cases, these morphology changes have been attributed to solid state transport without formation of a liquid phase.[ 81] While the bulk equilibrium phase diagram shows that a number of intermetallic phases are possible (Au7Ga 2, Au3Ga, Au2Ga, AuGa, or AuGa2), they have only been observed after cooling.j85] -[89] In contrast to the surface capillarity driven surface morphology mentioned above, solid state reactions between Au and GaAs often result in formation of elongated pyramidal pits (see Fig. 10a) bounded by {111 } planes and aligned in the [ 110] directions of GaAs after annealing above -3 50~ [90]-[92] Figure 10b is the corresponding schematic of Fig. 10a, and Fig. 10c is a bright field image of a {110} fractured cross section though a single pyramidal pit. Solid solutions of Au-Ga were found in the pyramidal pits. These solutions were sometimes separated from the GaAs substrate by an intermediate layer of Au-Ga compounds which probably formed upon cooling (i.e., they represent a divorced eutectic decomposition). The reactions pits form through solid-state dissolution of GaAs up to 450~ although a liquid reaction apparently increases their size and governs overall morphology above the melting temperature (--500~ of the Aurich solid solutions. The crystallographic orientation between the reaction products and the parent Au and GaAs has been attributed to the degree of misfit at the interface.j88][ 91] Others have suggested that the crystallographic dependence of GaAs dissociation kinetics may also influence their geometry.[ 92] In any case, formation of the pits on a GaAs surface during interaction with Au represents a very inhomogeneous reaction. The inhomogeneity of the interfacial reaction has also been attributed to the presence of native oxide which limits the ability of Au to react with the substrate.
104
Wide Bandgap Semiconductors
(a)
{111)As
[011"-] [o11 (b)
BaAs (100~
{ 111}Ga
(c)
Figure 10. Reaction "pits" from Au on GaAs after the Au has been chemically stripped. (a) Scanning electron micrograph of a pit. (b) Schematic representation of the crystallographic orientation of the pits. (c) Scanning electron micrograph of a fracture cross section through a pit (from Ref. 92).
Ohmic Contacts to II- VI and III- V Compounds
105
Upon reaction of Au with GaAs, the electrical characteristics of the interface have been observed to switch from Schottky to ohmic.[ 8~ To determine if this resulted from interfacial compound formation, Leung et al. [791and Lince et al.[941deposited Au-Ga phases (e.g., AuGa2) on GaAs. In general, these compounds did not result in ohmic contact formation because the surface of the GaAs remained pinned. In addition, the lack of or the limited interfacial reactions that were produced did not significantly modify the interfacial doping concentration, and Schottky contact behavior was maintained, although the quality of the Schottky contact (and barrier height) often was degraded. The chemical reactions between Au/GaAs were studied in detail by Mueller et al. to understand their effects upon the electrical properties of the Au/GaAs system.[8~ 811 It was shown that the Au/GaAs interfacial reactions were dependent upon the conditions of the initial interface. Surprisingly, an interface with a native oxide exhibited the most pronounced formation of reaction pits, consistent with the calculations of Mueller and Holloway showing that the phase stability of Au/GaAs is critically dependent upon the surface stoichiometry. Li and Holloway[771 showed that the Au/GaAs reactions could be modified extensively by supplying a flux of As to the surface during heat treatment. Liu and Holloway[ 921showed that the interfacial reactions pits underwent Ostwald ripening with time at temperature (i.e., some pits disappear while others grow during isothermal annealing). Appearance of the reactions pits results from dissolution of GaAs. Therefore, their disappearance indicates regrowth of GaAs, apparently from Ga and As in Au-Ga-As solid solutions. The reaction of Au with GaAs correlated with the electrical properties switch from rectifying to ohmic. The explanation for forming ohmic contacts was uncertain since the Fermi level was believed to remain pinned, and no dopant was added to the pure Au layer which could be incorporated during regrowth. The only possible dopant to be concentrated was the Si originally in the bulk GaAs. Segregation in the reactions pits of this Si dopant was detected by SIMS (see Fig. 11). Figure 11 is a SIMS image of m/e = 28 from a Si-doped GaAs after reaction with Au film at 450~ for 15 rain followed by chemical stripping of the Au film. Thus, the ohmic contact resulted from Si segregation in the pits causing a local n+region and TFE transport for ohmic behavior. Regrowth of the GaAs occurred as a result of the interfacial reaction, but did not play a role in the ohmic behavior. Ni/GoAs Metallization. Ni reacts uniformly with GaAs except where the native oxide-hydrocarbon interfacial contamination layer affects
106
Wide Bandgap Semiconductors
diffusion and compound formation.[95][96] Solid-state reactions between Ni and GaAs produced a hexagonal ternary phase after annealing at 100400~ for times of 5 min to 5 hours: Eq. (21)
xNi + GaAs = NixGaAs
g.
t Figure 11. SIMS imagefromm/e = 28 (i.e., Si§ showing segregation of Si in the reaction pits producedby Au on n-GaAs(fromRef. 93).
Diffraction analysis with TEM and XRD showed that NixGaAs could exhibit twinning and several epitaxial orientations with respect to the GaAs substrate. [97]-[106]The composition x was reported to vary between 2 and 4 without a change in crystal structure. Quantification of AES data yielded Ni2GaAs [96][97][99]while EDS and/or TEM measurement suggested NiaGaAs.[l~176176 1~ High spatial resolution EDS studies yielded NiE.4GaAs, as shown in Fig. 12.[1~ It is noted that compositions ranging from 2 to 4 were observed with the same NiAs-type hexagonal structure with varying Co/ao ratios.[a8][ 109] Studies of bulk material showed that the Ni-Ga-As ternary phase diagram exhibited five ternary phases with broad homogeneity ranges extending toward the binary phases.[ 1~ They all exhibited the hexagonal NiAs symmetry and were unstable in contact with GaAs.
Ohmic Contacts to II- VI and III- V Compounds
107
Figure 12. TEM micrograph of as-deposited GaAs/Ni2.4GaAs/Ge. Bright-field image of cross sectional view to show sharp interface between GaAs/NiE.4GaAs (from Ref. 108). While NixGaAs has often been observed in thin film reaction, it may not be an equilibrium phase. It was reported that NiaGaAs adopts a B8 structure with lattice parameters intermediate between those ofNi3.55Ga2.0 and NiAs B8 structures.[ 95][100]-[104] The hexagonal unit cell of NixGaAs has lattice parameters of ao~4A, co-~5A, similar to the Ni-As and Ni-Ga systems (NiAs:ao = 3.619/1,, %= 5.034A and Ni3Ga2:ao = 4.0A, %= 4.983A).E100]--[105] It has been speculated that NixGaAs is observed even though it may be metastable because of the epitaxial relationship with GaAs which would lower its nucleation barrier below that ofNiAs x and NiGay. [99]In addition, Ni is the mobile species at temperatures below-~300~ where Ga and As are relatively immobile, allowing formation of Ni~GaAs while avoiding separation of Ga and As.[1~176 1111Figure 12 shows an interfaeial morphology of a NiE.4GaAs/GaAs contact, which is rectifying with a barrier height of 0.9 eV.[ 1~ It should be noted that the interface is uniform in contrast to Au/GaAs as shown in Fig. 10e. While the NixGaAs phase is commonly observed in thin film systems at low temperatures, Ogawa showed that NiEGaAs decomposed into binary NiGa and NiAs upon annealing at 500~ for 5 min.[ 971Decomposition proceeded through NiAs precipitation in a matrix of Ni2GaAs at temperatures higher than 350~ After annealing at 500~ for 5 min, NiAs or As-rich phases were found near the GaAs and [3-NiGa near the surface. [97] However, Laval et al. and Guivareh et al. reported that Ni2GaAs epitaxed to (111) GaAs was more stable than on (100) and was still observed after annealing up to 600~ [99][106]On bulk GaAs at 600~ after 1 hour, NixGaAs decomposed into NiGa and NiAs.[ 1~ After the 600~
108
Wide Bandgap Semiconductors
anneal, [3-NiGa epitaxed to the GaAs substrate. Randomly oriented NiAs grains, about 0.7 mm in diameter, were observed.[ 99][1~ Sand et al. reported epitaxy for both NiAs and NiGa after annealing at 600~ for 1 hour.[ 11~ While the exact phases at the interface varied in all cases, Ni greatly improved the tmiformity of the reactions, presumably at least in part, because it forms both Ni-Ga and Ni-As compounds. With respect to electrical properties, Ni alone did not result in formation of ohmic contacts. The Schottky barrier height increased slightly (0.76 eV to 0.83 eV) upon formation of Ni2GaAs, then decreased upon decomposition into the binary phases.[ 99] Pd,/GaAs Metallization. Similar to Ni, Pd reacts with GaAs and rapidly and more uniformly penetrates the interracial native oxide, resulting in higher reactivity with GaAs. TEM micrograph images showed the formation of a PdxGaAs phase below the native oxide, where x is typically 2. [21][104][112]-[114] The exact stoichiometry of this reaction product varied, as did its crystallographic orientation with GaAs.[ 1121-[115]Below 300~ annealing partially decomposed the PdxGaAs into Pd-Ga and Pd-As binary phases, which caused the metal/GaAs interfacial morphology to become nonuniform.[ 116] Vacuum annealing above 450~ caused the loss of the Pd-As phases, presumably due to arsenic evaporation. Lin et al. reported that bulk diffusion couples of Pd (~0.6 mm thick)/GaAs showed three stable ternary phases at 600~ 116]They were characterized by broad homogeneity ranges, similar to Ni/GaAs. The thermodynamically stable phases with GaAs at 600~ were PdGa and PdAs 2. However, loss of As in an open system was expected to result in a PdGa/GaAs structure. Ge/GaAs Metallization. To demonstrate that regrowth of GaAs in the presence of Ge could lead to ohmic contacts, Li and Holloway[77][ 117] deposited layers of Ga, As, and Ge on GaAs and annealed the samples up to 500~ This resulted in the formation ofepitaxial GaAs doped with Ge and the formation of ohmic contacts, clearly demonstrating the feasibility of the solid phase epitaxial method of forming ohmic contacts to GaAs. In addition, they demonstrated that a flux of As onto the surface could stabilize GaAs against dissolution by Au, and could be used to regrow GaAs with dopant from a surface covered by Ge and Ga films. Refractory Metals/Go.As MetaUizations. High melting temperature metal elements such as Pt, Ti, and W are frequently used and studied for thermally stable contacts because the processing temperature can be as high as 800~ to anneal out damage by ion implantation.[ 1181Pt/GaAs produces Pt-Ga, Pt-As binaries upon annealing 400-500~ Similarly, a
Ohmic Contacts to II-VI and III-V Compounds
109
Ti/GaAs reaction produces Ti-As and Ti-Ga above 400~ [1191The melting temperatures of those reaction products are typically above 1000~ 12~ Electrically, these reaction products exhibit rectifying contacts. As a result, refractory metals (elements or compound) are more commonly used as constituents in multielement contact metallizations.[~2 ~1 More details will be presented below (in See. 2.3) in a review of the thermal stability and reliability of ohmic contacts. Multi-element/GaAs Contacts. Practical ohmic contact metallizations contain more than two elements. However, the prior knowledge obtained in single element/GaAs rnetallizations is useful in understanding complicated reactions between multielements and GaAs. Au/Ge/GaAs Metallization. The reactions and electrical properties of Au/Ge/GaAs contacts are quite different from the Au/GaAs system described above. An eutectic forms at 88-12 wt% Au-Ge with a melting temperature of 363~ However, even small concentrations of Ge (-~0.6 wt.%) drastically changed the morphology of Au-Ga intermetallic phases (e.g., Au7Ga2, Au3Ga ) and they form at progressively lower temperatures with increasing Ge concentration.[ ~22]In addition, a Au-Ge-As phase was observed to cover the contact after annealing at ~400~ with decreasing coverage at higher temperatures. Epitaxial regrowth of GaAs was reported after cooling to room temperature from higher temperatures. Longer time anneals caused the Au-Ge-As phase to disappear, perhaps from local melting, followed by solidification upon cooling.[ sS] For annealing above -~400~ a molten Au-Ge eutectic phase was detected followed by formation of Au-Ga compounds. These reactions resulted in an ohmic contact with a specific contact resistance of~l 0-5 ~-cm 2. Formation of ohmic contacts was attributed to incorporation of Ge during solidification of GaAs, leading to the formation of a heavily doped n + GaAs layer.liE3][TM]This is consistent with the observation ofepitaxially regrown GaAs after cooling to room temperature. The liquid phase could accelerate formation ofn + epitaxial GaAs, and Ge was responsible for both the low eutectic temperature as well as acting as the n-type dopant. While in-diffusion of Ge into GaAs was suggested, the n + layer was probably formed during cooling. Reduction in the barrier height from 0.77 eV to ~0.4 eV was also reported after annealing below the Au-Ge eutectic temperature, but this again probably resulted from surface doping by Ge. [1231[125] There were no reports of epitaxially regrown Ge films at the interface, therefore the reduced barriers did not result from formation of Ge at the interface (depicted in Fig. 8). Instead, regrowth of GaAs at the interface, doped with Ge, was responsible for the reduced barrier heights
110
Wide Bandgap Semiconductors
and ohmic properties. The reaction morphology was similar to Au/GaAs as shown in Fig. 10, even to the formation of pyramidal reaction pits, and even to lack of a uniform morphology. Ni/Ge, Pd/Ge, and Pd/Si/GaAs Metallizations. As shown by the above reviewed data, Au has only limited power to dissociate the lattice. Nickel and Pd are much more efficient at dissociating GaAs and bonding to both the Ga and the As to hold them in the thin film. As also pointed out, contacts containing Au show marginal thermal stability, irregular morphology, and degradation of contact resistance with time. The properties of Au-free metallizations were studied to determine if removal of Au would eliminate the problems. Bielement metallizations such as Pd/Si,[ 126] Pd/Ge, [127] and Ni/Ge[ TMwere tested. Without Ni or Pd, very little Ge was incorporated into GaAs after sintering at 450~ for 30 min.[128]-[13~ With a Ni overlayer on Ge to cause dissociation of GaAs, greatly enhanced incorporation of Ge resulted in ohmic contacts.[ 1281The Ge profile after annealing always followed the Ni distribution, suggesting a correlation between these elements during regrowth of GaAs.[ 13~ Marshall et al. showed that Ge/Pd/GaAs structures resulted in Pd dissociating the GaAs to form a ternary Pd2GaAs phase. This was followed by the formation ofa PdGe x intermetallic compound causing the release of Ga and As and regrowth of GaAs.[ 21][131][132]When regrowth occurs in the presence of a dopant (e.g., Ge), the dopant will be incorporated into the regrown layer and may result in ohmic contact formation. The concept of the regrowth mechanism was applied to explain ohmic behavior for Ni/Ge/GaAs metallizations.[ 21] In this model, Ni was postulated to react with GaAs to form NixGaAs, then decomposition of the ternary phase was driven by the lower free energy of formation of NiGe to result in: Eq. (22)
NixGaAs + xGe ~ xNiGe + GaAs(Ge)
where Ge incorporation is believed to result in n+-GaAs. Typically, the Ni and Pd germanides are formed by annealing above 400~ and they have a very smooth surface morphology and good thermal stability (up to 500~ due to their high melting temperature.[ 127]A minimum thickness ratio of Ni to Ge (29-3 8 at.%) was necessary to form ohmic contacts. Regrown GaAs was distinguishable from the single crystal wafer by a high density of stacking faults, microtwins, and precipitates.J21][ l~ The regrown GaAs was postulated to contain a high concentration of dopant (e.g., Ge with a concentration of ~ 1019 cm3). However, limited experimental data was presented to support the incorporation of Ge to form n+-GaAs.
Ohmic Contacts to II- VI and III- V Compounds
111
When the Pd/Ge layering sequence was reversed and samples of Pd/ Ge/GaAs were reacted below 400~ only a rectifying contact was observed. AES depth profiles showed that Pd-Ga and Pd-As phases were formed while Ge segregated near the surface to form PdGe and Pd2Ge.[ 129] The interfaeial barrier height was increased from -~0.7 eV to --0.8 eV by these phases. Reactions at 400-500~ were necessary to form ohmic contacts, and the Ge SIMS profile into GaAs correlated with the reacted layer.[ 13~ To explain these results, it was commonly postulated that Pd created Ga vacancies, which accelerated the diffusion of Ge into the nearinterface region to form n+-GaAs. The premise that diffusion of Ge caused formation of an n + layer was not directly supported by experimental data. Instead, incorporation of Ge during regrowth of GaAs was postulated by Marshall et al., without the need for solid state diffusion. As illustrated in Fig. 8, epitaxial Ge will potentially form nonalloyed contacts to n-GaAs. This was studied using Ge/Pd/GaAs metallization.[131 ][1321Upon annealing, Pd and Ge reacted to form PdGe while excess Ge was transported through the Pd layer to grow epitaxially on the GaAs substrate. To clarify the mechanism for ohmic transport, SIMS analysis was performed by removal of the vast majority of the GaAs substrate through etching of a parting A1GaAs layer. The concentration of Ge incorporated into the GaAs near-surface region was measured. [134] Concentrations ofGe of-~l x 1019cm-3 were correlated with the onset of ohmic behavior. Thus, epitaxial Ge was not the origin of the ohmic contacts; instead creation of an n § layer in regrown GaAs was concluded to have resulted in ohmic contact behavior. The characteristics of Si/Pd/GaAs were investigated and compared to those of Ge/Pd/GaAs. Again it was shown that an epitaxial Ge layer with a resultant low barrier heterojunction was not responsible for ohmic behavior.[ 126] Instead, a regrowth mechanism was suggested with excess Si reacting with PdaGaAs to produce the following: Eq. (23)
2Si + Pd4GaAs(Si ) ~ 2Pd2Si + GaAs(Si)
The critical product from this reaction is regrown GaAs doped with Si. This same mechanism was also tested in Ni/Si/GaAs which clearly showed that regrowth took place.[1351 Binary metallization of Au/Ni/GaAs is not discussed since this only results in formation of Schottky contacts. Without the presence of Ge, low resistance ohmic contacts cannot be formed.
112
Wide Bandgap Semiconductors
AuGeNi/GaAs-Diffusion Doping Model. AuGeNi contacts are the most heavily studied system for GaAs, partly because it is the oldest contact system. It was first introduced in 1967 by Braslau et al. to form ohmic contacts to GaAs-based microwave devices.[ 1~ The postulated mechanism of ohmic contact formation has undergone considerable evolution. The initial postulate was that a Au-Ge liquid eutectic formed and dissolved some GaAs which caused Ge to "diffuse" into the near surface region. Diffusion was possible because dissolution of Ga in Au would create a counter flux of Ga vacancies, which increased the transport of Ge by orders of magnitude.[ 1361The Ge created an n + layer and an ohmic contact. The role of Ni in this process was simply to prevent the AuGe melt from "bailing up" due to surface tension. Therefore, the three elements were evaporated to form a bilayer/substrate structure of Ni/AuGe/GaAs. Several investigators showed that with sufficiently high temperatures and long times, the sequence of the layers did not affect the final metallurgical or electrical properties.[ 137][138]A modified postulate was that Ni improved the surface morphology by improving the wetting of liquid Au-Ge to GaAs. It was observed that Ni and Ge accumulated at the interface with GaAs near the AuGe eutectic temperature, while Ga accumulated near the surface. [139]The evolution of the contact modeled continued when it was realized that the Au-Ge eutectic was not necessary and did not play a major role since Ge was gettered by Ni via a solid-state reaction.J14~ TM]The model evolved by recognition that Ni was correlated with the dissociation of GaAs, and the following reaction was suggested:[ 142] Eq. (24)
Au + Ni + GaAs --~ Au-Ga + Ni-As
Formation of nickel arsenides occurred at the GaAs interface and improved the surface morphology as illustrated in Fig. 13. [143]As shown in Fig. 13, without the first Ni layer on GaAs, Au/GaAs interfacial reaction dominated the interfacial morphology, which was not uniform as discussed above. However the reaction in Eq. 24 did not necessarily lead to formation of ohmic contacts; the barrier height, ~t~ increased from--,0.7 eV to ~0.9 eV as Ni accumulated at short times at the interface.j99][ 139]At longer annealing times, when ohmic behavior was observed, the surface morphology and contact resistance were both directly affected by the Ni to Ge ratio. Higher Ni concentrations led to better surface uniformity, while higher Ge concentrations led to lower contact resistance.[ 1441-[1471 The effect of Gewas still the creation of an n § surface layer by diffusion along a Ga vacancy flux.
Ohmic Contacts to II- VI and III- V Compounds
BE FORE ALLOYING
DURING HEATING ( BELOW 420*C)
i AutG,)'
Au
Ni
AF TER ALLOYING AT 440*C FOR 2~n
,
t
k'//
I
Au-Ge
SAMPLE A
GoAt
GoAt
"
~ A.(G',)
Au
N,
IJ-A. o , / / ' / A
Au (Ge,Go) ,,,m..
,,,.,Ira,.
(WITHOUT Ni FIRST LAYER)
113
4
Ni~Ge
i
Au- Ge
SAMPLE B
4' I AulGt,Go',
=Ni
/- ~" r .
(WITH 5nm Ni F I RST LAYER)
a,,,mm,
NixGoA$ NiAslGe) GoAs
GoAt
GoAs
Figure 13. Schematicillustration of the evolution of phases produced by the reaction of Au/ Ni/Ge layers on GaAs. Note that the first Ni layer on GaAs generally results in planar interface morphology. Generally,for a five elements system,the interface between reacted layers will not be planar (from Ref. 143).
An improved description of the evolution of the contact was suggested by the observation that Ni was always found to diffuse toward the interface and to react with GaAs. The Ge followed the Ni distribution. The Au was normally combined in Au-Ga compounds upon cooling. [85][138][139][141][142][148]-[150] Transmission electron microscopy (TEM) studies showed that annealing above 400~ for a few minutes followed by cooling to room temperature produced NiGe containing small concentrations of Ga and As, NiAs with Ge and Ga, and AuGa.[143][l 5~ The NiAs phase was in direct contact with GaAs resulting in a smooth interracial morphology (Fig. 13). A good ohmic contact with low contact resistance was attributed to a Ni2GeAs phase being in contact with n+-GaAs which resulted from diffusion of Ge from NiGe.[ 15~ In addition, the reaction was assumed to unpin E F, and the NiAs phases were suggested to be low barrier ohmic tunneling contacts where the contact resistance was a function of the NiAs(Ge) coverage at the interface. [23][143] At higher annealing temperatures, more NiAs phase with a highly oriented epitaxial
114
Wide Bandgap Semiconductors
relation was found at the interface.[ 85] The epitaxy was suggested to result from a close lattice matching between NiAs and GaAs. But even with NiAs at the interface, Ge diffusion to form n§ was necessary to explain ohmic behavior. Even though the Ga vacancy diffusion model for ohmic contacts was widely accepted, the activation energy for formation of contacts did not agree with the expected value for bulk diffusion. Therefore, Gupta and Kokle postulated that grain-boundary diffusion in the Au film was responsible for generation of Ga vacancies.[ 136] A Ga vacancy mechanism was also postulated to explain ohmic contact formation with Pd.[ 13~ Gupta and Kokle justified their grain boundary diffusion model based on the fact that contact resistance was high after either low or high temperature annealing, but low for intermediate temperatures.[ 136]This observation is generally true for any metallization leading to ohmic contacts on GaAs. Thus the "diffusion doping model" evolved to one in which it was presupposed that Ni dissociated the GaAs lattice and bonded to the As holding it at the interface. The Ga from this dissociation reacted with Au and formed Au-Ga solid solutions. Formation of AuGa x intermetallic phases was also believed to occur, especially during cooling after annealing. The formation of an n § surface region due to Ge diffusing along a counter flux of Ga vacancies was widely accepted as the mechanism of doping. Backside SIMS measurements in an alloyed AuGeNi system were used to demonstrate Ge doping of the near surface region,[ 152][153]but the profiles could not be analyzed in terms of a diffusion profile. In addition, the contact resistance correlated poorly with the detected Ge concentrations. Cross-sectional TEM showed that NiAs was in contact with the substrate. Based on these results, Bruce et al. suggested that n§ level was less important to contact resistance than the formation of the NiAs phase.[ 152]Further refinement of the model was clearly necessary. AuGeNi/GaAs-Solid Phase Regrowth Model. Because E F is pinned in GaAs (near mid bandgap at the surface), n § doping is critical to formation of an ohmic contact. The studies of interfacial reactions have been focused on the mechanism by which Ge occupies the Ga site to form n § GaAs. As pointed out above, while Au-Ga and Pd-Ga reactions have long been postulated to create Ga vacancies in GaAs, the data do not support this mechanism for Ge incorporation and doping. Therefore, the model of ohmic contacts formation in GaAs has evolved to the one of solid-phase regrowth, as discussed for Ni/Ge and Pd/ Ge metallization. Figure 14 is a simplified schematic to illustrate the evolution of phases in this model of ohmic contact formation with Au/Ge/
Ohmic Contacts to II- VI and III- V Compounds
115
Ni metallization. Between 100 and 300~ Ni reacts with GaAs to produce interfacial NixGaAs with low concentrations of Ge. Above this temperature range, the reaction normally results in "binary" phases consisting of NiGa x and NiAsy, both of which contain small concentrations of Ge. Simultaneous with Ni reacting with GaAs, Ni reacts with Ge to form NiGe, and Ge diffuses into Au. At longer times (Fig. 14b), the NixGaAs or NiGa and NiAs phases will be decomposed by formation of NiGe, whose free energy of formation is lower than that of either the ternary or the binary phases. The Ga and As released by this decomposition will lead to solid phase epitaxial regrowth of GaAs. Incorporation of Ge will cause the electrical properties to switch from rectifying to ohmic.
Ohmic contacts Au Ge Ni
GaAs
Au-Ge {de
Au-Ga
Ni-Ge
Ni-Ge
Ni-Ga-As-Ge
GaAs
(a) Interfacial reaction
GaAs
(b) Regrowth
Regrown GaAs (Ge) n+-GaAs
Figure 14. Idealized schematic of the evolution of phases during formation of an ohmic contact to n-GaAs using Au/Ge/Ni metallization. Even though planar interfaces are shown, non-planar reaction fronts are expected in five component systems.
Several key issues about the solid-phase regrowth model still need to be clarified to validate it. First, the evolution of phases at the interface needs to be clarified, and, since Ge is known to be an amphoteric dopant in GaAs, the mechanism(s) controlling selection of the Ga or As site in the GaAs matrix should be explained. A series of studies have been carried out
116
Wide Bandgap Semiconductors
in our laboratory to investigate the evolution of interfacial phases and their effect(s) on the formation of ohmic contacts. [74][75][77][93][108][118][154]-[157] Particular attention has been given to the reaction of Ni with GaAs and the evolution of the ternary hexagonal NixGaAs, since previous studies/21][135] showed that the contact properties were dependent on how this ternary phase evolved (see discussion of Ni/GaAs above). In previous experiments which demonstrated solid-phase regrowth, the entire layered structure was first deposited then annealed in one or two steps (e.g., first at 200-250~ prior to the final high temperature anneal).[ 135]As discussed above, multiple reactions occur simultaneously during this processing and it is unclear as to which reaction results in n§ doping. To better understand the reactions, Kim et al. deposited 650 A Ni films on GaAs and in situ annealed them at 300~ to form ~ 1300 A films of Ni2.4GaAs.[ 1~ This anneal was followed by in situ deposition of Ge and Ti films with various thicknesses, and evolution of the interfacial phases upon final annealing at 500~ was studied. Figure 15 is a schematic diagram of the experimental procedure. Ti rather than Au was used in the metallization to determine if the reaction sequence discussed above for the regrowth model (i.e., Ni dissociation of GaAs to form Ni-Ga-As-Ge phases which decompose to Ni-Ge and Ni-Ti plus epitaxial regrowth of Ge doped GaAs) was general. The general sequence of reactions was found for Ti/Ge/Ni metallization, supporting its general validity. In addition, the thickness of the Ge layer was varied from 300 A (thinner than the original 650 A Ni layer) to 750 A to control the degree of reduction of the Ni2.aGaAs phase to the NiGe phase, and thereby control the extent of GaAs regrowth. Ti was expected to assist the evolution of the interfacial phases at 500~ since Ni-Ti intermetallic compounds are formed with large, negative free energies of formation. By pre-reacting the Ni, the NiGe formed directly from Ni2.aGaAs + Ge, while the NiTi reaction occurred later after Ge had reacted with the Ni2.4GaAs layer. Separation of the reactions using this procedure was necessary since the overall metallurgical reactions at temperatures of -~500~ proceed rapidly over times of a few seconds.[23][148] In situ annealing to form Ni2.aGaAs allowed Kim to show that its evolution upon reactions with Ge and Ti proceeded along a route involving both transformation and decomposition.[ 1~ A flow chart showing the critical phases is shown in Fig. 16. In situ annealing at 300~ for 15 min produced Ni2.aGaAs, where the stoichiometry was determined using energy dispersive x-ray analysis (EDX) on a high resolution field emission STEM.[ lsS] After deposition of the Ge and Ti films on the Ni2.aGaAs and
Ohmic Contacts to II- VI and III- V Compounds
117
annealing at 500~ for 5 min, Ni2.4GaAs was decomposed to directly form NiTi and NiGe plus epitaxial doped GaAs as indicated by the dashed line, or decomposed into NiAs + Ni3Ga 2. This was followed by evolution into NiAs plus NiGeGa, and finally into TiNi + NiGe + GaAs(Ge), indicated by the solid line in Fig. 16.
Ge & Ti evaporation
Ni
I650A
Ni2.4GaAs
T-1300~
Ti Ge
Ni2.4GaAs
GaAs
Vacuum
I~ a t
nu = 2 910JScm4
GaAs
GaAs
annealing 500~
5rain
In situ annealing at 300~
Figure 15. Schematic of experimental procedure to investigate evolution of Ni2.4GaAs. See text for detailed explanation (from Ref. 108). GaAs + Ni
1
In situ annealed at 300"C, 15 minutes
Ni2.4GaAs /
/
/
Deposite Ge and Ti annealed at 500"C, 5minutes
~
!
I
NiAs + NisGa 2
I I ! ! I !, I
NiGeGa
! ! I
'
"
/
1
TixNi + NiGey + G a A s (Ge) Figure 16. Schematic of the evolution of phases on GaAs from an in situ anneal of Ni on GaAs at 300~ for 15 min followed by deposition of Ge and Ti and a subsequent anneal at 500~ for 5 min. See text for explanation (from Ref. 108).
118
Wide Bandgap Semiconductors
A representative set of data for the transformation and decomposition of Ni2.4GaAs phase are presented in Figs. 17-19. Figure 17 shows SIMS depth profiles of Ti/Ge/Ni2.4GaAs/GaAs. In Fig. 17a, the Ni, Ga and As signal are stable in the Ni2.4GaAs region, and the Ge and Ti layers are distinguishable at the surface. After annealing at 500~ for 5 min, the Ni and Ti signals are found together in the surface region, indicating formation ofNiTi x. The Ni2.4GaAs layer has been partially decomposed, but the Ge layer (250 A) was too thin to completely decompose this ternary film. The bilayer structure indicated by the SIMS data is shown by the cross section TEM mierograph in Fig. 18. The solid line in Fig. 18 indicates the location of the original interface between -~1300 A Ni2.4GaAs/GaAs. This interface obviously has moved towards the surface, indicating epitaxial regrowth of GaAs. In Fig. 18, a small arrow indicates trace of the decomposition of a Ni-As grain, resulting in the epitaxial regrowth of GaAs. Thus the area marked R in the figure is regrown GaAs. High spatial resolution EDX data shown in Figs. 19a and 19b were measured across an interface between regrown GaAs and a Ni-As (e.g., line scan 2 in Fig. 18) or Ni-Ga phase (e.g., line scan 1 in Fig. 18), respectively. These EDS results show that the Ni2.4GaAs transformed into Ni-As and Ni-Ga binaries. Since Ge is present in the Ni2.4GaAs and/or binary phases (4-12 at.% as shown in Fig. 19), it incorporated into the regrown GaAs to about 1020 cm-3.[ 158]This is sufficient to meet the requirements of TFE, as reported above, and therefore to achieve an ohmic contact. Electron diffraction analysis (e.g., from the structure shown in Fig. 18) showed several intermediate compositions for NiAs x and NiGay phases, most of which had crystal structures of the NiAs hexagonal type. They are believed to be intermediate phases along the reaction pathway followed to reach the equilibrium NiAs and Ni3Ga 2 phases.[l~ 159] Both NiAs• and NiGay phases must decompose to allow regrowth of Ge doped GaAs for ohmic contacts, and the decomposition products also depended critically upon the amount of starting elements. For example, in those samples where the amount of Ni was greater than the amount of Ge, the reaction to decompose Ni2.4GaAs or the binary phases did not go to completion. At times, this prevented the formation of an ohmic contact. When the ternary phase decomposed into the binary NiAs and Ni3Ga 2 phases, careful analysis with EDX and electron diffraction on the TEM Showed that Ge was incorporated predominantly at those regions where the binary phases decomposed and caused regrowth of GaAs. Figure 20 is a schematic summary highlighting the key information presented in
Ohmic Contacts to II-VI and III-V Compounds
119
Figs. 17-19 (i.e., the transformation of Ni2.4GaAs into NiAs x and NiGay binaries, and the decomposition of the binaries). The arrows in Fig. 20 indicate Ge indiffusion through the ternary/binary phases resulting in Gedoped regrown GaAs. Under some conditions, regrowth of GaAs occurred predominantly from decomposition adjacent to a NiGay binary phase at the interface, and ohmic characteristics were not observed for I-V data from these samples. In other cases, regrowth predominantly from a NiAs phase led to linear I-V data. Apparently, ohmic behavior was not uniform across the interface. The majority of the current was conducted across regions where the NiAs phase existed at the interface, consistent with the literature reporting a lower specific contact resistance as the percent of the interface covered by NiAs increased.[ 23][143]
(a) As-deposited le+5 ?
:' .....
I
GaAs
u~-a'--
"''"~'.,.'~""
"f
le+4
[-.. o~
10-2
< O u t.)
104
L) o~
104
1
I
I ..
6~
L
800
4O0
200
ANNEALING
TEMPERATURE
(~
(a) E E
I
I
I
SAMPLE-3
6 v u.I
o
l-. l.u
fls h0 < Z 0 0
0 500
.I 550
l
I
600
650
ANNEAL 1NG TEMPERATURE
700 ('C)
(b) Figure 21. (a) Specific contact resistance for conventional Au-Ge film (filled circle) or Gedoped isothermally regrown GaAs (from Ref. 80), (b) contact resistance from Ge/Ni/GaAs after annealing at various temperature for 5 minutes (from Ref. 21). Note U-shaped behavior of specific contact resistance with respect to annealing temperature.
Ohmic Contacts to II-VI and III-V Compounds 125
2.4
Summary
The formation of ohmic contacts to GaAs has been often studied and the model of their formation has undergone considerable evolution. Beginning with the empirical observation that Au-Ge/Ni formed a good ohmic contact to n-GaAs for Gunn diodes, the mechanism to explain this contact has evolved from one of surface tension and Au reaction, to one postulating doping due to a vacancy flux, to one of epitaxial regrowth after dissociation of the GaAs wafer. Each of the constituents of the contact, chosen empirically, contributes critically to the development of low resistance contacts. The role of Ni is to dissociate the GaAs lattice and to bond with both the Ga and As, often as a ternary phase, te keep them at the interface. Shortly after dissociation of the lattice, Ge reacts with the Ni bonded to Ga and/or As and forms NiGe, causing release of the Ga and As with Ge present to force regrowth of doped GaAs. Au helps drive the reaction towards completion by forming Au-Ga solid solutions. Based on this sequence of events, it is possible to develop rules for the formation of ohmic contacts to those semiconductors requiting heavy surface doping (e.g., when their Fermi level is pinned). The rules may be stated as follows: 1. The initial reaction between the semiconductor and metallization should dissociate the compound semiconductor. 2. The metallization should react with all elements in the compound semiconductor and hold them near the interface. 3. The metallization should have a subsequent reaction which leads to regrowth of the compound semiconductor. 4. A dopant or dopants should be present in the metallization. 5. The metallization should control the Fermi level during regrowth to ensure that dopant is incorporated on the proper lattice sites to result in a free carder density sufficient to yield low resistance ohmic contacts by tunneling or thermionie field emission transport. By following these rules, ohmic contacts should be possible for a large number of compound semiconductors.
126 3.0
Wide Bandgap Semiconductors O H M I C C O N T A C T S T O InP
InP has higher peak and saturation velocities of electrons than GaAs, rendering InP a potentially better material for high-speed and microwave devices.[ 178][179] However, InP is significantly more difficult to process since it decomposes at ~550~ versus GaAs at-650~ Due to the lower melting temperature of InP, metal/InP reactions are more rapid and morphology control can be more difficult. Since contact schemes for InP are similar to those being used for GaAs, it is worthwhile to review the interfacial and metallurgical characteristics and their effects on InP contacts. Katz [is0] reviewed the formation of ohmic contacts on InP based materials.
3.1
Single Element/InP Metallizations
Au/lnP Metallizations. Au is extensively used as a contact metal for InP, just as it is for GaAs devices. Upon annealing from just above room temperature to 400~ Au reacts with InP by In diffusing into Au to form Au-In solid solutions.[181]-[ 188] For example: Eq. (25)
Au + InP = Au-In + P
When the amount of indium in the Au films exceeds the solubility limits for solid solutions, a Au-In phase (Au3In) was formed.[ 182][1841[1851[187][189][19~ Eq. (26)
5Au + 3InP = Au3In + Au2P 3
It was suggested that phosphorus atoms either could leave the system or occupy nonlattice sites near the metal/InP interface.[ 185] It should be noted that Au did not react with As in Au/GaAs but that there was an extremely limited solubility of As in Au-Ga. The same may be true for P in Au-In. Annealing at 400~ for a few minutes changed the electrical property of the contact from non-ohmic to ohmic with-~l 0-4-10 "6~'~-cm2, which was attributed to formation of AuzP3 .[188][191] The surface morphology became rough when Au-In and AuzP 3 phases were formed.J182][ 185]Similar to Au/GaAs, the interface morphology was nonuniform due to formation of rectangular shaped reaction pits bounded by (111) planes.[185][Iss][ 1891The reaction pits were filled with Au-In.[ 1871Longer annealing times showed that Au-In solid solutions formed AuaIn upon cooling, and they were converted to Au9In4 as lateral spreading of the reaction zone took place.J188][19~
Ohmic Contacts to II-VI and III-V Compounds
127
At temperatures above -450~ Au2P 3 is dissociated and phosphorus evaporates as detected by a sharp evaporation peak at temperatures dependent upon the thickness of the Au film and independent of whether the substrate was n- or p-type.[ 184][187] Melting of the metallic layer was observed concurrent with the evaporation peak of phosphorus. For anneals above 450~ contact resistance increased by up to two order of magnitudes (to ~10 -4 ~-cm 2) upon dissociation of the Au2P3 phase.[191] Thus, the Au/InP and Au/GaAs interracial reactions are very similar to each other except that Au can form AUEP3 which decomposes above --450~ Ni/InP Metallizations. As deposited Ni/InP contacts exhibited rectifying behavior. The interfacial reaction in Ni/InP occurred upon deposition or after annealing below 250~ by Ni, producing amorphous NixInP, NiEP or NiaP.[192]-[196]Note that a rectifying ternary phase was formed near 250-300~ for Ni/GaAs, but it was crystalline. Annealing at higher temperatures above-300~ caused the amorphous phase to crystallize into a hexagonal NixInP (x-- 2.7), [192] monoclinic Ni2InP, and/or NiEP. [194][196]A low contact resistance (~10 -6 ~ - c m 2) was obtained with crystalline Ni2InP or Ni2P phases after annealing at 350~ and they were stable with time and temperature.[ 194][195][197][198]At 470 or 600~ for Ni-In-P, monoclinic Ni2InP was reported to be in equilibrium with InP at ~450~ after 3 month annealing.[ 1951This phase disappeared at 600~ At both temperatures, Ni2P, NisP 4, and NiP 2 were also in equilibrium with InP, but no Ni-In phases were found. [197] Ni2InP and InP had an epitaxial relationship with poor mismatch, and was suggested to be the cause of deviations from ideal I-V behavior after annealing at 5 0 0 ~ [199] It was suggested that the Ni/InP reaction proceeded from amorphization to segregation into Ni-P and Ni-In binary phases, and eventually recombination into the ternary phase.[ 1951With the Ni-P binary phases in contact with InP, contact resistances of 10-4-10 -6 f~-cm2 were measured. [~91 ][193][198] Pd/InP Metallizations. Pd/InP contacts were rectifying as deposited and an amorphous PdxlnP (x-- 3--4) ternary phase was reported. The PdxlnP remained amorphous but grew thicker at 175-225~176176176 Crys tallization into epitaxed cubic Pd2InP was observed upon annealing at 250~176176176 TM] Further annealing at 400~ caused decomposition of Pd2InP into Pdln and PdP 2. These two binary phases were stable after annealing at 500~ for 350 hours, but loss of P was detected when exposed to air or annealed at 650~ for 30 min. Thus Pdln and PdP 2 were reported to be the thermodynamic equilibrium phases with InP in a closed system at higher temperatures. The absence of a Pd-P phase was attributed to the
128
Wide Bandgap Semiconductors
sublimation of P due to its high vapor pressure.[2~1762~ Other bulk experiments to determine the Pd-In-P phase diagram identified several ternary phases, for example, Pd5InP, Pd3.sInP , and Pd5InEP2 (suggested to be PdEInP in the report by Ivey et al.). However, at 600~ Pd-P and Pd-In binary phases were found to be in equilibrium with InP. [203] It was noted that an important difference between Pd/InP reaction and Pd/GaAs reaction was that the initial PdxInP ternary phases produced by annealing at 300~ or below were amorphous, while PdxGaAs was hexagonal crystalline. Both systems have the same characteristics; the high temperature equilibrium phases in an open system are binary compounds, although the Pd-P and Pd-As phases are difficult to maintain because of evaporation. Ohmic behavior for Pd/InP was observed after annealing at ~300~ and was attributed to the presence of the PdIn phase.[ 2~ Particularly, PdIn/InP was ohmic as deposited with a contact resistance of 6 x 10.5 ~-cm2. [204]
3.2
Multi-element/lnP Metallizations
Pd/Ge/InP and Au/Pd/Ge/InP Metallizations. The use of Ge/Pd and Ge/Pd/Au for contact to InP followed the results for GaAs contacts. PdGe formed ohmic contacts to n-GaAs with an abrupt interracial morphology due to regrown GaAs.[ TM] The solid phase regrowth model has also been demonstrated for Ge/Pd/n-lnP,[ 2~ and was tested for the Ge/Pd/Z~ Pd/p-InP system.[ 2~ Regrown InP and PdGe and ohmic behavior were found after annealing above 400~ suggesting the following reaction:[ 2~ Eq. (27)
PdxInP(Ge, Zn) + Ge ~ InP(Ge, Zn) + PdGe
For Pd/Ge contacts, the relative amounts of Ge to Pd was a critical factor to lower contact resistance. When thin Ge layers were used (e.g., Pd/ Ge ratio > 2), the dominant metall~gical reaction products were PdP 2 and PdIn, and the contact resistance was -~10-5 ~-cm2.[ 2~ With Pd/Ge < 1, a contact resistance of--4 x 10-6 fl-cm 2 was obtained for annealing at 400450~ with only PdGe in contact with InP. Adding Au to Pd/Ge/InP alters the metallurgical reactions to form Au-In compounds at the metal/InP interface. Spatially, Au~0In3 was in contact with InP with an irregular interracial morphology. Aut0In 3, PdGe, and GeP were the reaction products after annealing at 325-450~ [207]Contact resistances of 2-4 x 10.6 ff)-cm2 was obtained at ~350~ The formation of GeP is different from GaAs since a GeAs phase was not reported.
Ohmic Contacts to 11-I/I and III-V Compounds
129
Au/Ni/InP and Au/Ge/Ni/lnP Metallizations. TEM images of Au/ Ni/InP after annealing at 250-400~ for 15 seconds revealed a layered structure of Aualn/NiP-NiP2/Au3In/InP.[ 2~ This indicates that Au diffused through the Ni layer and reacted preferentially with In, while Ni reacts with phosphorus at temperatures as low as 250~ (which is consistent with AES depth profiles). [2~ Gas mass spectrometry showed that the phosphorus loss was negligibly small, as compared to that from the Au/InP reaction metallurgical reactions near 400~ indicating that Ni captured P by forming Ni-P compounds.[ 21~ It was reported that these reactions resulted in ohmic contacts to n-type (doped to -- 1018 cm-3) without Ge[1991[21l] but non-ohmic contacts on p-InP.[ 2~ Presumably the ohmic contacts without Ge on n-InP results from dopant segregation to the interface during reactions, similar to Au/GaAs reactions.[ 93] Adding Ge to Au/Ni to form ohmic contacts to InP leads to the identical recipe of AuGeNi for n-GaAs contacts. With respect to the metallurgical reactions which lead to the ohmic contacts, below 250~ the formation of amorphous NixlnP and indiffusion of Au towards InP and outdiffusion of In toward the free surface are all observed.[ 212]Annealing at 250-350~ causes Ge to migrate into the Ni layer while Ni accumulates at the metal/InP interface, presumably dissociating the InP lattice to form amorphous NixlnP. Diffusion of Au and In through the intervening layers was also noticeable. [179][191][212][213]The contact resistance was reduced from 10-2 to 10-4-10-s f2-cm 2 during this reaction stage. [179][191][212] At 350~ 450~ where the lowest contact resistances were frequently measured, metallurgical reactions were more pronounced. Nickel accumulated at the InP interface as Ni2P, Ge reacted with Ni to form NiGe, and Au reacted to form Au-In compounds,[ 212][2141all of which are similar to the characteristics for AuGeNi/GaAs reactions.[179][2~ Spatially, the phases in contact with InP are Ni-P and Au-In, with Au-In and indium oxide phases at the metal/ambient interface. The Au-In phase at the InP interface causes irregular interfacial morphology. Elemental Ge coexisted with Au-In alloys at the metal/ambient interface rather than segregating at the metal/InP interface, again similar to GaAs with excess Ge.[ 212] 3.3
Mechanisms for Formation of Ohmic Contacts
As to the mechanism for ohmic contacts, the similarity to GaAs suggests strongly that the solid phase regrowth model should be applicable. In this model, again Ni dissociates the InP lattice and bonds to the P.
130
Wide Bandgap Semiconductors
A ternary phase may form at the interface, however, in the case of ternary decomposition to binary alloys, Au or a similar element may be necessary to hold the In at the interface. Formation of Au-Ni or NiGe then would cause regrowth of InP with incorporation of a dopant (Ge or Zn reported above). A significant difference between the two systems is that extremely low contact resistance (10 -7-10 -8~-em 2) was reported without using Ge,[ 215] particularly when Ni-P phases were formed at the interface with n-InP. The use of Ge is essential to formation of low resistance ohmic contacts to nGaAs. This difference with InP might result from stronger segregation of the dopant during metallurgical reactions, or since the Fermi level is only partially pinned for InP, formation of Ni-P may unpin the Fermi level and thereby significantly lower the barrier height. Thus, barrier lowering effects from unpinned Fermi levels may be an important factor in the formation of ohmic contact to InP.
4.0
O H M I C C O N T A C T S T O GaN
The situation for forming ohmic contacts to GaN is quite different from that for GaAs or InP in that Fermi level pinning does not occur. GaN devices are similar in this respect to the situation for ZnSe-based devices, as discussed below. Since the Fermi level is unpinned, formation of ohmic contacts can be predicted by the simple rule (Eq. 4) of the metal work function being less than the electron affinity of the semiconductor (~m < Xs) for an ohmic contact to n-GaN. [216][217] Since ~ G a N -- 4.14 eV, Foresi and Moustakis[ 218] showed that A1 (~gl - 4.08 eV) on n-GaN resulted in an ohmic contact, while Au contacts (~gu = 5.1 eV) were rectifying. Miller and Holloway also reported that Au formed rectifying contacts to nGaN.[216] Miller and Holloway studied Ag ( ( I ) g g = 4.3 eV) single component metallizations on n-GaN which resulted in weakly rectifying contacts in the as-deposited or heated to 500~ conditions. No interface reactions nor phase formation have been observed for either Au or Ag metallization on GaN epitaxial layers. However, Guo et al. showed that the Schottky barrier height for Pt and Pd varied with work function indicating a partially pinned surface level.[ 219] Wang et al. reported for the same system that the Schottky barrier height was the same for Pt and Pd.[ 220]These conflicting results clearly show that cleaning of the GaN surfaces is critical to successful and reproducible formation of ohmic contacts, as suggested by Murakami
Ohmic Contacts to 11-111and I I I - V Compounds
131
et al.[221]Thus, E r of GaN is regarded as unpinned, while partially pinned E F have been attributed to interfacial contamination. The characteristics ofmulticomponent contacts incorporating Ti are consistent with the conclusion of unpinned EF.[216][217][222]-[224]For pure Ti with a ~Ti - 4.3 eV, a weakly rectifying contact would be predicted. However, Ti will react with GaN to form TiN. With a theoretically predicted work function for TixN of 3.74 eV and reasonable electrical conductivities, this material should make a good ohmic contact to n-GaN. At the interface between a single or multicomponent metallization (e.g., Ti versus Ti/Pt/Au or Ti/A1/Ni/Au), TixN can be grown by interracial reactions during heating to temperatures ranging from 250~ (furnace anneal) to 900~ (rapid thermal anneal). The TiN interface phase is so thin that it is difficult to measure using the conventional thin film profiling techniques, such as AES, XPS, or SIMS. This interracial reaction product has only been suggested by limited depth profiling data and by ohmic electrical behavior after annealing. Very low specific contact resistances (~10-5-10 "7 f2-cm2) have been reported.j222][223]Ping et al. showed that A1 adjacent to n-GaN with a top Pd layer would also form an ohmic contact with a low specific contact resistance (10 -5 ~-cm 2) after heating to 650~ for 30 sec.[ 225] Even when the work function of the metallization is not appropriate, increased surface doping of n-GaN can be used to achieve ohmic contact. Miller and Holloway showed that incorporation of Si into a Au/Si/Ni metallization improved the performance.[ 2161Since the work function of both Au and Ni (5.2 eV) are greater than the electron affinity of GaN, neither should make an ohmic contact to n-type material. However, Si apparently helps create an n § surface layer which allows tunneling transport to result in an ohmic behavior. Another way to increase doping at the surface of n-GaN is to heat treat at high temperatures, resulting in the loss of nitrogen and formation of a shallow donor. As a result, Zolper et al. showed that high temperature heat treatment without a capping layer improved the properties of an A1/Ti ohmic contact, while it degraded the performance of a A u n t Schottky contact on n-GaN.[ 227] If the GaN surface was capped with an A1N layer before annealing, loss of N was prevented, the ohmic contact was worse, and the Schottky contact was constant. Lester et al.,[ 228] Cole et al.,[ 2291 Binari et al.,[ 23~ Lin et al.,[223] Revimov et al.,[ TM] and Smith et al.[232] all reported that formation of N vacancies during heat treatment or by interfacial phases increased the free electron concentration near the n-GaN interface and assisted in formation of ohmic contacts. Finally, Ingerly et al. heated Ptln 2
132
Wide Bandgap Semiconductors
surface layers on GaN and reported that decomposition of the surface layer resulted in formation of InGaN narrower bandgap material. A specific contact resistance of < 10-3 f2-cm 2 was reported.[ 233] In contact studies to n-GaN, thermal stability of the contacts has only been tested in a cursory fashion. Dt~bha et al. studied the contact resistance, topography, and composition profiles for Ti/Pt/Au, Au/Ge/Ni, A u ~ e and WSi x on GaN or In0.sGa0.sN thin films.[ 217] The WSi x contacts were found to exhibit excellent thermal stability (smooth surface morphology, no interdiffusion, and low contact resistance) even for heat treatments up to 800~ The minimum, specific contact resistance on In0.5Ga0.sN was 1.5 x 10-5 f~-cm 2. The Ti/Pt/Au, Au/Ge/Ni, and A u ~ e contacts showed much lower thermal stabilities, with Au/Be being unstable at temperatures as low as 400~ The instabilities were largely evident from changes in the surface morphologies resulting from voids or islands of metallization (from surface capillarity forces). There was no evidence for formation ofinterfacial phases, even though the interface width measured from AES depth profiles became broader, presumably due to roughening of the interface. In contrast to n-GaN, a metal or compound with a very large work function would be required to make an ohmic contact with p-GaN (electron affinity plus bandgap of 4.1 eV + 3.4 eV = 7.5 eV) as discussed above. Consistent with expectation, studies of the Schottky barrier height for Pt, Ni, Au, and Ti to p-GaN[ 214] showed that all contacts were rectifying and the barrier height decreased with increasing work function. In addition, Ishikawa et al. showed that the interfacial barrier height for contacts of Pt, Ni, Pd, Au, Cr, Ti, A1, and Ta also decreased as the work function increased. [221] Both studies are consistent with unpinned E F or only partially pinned E F. Since there are no metals with a work function >7.5 eV, ohmic contacts to p-GaN must be formed by either a compound with a very large work function/electron affinity, by doping the surface region to p++ with a lowered surface barrier, or by a graded surface barrier normally created during the last phases ofMBE, MOMBE, or MOCVD growth of the GaN epilayer. Nakamura et al. used Au or Au/Ni contacts to p-GaN for a demonstration of LEDs and electrically injected LDs based on GaN. [234][235] Khan et al. also reported using Au/Ni contacts on p-GaN. Other contacts metallizations to p-GaN include Ni by Sakai et al., [236] Au/Zn by Kuga et al.,[ 237] and Ti/Mo/Au by Goldenberg et al. [238] Trexler et al. studied the interracial reactions between Au, Au/Ni, Au/C/Ni, and Au/Cr metallizations on GaN doped with Mg (free hole density of 5 x 1016 to 2 x 1017cm 3)
Ohmic Contacts to II-VI and III-V Compounds 133 at temperatures up to 900~ [239] The Au contacts were rectifying and remained so even after heat treatment to temperatures up to 600~ There was very little interaction between the Au and GaN and no evidence for formation of an interfacial reaction phase or for diffusion of Au into the epilayer. For Au/Ni and Au/C/Ni metallizations, Ni dissociated the GaN lattice and led to a lower barrier height at the interface.[ 24~ After heat treatment at 600~ the Au and Ni had interdiffused and the width of the Ni/GaN interface had increased significantly, indicating dissociation of the GaN lattice. The reaction products of this dissociation are uncertain since there is low solubility of N in Ni, even though Ni will combine with Ga to form a solid solution and several binary intermetallic solid phases. While the Au/Ni contacts never achieved linear I-V ohmic behavior, the reverse bias breakdown voltage dropped from near 3 V to less than 0.5 V. Trexler et al. speculated that interracial carbon from contamination acted as a dopant and was incorporated into the lattice during this interfacial reaction.[ 24~ Abemathy et al. showed that C can act as an acceptor in GaN, thus it is reasonable to speculate that interfacial contamination detected by both AES and SIMS could lead to a higher free hole concentration at the interface.[ TM] Therefore, the contact structure of Au/C/Ni, where 10 nm of C was evaporated between deposition of Au and Ni, was tested to determine if increased doping and better ohmic contacts would be observed. As shown in Fig. 22, the current-voltage data for this contact scheme exhibited a higher resistance than did the AtffNi scheme. Obviously an evaporated film of C at the metal/semiconductor interface did not improve the ohmic behavior. This could result from a number of factors, including the fact that a monolayer interfacial C contamination layer should be sufficient to saturate the doping effects. In addition, interfacial contamination C is largely "adventitious" organic and hydrocarbon molecules which are similar to the C source used by Abemathy et al. to demonstrate p-doping of GaN. The C present from evaporation of a thin film is bound primarily as C sp 2 in a ring structure, which is notoriously poor as a doping source during thin film growth. Trexler et al. also studied the formation of Au/Cr contacts to p-GaN and found better linear I-V data than for Au/Ni or Au/C/Ni. Contacts were rectifying as deposited with a reverse bias breakdown voltage of ~1.5 V.[239] Upon annealing at 200~ or 400~ for 5 min, there was a slight decrease in conductivity across the contacts and samples. However after an RTA at 900~ for 15 sec., the I-V data were very linear with large current flow across the interfaces. Auger analysis showed very little reaction between
134
Wide Bandgap Semiconductors
the Cr and GaN at 200~ or 400~ but extensive dissociation of the GaN lattice and formation of a Cr-N-Ga ternary phase at the interface after the 900~ 15 see. RTA. Thus, Au/Cr appears to be a better contact metallization than does Au/Ni for p-GaN. While the specific contact resistance was not measured, it is expected to be very high compared to the values for ZnSe or GaAs (i.e., much higher than ~10 -5 f)-cm2). 2e-4
2e-4
le-4
~" 400
n0
300
-~m" un I n " "
m
,-O 200
(3' o"O" I
20
--"O
.. ,,..~~ -
[o]
30
%" m~
20
go
10
I
I
I
40
60
80
0
m
z
Co -4
-
SF 6 > CF 4 > CHF 3. When the minimal amount ofH 2 was added, the reactivity of the two fluorocarbon gases became comparable, as did the reactivities of the SF 6 and NF 3 plasmas. Moreover, when H 2 was added, the maxima in the etch rates corresponded to mixtures containing approximately 10% 0 2 in each case. It was also reported that the micro-masking effect can be eliminated with the use of a graphite sheet covering the powered electrode.[ 3~ However, side effects which have been reported include polymeric deposition and flaking within the reactor, contamination of the pumping equipment, and mass-loading effects. Casady et al. [31l showed that RIE using N F 3 at the relatively higher pressure of 150-350 mtorr can result in high etch rates and residue-free surfaces without the use ofH 2 or O 2 additives. They reported etch rates of approximately 150 nm/min at an rfpower density of 1.7 W/cm 2 and a pressure of 225 mtorr. At these higher pressures, the corresponding de biases were reported to be in the range of only -25 t o - 5 0 V, suggesting that these conditions may be advantageous for recessing down to active regions of SiC devices. When comparing studies reported in the literature, it appears that, under most conditions, [3-SIC films etched approximately 20-50% faster than 6H-SiC, although accurate experimental comparisons have not been made. It could be speculated that the differences in the etch rates are due more to differences in the dangling bond densities and the corresponding reactivities of the crystal faces, than to the different crystal structures. For example, each atom on a cubic (001) face has two dangling bonds; whereas only one dangling bond exists on a (111) face or similarly on the (0001) face of hexagonal SiC. RIE of (111) I3-SiC films has been reported by Wu et al.,[ 32] albeit under conditions where a direct comparison with reports on (001) oriented films cannot be readily made.
Dry Etching of SiC 163
60
'
_'
'
I
'
'
'
I
'
'
'
I
'
'
'
I
'
'
0
'
.
.
.
I
'
'
'
I
'
'
50~ E 40,
"'.~
3o
""
'E
..'.
-
"r0
-
96H-SiC . 4H-SiC
~ 6-3
,L 3C-SiC
uo 4 -
= 2j,
m ~
~
.
,, p-
A 9
14
19
24
29
34
39
44
Temperature (K)
(a) I~ Low
I~ Median
4"2
"
~---
"~2-
8 t-I--
D High
6H
4H SiC Pol~ype
3C
(b) Figure 4. Reported values from literature of thermal conductivity for SiC at (a) low temperature and (b) room temperature. Range of room temperature data is reported as high, median, or low. Taken from compiled data in Ref. 95.
Processing of Silicon Carbide for Devices and Circuits
189
Electrically active impurities in semiconductors are normally substitutional dopants occupying vacant lattice sites. Dopants for SiC include N, P and As (n-type), and A1, B, Be, Ga, O, and Sc (p-type), with A1 being the most common p-type dopant because it has the shallowest acceptor level.[ 17] Undoped SiC is typically n-type from residual nitrogen, and has a slight green tint in color for 6H-SiC. Color of the material depends upon the specific polytype, however. Donor activation energies are often found to vary over a wide range, depending upon measurement technique, material quality, polytype and dopant concentrations. Activation energies also vary depending upon the substitutional site occupied in the lattice (cubic or hexagonal). For n-type 3C, Hall measurements have yielded nitrogen activation energies from 18-48 meV. In 6H-SiC, two donor levels have been found depending upon the occupancy site. Site 1 (hexagonal site) is from 84-100 meV, and site 2 (cubic site) is from 125-150 meV.[ 17]In 4H-SiC, donor levels are 45 meV, and 100 meV for site 1 and site 2, respectively.[ 17] The fact that most dopant levels are deeper than those found comparably in silicon explains the partial carrier freeze-out found in SiC at room temperature, since the thermal energy (kT/q) is only---25.9 meV at 300 K. Since carriers are ionized by electric fields in many devices (most notably field effect transistors), lower temperature operation is also feasible. For example, SiC Junction Field Effect Transistors (JFET's) have been operated to temperatures as low as 77 K, because of field ionization of dopants.[ 41] For p-type A1 doped SiC, an average acceptor energy level of approximately 200 meV is found for all pol3~,pes. [17]Other p-type dopants such as boron have deeper acceptor levels (ranging from approximately 320 meV to 735 meV), and are not as commonly used because of increased concerns regarding carrier freeze-out. Boron's most common use has been in work involving deeper (>0.8 ~tm) p-type implants where boron is preferred over aluminum because of its lighter mass.
2.0
SILICON CARBIDE DEVICE PROCESSING
Now that a brief description of SiC material properties has been given, examining processing issues of the semiconductor for combined high-temperature, high-power, and high-frequency applications is now appropriate. SiC device processing has rapidly evolved since the commercial availability of SiC substrates in 1991. In this section, the major aspects of SiC device processing are discussed, beginning with bulk material growth.
190 2.1
Wide Bandgap Semiconductors Bulk SiC Growth
Historically, bulk growth of SiC has been perhaps the most significant problem limiting the usefulness of SiC in electronic applications. [43}-[59]Singlecrystal wafers of 6H-SiC have been available commercially only since 1991 (from Cree Research, Inc. )[101 and 4H-SiC wafers have only been available since 1994 (from Cree in 1994 and ATMI in 1995).[1~ 11] Research environments, but not commercial ones, have produced 3C-SiC wafers[ 211and 15R-SiC[ 1l] wafers. An excellent review of commercial SiC boule growth by seeded sublimation is given by V. F. Tsvetkov et al.[431 Single crystal boules grown from either a melt or solution would require excessive temperatures (>3200~ and very high pressures (> 100,000 atm) which precludes this growth method from being utilized. Thus, in the absence of other viable melt-growth techniques, physical vapor deposition via seeded sublimation is the most commonly used approach. The vapor phase of SiC (typically Si, Si2C, SIC2) is deposited upon a SiC seed crystal at high temperatures (>2000~ Examples of a 4H-SiC boule grown via the sublimation process is shown in Fig. 5 and example wafers shown after slicing and polishing are shown in Fig. 6. Typical wafer diameters are 3 5 mm, although Cree Research is currently offering the sale of 2 inch (~50 mm) diameter wafers beginning in 1997. One major problem with the sublimation growth technique has been the formation of mieropipe defects. Although defects such as mieropipes are not found in Lely platelets of SiC, this type of growth results in irregular shaped SiC substrates which are unsuitable for commercial SiC device produetion.[5~ 551Mieropipes are bulk defects (voids) which propagate the length of the boule from the seed crystal, and are also found to propagate through subsequent epitaxially grown SiC layers. Mieropipes have hexagonal cross-sections with diameters from about 0.1 ~tm to 5 ~tm. [43][46] Mechanisms causing the mieropipes have not been clearly identified in the literature, but 13 possible thermodynamic, kinetic, and technological mechanisms have been identified.[ 43] A good discussion of various defects (hexagonal pits, mieropipes, screw dislocations, hillocks, etc.) and possible causes is found in Ref. 56. In physical vapor transport grown 6H-SiC substrates, all mieropipe defects were positioned along the lines of super screw dislocations with Burgers vectors of at least four times that of the eaxis lattice constant.[ 461Mieropipe defect densities (MDD), found in densities of 1000/em 2 in the early 1990's, were reported as reduced to 3.5/em 2 at the research level on a 30 mm (1.18 inch) 4H-SiC wafer in 1995.[43]Typical commercial wafers in 1997 possess mieropipe defects densities ranging
Processing of Silicon Carbide for Devices and Circuits
191
from 50-200/cm 2. One fact often overlooked by device engineers is that the mieropipe defect density is non-uniform and often locally clustered on the wafer. The non-uniform density is a real benefit to device engineers seeking large-area regions for power devices or complex circuits since 1 cm 2 areas on wafers have often been found with zero micropipes. [44]Figure 7 illustrates an example of the non-uniform micropipe defect density found on a typical 4H-SiC substrate, similar to that reported in the literature.[ 441 While some areas are virtually defect-free, other areas have a large density ofmicropipe defects on the same wafer. Thus, quotes of average micropipe defect density across a wafer may not give a true representation of the substrate quality. Elimination of the micropipes found in bulk SiC is a critical issue for development of SiC power devices and larger-area integrated circuits[ 45] and is expected to happen by the year 2000.
Figure 5. Polished 38 mm diameter, -oriented 4H-SiC boule grown via seeded sublimation technology shown prior to slicing. Photo courtesy of Northrop Grumman.
192
Wide Bandgap Semiconductors
Figure 6. Examples of sliced and polished SiC substrates. Note that SiC is optically transparent and that substrate color is dependent upon doping type, concentration, and material polytype. Photo courtesy of Northrop Grumman. It should be noted that at least three different corporations in the United States (Northrop Grumman, Cree Research, and ATMI/Sterling Semi.) are currently producing SiC wafers via seeded sublimation, with other companies in Russia, Japan, and Europe also producing wafers (see Table 1). A brief, non-inclusive listing of other outstanding references to bulk growth of SiC are listed for the interested reader.[21][25][43t-[59]Work by H. M. Hobgood et al., [47] D. L. Barrett et a1.,[521[551 and G. Augustine et al. [44] provide superb discussions of SiC bulk growth using a sublimation-source
Processing of Silicon Carbide for Devices and Circuits
193
physical vapor transport system at Northrop Grumman, with results comparable to that of Cree's. Notable achievements include production of 6HSiC boules up to 60 mm (2.36") in diameter. It is estimated that 100 mm, high-quality wafers of reasonable cost will be required for high-power commercial SiC device production, while 50-75 mm wafers should suffice for low-power commercial products. [43]Growth was done at ~2300~ while the oriented SiC seed crystal was held at a lower temperature (~2200~ The major crystalline defects reported in the 4H-SiC substrates grown by physical vapor transport were micropipes (10 cm -2 on best wafers) and dislocations (104 cm -2 range). Room-temperature electrical conductivity of the substrates could be varied from less than 1 x 10 -2 f~-cm, n-type, to insulating (> 1015 ~-cm).
Figure 7. Example of the non-uniform micropipe defect density on a typical 4H-SiC substrate. Photo courtesy of Northrop Grumman.
194
Wide Bandgap Semiconductors
Other recent research has focused upon the use of tantalum coated crucibles and containers for sublimation growth in place of the traditional graphite crucibles. For example, growth rates of 1.5 mm/h have been achieved using Ta container material, with growth temperatures ranging from 1600-2100~ When comparing Ta and graphite crucibles for growth (using polycrystalline SiC source material and (0001) 6H-SiC Lely grown platelets as seed), it was found that at low temperature gradients ( 100 k~.cm), and even insulating (1011-1012 k~.cm) type behavior. Achieving semi-insulating 6H-SiC has been accomplished by using undoped and vanadium-compensated substrates. The resistivity of the undoped 6H-SiC semi-insulating substrates has a strong temperature dependence dominated by a single activation (ionization) energy of-~0.3 5 eV identified as the B acceptor level.[59]The temperature dependence of the vanadium-compensated 6H-SiC material is also strong but more complex, with at least three activation energies identified resulting from residual boron and the complex behavior of vanadium. Vanadium has been proposed as both a donor (ED =1.35 eV) and an acceptor (EA= 0.8 eV). These temperature dependencies should obviously be taken into account when examining limits to high temperature operation of high frequency SiC devices fabricated on semi-insulating SiC substrates.
2.2
Doping of Silicon Carbide
Epitaxial Growth of SiC. Doping in SiC for device fabrication is accomplished via epitaxially controlled doping and hot ion implantation. Temperatures required for diffusion are too high (greater than 1800~ for standard device processing because of the very high bond strength possessed by SiC. The two most common dopants used in SiC are nitrogen (n-type) and aluminum (p-type), as discussed previously, although boron is sometimes used for p-type implants. In the absence of diffusion, epitaxial and ionimplanted control of dopants are critical for the development of devices and IC's. Numerous high-quality publications on epitaxial growth processes exist (see for example, Refs. 22-23, 40, and 60-75). Silane and propane are typical source gases of Si and C, respectively. Typical growth rates for 6H-SiC homoepitaxy layers on Si-face n-type substrates are--3 ~tm/hour. For
Processing of Silicon Carbide for Devices and Circuits
195
example, a prototype horizontal flow epitaxial growth reactor used at Northrop Grumman Science and Technology Center for homoepitaxial 4H-SiC and 6H-SiC growth is shown in Fig. 8. Increasing the growth rate while maintaining polytype control, good surface morphology, uniform thickness, and accurate control of dopant levels is critical in producing thick (50-100 ~tm) blocking layers necessary for high-voltage (>5 kV) power devices. Growth rates of up to 6 ~tm/h have been demonstrated in vertical low pressure CVD systems manufactured by EMCORE, Inc.[TMIn this system, high speed rotation of the wafer carrier was employed to stabilize the gas flow in the reactor at a growth temperature of-1500~ The resulting epitaxial layers were characterized to have breakdown strengths of 2 MV/cm, electron mobility of 700 cm2/V.s in lightly doped 4H-SiC epitaxy, and a background concentration of approximately 2 x 1014 cm-3 (boron). Doping levels were controlled over a magnitude order range of three. Other work using hot-wall CVD reactors has produced impressive results, with 45 ~rn thick epitaxial layers grown and used in 100 ~tm diameter 4.5 kV diodes. [74]
Figure 8. A horizontal Vapor Phase Epitaxy (VPE) growth system used to grow SiC epitaxy at rates of~3 mm/h at temperatures of 1500-1600~ Photo courtesy of Northrop Grumman.
For homoepitaxial growth of 6H-SiC, the substrates used are normally silicon-face, 3.5 degree off-axis, while for homoepitaxial growth of 4H-SiC, silicon-face 8 degree off-axis substrates are generally used. The reason for the difference in off-axis angle for the different polytypes is based upon the greater step height in 4H-SiC. Using on-axis or non-optimized
196
Wide Bandgap Semiconductors
off-axis wafers or non-optimized growth conditions for growth of epitaxy results in the formation of triangular morphological defects, caused by inclusions of 3C-SiC which interrupt the step flow mode of growth. These defects have been shown to cause high leakage currents in diodes. These and other surface morphology defects such as Si-droplet formation, faceting, and hillocks are reported in greater depth elsewhere.[43][73] Nitrogen and aluminum doping have been investigated for many years; however, residual doping levels were too high for many devices. The discovery of the site-competition effect[ 611for Si-face SiC epitaxy enabled the reduction of residual doping levels to 1014c m "3 and intentional incorporation over the entire range of possible concentrations. An example of controlling the nitrogen doping of SiC via gas Si/C gas ratio during growth is seen in Fig. 9,[75] for the same reactor as pictured in Fig. 8. This discovery has opened the way for much of the device results shown in subsequent sections. The site-competition effect works by adjusting the Si/C source gas ratio in the growth reactor to control the amount of dopant incorporated into substitutional SiC crystal lattice sites.[63] The model is based upon N and C competition for C sites, with AI and Si competition for Si sites in the SiC lattice. This effect has also been observed for boron and phosphorous doping of SiC.[61]Growth on Cface substrates has behaved quite differently, and more work is still needed to fully understand all growth mechanisms[ 611and tie together behavior on both faces. Numerous industrial and university laboratories now produce homoepitaxial, device-quality growth of SiC.
~I - - - l x 1 0 ]u
E t-
.s ~ c-
o
o
lx10
18
I xlO 11
1• lx10
L,.
]~ 15 ,
,--,_ 9,9 lx1014 o
J
,h.
[] i
, / / ' 6 "~'-
lx10
0
0.2
0.4
0.6
0.8
1
h pu t SJ IC Ratio
Figure9. Controlof doping demonstratedby varyingthe Si/Cgas ratio. TakenfromRef. 75.
Processing of Silicon Carbide for Devices and Circuits
197
Ion Implantation. Ion implantation is also proving to be a vital component of device processing in the absence of a usable diffusion process. Ion implantation has several critical roles in device fabrication. First, high-dose, shallow depth implants are often used in order to reduce the contact resistance to SiC. This type of implant is of particular importance when forming a contact to a lightly doped layer. A common example would be to use implantation to form n § drain and source regions in an n-channel SiC MOSFET. Nitrogen implants are typically used to form n § regions while aluminum implants are normally done to form p+ regions. A second use for ion implantation would be to form the p-wells in CMOS and planar power MOSFET's. Obtaining a deep implant into SiC is particularly difficult. Typical implant systems have energy limits of around 400 keV, which is insufficient for implants of >0.8 ~tm depth. Higher energy implants using MeV range energies offer the promise of deeper implants, but the amount of crystal damage may be very high, and impossible to remove. Numerous papers may be found in the literature regarding dopant ion implantation into SIC.[ 76]-[94]Finding the optimum high-temperature implant (eg., 500~ to 1000~ and subsequent high-temperature anneal (eg., 500~ to 1700~ which will fully activate the dopants, prevent amorphization of SiC, and not cause damage to SiC epitaxial layers (remembering that SiC epitaxial layers are typically grown at temperatures 0.5 Ixm) than A1 because A1 has greater atomic mass than B and causes much more lattice damage. High-temperature ion implantation is preferred, since room temperature ion implantation results in amorphized material, while high-temperature implants do not. In Ref. 87, A1 and N implants were performed at 850~ and 700~ respectively, into Si-face, (100) 6H-SiC, followed by a 10-45 minute anneal at 1100~ to 1650~ in Argon. All implants were single energy implants ranging from 50 keV to 3 MeV. SIMS (secondary ion mass spectroscopy) analysis revealed no N redistribution after anneal, although a slight out-diffusion of A1 was reported. Also, due to its higher atomic mass, A1 implants caused more lattice damage than N implants as confirmed from RBS (Rutherford backscattering) data. Aluminum, because of its deep acceptor level and large mass, was not fully activated at anneal temperatures less than 1500~ indicating
198
Wide Bandgap Semiconductors
that higher anneal temperatures are necessary for A1. Si and C co-implantations were also performed in attempts to increase A1 activation, but C had no effect and Si co-implantation decreased dopant activation.[S7H ss]
101 '
al.
' J . Gar(Iha', et
' ...... w
.......
J. Bec. Ma1:.,25,885[1996) im plant t mz p. = 700 *(2 adnneed temp. = 161)0 *C anneal l i m e : 15 rain. 9 j \ j i i
-
C
i '5
io o
T. Kimot o, etal. j . R ec. M~., ~ , 2 3 ~ 1995) implant t r a p . = 23 *C ormeed temp. = 1500 anneal t i m e - 3Omin.
3
!
.- ........................... , ........................... .~. ...... ~ ...... . ................
_
mm
4-1 m
-
i
-
.
:
.
~
:
lib
41o
*
mi
N. mnoue,:~t d .
-'
m
.
er~
I
Inst. Phys. C m f . S
3
"
No. H2, 525 (199q implant tem p = 5 0 0 - 860 "C anneed tmzp = 1500 anneal tim e = 30 rain. , , , , ,,,I
1 0 "i
!
1 0 is
l
I
l
I
I Ill
l
I
l
1014 Implant
I
i
I Ill
1 0 ts Dose
I
1 0 le
l
I
l
l
l
Ill
1017
(cm -z)
Figure 10. Best reported sheet resistivity in n-type SiC as a function of implant dose. Implant and anneal temperatures (including references) are given. Figure courtesy of J. Cooper, Purdue University.
The effect of implant temperature is still not fully understood, but studies investigating it have revealed some key information. In one such study, [89] N implants at various temperatures into 4H-SiC and 6H-SiC ptype epi layers were examined. When the total dose exceeded 4 x 10 ]5 cm2, room temperature implants resulted in the creation of a totally amorphous layer, which remained heavily damaged even after high-temperature (1500~ anneals. Under the same high-dose conditions, the damaged regions could be annealed back to almost perfect crystallinity if a hightemperature implant was used. However, to achieve full activation of
Processing of Silicon Carbide for Devices and Circuits
199
implanted N, annealing temperatures in excess of 1500~ should be used. When comparing high-temperature implants from 500-950~ it was also found that the surface of the 950~ implanted material was Si depleted; graphitic C-C bonds were observed, while the 800~ implanted material retained the stoichiometric composition as verified by XPS (x-ray photoelectron spectroscopy). This is consistent with earlier research which has found that the surface of SiC at high-temperatures undergoes a preferential desorption of Si atoms resulting in graphitic layers formed on the surface. Thus, after ion implant and/or anneal temperatures which exceed-~950~ a sacrificial oxidation is normally performed to consume the (often conductive) graphitic surface. Another method to prevent loss of Si and maintain surface morphology has been to anneal in a silane (silicon) ambient. This has been demonstrated using an epitaxy reactor by ABB, Cree Research, and Mississippi State University with extremely promising results, as shown in Fig. 12.
1 0s
'
'
'
'
'
''''I
'
.
,
,
,
If,,|
.
,
.
,
li,,]
,
i
i
,
,,ij.
-"
." 9
T. K i ~ t o , ~ al.
-
~
J. Elec. Mat..25.879(1996) -
-
!
.=
!
i 1 i
C
1 0a
.............................
implant temp. = 23 ~ anneal tonnp. = 1500 ~
~ ............. .ann~..I
:
...
i i i
i9
:.
9 ...
:9 i
i :
:9
'5 m m
.m
o,m
~
: : i
i
.m
t ~ e = 30 min.
O
m .
"
:"9
-
O
101
!. . . . . . . . . .i.. .. .. . . . . . .j........ . . . . . . . . . . . . . . .!. . . . . . . . . .
3
.=
.."
9 9
10 0 1 0 ~s
, i ,,,,,,I
I , ,,,,,,I
1014
101~
Impl an t D o s e
I
I
I
I
IIIII
1016
I
I
I
I
IIII
1017
(cm- z )
Figure 11. Best reported sheet resistivity in p-type SiC as a function of implant dose. Implant and anneal temperatures (including references) are given. Figure courtesy of J. Cooper, Purdue University.
200
Wide Bandgap Semiconductors
N, c m "3
10 ~8
:....
N, cm "3
.:..
. . . . . . . . . . . . . .
. . . . . . . . . . . . .
:-..-..
::: : . . : .
-..::
9::..
:
.... ::
.:.....-.
10'*
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
10 '7 10 ~T I 0 ~6
10 ~=
t4
2.5
3.0
3.5
4.0
4.5
Depth, l~m
10 0.25
i
0.30
0,35
0.40
Depth, I~m
Figure 12. Implant activation of Al-implanted (p-type) SiC is demonstrated by CV extracted dopant density before and after anneal. Anneal was performed at 1600~ in an epitaxy reactor with silane overpressure. (Figure courtesy of Dr. S. Saddow, Mississippi State University.)
Other work has looked at implantation of hydrogen and boron into SiC. Hydrogen is important to study because it is present in high concentrations during the epitaxial growth process, and because it also plays a significant role in affecting the SiC/SiO 2 interface. In Ref. 94, boron was implanted at 700~ 350 keV and a dose of 1014 cm "2, and then activated by a 1700~ anneal in a SiC container with an argon atmosphere. The resulting junction depth (at-1018 cm -3) was 0.6 ~tm. Unlike N implantation, the activation anneal can result in appreciable diffusion of the boron implant species. Implanting hydrogen at the same dose (T = 80 K, energy of 80 keV) yielded a concentration of 5 x 1018cm -3 at the same junction depth, and subsequent low temperature photoluminescence has indicated the presence of defects incorporating implanted boron. Little has been reported on the use of MeV implants for the formation of p-type wells in planar CMOS and power device structures. However, the issue ofrecrystallization in MeV implanted 6H-SiC has been examined experimentally in Ref. 93. Wafers of 6H-SiC (0001) were implanted using a tandem accelerator at 160~ with 8 MeV Si 3+ ions (dose = 1 x 1016 c m "2) at a 7 ~ tilt from the [0001 ] direction to prevent channeling. The goal of this implant was to obtain a 1 ~tm deep amorphous SiC layer, and the implant was followed by a 30 minute, 1000~ implant anneal in nitrogen for recrystallization. Cross-sectional TEM micrographs revealed amorphous layers
Processing of Silicon Carbide for Devices and Circuits
201
between 2.6 and 3.4 ~tm from the substrate surface before anneal. Surprisingly, from diffraction pattems, the amorphous layer was completely recrystallized after the anneal, indicating a much lower temperature anneal is required than for keV implants which typically require anneals in excess of 1450~ Also noted was the presence of 3C-SiC material in the midst ofthe recrystallized 6H-SiC, which apparently grows epitaxially in the recrystallized region. Finally, while layer-by-layer growth of 6H-SiC occurred initially during the recrystallization, columnar growth of 6H-SiC followed along with epitaxial growth of 3C-SiC during the recrystallization process. The resulting mismatch between layer by layer growth and columnar growth resulted in stacking fault formation in the columnar 6H-SiC. To summarize SiC doping techniques, both ion implantation and epitaxial controlled doping should both prove vital processes for SiC device fabrication. Ion implantation has the advantage of selective doping, which is important for complementary logic structures, power device termination, and isolation. Ion implantation also results in a more planar device topology as compared to strict epitaxially controlled doping techniques, which have benefits in terms of photolithographic resolution, uniform final passivation coverage, and similar issues. Epitaxial doping has the advantage of not inducing lattice damage which eliminates the need for a high~ temperature anneal and the requisite creation of defects. Also, micropipe defects may affect ion implanted doping profiles by channeling the implanted species deeper into the SiC. Epitaxial regrowth after ion implantation, or annealing in a silane ambient to protect surface morphology, shows the vital link between implant and epitaxy required for advanced device fabrication. Al-implanted 4H-SiC in an epitaxy reactor with silane overpressure is shown in Fig. 12.
2.3
Etching of Silicon Carbide
Wet etching of SiC has not proven to be feasible from a practical device processing standpoint, as it requires molten salts (for example, NaOH-KOH at 350~ to be used at high temperatures. The difficulty encountered in etching SiC results from the high bond strength, a property which makes SiC useful for high-temperature operation, but a hindrance in fabrication. Nonetheless, numerous dry etches (primarily focused on Reactive Ion Etching processes) have been developed for the various polytypes of SiC. [96]-[120] Electron Cyclotron Resonance (ECR) etching has also been employed. [1111-[1141Most published RIE etches all make use of fluorinated gases (typically SF6, CHF3, CBrF3, CF4, NF3) to etch SiC, although etch rates
202
Wide Bandgap Semiconductors
of 1900 A/minute have been obtained in 6H-SiC using a chloride-based (C12/SIC14/O2) etch with SiO2 mask.J1~ RIE etch rates of 6H-SiC and 4H-SiC are typically slow in comparison to Si (300 A/minute to 2000 A/minute), with residue-formation problems commonly found, although not prevalent in all etches. A good overview of SiC etching is also found in the text, Properties of Silicon Carbide. [95] Etching of silicon in fluorinated gas has been found to occur by the reaction mechanism below: Eq. (2)
Si + 4F ---~SiF4
The removal of C has been debated in the literature, with some works indicating that addition of oxygen removes the C, as illustrated by Eqs. 3 and 4. Others claim that the C removal is via physical bombardment (see Eq. 5) or reactive chemistry between the fluorine and carbon. Depending upon the etch chemistry and process parameters (RF power, chamber pressure, electrode area and spacing, etc.), any one of the three mechanisms may actually dominate. Three of the most common fluorinated RIE mechanisms for removal of C proposed in the literature are listed below in Eqs. 3-5: Eq. (3)
C + xO -->CO o r
Eq. (4)
C + xF --~CF4 or CF 2
Eq. (5)
Physical (Ionic) Bombardment
CO 2
Several examples of residue-free RIE of 6H- and 4H-SiC can found in the literature.[96]-[99][l~176176176 For RIE with high etch rates and low surface damage, using high pressure (in comparison to the normal 10 to 100 mT pressures used in RIE) etch recipes seems to be prevalent. For example, in Ref. 97, high etch rates (-~1500 A/minute) were obtained using only NF 3 in a high-pressure (225 mT) RIE recipe. Using a commercial DryTek Quad 400 system, the RF power was 275 W, and the self-induced bias was only 25-50 V. A SEM of a typical sample after RIE is shown in Fig. 13.[97] Another group, using a Plasmatherm Series 790, experimented with 190 mT etch recipes using SF 6 or CF4, and N 2 and 0 2 additives in etching 6H-SiC.[ 1181The peak etch rate was found to be ---2600 A/minute for the following parameters: pressure: 190 mT; SF6 gas flow: 40 seem; 02 gas flow: 10.6 seem; and power: 300 W. Additionally, from atomic force microscopy, a 150 s etch was sufficient to actually remove residual damage leR from the as polished surface. Others have also reported high etch rates of up to 2200 A/minute at 82%O2/18~ r e m o t e plasmas at 330~ [1~176 This etch
Processing of Silicon Carbide for Devices and Circuits
203
produced minimal surface damage since the SiC sample was etched downstream (about 50 cm) from the microwave generated (400 W @ 2.45 GHz) plasma and the pressure was nearly 1 T. This particular etch was performed on the carbon face of 6H-SiC using an evaporated 500 nm thick aluminum mask. When performing the same etch on silicon-face 6H-SiC, higher percentages of 0 2 (~80%) were required to achieve the same high etch rates.
,i
t (a)
(b)
Figure 13. Scanning Electron Microscope(SEM) photographwith (a) 18.1 kX magnification of a 4H-SiC sample, and (b) 3 kX magnification of an etched profile for a 6H-SiC sample. Taken from Ref. 97.
Obtaining a very low etch rate is also important in many device fabrication unit steps. For example, in the fabrication of a bipolar transistor, when making contact to the very thin base region (as thin as 10 nm), it is critical to have a very slow and well-controlled etch recipe. Several etch recipes utilizing R/E at lower pressures (--10-100 mT) and high selfinduced bias (>300 V) can also be found in the literature.[961199]These etch processes often produce more surface damage,Ilia] but are critically needed in some processes. Perhaps because of the high self-induced bias, aluminum micromasking, which may be a result of aluminum being etched from the powered electrode or the mask material, is more common in these etch recipes. Covering the electrode with graphite and/or adding hydrogen to the fluorinated gas chemistry has been shown effective in preventing residue formation.[961[1~176
204
Wide Bandgap Semiconductors
While higher pressure etches have been used to obtain low surface damage, residue-free etching, and high etch rates, other approaches have also been utilized towards those goals. ECR etching has been shown to produce very smooth etched surfaces, and the degree of etch anisotropity can be controlled by the substrate bias.[ 113]ECR etching has the advantages of higher density plasma and independent control of the substrate bias in comparison to RIE etching. When comparing similar RIE and ECR processes using a CF4/O 2 mixture, it was found that the RIE process significantly damaged the SiC surface and also left A1, F, and O impurities on the sidewalls. The ECR process did not.[ 114]The surface damage was measured by comparing the Schottky barrier height of Pd Schottky diodes on ECR etched, RIE etched, and unetched SiC surfaces. However, this particular RIE process has a very high self-induced bias (500 V) in comparison to other SiC RIE processes;[99][ 118]this comparison may vary for different RIE processes. The ECR etching of single crystalline 4H-SiC cantilever structures using a combination of high and low substrate bias has been demonstrated.[ 113]Using a sputtered indium-tin oxide mask, results of ECR etching of 4H-SiC and 6H-SiC in a 1 mT, 17%O2/83%CF 4 plasma indicated a strong dependence of etch rate and etching isotropy on the position of the sample relative to the ECR plasma and the substrate bias. High substrate bias (100 V) produced smoother surfaces and higher etch rates (700 A/minute), but also exhibited trenching (deep trenches at the bottom comers of the etched patterns). Another form of etching has also recently been reported for SiC utilizing Inductively Coupled Plasma (ICP) etching. ICP creates a much denser plasma than traditional RIE by using two RF power supplies instead of one. ICP etching has shown to offer low surface damage, residue-free etching, and fairly high etch rates. Plasma densities in the 1 x 1011 to 1 • 10 ~2ions cm -3 range can be obtained, with the incident ion energy controlled independently by decoupling plasma generation power and bias power. Preliminary ICP etching of 6H-SiC in CF 4 and 0 2 has been investigated using a commercial PlasmaTherm ICP etching system.[ 1191 Using ITO masks, peak etch rates of 1500 A/minute were obtained at 900 W. Sample bias was varied from 10-60 V, with higher etch rates obtained for higher bias values. Using Schottky diodes, the ideality factor (1.1) was the same for both ICP etched and non-etched 6H-SiC. Additionally, the leakage currents of both sets of diodes was similar, indicating that the surface damage was minimal. A cross-sectional SEM of trenches etched into SiC using ICP etching is shown in Fig. 14.
Processing of Silicon Carbide for Devices and Circuits
205
Figure 14: SEM of SiC trenches etched using an Inductively Coupled Plasma (ICP) etch process. Photo courtesy of J. Zhao, Rutgers University. Photoelectrochemical etching (PEC) of SiC has produced very high etch rates. Several examples can be found on PEC as well, but ICP, RIE, and ECR etching currently appear more compatible with small-feature device processing (for example see Ref. 117). However, etch rates of up to 25 lam/min, for n-type 6H and 3C-SiC have been reported with dopantselective etch stops possible,[ 12~ which may be of use in deep trench or via etching. The dopant-selective etch stop could also be of great importance when etching back to make contact to very thin epitaxial layers of SiC, such as the base layer of a SiC bipolar junction transistor. The etch process is briefly described as a hole-catalyzed surface dissolution, with the holes supplied from bulk p-type SiC or from I_W photogeneration in n or p-type SiC. The transfer of holes into the electrolyte promote oxidation of the surface. The oxide is removed with the HF acid contained in the electrolyte.
206
Wide Bandgap Semiconductors
2.4
Oxidation and Passivation of SiC
The ability of SiC to oxidize and form SiO 2 has allowed compatibility with standard silicon-based fabrication. Obtaining a high-quality oxide with low interface state and oxide trap densities has proven to be challenging because of the carbon on the surface, as well as off-axis epitaxial layers which often have rough surface morphologies. Thermal oxidation rates are considerably slower than that of silicon, so typical oxidation temperatures for SiC are often higher than those ofSi (-~1050-1200~ Growth rates of SiO 2 on p-type 6H-SiC are shown in Fig. 15 [124] for both wet and dry oxidation, as well as silicon-face and carbon-face material. Growth on carbon-face SiC is much higher (-5 X) than the growth on silicon-face SiC. The slower oxide growth on Si-face material has been attributed to an interracial oxide layer (Si4C4.xO2 with X -2), identified by angle resolved xray photoelectron spectroscopic analysis, which forms on the Si-face 6HSiC surface and retards the oxidation rate.[1221[142]6H-SiC was polished and cleaned by HF and R/E, and then oxidized for 15 min. at 1100~ in dry O 2. The interracial layer, which has a thickness of about 1 nm situated between SiC and SiO2, is believed to be a reaction product of molecular oxygen bonded peroxidically to a SiC double layer. This interfacial layer can react with oxygen to form Si4C404, SiO2, and CO.[ 122]No interracial layer was found on the carbon-face material, but both silicon-face and carbon-face material had layers of graphite, CxHyOz, and Si4C404 formed on the surface of the oxide. The above discussion of oxide growth and quality is true for oxides grown on standard silicon-face (0001) and carbon-face (000]-) surfaces. The oxide growth rate and quality also varies considerably for other crystal orientations such as (1120) and (1]-00) a-axis orientations.[ 123]Generally, the growth rate on a-axis orientations is 3-5 times higher than that of the silicon-face c-axis SiC material, and the interface state densities are 4-10 times higher as well. This has important ramifications for UMOSFETs, VMOSFETs, and other trench gate MOSFETs in SiC. As an example, a cross-sectional SEM from a 4H-SiC UMOSFET is shown in Fig. 16. The thin white line along the edge of the U trench is SiO2, and as can be seen, this oxide is thicker along the sidewalls than on the bottom of the trench. The higher sidewall Dit also is detrimental to device performance by limiting channel mobility and thus increasing the on-resistance of the device. The high fields present in the device often cause breakdown of the gate oxide via field crowding at the comers of the UMOS trench. The vertical
Processing of Silicon Carbide for Devices and Circuits
207
sidewalls problems relating to variation of oxide growth and quality are also well documented elsewhere.[127][133][16~
2,5 1126 "C
1200 9
E =l. 2,0 c
1090 ~
1,5
Jg .o_
1050 ~
.~_ 1,0
1125 "C
X 0
1090 ~
0,5 1050 "C 0'00
100
200
300
400
oxidation time [hi
! O~ E ~2
Q
i' 0
l I ...
0
1125 "C ~ - - f ~ .
20
40
1090 '~:
1060"C
.....
4 ....
60
80
,c
100
120
oxidation time [hi
Figure 15. Oxidation rates in dry (upper) and wet (lower) 0 2 environments for p-type 6HSiC. Open symbols are for C-face and solid symbols are for Si-face. Taken from Ref. 124.
208
Wide Bandgap Semiconductors
Figure 16. Cross-sectional SEM of a completed 4H-SiC UMOSFET.The thin white region is identified as the gate oxide, with a thicker polysilicon layerdeposited over the gate oxide.
Photo courtesy of Northrop Grumman.
Numerous studies have been published about cleaning and oxidation procedures.[ 121]-[160]The goal of low (comparable with silicon) interface state densities is close to fulfillment. One cause attributed to high interface state density of oxides on SiC is that of C. The problem of C at the SiC-SiO 2 interface is compounded by the high temperatures required for reasonable oxidation rates. When exposed to these high temperatures, the surface of SiC often undergoes a preferential loss of Si, resulting in a graphitic surface layer increasing the carbon concentration.[ 155] Three distinct temperature regimes have been identified for 6H and 15R-SiC. Below 900 K (627~ no graphitization of the SiC surface is observed. Between 900 K (627~ and 1300 K (1027~ both Si-face (0001) and C-face (000T) SiC surfaces are terminated in a surface graphite layer. Above 1300 K (1027~ the C-face SiC graphitizes at a rate higher than the Si-face SiC. Both surfaces exhibit massive graphitization above 1027~ as a result of Si sublimation from the SiC surfaces.[ 155] Because of this problem, SiC wafers are otten loaded into oxidation fumaces at lower temperatures (300~ in order to see the response time of interface states lying deep (>0.6 eV) in the bandgap of SiC. Standard roomtemperature C-V analysis will not reveal the true interface state density. Care must also be exercised because of the large surface potential fluctuations present in SiC C-V measurements.[ 138]J. N. Shenoy et al.[ TM]reported interface state densities (Dit) near the 1 x 1011 per (cm2/eV) order of magnitude, and fixed oxide charge (Qox)of 1 x 1012 (C/cm 2) in thermally grown SiO 2 on 6H-SiC. Another method of reducing Bit and Qo, has been to perform post oxidation anneals in oxidizing environments.[ 156]On p-type (Al-doped) silicon-face 6H-SiC epitaxial layers,Dit of 1 x 1011 cm 2/eV and Qox of 1.0 x 1012 cm-2 was measured on the best oxide, which used a 1050~ oxidation (25 h) and a 3 h post-oxidation anneal in 0 2 at 950~ The interface trap density was extracted using the conductance technique, with measurements taken at 250~ Post-oxidation anneals in oxygen may have a secondary benefit, which would be densification of the SiO 2 to increase oxide reliability and strength. This is evidenced by the high breakdown strength of these oxides (up to 11.5 MV/cm). Time-dependent dielectric breakdown results on these oxides indicated lifetimes of 700 years at 2 MV/cm and 350~ when testing under a negative bias on p-type 6HSiC.[ 158]However, under positive bias on n-type 6H-SiC, the lifetime was significantly less, although no specific numbers were reported. Positive bias stressing was not performed on oxides grown on p-type 6H-SiC, due to the difficulty in forming inversion. The apparent unreliability of oxides under positive bias can also be supported by the finding of hole traps in oxides on SiC (which were not treated to the same annealing conditions).[157] The hole traps (in oxides grown at 1120~ in dry 0 2 with standard 1 h Ar anneal) were found to be related to oxygen vacancies in the oxide, and no direct contribution of impurities such as A1 or C to the trapping was observed. Additionally, thermally unstable interface states were generated during hole injection at the SiC/SiO 2 interface, similar to that of SiO 2 on Si. All of the reliability issues are further supported by the evidence of Fowler-Nordheim tunneling across oxides grown o n SiC. [133][160] That FN injection is more severe for the SiC MOS system is not surprising considering the wide bandgap and reduced transport barrier of the SiC MOS system as compared to silicon. For example, Fig. 17 compares the bandgap, electron affinity, and subsequent barrier for tunneling of silicon, 6H-SiC, and 4HSiC MOS systems via an energy band diagram. In 4H-SiC, the barrier for
210
Wide Bandgap Semiconductors
electrons in the conduction band is only 2.70 eV as compared to 3.15 eV in Si. Thus, the Fowler-Nordheim tunneling probability (exponentially dependent upon the electron barrier height) is much higher for SiC, especially at elevated temperature and bias where SiC has many targeted applications. In fact, for a 4H-SiC n-channel power MOSFET, it has been recommended that the maximum positive applied bias be limited to 1.5 MV/cm at room temperature[1601 which in tum limits the maximum gate bias applied under the on-condition as well as placing constraints on the doping of the p-type channel. Type 6H-SiC is also preferred for MOSFET devices since 4H-SiC has a higher Dit at the conduction band edge.
Z l
3.15
1.1,Si
2.95
T
2.70eV
,,. ~ - E o
? 2.85
3.25
'
,
,
i
3.2
3.05
6H-SiC 4H-SiCI 9.0 eV s!c2
4.75
Figure 17. The energy band offsets of 6H-SiC and 4H-SiC with respect to SiO 2. The barrier for electron injection from 4H-SiC into SiO 2 conduction band is significantly lower than silicon. Taken from Ref.4.
Other high-temperature, high-power operation and reliability issues remain to be addressed. Alternative insulators (such as nitrides and oxynitrides) are also being pursued for high-temperature device applications, just as in the silicon industry. One example involves thermal oxidation of SiC in N 2 0 . [152] Diffusion of CO through the oxynitride was attributed to be the limiting factor in oxidation, and Qoxwas reported to be on the order of 1 x 1012C/cm2. The oxidation rate was found to be initially parabolic with time, eventually switching to linear after some
211
Processing of Silicon Carbide for Devices and Circuits
time, just as the case for Si. It is believed that the introduction of nitrogen into the oxide (or oxynitride) should reduce the Dit, although no specific numbers were mentioned in this work. Other work involving the annealing of oxides with N20 has not proven successful initially, although anneals in NO have shown promise for improvement of oxides on p-type SiC. [159] A quantitative and practical assessment of using thermally grown gate oxides in SiC devices has also been done by researchers at General Electric. [154]Because doping of SiC devices is via implantation or epitaxial growth, most MOSFET devices are fabricated using a non self-aligned process. In fact, the drain/source implantation' s are normally done prior to gate oxide growth. Since the structure is non self-aligned, the gate oxide will partially overlap the implanted drain/source regions. However, thermally grown SiO 2 on implanted regions of SiC has a breakdown strength of only 1-6 MV/cm as compared to 10 MV/cm on non-implanted regions. Premature breakdown of SiO 2 grown on implanted 6H-SiC is shown graphically by Fig. 18. Premature breakdown of the gate oxide in SiC MOSFETs fabricated using this type of process is thus to be expected.
dlO
T = 380~ C 30
1
20
10
0
1
2
3
4
8
8
T
tl
g
10
S (My/era)
(a) Figure 18. (a) BreakdownofSiO2on implantedSiC surfaces.ReproducedfromRef.154. (b) Breakdown of S i O 2 o n non-implantedSiC surfaces. Reproducedfrom Ref.154.
212
4~
Wide Bandgap Semiconductors
--
T m 3110"C
w
~0"
i.i 10
0 0
1
2
$
4
5
8
?
II
I)
E (UV/~)
(b) Figure 18. (Cont'd.)
2.5
Ohmic Contacts to Silicon Carbide
Low resistivity ohmic contacts are essential for high-frequency operation. Additionally, high-temperature and high-power requirements maintain that the contacts must be reliable under extreme conditions. Once seen as one of the primary impediments to SiC technological development, the ohmic contact area has now matured rapidly.[161]-[189] An area still requiring extensive work, however, is that of ohmic contacts to p-type SiC. The p-type contact to SiC is made difficult because ofthe wide bandgap of SiC. As illustrated in Fig. 19 bythe thermal equilibrium metal-semiconductor band diagram (q~M < q~sic), the Schottky barrier (OB) to majority carrier transport must be reduced as much as possible in order to provide for an ohmic contact. Since the bandgap and electron affinity (c) of SiC are fixed, the remaining options for reducing the ~B are to choose a metal with a large work function (OM), and also to dope the ptype SiC as heavily as possible. P-type ohmic contacts to SiC often use some variation of A1/Ti alloys. A contact with specific contact resistance measured at 1.5 x 10-5 f~.cm 2 on Al-doped samples (NA = 2 x 1019 cm -3)
Processing of Silicon Carbidefor Devices and Circuits
213
has been reported by J. Crofton et al.[161]using an A1/Ti alloy. The specific contact resistance is a strong function of doping (seen clearly by Fig. 20, where the specific contact resistance is plotted as a function of doping for the A1/Ti alloy contact). Although A1 melts at ~660~ a 90:10 AI:Ti alloy (by weight) is a mixture of solid and liquid phase at temperatures of 9501150~ which are typical anneal temperatures used in the formation of ohmic contacts to SIC.[161]-[163] More recent experiments using a 90:10 AI:Ti alloy has yielded specific contact resistance ranging from 5 x 10 -6 to 3 x 10 "5 f2ocm 2 on p-type 6H-SiC doped at 1.3 x 1019 cm -3. On the same material, pure Ti (with a 1 minute, 800~ anneal) was also used to form ohmic contacts with specific contact resistance ranging from 2-4 x 10-5 f2.cm 2. Etching off the metals after annealing revealed that the Al-based contact spiked into the SiC, evidenced by pits in the SiC surface up to 2600 A deep, while the Ti contact exhibited little interfacial reaction. Thus, although Al-based contacts can yield exceptionally low specific contact resistances, the contact can suffer from poor reproducibility and aluminum oxidation during annealing (A1203). [164]
Eo
metal
~
p-type 4H-SiC '! Z = 2.70 eVNo
....... l !l
. . . . . . . . . . . . . . . . . . . . . . . . .
E F
Figure 19. Band diagram of metal and p-type 4H-SiC before and after making contact ignoring any effects of Fermi-level surfacepruning.
214
Wide Bandgap Semiconductors
10 -1
...... ' "
'--"
""-"1
9 '"-'~;;:"~
0 I
.....
w
. . . . .
"I
~ ..... " = " - ' ~ " ' ~
ClUW p - t l r p o 6 H - S t C
A
Net
.... "'"';""
H
10 - t
'
o.av ev
Zb -
e o 9
~,o
10
-3
0
p.
IIle ~ aIp
10
9 ,~
...-4
m
'~
O-S
0 0
10 "e 10 is
9
a A 9149
.... J_
10 is
__J__ 9
~__
9
10 ~7 doptnft
9A s---J
9
9~ . 9
10 TM lout 3
101o
9 A ,L . * * -
10 zo
)
Figure 20. Specific contact resistance as a function of doping on p-type 6H-SiC using an AI/ Ti alloy for ohmic contact. Taken from Ref. 161.
Because A1 is easily oxidized, other contacts such as boron carbide (for example, B4C), Ta, Ti, cobalt silicides[ 162]and Mo may prove to be superior for high-temperature ohmic p-type contact. As an example, refractory metal boride ohmic contacts to p-type 6H-SiC (doped at 1.3 x 1019 cm -3 with A1) have recently been reported.[ 1651Short anneals (2-10 min.) at 1100~ in vacuum (5 x 10-7 torr) yielded specific contact resistance of 8.2 x 10-5 f2.cm 2 for CrB 2 and 5.8 x 10-5 f2.cm 2 for W2B. Experiments with TiB 2 were also performed, but with varying results. Longer anneals lowered the specific contact resistance values further. Recent studies have indicated that formation of cobalt silicide (CoSi2) may prove to be a thermally stable, low-resistance ohmic contact with a specific contact resistance of 2 ~tm/min), well controlled, anisotropie, and smooth. The etch rates tend to be faster in pure C12 due to the strong chemical component of the etch mechanism and the high concentration of reactive C1 available in the plasma. However, the surface morphology can be rough. The addition of BC13 to the plasma chemistry reduces the etch rate due to less available reactive C1, but the etch tends to be smoother and more anisotropie (due to a higher physical component of the etch mechanism). Smooth anisotropie profiles have also been obtained for Ga-eontaining compound semiconductors etched in CH4/Hz-based plasma chemistries at much slower rates as compared to Cl-based plasmas. This is unexpected based on the information in Table 1 (the volatility of the Ga(CH3) 3 etch product is much higher than that for GaC13 and demonstrates the complexity of the etch process where redeposition, polymer formation, and gasphase kinetics can influence the etch rates). Etch rates for In-containing species in a RIE-generated C12 plasma at room temperature tend to range from a few hundred to-1000 A/min. The surfaces are typically rough with an overcut profile due to the low volatility of the InC13 etch products and preferential loss of the group-V species. However, at elevated temperatures (>130~ the volatility of InC13 etch product increases and the etch rates and surface morphology improve.[8~ 831 Recently, high rate InP etching (1.0 to 3.8 ~tm/min) has been reported in C12/Ar ECR-generated plasmas at 25~ [84][85]These results suggest that using a high-density plasma etch platform enables sufficient sputter desorption of the low volatility InC13 etch products before the surface is passivated. However, for room temperature etching of In-containing films, CHa/H2-based plasmas have often been preferred due to the formation of volatile (CH3)3In etch products.[ 55][86]-[89]InP etch rates obtained in RIEand ECR-generated CHa/H2 plasmas range from--500 to 1000 A/min with anisotropic profiles and smooth etch surfaces. GaN typically etches at much slower rates than GaAs. The high volatilities of the Ga- and the nitrogen-based etch products shown in Table 1 implies that the etch rates are not limited by desorption of the etch products. However, due to the strong bond energies of the III-V nitrides, the initial breaking of the group-III-N bond, which must precede the etch product formation, may be the rate limiting step.[ 9~ This topic will be discussed in more detail later in this chapter.
262
Wide Bandgap Semiconductors
T a b l e 1. Boiling Points for Possible Etch Products o f C o m p o u n d S e m i c o n ductor M a t e r i a l s E t c h e d in H a l o g e n - or CHa/H2-based P l a s m a s
Etch Products
3.1
Boiling Points (~
AICI3
183
AIF3
na
AII3
360
A1Br3
263
(CH3)3AI
126
GaCI 3
201
GaF 3
1000
GaI 3
sublimes 345
GaBr3
279
(CH3)3Ga InCl 3
55.7 600
InF 3
>1200
InI 3
na
InBr3
sublimes
(CH3)3In
134
PC13
76
PF 3
-101
PI 3
decomposes
PBr 3
173
PH 3
-88
AsC13
130
AsF 3
-63
AsI 3
403
AsBr3
221
AsH 3
-55
Gas Mixtures A l t h o u g h fast G a N etch rates h a v e b e e n observed in c h l o r i n e - b a s e d
plasmas, the source o f reactive C1 as well as the use o f additive gases h a v e not been discussed. V e r y often gas m i x t u r e s are used in a p l a s m a to increase etch rate, i m p r o v e anisotropy, increase selectivity, or p r o d u c e
Plasma Etching of lll-V Nitrides 263 smoother etch morphologies by adjusting the chemical:physical ratio of the etch mechanism. For example; 02, H 2, or N 2 have been added to halogenbased plasmas to modify the concentration of reactive species in the plasma or initiate the formation of sidewall polymers to enhance anisotropy. Additionally, inert gases such as Ar or He are often used to stabilize or dilute the plasma. Fluorine-containing gases are often added to improve etch selectivity of GaAs films over A1GaAs films due to the formation of non-volatile A1F3 etch products. The use of gas mixtures to etch the UI-V nitrides in high-density plasma reactors has had varying success and is summarized below.
3.2
Hydrogen
The addition ofH 2 to chlorine-based plasmas typically results in slower etch rates for GaAs and GaP since H 2 acts as a scavenger of reactive C1 and forms HC1. In Fig. 6, GaN etch rates are plotted as a function of %H 2 concentration for ECR- and ICP-generated C12/H2/Ar plasmas. ECR etch conditions were: 1 mtorr, 850 W ECR source power, 150 W rf-cathodepower with a corresponding de-bias of-190 + 20 V, 25 sccm C12/H2, and 5 sccm Ar. ICP etch conditions were: 2 mtorr, 500 W ICP source power, 95115 W rf-cathode-power with a constant dc-bias of-250 + 10 V, 25 sccm C12/H2, and 5 sccm Ar. Samples etched in the ECR were grown by metalorganic molecular beam epitaxy (MO-MBE), whereas samples etched in the ICP were grown by metal-organic chemical vapor deposition (MOCVD). GaN etch rates in the ECR and ICP increased slightly as H 2 was initially added to the C12/Ar plasma (10% H2). Using quadrupole mass spectrometry (QMS) in the ECR discharge, the C1 concentration (indicated by m / e = 35 peak intensity) remained relatively constant at 10% H 2. As the H 2 concentration was increased further, the C1 concentration decreased and the HC1 concentration increased as the GaN etch rates decreased in both plasmas, presumably due to the consumption of reactive C1 by hydrogen. GaN etch profiles showed a strong dependence on the %H 2 in the C12/H2/Ar ECR plasma (Figs. 7a-c). The etched surface was quite rough (Fig. 7a) in the C12/Ar plasma, possibly due to preferential removal of the GaC13 etch products or micromasking of the surface. The foot observed at the edge of the etched feature may be attributed to mask-edge erosion due to the aggressive attack ofphotoresist by reactive C1. As the H 2 concentration was increased to 20%, the etch became smooth and very anisotropic (Fig. 7b). At 60% H 2, the etch remained smooth and anisotropic with a slight foot at the base of the feature (Fig. 7c). For GaN samples etched in
264
Wide Bandgap Semiconducwrs
the C12/H2/ArICP-generated plasma (Figs. 7d-0, the features were anisotropic and smooth, independent of%H 2.
5000
4OOO
ICP
3000
2000
ECR
I000
0
t
i 0
i 20
i 40
! 60
~
i 80
i 100
% H 2 Concentration in CI2/H 2 Plasma
Figure 6. GaN etch rates as a function of%H 2 for ECR- and ICP-generated CI2/H2/Arplasmas.
I
ttt~/
ECR
(a)
(b)
(c)
|CP
(d)
(e)
(f)
Figure 7. SEM micrographs of GaN samples etched in either an ECR or ICP CI2/H2/Arplasma (a, d) 0 % HE, (b, e) 2 0 % HE, and (c, f) 6 0 % H E. T h e photoresist m a s k has b e e n r e m o v e d .
at
Plasma Etching of lll-VNitrides 265 In Fig. 8, BCI 3 was substituted for C12 and was used to etch GaN in both the ECR and ICP reactors. The increase in etch rate observed at 10% H 2 concentration in the ECR-generated BC13 plasma correlated with an increase in the reactive C1 concentration as observed by QMS. As the H 2 concentration was increased further, the C1 concentration decreased, the HC1 concentration increased, and the GaN etch rates decreased due to the consumption of reactive C1 by hydrogen. In the ICP reactor, GaN etch rates were quite slow and decreased as H 2 was added to the plasma, up to 80% where a slight increase was observed. The GaN etch rates were consistently faster in the C12-based plasmas as compared to BC13, due to the generation of higher concentrations of reactive C1.
2000
.=. E -2
i
i
i
1500 ECR 1000
Z
ICP
500
0 !
I
I
I
I
I
I
0
20
40
60
80
100
% H 2 Concentration in BCI3/H 2 P l a s m a
Figure 8. GaN
3.3
etch rates as a function o f % H 2 for ECR- and ICP-generated BCI3/H2/Ar plasmas.
Sulfur Hexafluoride
In Figs. 9 and 10, GaN etch rates are shown for ECR- and ICPgenerated C12/SF6/Ar and BC13/SF6/Ar plasmas as a function of %SF 6. ECR plasma conditions were: 1 mtorr, 850 W ECR source power, 150 W rf-cathode-power with a corresponding dc-bias o f - 190 + 20 V, 25 sccm C12/SF 6 or BC13/SF 6, and 5 seem Ar. ICP plasma conditions were: 2 mtorr, 500 W ICP source power, 95-115 W rf-cathode-power with a constant debias of-250 • 10 V, 25 seem C12/SF 6 or BC13/SF 6, and 5 seem Ar. (The pressure was increased to-~3 mtorr in the ICP reactor at > 60% SF6). With
266
Wide Bandgap Semiconductors
the substitution of SF 6 for H 2 in the C12-based plasma, the GaN etch rates were typically slower, by a factor of 2. In general, as the concentration of SF 6 was increased, the etch rates decreased independent of etch technique. As the %SF 6 was increased from 0 to 20 in the ECR, the C1 concentration (m/e = 35) decreased but remained significant; faster GaN etching at 20% SF 6 might be expected, based on the C1 concentration alone. However, formation of SC1 (m/e = 67) was observed at 20% SF6 which may be responsible for the reduced GaN etch rate due to consumption of the reactive C1 by S. At 30 and 40% SF6, the C1 concentration was greatly reduced and slow GaN etch rates resulted.
5000
,, I
I
I
,
,
I ~
~ECR 4000
ICP
....,
E 3000
2000 t.t.l t,.3
I000
-
,_, I
0
. /
.
20
I
I
40
60
0
% S F 6 C o n c e n t r a t i o n in CI2/SF6Plasm,'~
Figure 9. GaN etch rates as a function of %SF 6 for ECR- and ICP-generated CI2/SFe/Ar plasmas.
In Fig. 10, the opposite trend was observed for BC13, where the GaN etch rates were typically greater when SF 6 was substituted for H 2 in both the ECR and ICP. The GaN etch rate increased up to 20% SF 6 in the ICP and 30% SF 6 in the ECR, and then decreased sharply. In the ECR, the C1 concentration (m/e = 35) increased as the SF 6 increased to 30% and then decreased at 40%. This implies that at low SF 6 concentrations, the SF 6 enhanced the dissociation of BC13 resulting in higher concentrations of reactive C1 and faster etch rates. However, above --30% SF6, the sulfur appeared to consume reactive chlorine as the SC1 concentration increased and the etch rate decreased.
Plasma Etching of lll- V Nitrides 267
5000
,
l
i
"'C
i ~
4000 E
-2
,,...,
3000
2000
t.rd
1000
0
1
1
1
I
0
20
40
6()
_
i
~0
% SF Concentration ill BCIJSF Plasnia 6
Figure 10. GaN etch rates as a function o f % S F 6 for ECR- and ICP-generated BC13/SF6/Ar plasmas.
3.4
Nitrogen
GaN etch rates were also obtained for C12/N2/Ar and BC13/N2/Ar plasmas under the following ECR and ICP conditions: 2 mtorr, 850 W ECR source power or 500 W ICP source power, 110 to 170 W rf-cathode-power with a constant de-bias of-200 + 10 V, 25 seem C12/N2 or BC13/N2, and 5 seem Ar. Figure 11 shows GaN etch rates as a function of %N 2 concentration in both ECR and ICP C12 plasmas. As the %N 2 increased in the C12 plasma, the GaN etch rates decreased due to less available reactive C1. However, as shown in Fig. 12, GaN etch rates increased significantly as N 2 was added to the ECR- and ICP-generated BC13 plasma. Etch rates increased up to 40% N 2 and then decreased at higher N 2 concentrations. This trend was similar to that observed in ECR and ICP etching of GaAs, GaP, and !n-containing materials.[ 91]-[93]Ren et al. observed peak etch rates for In-containing materials in an ECR plasma at 75% BC13- 25% N 2. As N 2 was added to the BC13 plasma, Ren observed maximum emission intensity for atomic and molecular C1 at 75% BC13 using optical emission spectroscopy (OES). Correspondingly, the BC13 intensity decreased and a BN emission line appeared. It was suggested that at 75% BC13, N 2 enhanced the dissociation of BC13 resulting in higher concentrations of reactive C1 and C1 ions and higher etch rates. This may explain faster GaN etch rates observed as N 2 was added to the BC13/Ar plasmas. In this study, however,
268
Wide Bandgap Semiconductors
higher concentrations of reactive C1 and BN emission were not observed using OES in the ICP.
7000
I
I
I
I
I
-~
6ooo
ECR [
" - ~ ICP I "~
5000 4OOO
.~
3O00
2ooo r~ 1000 I
I
I
I
I
0
20
40
60
80
I
100
% N 2 Concentration in Cl2/N2Plasma Figure 11. GaN etch rates as a function of %N 2 for ECR- and ICP-generated CI2/N2/Ar plasmas.
4000
I~.~ i
ECR
3500
ICP
3000 *~
2500 2000 1500
z
r~
lO~ 500 "
I
I
I
I
I
I
0
20
40
60
80
100
-
% N 2 Concentration in BCl3/N2Plasma Figure 12. GaN etch rates as a function of %N 2 for ECR- and ICP-generated BC13/N2/Ar plasmas.
Plasma Etching of lll-VNitrides 3.5
269
Argon
Small amounts of Ar are often added to halogen-based plasmas to stabilize the plasma or to improve the sputter desorption efficiency of etch products from the substrate surface (thus increasing etch rate and anisotropy). [93]-[95] GaN etch rates are shown in Figs. 13 and 14 as a function of %Ar in ICP- and ECR-generated C12 and BC13 plasmas. The plasma etch conditions were: 2 mtorr chamber pressure, 30 sccm total flow rate,-~-250 V dc-bias, 25~ substrate temperature, and 500 W ICP source power or 850 W ECR source power. As Ar was added to either ICP or ECR C12 discharges, GaN etch rates decreased due to less available reactive C1 in the plasma. When BC13 was substituted for C12, etch rates were as much as 6 times slower than those obtained in the C12/Ar plasma due to lower concentrations of reactive C1 (see Fig. 14).
7000
i
t
I
!
t'
I -El- ~c~ I 6000 "~
5000 4000
t~ .m
3000
m Z
2000 IOO0
-
I
I
I
I
I
0
20
40
60
80
t
=
100
% Ar Concentration in Cl2/ArPlasma Figure 13. GaN etch rates as a function of%Ar for ECR-and ICP-generatedCI2/Arplasmas.
Based on the preceding observations, GaN etch rates as a function of plasma chemistry can be summarized as follows: (a) comparable in ECR and ICP reactors; (b) faster in C12-based plasmas as compared to BC13based plasmas; (c) sensitive to the gas mixtures and gas ratios; and (d) highly dependent upon the concentration of reactive C1 in the plasma.
270
Wide Bandgap Semiconductors 2000
I
I
I
I
L~
15oo
/
ICP
E ~
1000
Z
500
I
r~
I
0
I
I
I
I
20
40
60
80
II 100
% Ar Concentration in BCI3/Ar Plasma
Figure 14. GaN etch rates as a functionof%Ar for ECR-and ICP-generatedBCI3/Arplasmas.
3.6
Additional Halogens
Additional halogen-containing plasmas have been used to etch GaN including IC1/Ar and IBr/Ar. In Fig. 15, GaN etch rates are shown as a function of if-cathode-power in ECR-generated IC1 and IBr plasmas. Etch conditions were: 1.5 mtorr pressure, 1000 W ECR source power, 4 seem IC1 or IBr, and 4 seem Ar. The GaN etch rates increased with increasing if-cathodepower at similar rates up to 150 W. However, at 250 W if-cathode-power, the GaN etch rate in ICI/Ar increased to > 1.3 ~tm/min whereas the IBr etch rate decreased slightly. This is the fastest etch rate reported to date for GaN. Near-surface AES showed no loss of N in the Ga:N ratio at low rfcathode-powers. The high volatility of the GaI 3 etch products may have increased the chemical etch mechanism of the Ga and thereby minimized preferential loss of the lighter N atoms and maintained the stoichiometry of the as-grown film. [58][96][97]
3.7
Hydrocarbons
Pearton and coworkers were the first to etch GaN, A1N, and InN in ECR-generated CH4/H2/Ar plasmas.[ 55] Etch rates were < 400 A/min at 1 mtorr pressure, 1000 W ECR source power, and ~ -250 V de-bias. As stated earlier, CH4/H 2 plasmas have been used to etch compound semiconductor materials at etch rates much slower than those obtained for C1based plasmas. In Fig. 16, GaN etch rates are plotted as a function of rf-
Plasma Etching of l l l - V Nitrides 271 cathode-power for C12/Ar and CH4[H2/Ar ECR- and RIE-generated plasmas. These experiments were performed in the same chamber with either 1000 or 0 W of applied ECR source power. The pressure was held constant at 1.5 mtorr and the plasma chemistry was 5 seem C12 and 10 sccm Ar or 5 seem CH4, 15 seem H 2, and 10 sccm Ar. GaN etch rates were significantly faster in C12/Ar, independent of etch technique, possibly due to more efficient formation and sputter desorption of the GaC13 etch products as compared to Ga(CH3) 3. Redeposition or polymer formation on the etched surfaces may also be initiated by CHa/H 2 plasmas, thereby reducing the etch rates. Faster GaN etch rates in the high-density ECR were attributed to higher plasma densities, which enhanced the GaN bond breaking and sputter desorption of the etch products.
14,ooo 12,ooo
I
I
i
I
I
--~ -~cll
10,000 ~
8000
-~
6000
z
4ooo
r~
2000
0
I
I
I
I
I
50
100
150
200
250
300
if-cathode-power (W) Figure 15. GaN etch rate as a function of if-cathode-power for ECR-generated ICI/Ar and IBr/Ar plasmas.
In Fig. 17, InN etch rates are plotted as a function of if-cathodepower for C12/Ar and CH4/H2/Ar ECR- and RIE-generated plasmas. These experiments were performed under the same conditions discussed above. InN etch rates obtained in the ECR were fast (independent of plasma chemistry) and increased with if-cathode-power, due to more efficient bond breaking and/or sputter desorption of the etch products at high de-bias. In RIE, the etch rates were quite low and comparably independent of plasma
272
Wide Bandgap Semiconductors
chemistry. GaN etched much faster than InN in the RIE-generated C12/Ar plasma due to lower volatility oflnC13 as compared to GaC13.
7000 - - ~ - RIE Clz/Ar ECR CI21Ar ""A" "RIECH4/H2/Ar - ~ ECR CH4/I-12/Ar
6000 o,,.~
E
5000
i
4000 .=
;~
3000
S
..Q
.(3 s
2ooo
.
..~
..... ,., O..., .,, ........~jC. .....
1000 .z~ -
2 mtorr for ECR and > 5 mtorr for ICP), GaN etch rates decreased due either to higher ion energies, lower plasma densities, redeposition, or polymer formation on the substrate surface.
5.0
I O N E N E R G Y AND P L A S M A D E N S I T Y
Etch characteristics show a strong dependence on ion energy and plasma density. Ion energies influence the physical component of the etch, whereas plasma density can effect both the physical and chemical components. In general, etch rates increase as the ion energy increases due to improved sputter desorption of etch products from the surface as well as
274
Wide Bandgap Semiconductors
more efficient breaking of the group-III-N bonds. Anisotropy also typically improves due to the perpendicular ion energy which maintains straight wall profiles. Etch rates may decrease under high dc-bias conditions due to sputter desorption of reactive species from the surface before reaction occurs indicating an adsorption limited etch regime. As a function of increasing plasma density or source power, etch rates typically increase due to: (1) higher concentrations of reactive species which increase the chemical component of the etch mechanism, and/or (2) higher ion flux which increases the sputter desorption component of the etch mechanism. However, trends have been observed where etch rates stabilize or decrease at high plasma densities due either to saturation of reactive species at the surface or creation of an adsorption limited regime. The effects of ion energies and plasma densities are more obvious for high-density plasma systems, since ion energies and plasma densities can be more effectively decoupled as compared to RIE.
8000 7000
E
6000
o
~
4000
~
3000
~ r~
2~
_]+
ICP
i'
1000 0
0
I
I
I
I
I
2
4
6
8
10
Pressure (mTorr) F i g u r e 18. GaN etch rates as a function o f pressure in an ICP-generated C12/H2/Ar plasma at 25~ and in an ECR- generated C12/HE/CH4/Ar plasma at 170~
The III-V nitrides typically etch at much slower rates than more conventional compound semiconductors films, including GaAs and InP. As stated earlier, due to the high volatility of etch products formed in Cl-based plasmas and the strong bond energies of the III-V nitrides, the GaN bond breaking may be the rate limiting step of the etch.[ 9~ These tendencies may
Plasma Etching of l l l - V Nitrides 2 75 be observed in Fig. 19 where GaN etch rates are plotted as a function ofrfcathode-power for ECR, ICP, and RIE discharges. Plasma etch conditions were: 1 mtorr pressure, 10 seem C12, 15 seem H 2, 3 sccm CH4, 10 seem Ar, 1000 W ECR source power, and 500 W ICP source power. For ECR and ICP, the rf-cathode-power ranged from 1-450 W with corresponding dc-biases o f - 1 0 to -400 + 25 V. RIE rf-powers ranged from 50-450 W with corresponding de-biases of-270 to -950 + 50 V. Independent of etch technique, the GaN etch rate increased as the rf-cathode-power or ion energy increased, due to more efficient bond breaking of the GaN and/or more efficient sputter desorption of the etch products. GaN rates were-- 5 to 10 times faster in the high-density ECR and ICP etch systems as compared to R/E, due to a two step process directly related to the plasma flux. Initially, the high-density plasmas increase the bond breaking mechanism allowing the etch products to form and then produce efficient sputter desorption of the etch products. The etched surface morphology or rootmean-square (rms) surface roughness was evaluated and quantified using atomic force microscopy (AFM) as a function of plasma etch technique. It remained relatively constant and smooth (< ---3 nm), independent of ion energy and etch platform.
7000
I
I
--~
I
- ECR
6OOO .N
E
5ooo
.
- -A,- - ICP
/
v
4OOO
-~
3OOO
Zea
2OOO
J
.§
/
o 9 f
o~ .f
1000 0 I
I
I
I
0
1~
2~
3~
I
4~
5~
if-cathode-power (W) Figure 19. GaN etch rates as a function of rf-cathode-power as etched in ECR-, ICP-, and RIE-generated CI2/H2/CH4/Arplasmas.
276
Wide Bandgap Semiconductors
In Fig. 20, GaN etch rates and rms roughness are shown as a function of ICP-source power. During these runs, the if-cathode-power was held constant at 150 W, resulting in a decrease in de-bias (-285 to - 130 V) as the ICP source power was increased (250 to 1000 W). Lower debiases were attributed to increased plasma density and lower mean free path at higher ICP source powers. GaN etch rates increased as the ICP source power was increased from 250-500 W, due to higher concentrations of reactive species in the plasma or higher plasma flux which resulted in more efficient bond breaking and sputter desorption of the etch products. As the ICP source power was increased further, the GaN etch rates decreased, due either to lower ion energies or sputter desorption of the reactants at the surface prior to reaction. The rms roughness increased to greater than 65 nm at 750 W ICP source power (---150 V dc-bias), possibly due to preferential sputtering of N at higher plasma densities.
7000
I
I
I
I
70
I
m / 6000
60
/ /
E
5000
~
4000
-~
3000
so
I -
2ooo
40
30 ~ 0
m
z
~"
-
20 ~
/
r~
I
rrns
10
I
1000 I
I
I
I
I
200
400
600
800
1000
ICP Power (W) Figure 20. GaN etch rates and rms-roughness as a function o f l C P source power in an ICPgenerated CI2/H2/Ar plasma.
6.0
TEMPERATURE DEPENDENCE
As stated earlier, etch rates and profiles can be influenced by the volatility or desorption rate of etch products. Using a variety of clamping techniques along with backside heating or cooling procedures, the
Plasma Etching of llI-VNitrides 277 temperature at which the sample is maintained during the etch can be controlled. The substrate temperature can effect the desorption rate of etch products, the gas-surface reaction kinetics, and the surface mobility of reactants which earl influence etch results. Several reports are available which present temperature dependent etching of GaAs and InP. [80]-[82][9811991 GaAs etch rates generally increase at substrate temperatures above -~150~ due to one of the mechanisms stated above; high temperature etching of Incontaining species has been quite successful using C12-based chemistries, due to the higher volatilities of In-chlorides at temperatures above--130~ Etch rates for GaN and A1N are shown in Fig. 21 as a function of temperature. The GaN samples were etched in C12/I-I2/CH4/Arand C12/H2/ Ar plasmas while the A1N samples were etched only in C12/HJCH4/Ar plasmas. The etch rates for A1N showed a monotonic decrease of a factor of--2 as the temperature was increased from 30-170~ The GaN etch rate in the CI2/Hz/CH4/Ar plasma was relatively constant up to ~125~ and then increased slightly as the temperature was increased to 170~ When CH 4 was removed from the plasma chemistry, the rates decreased by---2040% and showed a monotonic increase as a function of temperature. Faster GaN etch rates at 170~ may be attributed to either higher volatility and desorption rates of etch products, or changes in the gas-surface reaction kinetics. In Fig. 22, the InN etch rates are shown as a function of temperature for the same plasma chemistries. The InN etch rate decreased by > 60% as the temperature was increased to 150~ for CI2]Hz/CH4/Ar plasma chemistry, however, the etch rate increased rapidly above 150~ A similar trend was observed in the CI2/H2/Ar plasma chemistry, with etch rates 20-50% slower than those obtained with CH 4 in the plasma chemistry. Faster GaN and InN etch rates observed with CH 4 present in the plasma chemistry, may be attributed to the formation of the group III-methyl etch product. The initial InN etch rate decrease observed in the C12]I-I2/CH4/Ar plasma chemistry may be due to competitive reactions to form either InC13 or (CH3)3In. As the temperature was increased above 150~ either the volatility or desorption rate of one of the etch products increased or the reaction kinetics became dominated by one of the surface reaction mechanisms. Figure 23 shows characteristic Auger spectra for GaN samples before and after ECR etching at 30 and 170~ in a C12/H2/CH4/Ar plasma. Prior to exposure of the GaN to the plasma, the Auger spectrum showed normal amounts of carbon and native oxide on the GaN surface. Following exposure to the plasma, within experimental error, there was no change in the stoichiometry of the GaN surface, with some residual atomic C1 present. Similar results were observed for the InN and AIN samples.
278
Wide Bandgap Semiconductors
2500
i
i
GaN with CH4
2000
-~N--~ rrl,
1500 GaN without CH 4 1000
-
-|
......... |
AIN with CH 4 500
-.|
-
0 0
I
I
I
50
100
150
200
Temperature (~ Figure 21. Etch rates of GaN and AIN as a function of temperature for ECR-generated C12/ H2/CH4/Ar or CI2/HE/Ar plasmas.
2500
i
i
i
InN
?
2000
-2 1500
t~ 9 ut _CH 4
1000
500 0
I
I
I
50
100
150
Temperature
200
(~
Figure 22. Etch rates of InN as a function of temperature for ECR-generated CI2/H2/CH4/ Ar or C12/HE/Ar plasmas.
Plasma Etching of llI-VNitrides 279
GaN
I
N
O
I
.... GaI
C
(b) 30~ Etch
I
i
Ud
1
Ga
c (c) 170~ Etch
J
CI
I C
I 200
N
I Ga
I
I
400
600
I 800
I 1000
I 1200
Electron Energy (eV) Figure 23. AES surface scans ofGaN (a) before exposure to the plasma, (b) at 30~ at 170~ in an ECR-generated CI2/H2/CHa/Ar plasma.
7.0
GROWTH
and (c)
TECHNIQUE
Since several GaN samples discussed in this chapter were grown by different techniques, it is important to identify any etch dependence on growth technique. In Fig. 24, GaN ECR etch rates are shown as a function of rf-cathode-power for samples grown by MO-MBE, rf-MBE, and MOCVD. The ECR plasma conditions were: 2 mtorr pressure, 22.5 sccm C12, 2.5 sccm H2, 5 sccm Ar, 30~ electrode temperature, 1000 W microwave power, and rf-cathode-powers ranging from 1-450 W with corresponding dc-biases of-25 to -275 + 25 V. GaN samples were etched simultaneously. As the rf-cathode-power and ion energies increased, the
280
Wide Bandgap Semiconductors
GaN etch rates increased (due to more efficient bond breaking of the GaN and/or more efficient sputter desorption of the etch products), independent of growth technique. Etch rates approaching 9000 A/min were obtained for the MO-MBE and rf-MBE GaN samples at 450 W rf-cathode-power (--275 V de-bias). A trend where the MO-MBE GaN etched faster than the rf-MBE GaN (which was faster than the MOCVD GaN) was observed. Faster etch rates correlated with higher rms-roughness for the as-grown GaN samples. The rms-roughness for the as-grown GaN samples were: 19.38 + 0.44 nm for MO-MBE, 3.12 + 0.84 nm for rf-MBE, and 1.76 + 0.29 nm for MOCVD.
10,ooo -~
--
I
I
I
I'
""
I
8000
o,,.
E ~"
6000 oO
J
e-_
2
4(~o
z
~
2000
0
IO0
200
300
400
500
rf-cathode-power (W) F i g u r e 24. GaN etch rates as a function o f if-cathode-power for M O - M B E , M O C V D , and If-MBE grown GaN in a CI2/H2/Ar ECR-generated plasma.
8.0
E T C H P R O F I L E , M O R P H O L O G Y , AND STOICHIOMETRY
Three of the more critical parameters used to evaluate the value of an etch process are etch profile, surface morphology, and the chemical composition of the etched surface. Maintaining surface stoichiometry and smoothness are critical for subsequent process steps including the formation of metal contacts, deposition ofinterlevel dielectric films, passivation, or regrowth. Maintaining etch profile is essential for dimensional control and reliability of metal step coverage.
Plasma Etching of l l l - V Nitrides 281 Figure 25 shows SEM micrographs of GaN and InN samples etched in an ECR-generated C12/H2/CHa/Ar, at 850 W of ECR source power, 1 mtorr, 170~ and 150 W rf-cathode-power. The GaN etch was approximately 5800 A deep and was anisotropic with reasonably smooth sidewalls and surfaces. The high anisotropy of the etch may be attributed to the possible formation of a sidewall polymer involving the methane, as previously reported by Constantine et al. with this plasma chemistry.[ 8~ A trench was observed at the base of the GaN feature which may have occurred due to the Si3N4 mask-edge erosion. The InN etch was somewhat rough with a sloped sidewall, also due to erosion of the mask-edge. The InN was etched approximately 1.12 lam deep, which was approximately 1000 A into the GaAs substrate. The surface roughness may have been due to etching GaAs under high temperature C12 plasma conditions or to preferential etching of the InN.
GaN I LIIII
....
InN LIIIi
(a)
(b)
Figure 25. SEM micrographsof (a) GaN and (b) InN etchedat 150 W rf-cathode-powerin an ECR-generatedC12/H2/CH4/Arplasma. Highly anisotropic, smooth GaN etching has been achieved in a C12/ H2/Ar ICP-generated plasma as shown in Fig. 26. The GaN, which was grown by MOCVD, was overetched by approximately 15%. The plasma conditions were: 5 mtorr pressure, 500 W ICP source power, 22.5 seem C12, 2.5 seem H 2, 5 seem Ar, 25~ electrode temperature, and 150 W
282
Wide Bandgap Semiconductors
if-cathode-power with a corresponding dc-bias of-280 • 10 V. Under these conditions, the GaN etch rate was -~6880 A/min with highly anisotropic, smooth sidewalls. The vertical striations observed in the sidewall were due to striations in the photoresist mask which were transferred into the GaN feature during the etch. The sapphire substrate was exposed during the overetch period and showed significant pitting possibly due to defects in the substrate or growth process. With optimization of the masking process, these etch parameters may yield profiles and sidewall smoothness which improve etched facet laser performance.
Figure 26. SEM mierograph ofMOCVD GaN etched in an ICP-generated CI2/H2/Arplasma.
As discussed earlier, surface roughness can be quantified using AFM. The rms roughness and AFM images for GaN and InN etched in an ECR-generated C12/H2/CH4/Ar plasma are shown in Figs. 27 and 28, respectively, as a function of if-cathode-power. The GaN rms roughness remained relatively constant as the if-cathode-power was increased from 0-150 W; however, as the if-cathode-power was increased further to 275 W, the rms roughness increased to ~85 nm. The data suggest that at high if-cathode-power preferential sputtering or micro-masking occurred which roughened the surface. The rms roughness for InN was greatest at 65 W if-cathode-power. This implies that the ion-bombardment energy was critical to balance the chemical and sputtering effect of this plasma chemistry in order to maintain smooth surfaces and reasonable etch rates. In general, III-V nitride rms-roughness was smooth over a wide range of etch parameters, but appeared to be very sensitive to high-density plasma conditions and high ion energies.
Plasma Etching of lll-V Nitrides 283
100
I
I
I
| I I I
80 E
~ ~:~
60
I
InN/~
///
40
0
1,-,
~
2o
!
!
i
0
100
200
300
if-cathode-power (W) Figure 27. RMS roughness for GaN and InN as a function of rf-cathode-power in an ECRgenerated CI2/H2/CH4/Ar plasma.
In many cases, preferential etching of the substrate can cause significant variations in the stoichiometry of the material (which in tum can effect device performance or post-etch process steps). Plasma species may also be adsorbed or implanted into the substrate, which can also reduce device performance. In Fig. 29, Auger spectra were taken to evaluate the near-surface stoichiometry of GaN samples (a) before and after ECR etching at 850 W applied microwave power and rf-cathodepowers of (b) 65 W, and (c) 275 W. Prior to exposure of the GaN to the plasma, the Auger spectrum showed a Ga:N ratio of 1.5 with normal amounts of adventitious carbon and native oxide on the GaN surface. Following exposure to the plasma, a general tendency for the Ga:N ratio to increase with rf-cathode-power was observed. The Ga:N ratio increased from 1.8 to 2.3 as the rf-cathode-power was increased from 65-275 W at a microwave power of 850 W. Additionally, the Ga:N ratio also increased as a function of ECR microwave power. Within experimental error, these trends imply that the GaN film was being depleted of N, perhaps due to preferential etching or sputtering of the lighter N-atoms due to higher ionbombardment energy and higher ion density. Auger spectra were also taken to determine the near-surface stoichiometry of GaN as a function of plasma chemistry following exposure to ECR-generated C12/H2/Ar, C12/
284
Wide Bandgap Semiconductors
SF6/Ar , BC13/H2/Ar , and BC13/SF6/Ar plasmas. In general, the Ga:N ratio increased as the %H 2 or %SF 6 concentration increased in either BC13 or C12. These trends imply that the GaN films were depleted of N perhaps due to preferential chemical etching of the N atoms with the addition ofH 2 or SF 6 to the plasma.
GaN
InN
a)
As-grown
3.2u,,, gm
lain
~.3Ilnl
b) rf=65
9.5r,,,,
W
lain
Jam
-~ t m'rl
c)
rf = 2 7 5 W
85nm
~m
.m
39nm
Figure 28. AFM micrographs for (a) GaN and InN as-grown, (b) GaN and InN etched at a if-cathode-power of 65W, and (c) GaN and InN etched at a if-cathode-power of 275 W in an ECR-generated C12/H2/CH4/Ar plasma. The Z-scale is 100 nm/division.
9.0
PLASMA INDUCED DAMAGE
As stated in earlier sections, the fabrication of high-density integrated circuits and optoeleetronie devices often requires pattern transfer with
Plasma Etching of lll- V Nitrides 285 highly anisotropic profiles and smooth surface and sidewall morphologies. These requirements are often achieved using plasma etch techniques where energetic ions are accelerated from the plasma to the wafer. However, when these energetic ions strike the sample, surface damage as deep as 1000 A can occur (100), causing degradation of electrical and optical properties of the device. Plasma-induced-damage can include defects or dislocations in the lattice, formation of dangling bonds on the surface, implanted etch ions, or deposition of material on the sample. Attempts to minimize the damage by reducing the ion energy below the damage threshold for compound semiconductors (< 40 eV) [49] or by increasing the chemical component of the etch often results in more isotropic profiles, significantly limits minimum dimensions, and reduces the etch rate.
As-grown
100 :
(a)
--
.
,
~.0
100
Si
65 W if-cathode-power
~
.
-
200 I
(b)
O v
C,:N-1.8__,!.]j
.
.
.
.
.
0
Si
O
IGa m
w
I-
-
200 275 W 200
-
if-cathode-power -
I-- . . . . .
-
-
I
__'
o
(c)
Ga:N =2.3
I|
o .,',4
t
Si
il
O II 500
I
"'' ,
I
Ga
I--I
1000
Kinetic Energy (eV) Figure 29. AES surface scans ofGaN (a) before exposure to the plasma, (b) at 65 W, and (c) 275 W rf-cathode-power, 1 mtorr, 170~ and 850 W microwave power in an ECRgenerated CI2/H2/CH4/Ar plasma.
286
Wide Bandgap Semiconductors
Since GaN is more chemically inert than GaAs and has higher bonding energies, higher ion energies may be used during the etch process with potentially less damage to the material. However, reports of plasmainduced-damage of the III-V nitrides have been limited. Pearton and coworkers have reported plasma-induced-damage results for InN, InGaN, and InA1N in an ECR-generated plasma (the damage increased as a function of ion flux and ion energy).[l~ In the following section, ICP and ECR plasma-induced-damage of GaN is evaluated using photoluminescence (PL) measurements as a function of rf-cathode-power and source power. Pure Ar plasmas were used to simulate the ion bombardment conditions created during plasma etching of the III-V nitrides. It is important to realize that the use of a pure Ar plasma creates a worse case scenario for plasma-induced-damage due to the lack of chemical interactions. With the introduction of reactive gases to the plasma for a given plasma power and density, the damage will be reduced when compared to a sputter mechanism, since damaged material is typically being removed at a higher rate, leaving a shallower damage depth. Prior to the plasma exposure, a 2 inch GaN wafer (grown by MOCVD) was mapped out to examine the uniformity of the PL emission. In Fig. 30, the PL spectrum at the center of the wafer taken with the low resolution grating is shown. The spectrum consisted of two distinct features. The dominant near band-edge resonance was seen at 3.472 eV. Emission resonances in this spectral region have been identified with recombination of a neutral-donor-bound exciton.[l~ 1~ The free exciton resonance, expected at an energy of~3.485 eV (102), was not clearly resolved. The broad spectral feature centered at approximately 2.21 eV was associated with emission from deep level impurities. In Fig. 31, the peak intensity of the near band-edge emission is plotted as a function of radial position on the two inch GaN wafer. The intensity dropped by --15% at a radial position of 0.5 inches and ---30% at a radial position of 0.8 inches. Toward the edge of the wafer, the intensity drop was significantly more rapid. The samples that were used in the etch studies were taken from the center of the wafer. The PL spectra were taken before and after etching for each sample. In Fig. 32, the % change in the peak PL intensity versus rf-cathodepower is plotted for both ECR and ICP etching. GaN samples were exposed to ICP- and ECR-generated Ar plasmas for 1 minute under identical plasma conditions while the rf-cathode-power was increased. The de-bias was --10 to 65% higher in the ECR under comparable conditions. For the ICP case, at relatively low rf-cathode-powers (1 and 50 W), the PL intensity slightly degraded, and as the rf-cathode-power was increased up to 250 W, increasing degradation in PL intensity was seen. Depth profiling
Plasma Etching of l l l - V Nitrides 28 7 of similar films at a rf-cathode-power of 1 W (-~ - 10 V de-bias) revealed no detectable removal of GaN; whereas, at 250 W etch (-300 V de-bias)-~770 A of GaN was lost during a 1 minute exposure. Distinctly different results were obtained for etching in the ECR plasma system. Etching with very low rf-cathode-power (1 W) resulted in > 80% increase in the PL intensity and virtually no sputter loss of GaN. Etching at higher rf-cathode-powers also improved the PL intensity, but to a lesser degree as the rf-cathodepower was increased. The highest rf-cathode-power (150 W) etch resulted in a very slight decrease in PL intensity and a GaN sputter rate of--820 A/min.
9
!
,
0
,
,
!
,
,
9
,
,
..,d
2.00
2.50
3.00
3.50
Energy (eV) Figure 30. PL spectrum from the GaN film at T=10 K. The effect of plasma density on the peak near band-edge PL intensity was also studied. GaN samples were exposed to ICP- and ECRgenerated Ar plasmas for 1 minute while the de-bias was held at -~ -65 V. The data were more scattered than the rf-cathode-power data for both ICP and ECR conditions. The ICP showed virtually no change in PL intensity at 250 W source power and then decreased by 30% as the ICP source power was increased to 750 W. The PL intensity decreased by only 10% at 1000 W ICP source power, which was an improvement of almost 20% over 750 W. In the ECR, an increase of--115% in PL intensity at 250 W ECR source power was observed. Similar to the trend observed as a function of rf-cathode-power, the PL intensity also improved at higher ECR source powers but at a lower rate. Sputter rates for GaN were /;~/>,/'///~/;~/',/'/,//J ~
sapphire
GaN Bu ffer
]
Mask InGaN ~
b.\\\\\\\\\\\
Co)
- GaN AIN
///.////.//,/////////////~
~
G aN Buffer
ECR Etch sapphire
InGaN ~
K\\\\\\\\\\"~I "~,.,.. GaN
~,,~,,,~.~,~,4,,...
(c) GaN Buffer
Wet Etch
sapphire
Figure 34. Schematic diagram of the process steps for the fabrication o f a GaN microdisk laser showing (a) the epitaxial growth structure, (b) ECR plasma etch, and (c) selective wet etch of the AIN.
The following ECR conditions were incorporated to remove-500 to 1000 A of the p-GaN: 1.0 mtorr pressure, 850 W ECR source power, 150 W if-cathode-power with a corresponding de-bias of~- - 185 V, 15 seem BC13, 20 seem H 2, and 5 seem Ar. The etch process must be well controlled and repeatable with smooth etch morphology and minimal damage for subsequent metallization of the source and drain ohmic contacts.
292
Wide Bandgap Semiconductors
a)
/
GaN/InGaN A1N
GaN/InGaN b)
A1N
Figure 35. SEM micrograph of the microdisk laser showing (a) the non-selective ECR plasma etch and (b) the selective wet etch of A1N.
11.0 C O N C L U S I O N S In summary, the utilization of high-density ECR and ICP chlorinebased plasmas has resulted in high rate (> 1 ~m/min), smooth, anisotropie etching of the III-V nitrides. The source of reactive C1 (C12, BC13, IC1, etc.) and the use of additive gases (Ar, H2, N2, and SF6)have several effects on the etch characteristics of III-V nitride films. Using Cl2-based plasmas typically results in high concentrations of reactive C1 which increases the chemical component of the etch mechanism and the GaN etch rate. The addition of H 2, N 2, or SF 6 to C12- or BC13-based plasmas appears to effect the chemical removal of the N atoms from the GaN as well as the concentration of reactive C1 in the plasma (which directly correlates to etch rate). Very smooth pattern transfer has been obtained for a wide range of plasma chemistries and conditions. The bond breaking mechanism for the
Plasma Etching of l l l - V Nitrides 293 III-V nitride bonds appears to be critical and perhaps the rate limiting step in the etch mechanism. The use of high-density plasmas results in improved etch results possibly due to a two step process directly related to the plasma flux. Initially the high-density plasmas increase the bond breaking mechanism allowing the etch products to form and then produce efficient sputter desorption of the etch products. ICP etching of the III-V nitrides in C12/H2 plasmas yields etch profiles and sidewall morphologies which may improve the yield and performance of etched facet lasers. Although work has been significant in this area over the past few years, identifying plasma conditions which maintain the stoichiometry of the as-grown films and minimize plasma-induced-damage are critical to the fabrication of high performance III-V nitride devices and must be further developed and better understood.
GateImplant
photoresist (a)
"~ISI_GaN ~ ~ ~ ~ photoresist
Ohmic Implant
Co)
Gatecontact ~
Gatecontact "Dryetch of p-GaN"
(c)
Ti/AIohmiccontacts
~txxXxXxl
LX%XXXX~I
(d)
I'
'1
Figure 36. Schematic diagram of the process steps for the fabrication of a GaN JFET (Zolper, Ref. 20) showing (a) selective area implant of the n-channel and p-gate, (b) refractory metal gate deposition, gate definition, and selective area implant of the source and drain, (c) a timed plasma etch to remove the p-GaN in the source and drain regions, and (d) formation of ohmic source and drain contacts.
294
Wide Bandgap Semiconductors
ACKNOWLEDGMENTS The author would like to acknowledge S. J. Pearton, C. R. Abemathy, J. C. Zolper, M. Hagerott Crawford, R. D. Briggs, C. B. Vartuli, R. F. Karlicek, Jr., C. Tran, M. Schurman, C. Constantine, C. Barratt, K. P. Killeen, J. Han, R. J. Hickman, and P. A. Barnes for their collaboration on this work. The author would like to also thank P. L. Glarborg, A. T. Ongstad, and L. Griego for their technical support. This work was supported by the United States Department of Energy under contract DEAC04-94AL85000. Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy.
REFERENCES 1. Nakamura, S., Mukai, T., and Senoh, M., Jpn. J. AppL Phys., 30: L 1998 (1991) 2. Nakamura, S., Mukai, T., Senoh, M., and Iwasa, N., Jpn. J. Appl. Phys., 31:L139(1992) 3. Foresi, J. S., and Moustakas, T. D.,Appl. Phys. Lett., 62:2859 (1993) 4. Binari, S. C., Rowland, L. B., Kruppa, W., Kelner, G., Doverspike, K., and Gaskill, D. K., Electron. Lett., 30:1248 (1994) 5. Nakamura, S., Mukai, T., and Senoh, M., Appl. Phys. Lett., 64:1687 (1994) 6. Nakamura, S., Senoh, M., Iwasa, N., and Nagahama, S., Jpn. J. Appl. Phys., 34:L797(1995) 7. Nakamura, S., Senoh, M., Nagahama, S., Iwasa, N., Yamada, T., Matsushito, T., Kiyoku, H., and Sugimoto, U., Jpn.. J. Appl. Phys., 35 :L74 (1996) 8. Matsuoka, T., Sasaki, T., and Katsui, A., Optoelectronic Devic'es and Technologies, 5:53 (1990) 9. Amano, H., Kito, M., Hiramatsu, K., and Akasuki, I., Jpn. J. Appl. Phys., 28:L2112 (1989) 10. Strite, S., andMorkoc, H.,J. Vac. Sci. Technol. B, 10:1237 (1992) 11. Kahn, M. A., Kuznia, J. N., Van Hove, J. M., Olson, D. T., Krishnankutty, S., and Kolbas, R. M., Appl. Phys. Lett., 58:526 (1991) 12. Tansley, T. L., and Egan, R. J., Phys. Rev. B, 45" 1094 (1993) 13. Khan, M. A., Bhattarai, A. R., Kuznia, J. N., and Olson, D. T., Appl. Phys. Lett., 63:1214(1993) 14. Pearton, S. J., Abemathy, C. R., Wisk, P., Hobson, W. S., and Ren, F., Appl. Phys. Leg., 63:1143 (1993)
P l a s m a E t c h i n g o f l l l - V Nitrides 15. 16. 17. 18. 19. 20. 21. 22. 3.
24. 25.
26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40.
295
Davis, R. F., Proc. IEEE 79:702 (1991) Nakamura, S., Senoh, M., and Mukai, T., Jpn. J. Appl. Phys., 30:L 1708 (1991) Akasaki, I., Amano, H., Kito, M., and Kiramatsu, K., Lumin., 48/49:666 (1991) Nakamura, S., Senoh, M., and Mukai, T.,Appl. Phys. Lett., 62:2390 (1993) Khan, M. A., Chen, Q., Shur, M. S., Dermott, B. T., Higgins, J. A., Burm, J., Schaff, W., and Eastman, L. F., Electron. Lett., 32:257 (1996) Khan, M. A., J. N. Kuznia, Bhattarai, A. R., and Olson, D. T., Appl. Phys. Leg., 62:1248 (1993) Khan, M. A., Shur, M. S., and Chen, Q., Electron. Lett., 31:2130 (1995) Zolper, J. C., Shul, R. J, Baca, A. G., Wilson, R. G., Pearton, S. J., and Stall, R. A.,Appl. Phys. Leg., 68:2273 (1996) Manos, D. M., and Flamm, D. L., Plasma Etching, Academic Press, San Diego (1989) Chapman, B., Glow Discharge Processes, Academic Press, New York (1980) Dautremont-Smith, W. C., Gottscho, R. A., and Schultz, R. J., Semiconductor Materials and Processing Technology Handbook, (G. E. McGuire, ed.), pp. 191, Noyes, Park Ridge, NJ (1988) Chu, T. L., J. Electrochem. Soc., 119:1200 (1971) Pankove, J. I.,J. Electrochem, Soc., 119:1118 (1972) Guo, Q. X., Kato, O., and Yoshida, A.,J. Electrochem. Soc., 139:2008 (1992) Pearton, S. J., Abemathy, C. 1L, Ren, F., Lothian, J. R., Wisk, P. W., and Katz, A., J. Vac. Sci. Technol. A, 11:1772 (1993) Sheng, T. Y., Yu, Z. Q., and Collins, G. J.,Appl. Phys. Lett., 52:576 (1988) Pauleau, T.,J. Electrochem. Soc., 129:1045 (1982) Taylor, K. M., and Lenie, C.,J. Electrochem. Soc., 107:308 (1960) Long, G., and Fuster, L. M.,J. Am. Ceram. Soc., 42:53 (1959) Barrett, N. J., Grange, J. D., Sealy, B. J., and Stephens, K. G., J. Appl. Phys., 57:5470(1985) Aita, C. R., and Gawlak, C. J., J. Vac. Sci. Technol. A, 1:403 (1983) Kline, G. R., and Lakin, K. M.,Appl. Phys. Lett., 43:750(1983) Mileham, J. R., Pearton, S. J., Abernathy, C. 1L, MacKenzie, J. D., Shul, R. J., and Kilcoyne, S.P.,J. Vac. Sci. Technol. A, 14:836(1996) Mileharn, J. IL, Pearton, S. J., Abemathy, C. R., MacKenzie, J. D., Shul, IL J., and Kilcoyne, S. P., Appl. Phys. Lea., 67:1119 (1995). Minsky, M. S., White, M., and Hu, E. L., Appl. Phys. Lett., 68" 1531 (1996) Youstey, C., Adesida, I., and Bulman, G., Electron. Letts., 33:245 (1997)
296
Wide Bandgap Semiconductors
41. Pearton, S. J., Abemathy, C. R., Ren, F., and Lothian, J. R., J. Appl. Phys.,
76:1210,(1994) 2, Adesida, I., Mahajan, A., Andideh, E., Asif Khan, M., Olsen, D. T., and 3, 4. 5.
46. 47. 48.
49. 50. 51.
52. 53. 54. 55. 56. 57. 58. 59. 0.
61.
Kuznia, J. N., Appl. Phys. Lett., 63:2777 (1993) Lin, M. E., Zan, Z. F., Ma., Z., Allen, L. H., and Morko~, H., Appl. Phys. Lett., 64:887 (1994) Ping, A. T., Adesida, I., Asif Khan, M., and Kuznia, J. N., Electron. Lett., 30:1895 (1994) Lee, H., Oberman, D. B., and Harris, J. S., Jr., Appl. Phys, Lett., 67:1754 (1995) Asmussen, J.,J. Vac. Sci. Technol. B, 5:328 (1987) Popov, O. A., High-density Plasma Sources, Noyes Publications, Park Ridge, NJ (1996) Lieberonan, M. A. ,and Gottscho, R. A., "Plasma Sources for Thin Film Deposition and Etching," (M. H. Francombe, and J. L. Vossen, eds.), Physics of Thin Films, Vol. 18, Academic Press, San Diego (1994) Constantine, C., Johnson, D., Pearton, S. J., Chakrabarti, U. K., Emerson, A. B., Hobson, W. S., and Kinsella, A. P.,J. Vac. Sci. Technol. B, 8:596 (1990) Pearton, S. J., Chakrabarti, U. K., Kinsella, A. P., Johnson, D., and Constantine, C., Appl. Phys. Lett. 56:1424 (1990) Murrell, A. J., Grimwood, R. C., O'Sullivan, P., Gilbert, M., Vanner, K., Ruddell, F., Davies, I., Hilton, K., Bland, S., and Spear, D., Technical Digest, Proc. 1992 GaAs IC Symposium, pp. 173 Cheung, R., Lee, Y. H., Lee, K. Y., Smith, T. P., III, Kern, D. P., Beaumont, S. P., and Wilkinson, C.D.W.,J. Vac. Sci. Technol. B, 7:1462 (1989) Ko, K. K., and Pang, S. W.,J. Electrochem. Soc., 141:250(1994) Pearton, S. J., Abemathy, C. R., Ren, F., Lothian, J. R., Wisk, P. W., Katz, A., and Constantine, C., Semicond. Sci. Technol., 8:310 (1993) Pearton, S. J., Abemathy, C. R., and Ren, F., Appl. Phys. Lett., 64:2294 (1994) Zhang, L., Ramer, J., Zheng, K., Lester, L. F., and Hersee, S. D., Mat. Res. Soc. Proc., 395:763 (1996) Humphreys, B., and Govett, M., MIJNSR 1 (1996) Vartuli, C. B., Pearton, S. J., Lee, J. W., Hong, J., MacKenzie, J. D., Abemathy, C. R., and Shul, R. J.,Appl. Phys. Lett., 69:1426 (1996) Vartuli, C. B., MacKenzie, J. D., Lee, J. W., Abemathy, C. R., Pearton, S. J., and Shul, R. J., J. Appl. Phys., 80:3705 (1996) Pearton, S. J., Abemathy, C. R., and Ren, F., Appl. Phys. Lett. 64:3643 (1994) Shul, R. J., Kilcoyne, S. P., Hagerott Crawford, M., Parmeter, J. E., Vartuli, C. B., Abemathy, C. R., and Pearton, S. J.,Appl. Phys. Lett., 66:1761 (1995)
Plasma Etching of llI-VNitrides
297
62. Shul, R. J., Howard, A. J., Pearton, S. J., Abemathy, C. R., Vartuli, C. B., Barnes, P. A., and Bozack, M. J., J. Vac. Sci. Technol. B, 13:2016 (1995) 63. Vartuli, C. B., Pearton, S. J., Abemathy, C. R., Shul, R. J., Howard, A. J., Kilcoyne, S. P., Parmeter, J. E., and Hagerott Crawford, M., J. Vac. Sci. Technol. A, 14:1011 (1996) 64. Shul, R. J., Ashby, C. I. H., Rieger, D. J., Howard, A. J., Pearton, S. J., Abemathy, C. R., Vartuli, C. B., and Barnes, P. A., Mat. Res. Soc. Symp. Proc., 395:751 (1996) 65. Shul,R. J., McClellan, G. B., Pearton, S. J., Abemathy, C. R., Constantine, C., and C. Barratt, Electron. Lett., 32:1408 (1996). 66. Shul, R. J., McClellan, G. B., Casalnuovo, S. A., Rieger, D. J., Pearton, S. J., Constantine, C., Barratt, C., Karlicek, R. F., Jr., Tran, C., and Schurman, M., Appl. Phys. Lett., 69:1119 (1996) 67. McLane, G. F., Meyyappan, M., Cole, M. W., and Wrenn, C., J. Appl. Phys., 69:695 (1991) 68. Meyyappan, M., McLane, G. F., Lee, H. S., Eckart, E., Namaroff, M., and Sasserath, J.,J. Vac. Sci. Technol. B, 10:1215 (1992) 69. McLane, G. F., Casas, L., Pearton, S. J., and Abemathy, C. R., Appl. Phys. Lett., 66:3328 (1995) 70. Adesida, I., Ping, A. T., Youtsey, C., Sow, T., Asif Khan, M., Olson, D. T., and Kuznia, J. N., Appl. Phys. Lett., 65:889 (1994) 71. Ping, A. T., Adesida, I., and AsifKhan, M.,Appl. Phys. Lett., 67:1250 (1995) 72. Gillis, H. P., Choutov, D. A., Martin, K. P., and Song, L., Appl. Phys. Lett., 68:2255(1996) 73. Gillis, H. P., Choutov, D. A., Steiner, P. A., Piper, J. D., Crouch, J. H., Dove, P. M., and Martin, K. P., Appl. Phys. Lett., 66:2475 (1995) 74. Gillis, H. P., Choutov, D. A., and Martin, K. P., JOM, 50 (1996) 75. Gillis, H. P., Choutov, D. A., Martin, K. P., Pearton, S. J., and Abemathy, C. R., d. Electrochem. Soc., in oress. 76. Leonard, R. T., and Bedair, S. M., AppL Phys. Lett., 68:794 (1996) 77. Chuang, T. J., Laser Microfabrication: Thin Film Processes and Lithography, (D. J. Ehrlich and J. Y. Tsa, eds.), p. 87, Academic, San Diego (1989) 78. Ashby, C. I. H., Properties ofGaAs, 2nd edition INSPEC, London, Ch. 20, 653-682 (1990) 79. Hayes, T. R., Indium Phosphide and Related Materials: Processing, Technology, and Devices, (A. Katz, ed.), Artech House, Boston, Ch. 8, 277-306(1992) 80. Constantine, C., Barratt, C., Pearton, S. J., Pen, F., and Lothian, J. R., Appl. Phys. Lett. 61:2899 (1992)
298
Wide Bandgap Semiconductors
81. Constantine, C., Barratt, C., Pearton, S. J., Ren, F., and Lothian, J. R., Electron. Lett. 28:1749 (1992) 2, Lishan, D. G., and Hu, E. L., Appl. Phys. Lett., 56, 1667 (1990) 83. Donnelly, V. M., Flamm, D. L., Tu, C. W., and Ibbotson, E., J. Electrochem. Soc., 129:2533 (1982) 4,, Thomas, S., III, Ko, K. K., Pang, S.W.,J. Vac. Sci. Technol. A, 13:894 (1995) 85. Lee, J. W., Hong, J., and Pearton, S. J.,Appl. Phys. Lett., 68:847 (1996) 86. Niggebrugge, U., Klug, M., and Garus, G., Inst. Phys. Conf. Ser., 79:367 (1985) 87. Keskinen, J., Nappi, J., Asonen, H., and Pessa, M., Electron. Lett., 26:1371 (1990) 88. Hayes, T. R., Dreisbach, M. A., Thomas, P. M., Dautremont-Smith, W. C., and Heimbrook, L. A., dr. Vac. Sci. Technol. B, 7:1130 (1989) 89. Cheung, R., Thoms, S., Beaumont, S. P., Doughty, G., V. Law, and Wilkinson, C. D. W.,Electron. Lett., 23:857 (1987) 0. Pearton, S. J., and Shul, R. J., III-Nitrides, Academic Press, in press. 91. Ren, F., Lothian, J. R., Kuo, J. M., Hobson, W. S., Lopata, J., Caballero, J. A., Pearton, S. J., and Cole, M. W., Jr. Vac. Sci. Technol. B, 14:1 (1995) 2. Ren, F., Hobson, W. S., Lothian, J. R., Lopata, J., Caballero, J. A., Pearton, S. J., and Cole, M. W., Appl. Phys. Lett., 67:2497 (1995) 3. Shul, R. J., McClellan, G. B., Briggs, R. D., Rieger, D. J., Pearton, S. J., Abemathy, C. R., Lee, J. W., Constantine, C., and Barratt, C., J. Vac. Sci. and Technol. A, in press (1996) 94. Shul, R. J., Howard, A. J., Vartuli, C. B., Barnes, P. A., and Seng, W., J. Vac. Sci. Technol. A, 14:1102 (1996) 5. Shul, R. J., Baca, A. G., Rieger, D. J., Hou, H., Pearton, S. J., and Ren, F., Mat. Res. Soc. Syrup. Proc., 421:245 (1996) 6. Vartuli, C. B., Pearton, S. J., Lee, J. W., MacKenzie, J. D., Abemathy, C. R., and Shul, R. J., J. Vac. Sci. and Technol B, 15:98 (1997) 7. Vartuli, C. B., Pearton, S. J., MacKenzie, J. D., Abemathy, C. R., and Shul, R. J.,J. Electrochem. Soc., 143:L246 (1996) 98. Contolini, R. J.,J. Electrochem. Soc., 135:929 (1988) 99. Pearton, S. J., and Hobson, W. S.,J. Appl. Phys., 66:5018 (1989) 100. Pang, S. W.,J. Electrochem. Soc., 133:784 (1986) 101. Pearton, S. J., Lee, J. W., MacKenzie, J. D., Abernathy, C. R., and Shul, R. J., Appl. Phys. Lett., 67, 2329 (1995) 102. Shan, W., Schmidt, T. J., Yang, X. H., Hwang, J., Song, J. J., and Goldenberg, B., Appl. Phys. Lett., 66:985 (1995)
P l a s m a E t c h i n g o f l l I - V Nitrides
299
103. Chen, G. D., Smith, M., Lin, J. Y., Jiang, H. X., AsifKhan, M. and Sun, C. J., Appl. Phys. Lett., 67:1653 (1995) 104. Gottscho, R. A., Preppemau, B. L., Pearton, S. J., Emerson, A. B., and Giapis, K. P.,J. Appl. Phys., 68:440 (1990) 105. Aydil, E. S., and Gottscho, R. A., Mat. Sci. For., 148/149:159 (1994) 106. McCall, S. L., Levi, A. F. J., Slusher, R. E., Pearton, S. J., and Logan, R. A., Appl. Phys. Lett., 60:289 (1992) 107. Abemathy, C. R., Pearton, S. J., MacKenzie, J. D., Mileham, J. R., Bharatan, S. R., Krishnamoorthy, V., Jones, K. S., Hagerott Crawford, M., Shul, R. J., Kilcoyne, S. P., Zavada, J. M., Zhang, D., and Kolbas, R. M., Solid-State Electron., 39:311 (1996) 108. Lee, J. W., Vartuli, C. B., Abemathy, C. R., MacKenzie, J. D., Mileham, J. R., Pearton, S. J., Shul, R. J., Zolper, J. C., Hagerott Crawford, M., Zavada, J. M., Wilson, R. G., and Schwartz, R.N.,J. Vac. Sci. Technol. B, 10:3637 (1996)
7 Ion Implantation Wide Bandgap Semiconductors
in
John C. Zolper
1.0
INTRODUCTION
,\
The wide bandgap semiconductors, group-Ill Nitrides (INN, GaN, and A1N), SiC, and diamond have long been recognized as being ideal materials for short wavelength light emitters or detectors and for highpower, high-temperature electronics.[1]-[ 3] The group III nitride material system has generated considerable excitement in the semiconductor research community with the success in fabricating high performance light emitting diodes (LEDs), lasers, and transistors.J4] -[12] The interest in SiC has mostly been fueled by its potential in the electronics arena duc to its high breakdown field and high thermal conductivity making it useful for high-power or high-temperature operation.[13]-[ 16] Diamond also is expected to bc most applicable to electronic applications (transistors or detectors) due to its predicted excellent transport and thermal properties which arc superior to the group-HI Nitridcs and SIC.[1Ill7] Details of the attractive properties of these materials arc described in other chapters of this book. The wide bandgap II-VI semiconductors (e.g., ZnSe, ZnS, etc.) are not included in this chapter since their low melting point (-300~ makes them unattractive for electronic device applications where implantation is expected to have the most impact. Their low melting point will also make 300
lonlmplantation 301 it difficult to perform the required thermal annealing to activate implanted dopants in the II-VI materials. For this reason, little work has been reported on implantation in the wide bandgap II-VI semiconductors.[ 18] In this chapter, the status of ion implantation doping and isolation in group-III Nitride, SiC, and diamond semiconductors is presented. Ion implantation is a process whereby doping or compensating impurities are injected into a semiconductor by a high energy accelerator.[ 19]Implantation has come to be the dominant doping technology in silicon and gallium arsenide microelectronics, although there was significant technological development required to achieve this success. This success was due to the ability to precisely control the doping concentration and profile as well as the ability to minimize the processing thermal budget. Therefore, with improved wide bandgap semiconductor starting materials and improved understanding of the ion implantation process, this technology can be expected to play a critical enabling role in the maturation of these promising materials into sophisticated, manufacturable devices. In the following sections, the material and process technologies related to ion implantation in wide bandgap semiconductors are presented. First, the use of implantation to produce high resistivity, isolating regions is discussed. Second, ion implantation doping to achieve n- and p-type material is described. Third, results are given for the redistribution or diffusion of implanted impurities that occurs during the high temperature implantation activation annealing process. Fourth, a review is given of implantation-induced crystal defects and the removal of implant damage by thermal annealing. Fifth, examples are presented of devices in each material system that have employed implantation isolation or doping. Finally, areas for future work are outlined and conclusions are drawn. In each of the sections, a discussion is given for the three materials considered in the chapter: group-III nitrides, SiC, and diamond.
2.0
IMPLANTATION ISOLATION
Implant isolation has been widely used in compound semiconductor devices for inter-device isolation (as in transistor circuits) or to produce current channeling in lasers.[2~ 22] The implantation process can compensate the semiconductor layer either by a damage or a chemical mechanism. For damage compensation, the resistance typically goes through a maximum
302
Wide Bandgap Semiconductors
with increasing post-implantation annealing temperature as the damage is annealed out and hopping conduction is reduced. At higher temperatures, the defect density is further reduced below that required to compensate the material, and the resistivity decreases. For chemical compensation, the post-implantation resistance again increases with annealing temperature, showing a reduction in hopping conduction; but it then stabilizes at higher temperatures as a thermally stable compensating deep level is formed. Typically, there is a minimum dose (dependent on the doping level of the sample) required for the chemically active isolation species to achieve thermally stable compensation.[ 23] Thermally stable implant isolation has been reported for O-implanted n- and p-type A1GaAs, where an A1-O complex is thought to form. A C-N complex is postulated for N-implanted C-doped GaAs and A1GaAs.[23]-[25] With this background, implant isolation properties of wide bandgap semiconductors are reviewed.
2.1
Group IIl Nitrides
GaN. The first report of the use of implantation to reduce the free carrier concentration in GaN was by Khan, et al. Their work involved Be and N implants in GaN and A1GaN to achieve compensation of native donor defects to enhance Schottky barrier formation.j26][ 27] It was shown that shallow implants into samples with a high concentration of native donors reduces the donor concentration and allows the formation of improved Schottky contacts. Since the samples were annealed at 1100~ after implantation, the compensation was most likely due to activated acceptors in the case of Be and perhaps due to a reduction in the concentration of N-vacancies that acted as donors (in the case of N-implantation). A study on the thermal characteristics of compensation via Nimplantation in n- and p-type GaN was first reported by Pearton.[ 28] As shown in Fig. 1, N-implantation effectively compensates both p- and n-type GaN when implanted to achieve an implanted N density of-4 x 10 TMcm -3. For both doping types, the resistance first increased with annealing temperature, then reached a maximum before demonstrating a significant reduction in resistance (after a 850~ anneal for n-type and a 950~ anneal for p-type GaN). This behavior is typical of implant-damage compensation. The defect levels estimated from Arrhenius plots of the resistance/temperature product are 0.83 eV for initially n-type and 0.90 eV for initially p-type GaN.[ 28] These levels are not at midgap, but are sufficiently deep to realize a sheet resistance > 109 ~/EI. The implantation has also been reported to
Ion Implantation 303 effectively isolate n-type GaN, with the material remaining compensated to over 850~ 29] Interestingly, H-implant compensation of n-type GaN was reported to anneal out a t - 4 0 0 ~ with an anomalous dependence on implant energy.[ 291The reason for this is presently not known. In light of this result, however, H-implantation in GaN will require further study since H is often the ion of choice for photonie device isolation applications that require deep isolation schemes. Moreover, both He- and N-implant isolation appear to rely solely on implantation damage without any chemical compensation effects analogous to those in the O/A1GaAs case.J2~ 24] However, the implantation-induced defects in GaN are more thermally stable than other III-V semiconductor materials, such as GaAs or InP, where damage levels begin to anneal out below 700~ 2~ This may be a result of the higher bandgap of GaN or the more polar nature of the lattice causing more stable defects. Furthermore, recent results for very low dose (10 l~ to 1011 cm -2) N-implantation in lightly n-type GaN (as-grown ---5 x 1016 em-3) suggest that, for low doses, N-implantation may actually increase the free electron concentration by increasing the number of N-vacancies or interstitials.[ 3~ Clearly, further work is required to fully understand the nature of implantation damage in GaN.
1010 - I F p-type 109
n-type
10 8 107 10 6 105 , ~ , , , ~ 300 400 500 600 700 800 900 1000 annealing temperature (~ Figure 1. Sheet resistance versus annealing temperature for N-implanted initially n- and ptype GaN. The N was implanted at multiple energies to give an approximatelyuniform ion
concentration of 4 x 10Is cm-3 across ~500 nm (after Ref. 27).
304
Wide Bandgap Semiconductors
InGaN. Implant isolation of the In-containing nitrides (INN, InGaN, and InA1N) was first reported using F-implantation.[ 31] That work showed that InN did not demonstrate significant compensation while the temaries (InGaN and InA1N) increased in sheet resistance by roughly an order-ofmagnitude after a 500~ anneal, Data from a more extensive study of InxGal.xN implant isolation for varying In-composition using N- and Fimplantation is summarized in Fig. 2.[ 32] The InGaN temaries only realize a maximum of a 100 fold increase in sheet resistance independent of ion species after a 500~ anneal. Pure InN shows a higher increase of 3 ordersof-magnitude, but still only achieves a maximum sheet resistance of 104 f2/EI. This may be high enough for some photonic device current-guiding applications, but is not sufficient for inter-device isolation in electronic circuits. The damage levels created by N-implantation are estimated from an Arrhenius plot of the resistance/temperature product to be a maximum of 390 meV below the conduction band.J32] The defect level is high in the energy gap, not near midgap as is ideal for implant compensation. The position of the damage level is analogous to the defect position reported for implant compensated n-type InP and InGaAs, but different from the damageassociated, midgap states created in GaAs and A1GaAs.[23][24][33]
108 , , , , , . , , . , . , . , . , , , , 107 106
91
105
D
104 r
Q.
103
102 l 01 10~
--0- as-grown - t ~ F-isolation N-isolation ,
,
I
20
,
,
,
I
,
,
,
I
t
40 60 percent In
i
,
i
80
,
,
,
i
100
Figure 2. Maximum sheet resistance versus percent In for InGaN either as-grown or implanted with F or N and annealed at the temperature for maximum compensation for each composition (ion concentration - 5 x 1019cm-3) (after Ref. 34).
lonlmplantation 305 InAIN. As shown in Fig. 3, In0.75A10.25N, in contrast to InGaN, can be highly compensated with N- or O-implantation with over a three orderof-magnitude increase in sheet resistance after a 600 to 700~ anneal while F-implantation produces only one order-of-magnitude increase in sheet resistance.[a1][34][35] The compensating level in InA1N is also high in the bandgap (with the deepest level estimated from Arrhenius plots as being 580 meV below the conduction band edge in high dose N-isolated material). However, it is sufficiently deep to achieve highly compensated material.[34][35] The enhanced compensation for N- and O-implantation in InA1N suggests some chemical component to the compensation process (as compared to F-implantation). For N-implantation, a reduction in N-vacancies (thought to play a role in the as-grown n-type conduction), may explain the enhanced compensation. For O-implantation, the enhanced compensation may be the result of the formation of an O-A1 complex as is thought to occur in O-implanted A1GaAs.[23][24]
10 lo 10 9 10 8 [Z]
ck
10 7 10 6
105 10 4 103 200
I
I
400
!
I
600
I
I
I
800
.
1000
annealing temperature (~
Figure3. Sheet resistance versus annealing temperature for O-, N-, or F-implanted Ino.7sAlo.25N (ion concentration- 5 x 10 ts cm "3) (after Ref. 34).
306
Wide Bandgap Semiconductors
Figure 4 schematically summarizes the present knowledge of the position in the bandgap of the compensating implanted defect levels in group-III nitride materials and compares these to those in GaAs and InP. Although the levels are not at midgap, as is ideal for optimum compensation as occurs in GaAs and p-type InP, the levels are sufficiently deep to produce high resistivity material (with the exception of InGaN).
n,p GaN n-InGaN (47% InN)
n-InAIN (75% InN) n: 830 meV (He 760 meV)
n&p-GaAs
n&p_InP
~-390meV
~~580meV
EC
l p: 900 meV EV Eg =
1.42 eV
1.35 eV
-2.5 eV
-2.5 eV
3.39 eV
Figure 4. Schematic representation of the position in the energy gap of compensating defect levels from implant isolation in GaAs, InP, Ino.47Gao.53N,Ino.7sAlo.25N,and GaN.
SiC. Implant isolation of SiC layers is also of technological importance to allow the fabrication of planar electronic devices and circuits (limited work has been reported in this area). Nadella reported on the properties of H-implanted, n-type 4H-SiC and achieved resistivities as high as 2 x 106 ~_cm.[ 36] The H-implanted samples demonstrated a temperature activated conduction with characteristic energy of 0.41 eV suggesting that the compensating defect is in the upper half of the bandgap. This may limit this approach for devices that need to operate at elevated temperatures since carriers will be thermally activated out of the levels. Chemical compensation has been achieved in SiC by doping with vanadium either during epitaxy or by implantation. [371-[39]Vanadium was shown to act as a deep donor 1.35 eV below the conduction band edge that effectively compensated residual boron in bulk SiC.[38]V-implantation into
Ion Implantation 307 initially p-type (boron doped) SiC achieved resistivities as high as 1012 ~')cm after annealing up to 1500~ For N-doped, n-type SiC, the same Vimplantation scheme resulted in resistivities of 106 ~-cm that, although lower than the p-type starting material, is still sufficiently high for most device applications. In contrast to the trend for the resistivities, the activation energy for conduction was estimated to be 0.76 eV for the n-type starting material, and 0.1 eV for the p-type starting material. The lower activation energy for the p-type material was not consistent with a 1.35 eV V-donor level and was attributed to alternative leakage paths, such as on the mesa edge.[ 39]This may indeed be the case since V-implantation is now being applied to electronic devices as discussed in Sec. 6.2.[ 4~ Diamond. To date, the primary challenge for diamond devices has been realizing n- and p-type doping. Therefore, little work has been done on implantation isolation in this material. One study by Kalish and coworkers examined the compensation properties of He- and H-implants on p-type diamond.[41] Type IIa diamond films were doped with boron with a multiple energy implantation scheme followed by a two step anneal at 1050~ and 1350~ 42] The films were then implanted with He at an energy of 320 keV or H at an energy of either 30 or 320 keV. Figure 5 shows the effect on resistance versus implanted dose for the three implantation schemes. All three approaches effectively compensated the film for a sufficiently high dose. The dose dependence is explained by the difference in defect profile generation for the different ions and energies. Figure 6 shows resistance versus damage density for the different implants and demonstrates that complete compensation occurs at roughly the same damage level (within a factor of two for all three approaches). This is evidence that damage profile is the key element in compensation and not a chemical compensation by H or He. The damage could be annealed out, and the initial conductivity restored, by annealing the samples at 1350~ but not at lower temperatures (600~ This work also suggests that p-type diamond may be susceptible to radiation damage that will alter the electrical properties of active devices.[ 41]
308
Wide Bandgap Semiconductors
10 TM
~ 101o
o~ Z
109
VHe
ni
u') rY
I0 e
H
107 ,
101~
10"
10 '2
DOSE
10 I~
,
, IA,,~
10"
10 '5
(ions/cm2)
Figure 5. Resistance of B-doped diamond layers under irradiation with 320 keV He, 320 keV H, or 30 keV H. The resistance of a undoped layer measured under the same conditions is shown for comparison (after Ref. 41).
1012
" "Tl'"al" ' ' 'i .... I
' ' '1 .... I
' ''i""l
' '~i .... I
1011
E o
10 l~
Ld 0 z
109
Unirradioted
j
U3 m 10e
9320keY 9320
He
"
keY H
0 undoped
9 30
10 7 101~ DAMAGE
101.
1015
DENSITY
1016
keY H 1017
1010
( v a c a n c i e s / c m 3)
Figure 6. The data of Fig. 5 replotted after converting the dose into damage density expressed as vacancies/cm 3. The damage density required to completely remove the conductivity due to the B dopant is about 5 x 1016/cm3, which is comparable to the concentration of acceptors in the B-doped layer (after Ref. 41).
lonlmplantation 309 3.0
IMPLANTATION DOPING
3.1
Carrier Ionization in Wide Bandgap Semiconductors
Before reviewing the work on implantation doping in GaN, SiC, and diamond, it is constructive to review the physics of free carrier ionization. The assumption often taken in silicon and GaAs, where ionization energies are typically < 20 meV, of compete carrier ionization at room temperature does not apply for wide bandgap semiconductors due to the high ionization energies. Table 1 contains the accepted values for ionization energies for the most commonly used dopants in GaN, SiC, and diamond. The relationship between these ionization energies and the free carrier concentration is reviewed in the following section.
Table 1. Summary of Ionization Energies of the Most Common Dopants in GaN, SiC, and Diamond
GaN a
3C-SiC b
4H-SiC b
6H-SiC b
Diamond c
donors
Si (25), 0(29)
N(28)
N (45, 100)
N (80, 130)
N (1700)
acceptors
Ca (169), Mg (170), Zn (230)
B (350), Ai (200)
B(350), Al (200)
B (350), Al (200)
B (370)
a j. C. Zolper, et al., J. Electron. Mat., 25:839 (1996); J. C. Zolper, et al.,Appl. Phys. Lett., 68:1945 (1996); S. Strite and H. Morkoc, J. Vac. Sci. Tech. B, 10:1237 (1992) b G. Pensl and W. J. Choyke, Physica B, 185:264 (1993) c R. J. Farrer, Solid State Corn, 7:685 (169); A. T. Collins and A. W. Willimas, J. Phys. C.
Solid St. Phys., 4:1789 (1971)
310
Wide Bandgap Semiconductors The free electron density (n) can be expressed as: [43]
n=N ex~_(E _Ef)]kT
Eq. (1)
and the free hole density (p) can be expressed as:
Eq. (2)
p= N ex~-(Ekf E)]
Ec(v)
where is the conduction (valence) band minimum (maximum) energy. For n-type material, the position of the Fermi level can be solved for from the following expression for the density of ionized donors:
(Ef)
N~d -Nd 1-
Eq. (3)
1
1 (E~.Z.~" ~
l + -geXp~
kT )
while for p-type material the density of ionized aceeptors is:
Eq. (4)
N~=N
1 1+ g e x p ( E :
f,]
and the conduction (valance) band density-of-states, No(v),is defined as" 3/2
Eq. (5)
Mc
Nc(~)-212zcmhz(h)kT ) M~
where is the number of equivalent minima in the conduction band and mde is the density-of-states effective mass. The electron or hole ground
Ion Implantation 311 state degeneracy is expressed by g. The donor (acceptor) ionization energy (Ed(a)) is listed in Table I. Other terms in Eqs. 3-5 have their usual meaning. To simplify the discussion, we assume that the material does not contain significant compensating impurities. The key observation from the above equations is that the free electron and hole concentrations are exponentially dependent on the carrier ionization energy as it determines the position of the Fermi level. To illustrate this point, a simple exponential dependence or Boltzmann statistics of free carrier concentration (n,p .~ N d , a exp (Ed,a/kT)) at room temperature (23~ and 300~ versus carrier ionization energy is shown in Fig. 7. Nd, a w a s set to 100 meV for Fig. 7 so that the result can be displayed as percent ionized. For an ionization energy of 170 meV, as is reported for Mg acceptors in GaN, only---0.14% of the substitutional Mg will yield free holes at room temperature.[ 4][35] This increases to -~3% at 300~ This relationship should be kept in mind when considering the effectiveness of implantation doping in these materials since proper lattice occupation of the dopant (i.e., substitutionality) alone does not control the measured free carrier concentration.
100
i
I
\-.
o~
i
1
-
I L
~23
v
--
10
~
"300 ~
,irN ,,==,,
o o,=,~ r ro
%
L_
Q.
0.1
%
0.01
,
0
i
,
50
100
,.l
~i
150 200 250 300 350 400
ionization energy (meV)
F i g u r e 7. Percent of ionized carriers versus ionization energy at 23 or 300~ simplified exponential or Boltzmann statistics; n , p - Nd, a exp (Ed, a/kT).
based on a
312
3.2
Wide Bandgap Semiconductors
Implantation Activation Temperature
Selective area ion implantation doping can be used to form highly doped contact regions in lasers and FETs or to create precisely doped transistor channels. The ability to precisely control the doping level and spatial location enables many high-performance device designs. To achieve activated implanted dopants, the ability of the material to withstand the required annealing process must be accessed. Table 2 compares the melting point of wide bandgap semiconductors with more mature semiconductors (Si and GaAs), to the temperatures commonly reported for achieving activation of implanted dopants in these materials.[ 44]For Si and GaAs, the activation temperature is roughly two-thirds of the melting temperature. However, for the higher melting point materials (SiC and GaN), this temperature is closer to 50% of the melting point. The temperature reported for dopant activation in diamond is at an even lower fraction of the melting point (--0.4). While this may suggest that SiC and GaN are actually more stable at the activation temperature, this is not the case since both materials sublime well below their melting point. In the following sections, the status of activating implanted dopants in the wide bandgap semiconductors is reviewed. Details of various annealing schemes are also presented.
Table 2. Comparison of Semiconductor Melting Points (Tmp) to the Temperature Required to Activate Implanted Dopants (Tact) (after Ref 44)
T,.p(~ GaSb InP GaAs Si SiC c GaN c diamond
707 a 1057 a 1237 a 1410 a 2797 a 2518 b 4000
T,a (~ 500--600 700-750 750-900 950-1050 1300-1600 --1100 .-1300 c
rac~/Tmp 0.71-0.85 0.66-0.71 0.61-0.73 0.67-0.74 0.46-0.57 0.44 0.33
a Robert C. Weast, Handbook of Chemistry and Physics, CRC Press, Boca Raton, FL, p. E92-93 (1983) b j. A. Van Vechten, Phys. Rev B, 7:1479 (1973) r May include defect and impurity conduction[ TM
Ion Implantation 313 3.3
GaN
In the early 1970's, Pankove and co-workers used ion implantation to characterize the photoluminescence spectra from an array of dopant species in GaN.[ 45] In this work, the energy levels of the common III-V semiconductor acceptors (C, Be, Mg, Zn, and Cd) were first determined, as summarized in Table 3. Magnesium was reported as having the shallowest acceptor level (-~240 meV), while Zn (with an energy level of~580 meV) had the strongest luminescence intensity.[46] To remove the implantation damage, these samples were annealed for 1 hr in flowing NH 3. No electrical properties of the implanted species were reported. These layers were most likely compensated by H that was generated from the decomposition of the NH 3 present during growth and during the annealing process. As discussed later, annealing in a non-hydrogen containing ambient is required to achieve electrical activity of acceptors on GaN.
Table 3. Photoluminescence Peak Position, Distance from the Conduction Band Edge, and the Rank Order of Luminescence Intensity for GaN Implanted with the Species Listed and Annealed in NH 3 at 1050~ for lh. (Data after Pankove in Ref. 45.)
Species
Peak Position
Bandedge-Peak
(eV)
(meV)
3.21
240
B
3.2
250
P
2.88
570
Zn
2.87
580
Cd
2.70
800
As
2.58
920
Ca
2.50
1000
ng
2.43
1070
C
2.17
1280
Be
2.16
1290
Mg
Rank Order of PL Intensity
314
Wide Bandgap Semiconductors
Figure 8 shows the evolution of sheet resistance versus annealing temperature for Si-implanted (200 keV, 5 x 10 TM c m -2) and unimplanted GaN.[ 28] The samples were annealed for 10 s in flowing N 2 in a SiC coated graphite susceptor. It is critical to use a hydrogen free ambient to avoid hydrogen passivation of the dopants and thus to achieve electrical activity. This discovery is in contrast to earlier annealing studies that used an NH 3 ambient to stabilize the GaN surface but allowed the generation of atomic hydrogen to passivate dopant species. [45][46] As seen in Fig. 8, electrical activity starts to occur at 1050~ as evident by the drop in sheet resistance, and further increases at 1100~ The ionization levels of implanted Si have been estimated from Arrhenius plots of either the sheet resistance or the carrier density to be 25 and 62 meV, respectively.[ 35] These values are in the range reported for epitaxiai doping of GaN with Si. [47]
10 6
105
D t~
a.
104 unimplanted --tl- Si-implanted
103 600
I
700
I
800
I
900
I
I
1000 1100 1200
annealing temperature (~ Figure 8. Sheet resistance versus annealing temperature for GaN either unimplanted or implanted with 28Si (200 keV, 5 x 1014cm-2). The implanted sample demonstrates enhanced ntype conduction after a 1050~ anneal with further improvements at 1100~ (aider Ref. 27).
Figure 9 shows an Arrhenius plot of the sheet electron concentration versus annealing temperature for Si-implanted GaN (200 keV, 5 x 1014 cm2).
lon lmplantation 315 A carrier activation energy (not to be confused with the ionization energy) can be estimated as 6.7 eV from the region of increasing electron concentration by examining Fig. 9.[35]This high activation energy is not consistent with a simple hopping process whereby the Si atoms occupy the nearest Ga vacancies to become electrically active. Such a hopping process is thought to occur during GaAs and InP implantation activation with activation energies in the range of 0.4 to 1.9 eV. [48] The high activation energy for Siimplantation in GaN can be explained in terms of a substitutional diffusion process. In this case, the implanted Si would occupy substitutional Ga-sites at relatively low temperatures, but remain electrically inactive (compensated) by implantation induced point defects. The Si would then become electrically activated only when these defects diffuse away and re-occupy an appropriate lattice site. An example of such a defect reaction is given in Eq. 6: Sio a _ N +Oa ]0 ~ Sioa + NON + Ga 0Ga
Eq. (6)
1014 r J
o I::I O
9 i,,,,,i
1013 a
= 6.7 eV
o I::I
1012 m I
lO 11 . . . . 0.7
~ .... 0.8
I I
0.9
1
,
1.1
1000/T (K l ) Figure 9. Arrhenius plot of the sheet electron concentration for Si-implanted (200 keV, 5 x 10TM cm-2) GaN versus annealing temperature. The extracted activation energy for donor formation is 6.7 meV which is consistent with an inter-diffusion process (after Ref. 34).
316
Wide Bandgap Semiconductors
Therefore, the activation energy corresponds to the inter-diffusion coefficients of Ga and N in GaN, not the energy for Si to become substitutional. The energy of 6.7 eV, estimated from Fig. 9, is in the range reported for Ga and As inter-diffusion in GaAs and, therefore, is consistent with an inter-diffusion process.[ 49] Ion implantation has also been used to estimate the ionization energy of O in GaN.[ 5~ Determining the electrical nature of O-impurities in GaN is of interest, since O has long been suspected as playing a contributing role in the background n-type conductivity of GaN.[51][52] Figure 10 shows an Arrhenius plot of the resistance/temperature product for unimplanted and O-implanted (70 keV, 5 x 1014cm -2) GaN after annealing. The O-sample is n-type and displays an estimated ionization energy of-~28 meV while the unimplanted sample remains highly resistive with an effective ionization energy of 335 meV.
1013
, '
'
'
!
'
'
'
i
'
'
I
'
'
'
I
'
'
'
I
'
'
unimplanted, 1100 ~ - I - O-implanted, 1050 ~
1012
= 335 meV
O4
~:
'
lO ll
v
~.~
1010
Q.
109 I
108
J
2
,
i
I
4
i
i
i
i
6
,
,
,
I
8
,
,
,
I
,
10
,
,
I
,
12
,
,
t4
1000/T ( K ~) Figure 10. Arrhenius plot of the resistance/temperature product for unimplanted, annealed (1100~ and O-implanted (70 keV, 5 x 10TM cm-2), annealed (1050~ GaN. The extracted ionization level for O is 28 meV while that for the unimplanted sample is 335 meV (after Ref. 47).
Figure 11 shows the evolution of sheet resistance versus annealing temperature for Mg (180 keV, 5 x 1014cm-2), Mg+P (180/250 keV, both
_Ion Implantation 317 5 x 1014cnl2), and unimplanted GaN.[ 28] The Mg-only samples remained n-type up to 1100~ while the Mg samples co-implanted with P converted from n-top type after a 1050~ anneal. The effect of the P co-implantation may be explained by a reduction of N-vacancies, or an increase in Ga-vacancies, leading to a higher probability of Mg occupying a Ga-site. Co-implantation of P has also been shown to be effective in enhancing activation and reducing diffusion for p-type implantation in GaAs. [53][54][55]The ionization levels of implanted Mg has also been determined, from an Arrhenius plot of carrier density, to be 171 meV and is consistent with the value reported for epitaxial Mg-doped GaN.[35][4]
10 7
106
D
a- ~
10 5
104 600
unimplanted ~Mg "-II-- Mg+P(n) Mg+P(p) I
700
I
800
I
900
I
1000 1100 1200
annealing temperature (~ Figure 11. Sheet resistance versus annealing temperature for GaN either unimplanted or implanted with 24Mg (180 keV, 5 x 10TM cm-2) or 24Mg+alp (180 keV/250 keV, both 5 • 1014 cm2). The Mg+P implanted sample converts from n-to-p type after a 1050~ anneal while the Mg-only sample remains n-type even after a 1100~ anneal (after Ref. 27).
Since the ionization level of Mg in GaN is much greater than kT, less than 1% of the Mg-aeeeptors will be ionized at room temperature as described in Sect. 3.1. Therefore, it would be desirable to identify an aeceptor species with a smaller ionization energy. Since Ca has been
318
Wide Bandgap Semiconductors
theoretically, suggested to be a shallow acceptor in GaN, ion implantation was used to determine the ionization energy of Ca in GaN.[5~ Figure 12 shows the evolution of sheet resistance versus the annealing temperature of Ca (180 keV, 5 x 1014 cm-2), Ca+P (180/130 keV, both 5 • 1014 cm-2), and unimplanted GaN. Both the Ca-only and the Ca+P samples converted from n-to-p type after a 1100~ anneal with a further increase in p-type conduction after a 1150~ anneal. The fact that P co-implantation is not required to achieve p-type conductivity with Ca can be understood based on the higher mass of the Ca-ion, as compared to Mg, generating more implantation damage and therefore more Ga-vacancies. This explanation is supported by the higher activation temperature required for conversion from n-to-p type for the Ca-implanted samples compared to the Mg+P implanted samples (1100 versus 1050~ The ionization level of Ca was estimated (from the Arrhenius plot shown in Fig. 13) to be 169 meV, which is equivalent to that of Mg.[ 5~ Although the ionization level of Ca is not smaller than that of Mg, Ca may be preferred for forming shallow implanted p-regions in GaN due to its heavier mass and resulting smaller projected range and straggle than Mg for a given energy.
107 _ ,
..... , . . . . .
, ....
, ....
, ....
10 6 n-to-p
V1 105 unimplanted ---{S}-Ca: n-type Ca+P: n-type Ca: p-type Ca+P: p-type
104
103 950 .
.
.
.
1
.
1000
,
,
~
I
.
1050
.
.
.
I
.
1100
.
.
.
I
.
1150
.
.
.
1200
annealing temperature (~
Figure 12.
Sheet resistance versus annealing temperature for GaN either unimplanted or implanted with 4~ (180 keV, 5 x 10 TMcm -2) or 4~ (180 keV/130 keV, both 5 x 10 TMcm2). The implanted samples convert from n-to-p type after a 1100~ anneal (after Ref. 47).
lonlmplantation 319
1013
'
.,
'
'
I
'
'
"
i
'
'
'
i
'
'
+--
0
o
1012 E = 169 meV
0
""'~
a
o O a= o o r ~lU
II
, , ,
2.8
I
3
,
,
,
I
,
,
,
3.2
,
. . . .
3.4
3.6
1000/T (K 1) Figure 13. Arrheniusplot of the sheet hole concentration for Ca-implanted(180 keV, 5 x 1014cm-2)GaN annealedat 1150~ The extractedionization level for Ca is 169 meV (after Ref. 47).
3.4
SiC
Although reports of implantation doping of SiC date back at least to 1969,[57]-[59] significant advances in understanding of implantation damage accumulation and dopant activation occurred with work by Edmond (circa 1988)on the use of elevated temperature implantation.[6~ The key result of Edmond's research was that elevated temperature implantation markedly enhanced activation of implanted dopants by limiting the buildup of implantation damage via in-situ annealing. This has led to elevated temperature implantation being widely employed for doping of SiC, as discussed in this section. Implantation of the most common donor species in SiC, N, will be presented first, followed by discussion of the most widely used p-type dopants, B and A1. Hiranoet al. reported on the activation of implanted N in 3C-SiC layers as a function of implant temperature and N-dose.[ 62]As seen in Fig. 14, for a sample annealed at 1200~ for 20 min (furnace anneal, FA), the electron concentration was significantly enhanced for N-implantation (N2 at 30 keV and a dose of 3 x 1015 cm -2) by implanting at 400~ as compared to 200~
320
Wide Bandgap Semiconductors
(or at room temperature, RT). This effect was ascribed to less damage accumulation, which is supported by damage measurements by Rutherford Backscattering (RBS) discussed later in this chapter. Hirano also showed that there is a dose saturation level for activation near 3 x 1015cm -2 for their implantation conditions. For their conditions, they achieved an electron concentration near 5 x 1019 cm -3 as determined by C-V measurements. For room temperature (RT) N-implants Kimoto reported a minimum sheet resistance of 770 DJr'l (that corresponded with a saturation in N-donors) at a dose of 8 x 1014 c m -2 for furnace annealing at 1500~ 631Hirano also reported using 10 s, 1100~ rapid thermal annealing for both RT and 400~ implants. He reported an order-of-magnitude (to 1 x 1020 cm-3) increase in the electron concentration for the elevated temperature implant, as determined from C-V measurements.[ 62] Significantly, the RTA sample had a higher electron concentration than the 1200~ FA-sample implanted under the same conditions. The high concentration achieved for the RTA sample was reported to be near the highest n-type doping reported for implanted or epitaxial SiC at the time. The success using RTA processing and the resulting high doping level suggests this will be an effective process technology for device applications. 2
10
I
I I M P L A N T S "3.0 x 1015/cm 2 400~ O 9 ~ 200~ ~ 9 ~ 0 0 ~ 20min
f ~,
.
u
n 0~176176 o 0o00 o
r~
0 (:~
,
\ _ i 03,~
o
Z 101 " o a O
~SIMS
a
aa
"
A
,.-A
o
Oo0
A A
.o
%
0
:~
~"
.
U
9
.-
-
" -
-
g
1
~
-
"I
0
-
17 10 0
I 500 DEPTH
I 1000 (,~)
101 1500
Figure 14. Carrier concentration and mobility profiles for n-type layers formed in 3C-SiC by implanting 30 keV N 2 ions to a total dose of 3.0 x l 015 cm -2 at 200~ (triangles) and 400~ (circles) and by subsequent annealing at 1200~ for 20 min (after Ref. 62).
lon Implantation 321 J. N. Pan et al. also studied elevated temperature N-implantation in SiC and, in particular, the details of the annealing process.[641 As shown in Fig. 15, for implants performed at 650~ the optimum time for the anneal is a strong function of the anneal temperature. The minimum anneal time used in this study was 5 min and was limited by the furnace annealing apparatus. It is interesting to note, however, that the sheet resistance appears to have not yet reached a minimum value at 5 min for the 1200~ samples. This also suggests that a rapid thermal anneal (t < 60 s), as reported by Hirano, may be effective to activate these types of implants.
4000 ....... i ..... 0"
,
......
,._iii
_i.ii.i.
3000
........
C v >.,
.....
2000
]
[
a
o m
n" ,-.9 r O3
1000 900 800 700 ....................................................
60O
500
|
I
,
,
, ,
|l|'
io
,
.
.
.
.
.
.
.
]
ioo
.
.
.
.
.
.
.
.
.
.
.
Iooo
.
.
.
.
Ioooo
Anneal Time (minutes)
Figure 15. Sheet resistivity as a function of anneal time for 900, 1050, and 1200~ anneal temperatures for N-implanted (3.8 x 1015cm-2at 30 keV plus 7.1 x 1015cm-2at 70 keV) 6HSiC (after Ref. 61). P-type implantation doping of SiC has been achieved with both B and A1. One of the early reports was for Al-implantation in n-type SiC by Gudkov.[ 65] Gudkov reported difficulty in maintaining the surface stoichiometry during the activation anneal, limiting the ability to quantify the acceptor activity. Rao made a significant contribution in this area by studying both A1 and B-implantation in 6H- and 3C-SiC. [66] Rao showed the importance of using elevated temperature implantation as developed by
322
Wide Bandgap Semiconductors
Edmond to achieve p-type activation. In this case, A1 was implanted at 850~ and annealed at 1400~ to produce p-type conductivity as determined by Hall and C-V measurements. These layers had--1% of the implanted Al-dose produce ionized acceptors at room temperature, due to the high ionization energy of A1 as discussed in See. 3.1. This means that 100% of the implanted A1 impurities occupied substitutional lattice sites. In this same study, p-type conductivity was not reported for B-implanted SiC. The lack of hole conduction for B was attributed to the still larger ionization energy of B in SiC, compared to A1, or a lack of suitable vacancies being formed during the implantation process for the B to occupy substitutional sites.[66] The difficulty in achieving p-type SiC by B-implantation was also reported by Kimoto, where B-implantation resulted in a highly resistive layer with an indeterminate carrier type. [67] In that same study, Kimoto was successful in producing p-type SiC with room temperature Al-implantation and annealing in an rf-induction furnace at 1500~ The apparent contradiction between Rao's report for the requirement for elevated temperature and that ofKimoto' s room temperature Al-implantation was postulated to be due to differences in the annealing dynamics between Rao's fiwnace anneal and Kimoto' s more rapid annealing scheme.[ 67] Finally, Rao has studied the effect of co-implantation of Si and C on the activation properties of Al-implanted 6H-SiC.[ 68]Co-implantation is often used in compound semiconductors to maintain the local crystal stoichiometry and promote the preferred site occupation of the implanted dopant.J53]-[55]In this case, the co-implantation of either Si or C did not enhance the formation of Al-acceptors. In fact, Si co-implantation resulted in highly resistive layers, either due to compensation by resulting carbon vacancies or other impurity/ defect complexes.[ 68]This area of co-implantation may warrant further study since there is often a dose dependence between the dopant and co-implantation species to achieve optimum electrical results.
3.5
Diamond
Diamond has long been identified as a promising semiconductor. However, its use has been limited by the inability to grow semiconductor grade synthetic diamond and by the difficulty in doping this very stable material. Ion implantation has been one of the most successful approaches to doping diamond due to the ability to create non-equilibrium defects during the bombardment process.j69][7~ A process described as cold-implantation-rapid annealing (CIRA) has been developed and applied to p-type,
Ion Implantation 323 boron doping. This involves implanting the sample at low temperature, typically liquid nitrogen temperature (77 K), to "freeze in" point defects that can then interact with dopant species during subsequent annealing. The annealing is done with a two step process, with a first anneal at ~1000~ and a second at ~1300~ 71] This approach has been successful in achieving p-type doping with a maximum hole mobility of 385 em2/Vs.[ 72] Successful boron doping has also been realized by combining the CIRA (77 K implantation and 1373 K rapid thermal annealing) process with carbon co-implantation.[73] In this study, the B-dose was varied between 1 and 10 x 1014c m "2 while the C-dose ranged from 3 to 20 x 1014 e m "2. The co-implanted sample with the optimum annealing sequence achieved a resistivity of 100 ~-cm with an estimated activation energy of conduction for boron of 0.1 eV. Optical data also support a high fraction of the boron being substitutional acceptors.[ 73] A second co-implantation approach to controlling the vacancy distribution and enhancing dopant activation is the use of a low dose preimplantation scheme that has been named "Low-Damage-Drive-In" Implantation (LODDI). In this approach, the non-dopant or co-implantation species creates the required defect distribution for the dopant atom. Although the annealing sequence depends on the dopant species employed, the key aspect in the process is the low dose required of the pre-implant. By combining He pre-implantation and various annealing steps with B-implantation, this approach produced a hole concentration of 4.0 x 1013 c m 2 and a low field mobility of 1953 c m E / V s . [69] Figure 16 shows results for a Hedamage implant (5 x 1010 cm -2) followed by a 600~ anneal, then an elevated temperature (400~ doping B-implant. During the B-implant, the B is thought to diffuse interstitially into the undamaged, underlying diamond layer. Finally, a carbon amorphizing implant is performed to allow a subsequent etch removal of the amorphous damaged region that is above the interstitially B-doped diamond. At this point, the diamond was subjected to a series of anneals at 1550~ Hall characterization of this sample gave a hole concentration of 4.0 x 1013 c m -3 with a low field mobility of 1953 cmE/Vs.[69] A similar approach for P-implantation in diamond using C as the damage implant species compared it to an Ar-implanted sample that should have the same damage profile. The results, shown in Fig. 17, demonstrated a lower resistivity for the P-implanted samples as compared to the Ar-sample, with the highest dose P-sample having the lowest resistivity. [69] The P-samples were implanted at 100 and 400~ at doses of 5 or 50 x 1016 cm -2 at each
324
Wide Bandgap Semiconductors
temperature. The Ar-implant was implanted at the sample temperatures and a dose of 4.2 x 1016 cm -2 (the same damage profile as the lower dose P-sample should be evident).[ 69]The temperature variation of the resistance for these Pimplanted samples was indicative of a highly compensated, low density of dopant. The activation energy for conduction was estimated at 0.3 eV for a P-dose of 1 x 1016 cm -2 and 0.25 eV for a dose of 1 x 1017 cm -2 suggesting that the P ionization energy is not more than 0.25 eV below the conduction band.[ 69] Although this P-implantation doping result is encouraging, significantly more work is needed to understand the actual doping and conduction mechanisms resulting from this process.
6.5 o o
ffl
o
E
o
cO
o o
W
0 0
Z < I--
0
m_ 5.5
0 0
0
UJ
0 0 o
holes=4.0x1013
0 o
UJ LIJ 7"
cm-3
0
/
O0 ~
mobility = 1953 cm 2 N - s
o _J
4.5
9
1.6
I
2
I
2.4
I ,
2.8
!
3.2
INVERSE T E M P E R A T U R E (I 000/3)
Figure 16. Sheet resistance of a diamond layer doped by B ion implantation using the LODDI process. This layer had a room temperature Hall mobility of 1953 cm2/Vs (after Ref. 66).
Ion Implantation 325
15
A
o"
E tO
Argon ions "
o Z . i-. z 1018
600~
UJ
t2
NO A N N E A L
=i
O
i.=
, I/1 r" ID
_=
=
I
15oo
~
,
I
155o
,
I
18oo
,
I
185o
Wavelength (nm)
Figure 12. Photoluminescence spectra of Er 3+ ions in AIN semiconductor host material as a function of measurement temperature. The sample film was grown by MOMBE.[521
374
Wide Bandgap Semiconductors
Photoluminescence excitation (PLE) spectra were also measured for Er-doped A1N layers grown by MOMBE.[ 47] For these measurements, the same OPO system was used as the source and the PLE signal was recorded as the ratio between the PL intensity detected at 1.54 m and the excitation power. The PLE spectra taken at 15 K from two different spots on the same Er doped A1N sample are shown in Fig. 13. Each spectrum contains one or more sharp lines superimposed on a broad PLE signal. The sharp lines may be due to a direct optical excitation mechanism via an intra-4ftransition (e.g., 4115/2 ~ 2Hll/2).The broad features may be indicative of a photocarrier mediated process. These PLE data give a partial explanation of why it was possible to excite RE doped GaN and A1N films with below bandgap optical radiation. The RE ions can be excited either through direct optical pumping into one of the 4flevels or through a carrier-mediated process.
I 41tS/2->
'
I
4F~2,SS2
'
I
'
I
'
I
,
I
~
I
4111i~ -=" 2HI1/2
4115/2 -> 4F1/2
,
I
450
L
I
,
5oo sso 600 Excitation Wavelength (nm)
I
6so
Figure 13. Photoluminescenceexcitation spectraof Er3+ions from two different spots on an A1N film that was grown by MOMBE.[47]
Rare Earth Impurities 3 75 Time-resolved PL measurements at various wavelengths gave further evidence of two excitation mechanisms occurring for the Er-doped A1N layers grown by MOMBE.[ 47] Two distinct PL decay patterns at 15 K were observed as shown in Fig. 14. The Er 3+PL, due to direct optical excitation at the absorption peaks (494, 525, and 653 nm), was longer-lived than that which occurred in the vicinity of the peaks (497, 537, and 640 nm). The PL decay transients were found to be non-exponential at all excitation wavelengths. Following a fast initial decay with a lifetime of about 50 s, the PL decayed with a long lifetime of about 0.83 ms. The PL, due to direct optical excitation at one of the absorption peaks, contained less fast-decay components than the PL which occurred via the carrier-mediated process. The de-excitation of RE PL depends upon the non-radiative recombination processes associated with the local RE ion environment. Consequently, the difference between the two types of measured PL decay indicates that two different subsets of Er 3+ sites are excited in these experiments. At least two Er 3+ centers are present in the Er-doped AIN layers grown by MOMBE. The long-lived Er 3+sites are excited primarily through direct optical excitation of one of the 4flevels; the shortlived Er 3+ sites are excited through an indirect via the carrier-mediated process.
10
I0
497 nm
e-
"e 10"
C:
10~
c .J
o.o
0.5
1.o
1.5
"rime(ms) Figure 14. Timedecayofthe 1.54mmemissionofEra§ionsin anA1Ngrownby MOMBE.The curves illustratetwo differentlifetimesdependinguponthe wavelengthofthe opticalsource.t47]
376
Wide Bandgap Semiconductors
MacKenzie et al. also succeeded in growing GaN films doped with Er atoms using MOMBE techniques.[ 53]The GaN films were grown on sapphire substrates and Er densities in the 1018 to 1019 cm -3 range (as measured by SIMS) were achieved. The Er doped films were optically excited at 488 nm and infrared spectra were measured at 13 and 300 K, as shown m Fig. 15. The room temperature PL of the Er 3+ ions in the GaN film was not nearly as intense as that from the A1N film shown in Fig. 15. The GaN films contained only a background level of O which may have influenced the lower PL intensity.
5
'
I
'
I
"
I
'
Er: GaN
i
~. =488nm (pulsed) 15K
30 ,
1450
,
I
1500
,
I
~
1550
I
1600
,
1650
Wavelength(nm) Figure 15. Photoluminescence spectra, measured at 15 and 300 K, of Er3+ ions in a GaN/ sapphire host.The sample film was grown by MOMBE.t531
Choyke et al. observed intense PL spectra in the region near 1.54 ~tm from Er-implanted samples of 4H, 6H, 15R, and 3C SiC crystals.[ 54] The samples were implanted to a fluence of about 1013 Er ions/cm 2 using four implant energies. Annealing was done in a SiC cavity at 1700~ The films were excited by an argon laser at 488 nm and PL spectra were measured at 2 K and 300 K (as shown in Fig. 16) for each of the polytypes. The PL spectra at 2 K are quite similar for the 6H, 4H, and 15R polytypes with a main line at 1.534 ~tm and 14 or more smaller peaks at longer wavelengths. The PL spectrum for the 3C sample contains two major peaks (one at 1.528 and the
Rare Earth Impurities 3 77 other at 1.534 ~tm) in addition to the smaller peaks. The integrated PL intensity varies little over this temperature range from 2 to 300 K. Moreover, even at elevated temperatures, strong emission from the Er ions can be detected. Figure 17 shows a comparison of the relative PL intensity spectra of Er 3+ ions in 6H-SiC, measured at 297, 391, and 518 K. This development of the PL spectrum indicates that the Er luminescence response is essentially flat from 2-400 K, after which it drops o f f b y a factor of about ten a t - 500 K.
0.82 l"
0.80
I
'
i
",==-- PHOTON ENERGY (oV) 0.78 0.76 0.82 'i'"
"
0.80
0.78
0.76
I
2K
i ,A,A, [=~
,
9
,..,.,
' i
!
,,
-
i
I
l
,,l,
I
r~
i,k ,iJ z
IJ.I
o z uJ r,.) r,/) uJ
z_
r
i
I
4H SIC i
,
I li
_1
o o
'
3O0K
z
m.,
"
i
15R SIC i
i, - 1
w
.
.
~. " ' - ~
~=
I
Ill 13:
300K
3C SIC o
t
15000 15400 15800 16200
.
.
I
_
l
15O0O 154OO 158OO 152OO
WAVELENGTH (A)
Figure 16. The relative photoluminescence intensity spectra, measured at 2 and 300 K, of Era+ ions in SiC polytypes of6H, 4H, 15R, and 3C. The sample films were implanted with Er+ ions and annealed at 1700 8917654]
3 78
Wide Bandgap Semiconductors
(a) .
.
.
.
(b)
6.sic
.
297 -
~
^
Z (/) UJ L..
T
v
.
zLU
, ,
(~ LU Z
A
'
--
~
--
UJ
C)
. . . . . .
-
"
-
' .....
'
9 0r
0
LU
ffl
j
zW
~
.J
~
_~ uJ
0
(/) Z
->
om'~ 0
100
UJ
Z
m
6H SiC
g3 -
K
~
...........
" ..... BK
N z
UJ
>
W
m 15000
15400 15800 16200 WAVELENGTH (A)
0
,
300
.
.
i
.
L
9
350 400 450 500 TEMPERATURE (K)
Figure 17. (a) Comparison of the relative photoluminescence intensity spectra, measured at 297, 391, and 518 K, of Er 3+ ions in 6H-SiC. (b) The relative integrated intensity of 6H-SiC implanted with Er § ions and heated from 297-518 K.[ s4]
Porous-Si may be another promising host for Er since porous Si appears to have a significantly larger bandgap than bulk Si and is rich in oxygen concentration.[ 2] Wu et al. studied porous-Si samples that were implanted with Er ions.[ sS] The samples were prepared using a wet anodization technique of p-type Si wafers and then implanted with Er + ions to a fluence of about 1015 ions/cm 2 at an energy of 380 keV. Annealing was done at 650~ for 30 minutes. The treated regions of the wafer were excited by an Ar + laser at 488 nm and the overall room temperature luminescence was recorded. As shown in Fig. 18, the PL spectrum consisted of a broad visible emission band and a narrow emission band ranging from the center at-1540 nm. The broad visible and near IR bands originated from the porous-Si host. The narrow band ( a t - 1.54 ~tm) was due to the Er 3+ transitions between the 4113/2 and the 4115/2manifolds. High resolution PL spectra, measured at 15, 150, and 300 K, of the narrow band at-- 1.54 ~tm are shown in Fig. 19.[56] The relatively large linewidth of this signal at 15 K suggested that there existed
Rare Earth Impurities 3 79 multiple Er sites within the porous-Si material. The temperature quenching of the integrated PL intensity in going from 15 K to 300 K decreased by less than a factor of two. This result supported the increased bandgap thesis ofporousSi and the use of wide gap semicondutors as hosts for RE ions. The Er 3+ PL decay transients were measured at 15 and 300 K and found to be nonexponential (see Fig. 20). [56] The decay transients were fitted with a double expontial curve, a fast initial decay (~ 140 s), and a long lifetime component of -1.3 ms. Based on these decay studies, the occurrence of two distinct Er sites within the porous-Si material seems likely. '
,
Er:Porous
9
,
---'--
', '
',
//'-r--
.....
"r-
Si
(n t--
.6 t..
.b,
Sample
A
~ ~ ~ . . _ .
_~
tn riD t~ ...I I:L
Sam ple.~,,~"'---'"'~,~~
!
600
j
I
__~
700
.
I
800
i
//
1
1500
,L . . . .
.L_
llJO0
Wavelength (nm) Figure 18. Room temperature photoluminescence spectrumof porous Si implantedwith Er3+ ions. The visible PL spectrum is fromthe porous-Si host, whereas the spectrumnear 1.54 I~m is from the Era+ions.[5s]
Another approach to RE doping of porous Si was reported by Kimura et a1.[5711581Samples were prepared using anodic etching of p-type Si wafers. Subsequently, these wafers were placed into an ethanol solution containing either ErC13 or YbC13. The RE ions were electrochemically incorporated into the porous Si layer by applying a negative bias. Room temperattwe luminescence was observed for both the Er 3+ions at-~1.54 ~tm and the Yb 3+ ions ate1.0 ~tm, after annealing at room temperatures above 900~ An Ar + laser at 514.5 nm was used to excite the RE3+e ions. As shown in Fig. 21, the PL spectrum from the Y b 3+ ions consisted of a broad emission band with a small peak at --0.98 ~tm.[5s] At 20 K, this peak, which can be attributed to the 2F5/2 --->2F7/2 transition of Yb 3+, becomes much more pronounced. The
380
Wide Bandgap Semiconductors
luminescence from the Yb 3+ ions was enhanced after annealing in H2; whereas, the luminescence from the Er3§ ions required annealing in 0 2.
m
3
U) C:
1500
1550
1600
Wavelength (nm) Figure 19. High resolution photoluminescenee spectra of porous Si implanted with Er+ ions, measured at 15, 150, and 300 K.[ 56]
A
C: c "O m
0.01
0
I
J
I
Time (ms)
Figure 20. Time decay of the 1.54 lam emission of porous Si implanted with Er3+ ions, measured from 15 to 300 K. Is6]
Rare Earth Impurities 381
I
RTA(1100%)
II
I
In H t
A
(n m
=3 .a
20K
m m
(n (l) r
4b,,r e~m
R .T. a.
0.6
0.8
1.0
1.2
1.4
1.6
W avelength (l~m) Figure 21. Photoluminescence spectra, measured at 20 and 300 K, of porous Si doped with Yb + ions. The sample films were annealed at 1100 89 for 30 sec in H2.lssl
Wang and Wessels studied the luminescence of Er 3+ions in GaP films and observed strong emissions from 12-295 K.[381The films were prepared by atmospheric pressure metalorganie vapor phase epitaxy (MOVPE) and doped during growth by sublimating Er(thd)3 and tranporting the vapor to the reaction zone with H 2 gas. The films were excited by an Ar +laser at 488 nm and the luminescence was recorded. Figure 22 shows the high resolution PL spectrum of the Er doped GaP epilayer measured at 12 K. The spectrum consists of a series of emission lines that may be due to weakly split excited levels or to different Er 3+ sites in the semiconductor host. The origin of these peaks was further investigated by Culp et al. [sg]They examined the luminescence properties of Er doped GaP as a function of temperature, applied hydrostatic pressure and excitation wavelength. They concluded that several Er 3+ sites contributed to the luminescence. Additionally, the PL intensity was significantly stronger when the Er doped GaP films were excited with below bandgap radiation. This dependence of the PL spectra on the excitation wavelength is shown in Fig. 23. This result was attributed to a competition between Er 3+excitation and non-radiative deep level recombination.
382
Wide Bandgap Semiconductors
m
w
,
: GoP:Er :
t-*i no5
9
l::,,O.
'
0.77
0.79
O.81
Emission energy (eV)
0.1~
Figure 22. High resolution photoluminescence spectrum of Er doped GaP measured at 12 K. The sample was prepared by MOVPE and doped with Er during growth.t3s]
4113/2--> 4115/2
.
1450
.
.
.
I
"
1500
"
I
I
I
.
1550
9
9
,
I
9
-
9
1600
Wavelength (nm) Figure 23. Photoluminescence spectra of Er doped GaP as a function of excitation wavelength. The sample was prepared by MOVPE and doped with Er during growth.[59]
Rare Earth Impurities 383 5.0
E L E C T R I C A L A C T I V A T I O N O F R E 3+ I O N S
The use of semiconductors as the host material offers the possibility of developing optical amplifiers and light emitting devices based on electrical excitation of the RE 3§ ions. Such a development would lead to a new class of compact optical sources for optical fiber communication and display systems. These sources could be designed to emit light at different wavelengths and to be integrated with electronic circuits. Most experiments concerning R E 3+ ions in semiconductor materials have centered on optical excitation. The few investigations of electrical excitation have dealt primarily with Er doping and have reported low optical emission efficiency. This is a major obstacle facing the development ofelectroluminescent (EL) devices based on RE doped semiconductor materials. A better understanding of the electrical excitation mechanisms could help to improve this situation. A schematic diagram of a typical semiconductor structure used for electrical excitation of the RE 3+ions is given in Fig. 24.[6~ Basically, it consists ofa p-i-n structure with the RE ions (Er in this illustration) incorporated in the undoped layer. In the forward bias condition, minority carriers are injected into the active layer and the RE ions are excited due to electron-hole pair recombination. As the electron-hole pairs recombine in the vicinity of the RE ion, a portion of the recombination energy is available for transfer to the ion. This is the standard configuration that has been used for the Er doped semiconductor light emitting diodes. When the p-i-n structure is operated in the reverse bias condition, it can also function as an EL device. Majority carriers (electrons) are then injected into the active region and the RE ions are directly excited by collisions with the high energy electrons. This process is known as impact excitation. Chang and Takahei reported on the electrical excitation of Er 3+ions in GaAs epilayers grown by MOCVD under both forward and reverse bias conditions.J61 ] When the Er-doped epilayer was excited using above bandgap radiation, a typical PL spectrum consisting of a set of emission lines near 1.54 ~tm was observed. After processing into a device structure and under forward bias conditions, the EL spectrum was very similar. There was little difference between the PL and EL spectra. In both cases, the Er 3+ ions were apparently excited by electron-hole pair recombination and subsequent energy transfer from the GaAs host. When the device was operated under reverse bias conditions, a different spectrum was observed. The different spectra indicate that the Er 3+ centers excited by electron-hole pair recombination may be different than those excited by direct impact of electrons.
384
Wide Bandgap Semiconductors
//
kk
::::::::::::::::::::::::ibium":i::::::::::::::::::::::::: Oxide
"////////////A
Oxide
l,+
|l
i
Figure 24. Schematic diagram of a typical semiconductor structure for electrical excitation of Er 3+ ions which are incorporated in a p-i-n structure.[ 6~
Torvik et al. were the first to demonstrate room temperature operation of a LED emitting at 1.54 I~m based on III-V nitride semiconductors.[62]The device consisted ofa metal/i-GaN/n-GaN (m-i-n) structure. The GaN layers were grown by CVD on an R-plane sapphire substrate. The active region of the device, the i-GaN layer, was co-implanted with Er § and O § ions and annealed (as in their prior studies).[ 491 The topography of the i-GaN top surface consisted of sharp ridges and valleys due to growth conditions.This surface topography permitted high field injection of electrons from the metal, under reverse bias. Apparently, the excitation of Er 3+ centers was by impact of energetic electrons. Typical current-voltage characteristics of the device are shown in Fig. 25. When the device was operated under forward bias, no Er 3+ related luminescence was observed. PL emission from the Er doped layer was obtained using a diode laser at 980 nm. At room temperature, the EL intensity was about 3 times stronger than the PL (see Fig. 26). They estimated that the electrical excitation cross section for Er in GaN was ~ 5x 1016cm-2 which is about 105 greater than the optical excitation cross section for 4115/2 - ~ 4I ~1/2transitions. [63]T h e external quantum efficiency, measured at 126 V and 318 A, was estimated to be ~ 10-7.
Rare Earth Impurities 385
50
L /
-50
~" .1oo ~L v
= -150
/
= -200 U
-250
/
I
Room Temperature
I
~~176 I -350 -150
-100
-50
I
0 50 Voltage (V)
100
150
F i g u r e 25. Current-voltage data, measured at room temperature, of an Er-doped GaN m-i-n light emitting diode.[ 62]
PL (135 mW at 980 nm)
1.5
3X Room Temperature
C
EL (318 I~A)
_J IX
0.5 ,0x
0 ', ": : : : : : , 850
: ' l " I-'-7",. 1050 1250 1450 Wavelength (nm)
: :,"~ 1650
F i g u r e 26. Electroluminescence and photoluminescence spectra, measured at room temperature, o f an Er-doped GaN m-i-n light emitting diode under reverse bias operation.t 62]
386
Wide Bandgap Semiconductors
Yoganathan et al. produced an EL device, based on 6H-SiC material, that showed room temperature operation under forward bias conditions.[ 64] The device consisted of an n-type SiC epilayer grown on a polished slice of a 6H-SiC boule. The epilayer was Er-implanted at four different energies to obtain a fiat Er dopant profile. After annealing the samples at 1700~ a ptype epilayer of 6H-SiC was grown to form an Er-doped p-n junction. The PL and EL spectra near 1.54 lam were measured at temperatures from 2-520 K. As shown in Fig. 27, the PL and EL spectra were nearly identical, indicating that the same Er 3§ centers were responsible for both emissions. There also appeared to be a correlation between the emission intensity and the shallow nitrogen donor concentration.
Wavelength (A) 16200 15800 15400 15000 :
'H' , ' ' '20, A ' resolution '~11~ ' i ' "
-~ u}9
=
6
.
SIC:Er3+ (p/n/n +)A
L
l
~
I
I
l
I
'
I
il
I
I
I
"qlll
6HSIC2E9r53K(pIn+)/n ..1
~
om
EL
0.775 0.800 0.825 Photon Energy (eV).---
Figure 27. Electroluminescenceand photoluminescencespectra, measuredat room temperature, of an Er-doped 6H-SiC p-n light emittingdiode under forward bias operation.t64]
Wang & Wessels fabricated an Er-doped light emitting diode based on GaP material. [65] They observed strong room temperature EL emission at 1.54 ~tm when the p-n diodes were operated under forward bias. The EL intensity was only weakly temperature dependent over a temperature range of 20-300 K. Ford and Wessels continued this work and examined the saturation of the EL intensity as a function of applied current. [66]The Er doped films
Rare Earth Impurities 387 were prepared by atmospheric pressure MOVPE and doped during growth by sublimating Er(thd)3 and transporting the vapor to the reaction zone with H 2 gas. The films were co-doped with Se to increase the net carrier concentration to-~ 1018 cm -3. The p-n junctions were made by diffusing Zn from a ZnP source into the epilayer. Figure 28 shows the room temperature EL spectra of one of the Er-doped diodes at three different current densities. At low current densities, the EL intensity is linearly dependent upon input power. However, at current densities of~ 10 A/cm 2, the EL intensity begins to saturate. While the precise cause is not determined, it may be due to a saturation of the optically active Er 3§ centers.
4oo
Room Temperature Er:GaP diode A
350'
I
1:If = 19 A/cm 2 2: If = 9 A/cm2 3: If = 5 A/cm 2
300 250g)
200"
150" 10C 50
ot 0.7
"'
'
'
I
. . . .
0.75
'I ~ '
0.8
"-~--i
"' '
0.85
'
'
0.9
Emission Energy (eV)
Figure 28. Electroluminescence spectra, measured at three different current densities, of an Er-doped GaP light emitting diode under forward bias operation.t661
In addition to the above efforts, strong EL emission from Er 3+ centers has been reported in a variety of low gap semiconductors. However, with the lower gap semiconductors, such as Si and GaAs, thermal quenching of the luminescence was a serious problem. Use of impurity elements such as O, F, C, and N alleviated this problem and room temperature Er doped Si LEDs have been demonstrated.[67][68][69]Priolo et al. measured the room temperature EL emission from Er doped Si diodes under both forward and reverse bias operation.[ 7~ The EL intensity under reverse bias was nearly 12 times
388
Wide Bandgap Semiconductors
higher than that under forward bias. Apparently, the excitation mechanisms were very different in the two cases. Under forward bias, electronhole recombination seemed to dominate. Under reverse bias, impact excitation was the main excitation process. The discovery of visible emission from higher excited 4f levels of RE ions incorporated in GaN was reported by Steckl and Birkhahn.[ 71] They achieved in-situ Er incorporation during GaN growth by MBE on sapphire substrates. Under UV laser excitation, the GaN:Er films exhibited intense green PL from the 2H 11/2 and 4S3/2 states. EL devices were fabricated on the GaN/Si structures using indium-tin (90%-10%) oxide (ITO) rectifying contacts. Visible EL emission at the wavelengths of 537 and 558 nm was measured. Near-IR EL emission was also observed with three well-defined peaks between 1.51 and 1.56 ~tm. Visible emission from RE doped GaN films may lead to novel optoelectronics applications.[ 72] As these results show, more research is required to better understand the excitation mechanisms for Er 3§ ions in semiconductor materials and to develop more efficient EL devices. However, many different experiments indicate that wide gap semiconductors offer a distinct advantage for reducing the thermal quenching of both PL and EL emissions from the RE 3§ centers.
6.0
SUMMARY
Substantial progress has been made in the past ten years in understanding the optical properties of RE-doped semiconductors. Luminescence of R E 3+ ions in many different semiconductors has been observed. Experiments show that the use of wide gap semiconductors reduces the thermal quenching of the luminescence. However, use of wide gap semiconductors alone may not offer the complete solution. The location of the RE ion in the crystal and its local chemical environment appear to be extremely important in determining the optical properties of the R E 3+ centers. Difficulties remain concerning the incorporation of RE atoms in wide gap semiconductor materials and the proper processing conditions necessary for optical activation. Prototype EL devices have been fabricated based on both low gap semiconductors and wide gap semiconductors. In general, the emission efficiencies need to be greatly improved. As with the PL experiments, the use of wide gap semiconductors leads to less thermal quenching of the electroluminescence. However, O doped Si does lead to state-of-the-art EL devices. Only a few EL devices based on wide gap semiconductors have been fabricated
Rare Earth Impurities
389
and tested. Further research to optimize the design of EL devices, based on both Si and wide gap semiconductor materials, is clearly needed. Due to the intense current research in SiC and III-V nitride semiconductors, improvements in the crystal quality and in the processing technology of these materials are very likely to occur. Such advances will assist efforts to develop EL devices based on wide gap semiconductors doped with RE ions. If the RE 3+related luminescence can be controlled and the efficiency of EL devices increased, then a new class ofoptoelectronic components for optical communications and display systems will be possible.
ACKNOWLEDGMENT The author expresses his appreciation to U. H6mmerich for valuable discussions in preparation of this chapter.
REFERENCES 1. Sze, S. M., Physics of Semiconductor Devices, John Wiley & Sons (1981) 2. Canham, L. T., Appl. Phys. Lett., 57:1046-1048 (1990) 3. Bhattacharya, P., Semiconductor Optoelectronic Devices, Prentice-HaU (1994). 4. Saito, S., Imai, T., and Ito, T., J. Lightwave Technol., 9:161 (1991) 5. Walker, G. R., Walker, N. G., Steele, R. C., Creaner, M. J., and Brain, M. C., d. Lightwave Technol., 9:182 (1991) 6. Mollenauer, L. F., Evangelides, S. G., and Haus, H. A., J. Lightwave Technol., 9:194 (1991) 7. Poole, S. B., Payne, D. N., Meats, R. J., Ferrmann, M. E., and Lanning, R. I., J. Lightwave Technol., 4:870 (1986) 8. Desurvire, E., Simpson, J. R., and Becket, P. C., Opt. Lett., 12:888 (1987) 9. Rare Earth Doped Semiconductors I," Materials Research Society Proc, Vol. 301, (G. S. Pomerenke, P. B. Klein, and D. W. Langer, eds.) (1993) 10. Rare Earth Doped Semiconductors II," Materials Research Society Proc., Vol. 422, (S. Coffa, A. Polman, and R. N. Schwartz, eds.) (1996) 11. Hiifner, S., Optical Spectra of Transparent Rare Earth Compounds, Academic Press (1978) 12. Koechner, W., Solid State Laser Engineering, Springer Verlag, 3rd Ed. (1992)
390
Wide Bandgap Semiconductors
13. Ermen, H., Schneider, J., Pomrenke, G., and Axmarm,A., Appl. Phys. Lett., 43(10):943-945 (1983) 14. Zavada, J. M. and Zhang, D., Solid-St. Electron., 38:1285 (1995) 15. Schetzina, J. F., Mat. Res. Soc. Syrup. Proc., 395:123-134 (1996) 16. See MRS Bulletin, 22(3) (1997) for an overview of SiC materials and devices. 17. Haase, M. A., Qiu, J., Depuydt, J. M., and Cheng, H., Appl. Phys. Lett. 59:1273-1275 (1991) 18. Eason, D. B., Yu, Z., Hughes, W. C., Roland, W. H., C. Boney, C., Cook, J. W., Jr. and Schetzina, J. F., Appl. Phys. Lett., 66:115-117 (1995) 19. Nakamura, S., Mukai, T., and Senoh, M., Appl. Phys. Lett., 64:1687-1670 (1994) 20. Nakamura, S., Senoh, M., Nagahama, S., Iwasa. N., Yamada, T., Matsushita, T., Sugimoto, Y., and Kiyoku, H., Appl. Phys. Lett., 70:868-870 (1994) 21. See MRS Bulletin, Vol. 22, No. 2 (1997) for an overview of III-V nitride materials and devices. 22. Casey, H. C., Jr., and Trumbore, F. A., Mater. Sci. and Eng., 6:69-109 (1970) 23. Favennec, P. N., L'Haridon, H., Salvi, M., Moutonnet, D., and GuiUou, Y. L., Electron. Lett., 25(11):718-719 (1989) 24. Neuhalfen, A. J., and Wessels, B. W., Appl. Phys. Lett., 60(21):2657-2659 (1992) 25. Taguchi,A., Takahei, K., andHoffkoshi, Y., J. Appl. Phys., 76(11):7288-7295 (1994) 26. Favennec, P. N., L'Haridon, H., Moutonnet, D., Salvi, M., and M. Gauneau, Jpn. J.Appl. Phys., 29(4):L524-526 (1990) 27. Michel, J., Benton, J. L., Ferrante, R. F., Jacobson, D. C., Eaglesham, D. J., Fitzgerald, E. A., Xie, Y., Poate, J. M., and Kimerling, L. C., J. Appl. Phys., 70(5):2672-2678 (1991) 28. Colon, J. E., Elsaesser, D. W., Yeo, Y. K., Hengehold, R. L., and Pomrenke, G. S., Mat. Res. Soc. Symp. Proc. 301:169-174 (1993) 29. Pearton, S. J., Mat. Sci. Rep., 4:313 (1990) 3 0. Wilson, R. G., Schwartz, R. N., Abemathy, C. R., Pearton, S. J., Newman, N., Rubin, M., Fu, T., and Zavada, J. M., Appl. Phys. Lett., 65(8):992-994 (1994) 31. Wilson, R. G., Stevie, F. A., and Magee, C. M., Secondary Ion Mass Spectrometry: A Practical Guide for Depth Profiling and Bulk Impurity Analysis, Wiley (1989) 32. See chapter "SIMS Analysis of Wide Bandgap Semiconductors" in this volume.
Rare Earth Impurities 391 33. Smith, R. S., Muller, H. D., Ennen, H., Wennekers, P., and Mailer, M., Appl. Phys. Lett., 50:49-51 (1987) 34. Nakagome, H., Takahei, K., and Homma, Y., J. Cryst. Growth, 85:345 (1987) 3 5. Charasse, M. N., Galtier, P., Huber, A. M., Grattepain, C., Chazelas, J., and Hirtz, J. P., Electron. Lett., 24:1458 (1988) 36. Zhang, T., Sun, J., Edwards, N. V., Moxey, D. E., Kolbas, R. M., and Caldwell, P. J., Mat. Res. Soc. Symp. Proc. 301:257-262 (1993) 37. Nakata, J., Taniguchi, M., and Takahei, K., Appl. Phys. Lett., 61:2665-2667 (1992) 38. Wang, X. Z., and Wessels, B. W.,Appl. Phys. Lett., 64(12): 1537-1539 (1994) 39. MacKenzie, J. D., Abemathy, C. R., Pearton, S. J., H6mmerich, U., Wu, X., Wilson, R. G., Zavada, J. M., and Schwartz, R. N., Appl. Phys. Lett., 69(14):2083-2085 (1996) 40. Zavada, J. M., Wilson, R. G., Schwartz, R. N., MacKenzie, J. D., Abemathy, C. R., Pearton, S. J., Wu, X., and H6mmerich, U., Mat. Res. Soc. Symp. Proc., 422:193-197 (1996) 41. Horiguchi, H., Kinone, T., Saito, R., Kimura, T., and Ikoma, T., Mat. Res. Soc. Syrup. Proc., 422:81-86 (1996) 42. Baumann, I., Beckers, L., Buchal, C., Brinkrnan, R., Dinand, M., Gog, T., Holzbrecher, H., Fleuster, M., Materlik, M., Muller, K. H., Paulus, H., Sohler, W., Stolz, H., vonder Osten, W., and Witte, O., Appl. Phys. A, pp. 33-44 (1997) 43. Kozanecki, A., Chan, M., Jeynes, C., Sealy, B., and Homewood, K., Solid State Commun., 78(8):763-766 (1991) 44. Tang, Y. S., Zhang, J., Heasman, K. C., and Sealy, B. J., SolidState Commun., 72(10):991-993 (1989) 45. Witte,O., Stolz,H., andvonderOsten, W.,J. Phys. D, Appl. Phys., 29:561-568 (1996) 46. Kim, S., Rhee, S. J., Tumbull, D. A., Reuter, E. E., Li, X., Coleman, J. J., and Bishop, S. G., Appl. Phys. Lett., 71:231-233 (1997) 47. Wu, X., H6mmerich, U., MacKenzie, J. D., Abemathy, C. R., Pearton, S. J., Schwartz, R. N., Wilson, R. G., and Zavada, J. M., Appl. Phys. Lett., 70:2126-2128(1997). 48. Klein, P. B., and Pomrenke, G. S.,Electron. Lett., 24(24): 1502-1503 (1988) 49. Torvik, J. T., Feuerstein, R. J., Qiu, C. H., Leksano, M. W., Namavar, F., and Pankove, J. I., Mat. Res. Soc. Symp. Proc., 422:199-204 (1996) 50. Silkowski, E., Yeo, Y. K., Hengehold, R. L., Goldenberg, B., and Pomerenke, G. S., Mat. Res. Soc. Symp. Proc. 422:69-74 (1996) 51. Pearton, S. J., Abemathy, C. R., MacKenzie, J. D., Schwartz, R. N., Wilson, R. G., J. M. Zavada, and R. J. Shul, Mat. Res. Soc. Symp. Proc., 422:47-56 (1996).
392
Wide Bandgap Semiconductors
52. Wu, X., H6mmerich, U., MacKenzie, J. D., Abemathy, C. R., Pearton, S. J., Wilson, 1L G., Schwartz, R. N., and Zavada, J. M., J. Lumin. 72-74:284-287 (1997) 53. MacKenzie, J. D., unpublished. 54. Choyke, W. J., Devaty, R. P., Clemen, L. I., Yoganathan, M., Pensl, G., and Hassler, C., Appl. Phys., 65(13):1668-1671 (1994) 55. Wu, X., H6mmerich, U., Namavar, F., and Cremins-Costa, A. M., Appl. Phys., 69(13):1903-1905 (1996) 56. Wu, X., White, R., H6mmerich, U., Namavar, F., and Cremins-Costa, A. M., J. Lumin., 71:13-20 (1997) 57. Kimura, T., Yokoi, A., Horoguchi, H., Saito, R., and Ikoma, T., Appl. Phys., 65:983-985 (1994) 58. Kimura, T., Hosokawa, I., Nishida, Y., Dejima, T., Saito, R., and Ikoma, T., Mat. Res. Soc. Symp. Proc., 422:149-154 (1996) 59. Culp, T. D., Wang, X. Z., Keuch, T. F., Wessels, B. W., and L. Bray, K., Mat. Res. Soc. Symp. Proc., 422:279-284 (1996) 60. Ren, F. Y. G., Michel, J., Sun-Paduano, Q., Zheng, B., Kitagawa, H., Jacobson, D. C., Poate, J. M., and Kimerling, L. C., Mat. Res. Soc. Symp. Proc., 301:87-95 (1993) 61. Chang, S. J., and Takahei, K., Appl. Phys. Lett., 65:433-435 (1994) 62. Torvik, J. T., Feuerstein, R. J., Pankove, J. I., Qiu, C. H., and Namavar, F., Appl. Phys., 69(14):2098-2100 (1996) 63. Payne, S. A., Chase, L. L., Chase, L. K., Smith, L. K., Kway, W. L., and Krupke, W. F., IEEE J. Quantum Electron., 28:2619 (1992) 64. Yoganathan, M., Choyke, W. J., Devaty, R. P., Pensl, G., and Edmond, J. A., Mat. Res. Soc. Syrup. Proc., 422:339-344 (1996). 65. Wang, X. Z., and Wessels, B. W., Appl. Phys. Lett., 65(5):584-586 (1994) 66. Ford, G. M., and Wessels, B. W., Mat. Res. Soc. Symp. Proc., 422:345-350 (1996) 67. Franzo, G., Priolo, F., Coffa, S., Polman, A., and Camera, A., Appl. Phys. Lett., 64:2235-2237 (1994) 68. Zheng, B., Michel, J., Ren, F. Y. G., Kimerling, L. C., Jacobson, D. C., and Poate, J. M., Appl. Phys. Lett., 64:2842-2844 (1994) 69. Lombardo, S., Campisano, S. U., van den Hoven, G. N., and Polman, A., J. Appl. Phys., 77:6504 (1995) 70. Priolo, F., Coffa, S., Franzo, G., and A. Polman, A., Mat. Res. Soc. Symp. Proc. 422:305-316 (1996) 71. Steckl, A. J., Birkhahn, R., Appl. Phys. Lett., 73:1702 (1998) 72. See MRS Bulletin, 24(9) (1999), for a description ofphotonic applications of rare earth doped materials.
9 SIMS Analysis of Wide Bandgap Semiconductors Robert G. Wilson
1.0
INTRODUCTION
Most semiconductors, including wide bandgap semiconductors, are used in the fabrication of devices for electronics or optics or optoelectronics, or electro-optics. Impurities may be intentionally introduced into the these materials to change their optical or electrical properties. Native or unintentional impurities may also affect the optical or electronic properties of these materials, and, in some cases, cause deleterious or uncontrolled effects. Secondary ion mass spectrometry (SIMS) is used to measure the presence, concentration, and depth distribution of these intentional or unintentional impurities, and changes in their depth distributions with processing. Wide bandgap materials are generally insulators; therefore, wide bandgap semiconductors may actually often be insulators. SIMS analysis of insulators poses a special issue of charging of the dynamic surface being sputtered by the incident primary ion beam. This subject is discussed in Ref. 1, but is summarized generally in the following. The charging issue is 393
394
Wide Bandgap Semiconductors
more serious for instruments with high secondary ion extraction voltage, or voltage placed on the target (sample for analysis), for example a CAMECA sector magnet instrument. Charging is less serious for low extraction voltage instruments such as quadrupole instruments. Both approaches use electron flooding of the target surface (directed beam or free electrons) to compensate the positive charge from the incident ion beam. In both approaches, the sputtering rate used to depth profile insulators is reduced from that used to profile semiconductors or conductors. In addition, a thin (30 nm) layer of a conductor may be deposited on the analysis surface, provided that it does not cause interfering masses for the desired analysis. Automatic variation of the potential placed on the target surface may be used with the CAMECA, to partially compensate for the change in potential caused by the incident ion beam (up to about 125 V). Insulating layers thinner than about 0.1 mm on a semiconducting substrate generally do not cause any difficulty using a CAMECA instrument. Analysis of semiconducting layers on bulk insulators like sapphire generally do cause difficulties using a CAMECA. In the material that follows: SIMS = secondary ion mass spectrometry; RSF = (SIMS) relative sensitivity factor, the calibration factor to convert secondary ion signal (counts/s) to atom density (cm-3), via the matrix signal measured under identical instrument conditions; wrt = with respect to; and E is often used to represent element, O for oxygen, M for a matrix element, etc., when secondary ions are discussed.
2.0
WIDE BANDGAP MATERIALS DISCUSSED HERE
A selected representative group of materials is discussed here, materials that are of interest today in the fields of electronics and optoelectronits. Those materials are diamond (5.4 eV), SiC (2.9 eV), ZnSe (2.7 eV), LiNbO 3, and the group III nitrides, e.g., A1N (6.1 eV) and GaN (3.4 eV).
3.0
S E C O N D A R Y I O N M A S S S P E C T R O M E T R Y (SIMS)
SIMS is a sensitive analytical technique (sputtering) that can detect all elements in all solid materials, with detection limits from 1013to 1016 cm -3,
SIMS Analysis 395 and to depths of many micrometers (which is compatible with the depths of dopants and the thicknesses of electronics and optoelectronics devices and other microelectronics structures). Depth profiles of intentional and unintentional impurities were measured in this work using both sector magnet CAMECA SIMS instruments (4fand 5f) and quadrupole instruments (PHI 6600) at Charles Evans and Associates. Oxygen primary ion bombardment was used to measure positive secondary ions, and cesium primary ion bombardment was used to measure negative secondary ions and, in limited cases, also Cs molecular ions. Additional details of the SIMS analysis techniques carried out in this work are described in Ref. 1. SIMS was used to measure changes in the depth distributions of elements grown in, implanted, or introduced unintentionally during processing steps.
4.0
SIMS ISSUES
Mixing of atoms in the dynamic surface being sputtered by the incident energetic ions and the equilibration/stabilization time/depth are issues that must always be addressed. Depth resolution in SIMS profiles is important for defining the structure of sharp interfaces and superlattices. Several factors affect depth resolution. Interface broadening is caused by surface topography and nonuniformity of layer thickness, and by ion mixing, which occurs within the penetration depth of the primary sputtering ions. If the layer thicknesses are uniform and the surface topography is good, then the SIMS experimental conditions become the determining factors. The mixing thickness decreases with decreasing primary ion energy and increasing angle of incidence (to a point). Reducing the primary ion energy is necessary to achieve the best depth resolution. Of the two most commonly employed primary ion species (oxygen and cesium), oxygen produces less ion mixing. The lowest practical oxygen ion bombardment energy is then used to produce the best depth resolution. For quadrupole instruments, this energy may be less than 1 keV. For CAMECA sector magnet instruments, this energy is often 1.5 keV/O (3 keV for 02). Another factor in achieving good depth resolution is sputtering rate combined with secondary ion collection time. The sputtering rate and the collection time must be adjusted to create enough data points to define a layer or interface accurately.
396
Wide Bandgap Semiconductors
Often, information is desired from SIMS profiling of unwanted impurity species (elements) in various materials. Some of the common impurity species are from the ambient vacuum or heated components of materials growth machines. The same ambient vacuum species exist in a SIMS instrument. The lower the ambient vacuum, the lower the sputtered secondary ion intensities of these species. The higher the sputtering rate during SIMS profiling, the lower the adsorbed density of these ambient species and the lower their sputtered secondary ion intensities. Thus, for the lowest backgrounds of these species, or the best detection limits, the lowest practical vacuum and the highest practical sputtering rate should be employed. Note that improved background and improved depth resolution cannot be achieved simultaneously because they vary in opposite dependence on sputtering rate. Often, separate profiles must be measured at very different sputtering rates to achieve depth profiles with good depth resolution, and with good detection limits or backgrounds for certain species (elements). These elements include H, C, N, O, Si (N 2 and CO can produce signals that interfere with Si), and all elements that may have a molecular interference when any of these elements are combined with the masses of the matrix materials (which may be dozens of masses in some cases). The sputtering rate in SIMS depth profiling of multilayer/ multimaterial structures is again an important issue. For fixed SIMS profiling conditions, the sputtering rates of all materials are different. Thus, the sputtering rate changes whenever an interface between different materials is crossed. The depth scale of a SIMS profile is usually obtained by measuring the crater depth at the end of each profile. This depth, divided by the sputtering time, yields the average sputtering rate. If this rate is applied uniformly to the profile, inaccurate layer thicknesses result if different materials are sputtered. To obtain accurate layer thickness, the sputtering rates of each and every material in the structure must be measured or otherwise known (from other work or published sputtering rates~for similar sputtering conditions). Then the total depth profile must be divided into layers of each different material and the appropriate sputtering rate applied to each layer. This capability is provided in the Charles Evans and Associates replot software used in this work, and by many other CAMECA instrument users.
SIMS Analysis 397 5.0
QUANTIFICATION
SIMS is a powerful analysis technique, but to be quantitative, a quantification system is required that relates the measured relative intensities of the secondary ions to absolute concentrations of the elements in the host matrix, as well as the accurate relative amounts of the matrix elements (host material) themselves. Different elements exhibit different relative secondary ion yields depending generally on their ionization potentials for positive secondary ions, and on their electron affinities for negative secondary ions. In the first case, sputtered atoms must lose an electron as they leave the sputtered surface (i.e., be ionized). In the latter case, they must take on an electron as they leave the sputtered surface, related to their electron affinity. Experimental work has shown that the secondary ion yields are not in all eases simply related to the ionization potential or the electron affinity of the elements; for some elements, there are other special features of the process that effect/alter the values of the secondary ion yields. All of this information is accounted for by determining the relative sensitivity factors (RSFs) for all of the elements in any or all host materials or matrices. This study has been largely completed and data for many materials have been compiled, including the materials discussed here. When RSFs have not been published for specific element-matrix combinations of interest, implanted standards can be prepared specifically for that purpose, or the systematics of RSFs in related materials can be used. Tabulations of SIMS RSFs have been published for diamond,[ 21but not for SiC, ZnSe, LiNbO 3, nor the III-nitrides, the materials chosen for discussion here. We have generated tables of SIMS RSFs for these materials using ion implanted standards, and they will be published in the future. For the purpose of this work, specific RSFs for the element-matrix combinations discussed here fi'om that work have been used, and some of them are listed here.
6.0
DIAMOND
Natural diamonds, synthetic diamonds, and CVD diamond films were characterized using SIMS, with a Cs beam and negative secondary ions for H, B, N, O, Si, and P, and an 0 2 beam and positive secondary ions for B, Fe, Mo, Ta, W, and Re. High mass resolution was used to profile N as 14N12Cbecause of the interference from 13C2-.B and P are intentional dopants for
398
Wide Bandgap Semiconductors
device applications. H, N, and O are unintentionally incorporated during growth from the ambient. Si, Fe, Mo, Ta, W, and Re may be unintentionally introduced from the growth apparatus or substrate. H can be electrically active in diamond films. Impurity densities were determined from ion implanted standards of the subject impurities. Natural IIa and IIb diamonds, HPHT synthetic diamonds, and diamond thin films grown using CVD techniques, undoped and doped with either B or P, obtained from a variety of sources, were analyzed using SIMS. Depth profiling was done using CAMECA sector magnet instruments and 8-keV 02 or 14.5-keV Cs primary beams for positive or negative secondary ions, respectively. The full 128-eV ban@ass was used except for high mass resolution measurements and measurements made using voltage offset. Bulk impurity mass surveys were made using an offset voltage to suppress interfering molecular ions. Quantification standards for all elements studied were implanted using a 400-keV, post acceleration mass separation custom ion mass spectrometer, in which the ion beam is mass separated at full energy using a double focusing magnet. The results of impurity analyses for six thin CVD diamond films are shown in Table 1, two doped with P (A and B), two undoped (C and D), one doped with B (E), one synthetic bulk crystal, and five natural crystals. The significance of these data depends partly on the detection limits (DL) for these elements in diamond, measured using SIMS. All of the CVD films and the synthetic crystal are seen to contain densities of H well above the DL, while natural diamonds 1, 2, and 3, have concentrations less than the DL. The last two natural crystals were selected because they were believed to have high H concentrations, which was verified by these analyses. Natural crystals 1 and 2 were selected because they were believed to have low concentrations of all impurities, including N, which was verified by these analyses as having less than the DL. Natural crystal 3, the synthetic crystal, and all six CVD films have significant N concentrations. The DL for O was high enough that no O concentrations greater than the DL could be detected in any of these diamond materials. Significant Si concentrations were measured in all six CVD films, but not in the synthetic nor natural crystals. The B-doped CVD films contained about 1 and 6xl 019 cm -3 B. The gray natural crystal was seen to contain significant concentrations of Na, A1, K, and Ca, and possibly other impurities. The yellow-brown natural crystal had fewer impurities. Fe was not detected in any of these diamond samples at densities greater than the DL of about 1 • 1014 cm -3 under routine analysis conditions.
Table 1. Impurities in Selected Commercial Diamond Films and Selected Natural Crystals (determined from SIMS analyses using the RSFs fi-omTable 2, determined from implanted standards) Synth.
Nat.
Nat.
Nat.
Nat.
Xtal
Xtal
Xtal
Xtal
Xtal
Xtal
Sumito
IA
IA
IIA
yellow
gray
8E20
9.7E18
4E18
Element
D square 0 2 SIMS Cs SIMS
D oval
D irreg.
OzSlMS
Cs SIMS
O z SIMS
Cs SIMS
2.7%
3.6%
450
3.6
H
150
2000
1900
4100
Li
0.001
310?
0.0037
0.83
B
0.09
N
90
2.8% 740
100
O
2.4
F Na
28
0.11
4.7
21
0.058
Si
(4.8)
CI
610
35 50
160
2.8%
3.5%
180
1300?
46
3
Ca
0.045
32
220
K
0.021
0.75
39
Cr
0.045
10
Fe
0.19
Ti
1.6
25 11 100
10
30
Co 0.038
100
70
42
7.5 2.3
Br Sr
1400
56
AI
Cu
500 110
11
Mg
Ni
780
0.007
Mo
90
Ba
0.18
0.022
SIMS Analysis 405
Acknowledgments. Some natural crystals were provided by Michael Seal of Drukker International, now of Sigilum. The synthetic crystal was a commercial Sumitomo diamond via Cal Tech. CVD films were provided by various commercial suppliers that will not be named to avoid any comparisons of developmental materials that may not be representative of what can be obtained presently.
7.0
SiC
Ion implantation is often used to dope SiC because diffusion temperatures are very high for SiC or diffusion does not work. SIMS can be used to measure the resulting depth distributions in bulk SiC or in SiC epilayers grown on various substrates, especially SiC of the opposite conduction type. Implantation has been studied in SiC for several applications, a few of which are noted below. One application is surface hardness; B and N[ 3] and Cr [4] have been studied. Doping and annealing is another application. See Ref. 5 for a study of Ga. B and A1 were studied,[ 6] and are the commonly used p dopants in device fabrication. N, P, Sb, and Bi were studied some years ago.[ 7] N is the most commonly used n dopant in device fabrication. Elevated temperature N implants in SiC are reported in Ref. 8. Er may be implanted into SiC to produce stimulated photoemission at 1.54 ~tm.[9] SIMS is the standard technique used to measure the depth distributions of these dopants in SiC. The implantation of B, A1, and Ga as potential p-type dopants, N as the primary n-type dopant, F as a possible isolation species, and Ti to quantify Ti that might diffuse into the SiC epilayer grown on TiC substrates have been studied. Only B and N are small enough to fit into the SiC lattice as substitutional dopants. The interface between a SiC epilayer and the TiC substrate measured using SIMS is illustrated in Fig. 5. In this case, the SiC layer was doped with B at about 7 x 10s8 em-2. The depth profile of A1 implanted in bulk cxSiC is shown in Fig. 6, where high mass resolution SIMS was used to determine carefully whether any I1B160 was present at the mass of 27A1. The depth profile for A1 implanted into a [3SIC epilayer grown on TiC is shown in Fig. 7. Depth profiles comparing B and A1 implanted in a [3SIC epi layer on TiC are shown in Fig. 8. A depth profile for N implanted into a [3SIC epilayer at 4 x 1017 c m -2 is shown in Fig. 9 for the measurement of negative secondary ions. The effect of the formation of SiN or CN at the
406
Wide Bandgap Semiconductors
peak of the distribution is also seen in Fig. 9 via the change in Si and C matrix signal intensities (change in sputtering rate and/or ion yield~the matrix effect). Corresponding results for the measurement of the same sample shown in Fig. 9 but using positive secondary ions and very high mass resolution are shown in Fig. 10. Depth profiles for Ga implanted at three fluences are shown in Fig. 11. A profile for Ti implanted into BSiC at a high fluence to be comparable with Ti atom densities that might diffuse during growth into the SiC epilayer from the TiC substrate measured using various positive secondary ions is shown in Fig. 12.
952s-2e-o43
1021 F
J
i
i
1020~A 1019k . . . . . . . . . .
~ .......
,..,
~: 1018-
' I
! /~
106
---~"/J~" -
105 0
2_ 104 0Z
f
(n
Z MJ
0 9 1017~"-
- 107
~
-
""
>" n-
-
[
- -=1 0 3 < Z 0
~
-
o
-
< 1016
0
/
102 (n'~'
SOn l 1015 1014 0
101
1
2 3 DEPTH (~m)
4
5
100
Figure 5. Depth profile of a SiC/TiC interface.
To summarize the application of SIMS to SiC: N and Ti must be measured using high mass resolution in SiC because of serious interference; the measurement of B, A1, Ga, and F is straight-forward. Important numbers for the use of SIMS analysis in SiC are the SIMS relative sensitivity factors, which have been determined and are given in Table 4.
SIMS Analysis 407
I 1020 ~ Rm = 0.250 lJm
I
i
10 TM
-~,10 TM o
~ 1017 zw t: lO16
=o F
i
os,
1015 ~-- ,,.,..'~. BO + 1014/ 0
I 0.5
~, I 1.0 DEPTH
(gin)
t 1.5
2.0
Figure 6. High mass resolution depth profile of AI implanted in SiC.
a525-26.o42
1020 I
I AI IONS I 200 keV AI 3.0 x 1014 cm-2 I~ SiC EPI SUBSTRATE _ RANDOM ORIENTATION Si (ARB UNITS) ROOM TEMPERATURE
10TM
V
A
?
E
~
IMPL4NT
NO ANNEAL
1018 SIMS: OXYGEN PRIMARIES POSITIVE SECONDARIES _
z ca :i lO17 SIC
TIC
AI_
10TM
lO15
I 0
0.5
1.0 DEPTH (l~m)
Figure 7. Depth profile of AI implanted in SiC.
1.5
2.0
408
Wide Bandgap Semiconductors
g525-~
10TM
I
I A I ~ 10TM E o
I
I
I
B IONS 100 keY AI IONS 200 keY 2.0 x 1013 cm-2 p SIC EPI SUBSTRATE RANDOM ORIENTATION ROOM TEMPERATURE IMPLANT NO ANNEAL SIMS: OXYGEN PRIMARIES DARIES
(n z 1017 w O
_
I
m
10TM
10TM
l
0
I
0.2
I',,
~
0.4 0.6 DEPTH (pm)
I
0.8
1.0
Figure 8. Depth profiles of B and AI implanted in SiC.
1023
~,
o,
X O
or~o-,e~
CI~ ~ N IONS C( e f ' / ~ 150 keV it',ON / N ~ 4.0xl0'Tcm "2 1022 ,4 Si2N t I~SIC EPI SUBSTRATE l/ ~ RANDOM ORIENTATION // I ROOM TEMPERATURE IMPLANT // ~ NO ANNEAL f/ | SIMS: ~/ ~ OXYGEN PRIMARIES 1021 ,,""~, t POSITIVE SECONDARIES ,,'"f ~ 1 HIGH MASS RESOLUTION
GaN
1017
r~
10 TM
101S ,
0
,
:
,
2000
4000
6000
9
,
8000
D E P T H (.~) Figure 3. Hydrogen ion profiles in SiC resulting from multiple energy proton implantation.
'~
1010 t
O
101s
4xlOlZcm"2,lOkeV 5xlO:Zcm"2,25keV 6xlOlZcm"2,$OkeV 8xlOt2cm"2,80keY lOxlOtZcm"2, lOOkeV sum --->SIC ~ .I.
101r O
~
101e
~101s ()
TO00 "10'00"2000"30'0040'00"d00"60'00"7C DEPTH 1.~1
Figure 4. Hydrogen ion profiles in GaN resulting from multiple energy proton implantation.
436
Wide Bandgap Semiconductors
2.0x10 s
A
800~ temperatures. The activation energies for dopant reactivation are of the order of 2.5 eV in bulk samples, where hydrogen retrapping is expected to be significant. Hydrogen plays an important role in semiconductors, passivating the electrical activity of shallow and deep level impurities.[ 81] Unintentional incorporation of hydrogen can therefore lead to uncontrolled variations in conductivity of a semiconductor, and this effect has been observed for virtually all group IV, III-V, and II-VI materials. The properties of hydrogen in Group III nitrides and their alloys are recently attracting attention
446
Wide Bandgap Semiconductors
because of the discovery that post-growth annealing is required to attain high conductivity.[ 82] This problem will be especially evident in materials grown using NH 3 as the nitrogen precursor. Past studies in other III-V semiconductors have shown that metalorganics such as trimethylgallium [Ga(CH3)3] can also produce significant hydrogen incorporation.[ 83] It is also well established in conventional III-V materials, that hydrogen can be incorporated during virtually any processing step, since most of the gases and liquids that are used contain at least some hydrogen.[ 84][85] Combined with the high diffusivity of atomic hydrogen in semiconductors, it is then quite common for hydrogen to disrupt the electrical properties of nearsurface regions. For photonic devices such as light-emitting diodes or heterostructure diode lasers, the effect of hydrogen in ternary and quaternary nitrides will also need to be understood, because these will comprise the active or cladding layers. Metal organic Chemical Vapor Deposition (MOCVD) of GaN occurs by the reaction: Eq. (4)
(CH3)Ga + NH 3 --->GaN + CH 4 1' + H 2 1"
While the growth temperature is generally around 1040~ there is a long cool-down under a NH 3 ambient during which atomic hydrogen can easily diffuse into the GaN.[ 81]Undoped GaN is generally n-type due to the presence of native defects or possible impurities, and thus, since this is relatively uncontrolled, the effects of any hydrogen passivation will not normally be noticed. When the GaN is intentionally doped with Mg from the Cp2Mg (cyclopentadienyl magnesium) source, it is invariably found that the material must be post-annealed at >700~ in order to dissociate neutral hydrogen-magnesium pairs that prevent achievement of p-type conductivity:[ 82] Eq. (5)
( M g - H) ~ ~ M g + H +
Similar behavior is observed to a greater or lesser extent in virtually all other semiconductors grown by gas-phase techniques.j86]-[ 1~ However, the high growth temperature (leading to a longer cool-down time than more conventional materials) and the high thermal stability of hydrogen bonding in GaN means that unintentional passivation is likely to be significant. There are many opporttmities for atomic hydrogen to diffuse into semiconductors during processing, since essentially all of the chemicals
Hydrogen in Wide Bandgap Semiconductors 447 that come in contact with the surface contain hydrogen. [102]In Si and III-V semiconductors, such as GaAs and InP, it is well established that hydrogen is among the fastest diffusing impurity,J1~ 1~ and that its solubility at temperatures _1100~ has been realized.[ 14~ While the activation efficiency of Ca in both implant schemes was ~100%, temperature-dependent Hall measurements showed that the ionization level of Ca was--168 meV, similar to that of Mg. The Ca atomic profile was thermally stable to temperatures up to 1125~ Since Mg has a substantial memory effect in stainless steel epitaxial reactors (or in gas lines leading to quartz chamber systems), Ca may be a useful alternative p-dopant for an epitaxial grown laser diode or heterojunction bipolar transistor structures in which junction placement, and hence control of dopant profiles, is of critical importance. In considering Ca-doped GaN for device applications, it is also necessary to understand the role of hydrogen. There is always a ready supply of atomic hydrogen available from NH3, the metalorganic group III source [typically (CH3)3Ga ] or from the gaseous dopant source when using chemical vapor deposition techniques. We have also found that Ca acceptors in GaN are also readily passivated by atomic hydrogen at a low temperature (250~ but they can be reactivated by thermal annealing at < 500~ for 1 min in lightly-doped (3 x 1017cm -3) materials. As the carrier density is restored by such annealing treatments, there is a corresponding decrease in hole mobility, indicating that there is a true passivation and not just compensation of the Ca acceptors by the hydrogen. Nominally undoped (n 400~ hydrogen is dissociated from the complex, leading to reactivation of the dopant. The hydrogen does not leave the crystal at this temperature but probably
Hydrogen in Wide Bandgap Semiconductors 467 associates with other hydrogen atoms to form molecules and larger clusters. At much higher temperatures (>800~ in GaN), these clusters are evolved from the sample.
(b)
(a)
:
S'
o
e
o
o
o
' Si
'
0 0
(DH
0 Q 0
Mg.
Figure 26. Schematicrepresentationof hydrogen-dopantcomplexesin GaN.
The ternary samples were exposed to an ECR D 2 plasma for 30 min at 250~ Figure 27 shows the fraction of passivated donors remaining in both InA1N and InA1GaN as a function of post-hydrogenation annealing temperature. Both samples displayed a decrease in carrier concentration of approximately an order of magnitude after hydrogen plasma exposure, consistent with previous reports.J15~ On subsequent annealing, the passivated donors begin to reactivate around 400~ and by 500~ 78% of the
468
Wide Bandgap Semiconductors
lost carriers were restored in InA1N and 66% in the InAIGaN. The reactivation o f the donors was fit to the relation:[ 81]
No = l - exp - tv exp - ~ -
Eq. (11)
N
for time t, v is the attempt frequency (assumed to be 1014S"1) and E d is the activation energy for reactivation. The recovery of the donor activity occurred over a slightly broader temperature range than generally observed for other passivated dopants in binary semiconductors, and is consistent with the presence ofa Gaussian distribution of activation energies.[ 811This may be due to nitrogen vacancies with different numbers of specific group III neighbors surrounding them (i.e., 2 In and 2 A1, against 1 In and 3 A1). Assuming a Gaussian distribution of activation energies, values for E d around 2.4 eV were obtained, with a full width at half m a x i m u m of--0.3 eV. The fact that reactivation of these donors occurs around 500~ means that the apparent thermal stability is similar to that ofpassivated M g acceptors in GaN.
I
r,~
o z o
,. I
I
I
I
1.00
0.75 < ;>