Mechanism for Ohmic contact formation on Si3N4 passivated AlGaN

APPLIED PHYSICS LETTERS 89, 201908 共2006兲

Mechanism for Ohmic contact formation on Si3N4 passivated AlGaN / GaN high-electron-mobility transistors B. Van Daelea兲 and G. Van Tendeloo EMAT, University of Antwerp, Groenenborgerlaan 171, 2020 Antwerpen, Belgium

J. Derluyn, P. Shrivastava, A. Lorenz, M. R. Leys, and M. Germain IMEC, Kapeldreef 75, 3001 Leuven, Belgium

共Received 21 August 2006; accepted 4 October 2006; published online 15 November 2006兲 Recent experiments have shown that in situ passivation by Si3N4 of AlGaN / GaN high-electron-mobility transistors results in improved electrical characteristics. Transmission electron microscopy techniques have been applied to study the metal contact formation on top of passivated AlGaN / GaN structures. Contrary to unpassivated AlGaN / GaN, the AlGaN top layer is not consumed by a typical Au/ Mo/ Al/ Ti metal stack. Instead, a thin AlN interface layer is formed, being the key factor in the Ohmic contact formation. The formation of this AlN is believed to be due to extraction of N atoms out of the AlGaN. The resulting N vacancies, electrical donors, create a conducting channel through the AlGaN. © 2006 American Institute of Physics. 关DOI: 10.1063/1.2388889兴 Although the formation of Ohmic contacts on GaN is well understood, little is known about the Ohmic contact formation on AlGaN / GaN high-electron-mobility transistors 共HEMTs兲. The Ohmic contact of an Al/ Ti-based metal stack on GaN is achieved by the extraction of N atoms out of the GaN by Ti,1,2 forming a thin TiN interface layer. As N vacancies in GaN are electrically active donors,3,4 a tunnel contact is created. Previously we have shown,5 by studying Ti contacts on AlGaN / GaN, that the reaction of Ti with AlGaN is not able to create the electrically active N vacancies in AlGaN / GaN. Al, present in AlGaN, is able to slow down the reaction of AlGaN with Ti resulting in the fact that the N atoms stay in the III-nitride. The observed reaction is the transformation of Ti/ AlGaN into Ti1−xGax / Al+ Ti+ N by exchanging Ti and Ga. This process evidently cannot lead to a good contact since the two-dimensional electron gas vanishes as soon the AlGaN transforms to Al+ Ti+ N. Recent results have shown that AlGaN / GaN HEMT properties, both structurally and electrically, are greatly improved by the introduction of an in situ Si3N4 passivation layer.6 Moreover, leaving a Si3N4 passivation layer under the metal contacts results in lower contact resistivities than in the test cases where the passivation layer had been removed or was not deposited at all. The above mentioned experiments prove that the introduction of the Si3N4 passivation layer is improving AlGaN / GaN HEMT performance. The reason why a passivation layer is necessary to improve the two dimensional electron gas properties is well known.7 However, the role of the Si3N4 passivation layer on the formation of improved Ohmic contacts has not yet been discussed. In this letter, the role of a Si3N4 passivation layer on the Ohmic contact formation on AlGaN / GaN HEMTs will be presented and a model will be described. AlGaN / GaN HEMT structures have been grown on c-plane sapphire in a Thomas Swan close-coupled showera兲

Present address: IMEC, Kapeldreef 75, 3001 Leuven, Belgium.

head reactor. After growth of the AlGaN / GaN active area, a 3.5 nm thick in situ Si3N4 passivation layer has been deposited. Further growth details are described in 共Ref. 6兲. Au/ Mo/ Al/ Ti 共50 nm/ 25 nm/ 40 nm/ 20 nm兲 metal layers were deposited on top of the Si3N4. The samples were subsequently thermally annealed using two different recipes. The samples of the first series 共“A”兲 were subjected to a rapid thermal annealing 共RTA兲 step of 1 min in nitrogen atmosphere, at temperatures of 775, 800, and 825 ° C. For the samples of the other series 共“B”兲, the same RTA step was preceded by a 15 min preannealing step at 550 ° C in forming gas atmosphere to deoxidize the furnace. The contact and sheet resistances have been determined using the transmission line method. A Schottky to Ohmic transition has been observed when the contact structures were annealed at 800 ° C or higher 共following both A and B procedures兲. The contact resistances of these specific samples annealed at 800 ° C were 2.1 and 0.91 ⍀ mm for series A and B, respectively. Transmission and scanning transmission electron microscopy 共TEM+ STEM兲 techniques have been performed to study the microstructure structurally and chemically. Additionally, a series of samples with simplified metal stacks 共Al/ Ti兲 was processed, annealed under the conditions of series B.

FIG. 1. HAADF image of the complete metal contact layer 共sample: series A, annealed at 800 ° C兲. Mo3Al grains 共gray兲 are formed in Au4Al and TiN can be retrieved near the AlGaN interface.

0003-6951/2006/89共20兲/201908/3/$23.00 89, 201908-1 © 2006 American Institute of Physics Downloaded 16 Nov 2006 to 146.103.254.11. Redistribution subject to AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

201908-2

Van Daele et al.

Appl. Phys. Lett. 89, 201908 共2006兲

FIG. 2. HAADF image of the metal/AlGaN interface 共sample: series A, annealed at 775 ° C兲. At the TiN / AlGaN interface Au 共bright layer兲 can be detected.

FIG. 3. HAADF image of the metal/AlGaN interface 共sample: series A, annealed at 800 ° C兲. At the TiN / AlGaN interface AlN 共dark layer兲 can be detected.

In general 共see Fig. 1兲, after the annealing 共whatever the temperature兲, the microstructure of Au/ Mo/ Al/ Ti/ Si3N4 / AlGaN / GaN contacts can be described as follows. Near the AlGaN / GaN, which remains intact, an interfacial layer has been formed. This interface reaction will be described later. On top of the interface layer, a ±10 nm thick TiN layer can be retrieved. The TiN is build up out of small 共±10 nm兲 grains which are orientated close to the typical 关110兴TiN 储 关112គ 0兴GaN, 共111兲TiN 储 共0001兲GaN epitaxial orientation. The TiN layer also contains some Au and Al. At the top of the contact, large Mo-rich 共containing Al and Si兲 polyhedrons can be found in a Au–Al matrix. This is consistent with the results on Au/ Mo/ Al/ Ti contacts reported by Wang et al.8,9 TEM shows that the polyhedrons consist of Mo3Al or Mo6共AlSi兲 surrounded by Au4Al. The Si concentration in the Mo grains varies: high concentrations up to 20 at. % have been encountered. The differences between series A and B are that in A the Au4Al is an incommensurately modulated cubic high-temperature phase, whereas in B it takes a metastable tetragonal form. It is noteworthy that in series B samples the TiN layer is not everywhere in contact with the AlGaN: in some places a Au4Al metal layer can be found in between the TiN and the AlGaN. Further on, this will be called a floating TiN layer. This effect is the most pronounced at annealing temperatures of 800 ° C and more. As N atoms could only be found in the TiN and in the perfect III-nitride below, it is very likely that the N in the TiN is the N provided by the former Si3N4 passivation layer 共otherwise N is lost out of the sample兲. Si does thus not stay in the passivation layer but forms an alloy with Mo and Al. The AlGaN layer was found to be almost intact. We note that at the threading dislocations, TiN inclusions in the 共Al兲GaN have been found. Typically these TiN inclusions are surrounded by a very thin Au layer 共see also Refs. 8 and 10兲. As these inclusions have been observed in equal amounts at all annealing temperatures investigated here, one can conclude that they cannot be responsible for the observed Schottky to Ohmic transition. At the III-nitride interface in a contact of series A annealed at 775 ° C 共still a Schottky contact兲 a bright layer below the TiN layer shows up in high-angle annular darkfield 共HAADF兲 images 共Fig. 2兲. This line has been identified as Au. Although some Al can be detected in this thin interfacial layer, TEM reveals that it has the fcc structure of elemental Au, not the structure of Au4Al 共which is found at the top of the TiN兲. In the sample from series B, annealed at

775 ° C, the Au layer at the interface is thicker than that observed in series A. An Ohmic contact annealed at 800 ° C is shown in Fig. 3 共HAADF image兲. In here, one can detect a dark line on the AlGaN. High-resolution TEM revealed that this is pseudomorphically grown AlN at the interface between AlGaN and TiN. The AlN has the tendency to form pyramidal, dotlike structures. These pyramidal dots are largest in samples of series B 共Fig. 4兲, especially at regions were the TiN layer is “floating.” The microstructure of the samples annealed at 825 ° C is identical to the one observed in the samples annealed at 800 ° C. Annealing at 825 ° C leads to a slightly increased contact resistance. We propose that the complete contact formation process can be understood as follows. Al melts at 660 ° C and alloys with the surrounding metals, preferentially Au and Mo. The Mo alloy does not form a Au diffusion barrier, but balls up into polyhedrons. TiN is formed as reaction product of Ti and Si3N4. Si diffuses into the Mo grains. The TiN formation takes place after the melting of Al because in an Al/ Ti/ Si3N4 / AlGaN / GaN test structure, Al3Ti/ Si+ Al + N / AlN / AlGaN / GaN is found after annealing. This means that alloying between Al and Ti takes place before the TiN formation. Au 共or Au4Al兲 finally diffuses through the TiN 共Au and Al have always been found in the TiN兲. The fact that the TiN is floating or not is related to the Au4Al phase formed. Once the activation temperature for AlN formation is reached, downwards diffused Al can start its reaction with the III-nitride. Note that extra experiments have shown that

FIG. 4. Bright-field TEM image, taken in GaN共0002兲 two-beam diffraction condition 共series B, annealed at 800 ° C兲. At the metal/AlGaN interface, large AlN pyramidal dots can be retrieved. The TiN layer 共bright兲 is floating in the Au4Al matrix. Downloaded 16 Nov 2006 to 146.103.254.11. Redistribution subject to AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

201908-3

Appl. Phys. Lett. 89, 201908 共2006兲

Van Daele et al.

Ti is necessary in the contact formation process. Leaving it out of the metal scheme produces very bad contacts, which can be understood by the fact that the Si3N4 must be consumed. The presence of Al and the formation of AlN at the interface are annealing temperature dependent and can be linked with the observed Schottky to Ohmic transition at an annealing temperature of 800 ° C. The formation of pseudomorphic AlN at the interface is the key factor for the Ohmic contact behavior. As the contact is Ohmic, even with an intact AlGaN, it is possible to state that a conducting channel through the AlGaN has been created. The mechanism is rather simple when assuming that the N of AlN is extracted out of the AlGaN, since there is no difference in the TiN thickness whatever the annealing temperature. Another indication that the N in AlN is not provided by the former passivation layer is given by the fact that otherwise no AlN could have been formed below the floating TiN parts in the samples of series B. This means that the same contact formation mechanism as on n-GaN holds with the difference that this time Al extracts N out of AlGaN and not Ti. Physically this process is triggered by the fact that the enthalpy of formation of AlN is larger than that of GaN 共respectively, −318.1 and −110.9 kJ mol−1兲. Ti cannot extract N out of AlGaN 共Ref. 5兲 because its enthalpy of formation 共−265.5 kJ mol−1兲 is smaller than that of AlN. The activation barrier requires temperatures of 800 ° C and more. The deactivation of the Ti– AlGaN / GaN reaction has its origin in the introduction of the Si3N4 passivation layer, which acts as a N source to the Ti. Although AlN is a high band gap material, it is very likely that it is conducting due to the presence of N vacancies as a result of an incomplete reaction. In this picture, the increase of contact resistance at annealing temperatures higher than 800 ° C can be explained, although no differences could be observed with TEM. A higher annealing temperature results in a more stoichiometric AlN, resulting in less conduction. The possible presence of remaining Si in the Al共Ga兲N is not believed to be important as Si forms a deep level in AlN 共and is therefore not electrically active兲. The formation of an AlN interfacial layer has also been obtained by Mohammed et al.11 by adding Si in the metallization scheme. For the Ohmic contact formation itself, the important issue thus is the deactivation of the TiN formation either by SiN of Si in the metallization scheme. The use of SiN is beneficial above Si in the metallization scheme as it also passivates the AlGaN / GaN.

The contacts in series B have a lower contact resistance than those of series A. Differences between both series were the different Au4Al phase, the floating TiN layer in series B, and the larger AlN pyramidal dots in series B 共Fig. 4兲. The larger AlN pyramidal dots in the best contacts are consistent with the improved contact resistance 共higher N-vacancy doping兲. The exact Au4Al phase determines the diffusion kinetics throughout the reaction and therefore also determines the morphology of the TiN layer. In conclusion, it has been shown that it is possible to form state-of-the-art Ohmic contacts to AlGaN / GaN HEMT structures without destroying the AlGaN. The key factor to the formation of a conducting channel through the AlGaN is the formation, by N extraction, of AlN pyramidal dots by the reaction of Al with nitrogen extracted from the AlGaN. The dots are also not completely stoichiometric AlN; annealing at too high temperatures 共825 vs 800兲 renders the AlN less conducting, thereby slightly increasing the contact resistance. The introduction of an in situ Si3N4 passivation layer is responsible for the switching of the III-nitride reaction from Ti to Al. This work has been performed within the framework of IAP V-1 and was supported by the European Space Agency 共ATHENA project, ESTEC Contract no. 14205/00/NL/PA兲. One of the authors 共B.V.D.兲 is grateful to the Fund for Scientific-Research-Flanders 共F.W.O.-Vlaanderen兲. J. K. Kim, H. W. Wang, and J. L. Lee, J. Appl. Phys. 91, 9214 共2002兲. S. Ruvimov, Z. Lilienthal-Weber, J. Washburn, K. J. Duxstad, E. E. Haller, Z. F. Fan, S. N. Mohammad, W. Kim, A. E. Botshkarev, and H. Morkoç, J. Appl. Phys. 69, 1556 共1996兲. 3 J. Neugebauer and C. G. Van De Walle, Phys. Rev. B 50, 8067 共1994兲. 4 D. C. Look, D. C. Reynolds, J. W. Hemsky, J. R. Sizelove, R. L. Jones, and R. J. Molnar, Phys. Rev. Lett. 79, 2273 共1997兲. 5 B. Van Daele, G. Van Tendeloo, W. Ruythooren, J. Derluyn, M. R. Leys, and M. Germain, Appl. Phys. Lett. 87, 061905 共2005兲. 6 J. Derluyn, S. Boeykens, K. Cheng, R. Vandersmissen, J. Das, W. Ruythooren, S. Degroote, M. R. Leys, M. Germain, and G. Borghs, J. Appl. Phys. 98, 054501 共2005兲. 7 T. Prunty, J. Smart, E. Chumbes, B. Ridley, L. Eastman, and J. Shealy, IEEE Conference on High Performance Devices, Ithaca, NY, 7–9 August 2000, edited by M. G. Adlerstein, Vol. IV-6, p. 208. 8 L. Wang, F. M. Mohammed, and I. Adesida, Appl. Phys. Lett. 87, 141915 共2005兲. 9 L. Wang, F. M. Mohammed, and I. Adesida, J. Appl. Phys. 98, 106105 共2005兲. 10 M. W. Fay, G. Moldovan, P. D. Brown, I. Harrison, J. C. Birbeck, B. T. Hughes, M. J. Uren, and T. Martin, J. Appl. Phys. 92, 94 共2002兲. 11 F. M. Mohammed, L. Wang, and I. Adesida, Appl. Phys. Lett. 88, 212107 共2006兲. 1 2

Downloaded 16 Nov 2006 to 146.103.254.11. Redistribution subject to AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp