Comparative Study of AlGaN/GaN HEMTs on Free-Standing Diamond and Silicon Substrates for

Thermal Effects

Manuel Trejo, Kelson D. Chabak, Brain Poling , Ryan Gilbert , Antonio Crespo, James K. Gillespie, Mauricio Kossler, Dennis E. Walker, Glen D. Via, Gregg H. Jessen

Air Force Research Laboratory Sensors Directorate Dayton, OH, U.S.A

Daniel Francis, Firooz Faili, Dubravko Babić, Felix Ejeckam

Group4 Labs, Inc. Freemont, CA, U.S.A.

Abstract—In this work, we compare for the first time the performance results of AlGaN/GaN HEMTs processed on a free-standing chemical vapor deposition (CVD) polycrystalline diamond substrate and a silicon substrate with nominally the same epitaxial AlGaN/GaN layers both grown by metal-organic chemical vapor deposition (MOCVD). The objective of this work is to compare the small signal and DC trends of the transistors fabricated on the different substrates as a function of temperature. Wafer scale results were obtained from both wafers for 2 x 150 µm devices with gate lengths of 0.18µm and 0.20µm for the silicon and CVD diamond wafers respectively.

Keywords-HEMT; AlGaN; CVD Diamond; thermal

I. INTRODUCTION High power RF applications have taken advantage of

AlGaN/GaN HEMT technology to meet performance goals. Power output densities surpassing 40 W/mm [1] have been achieved along with excellent frequency response [2-3] making AlGaN/GaN HEMTs highly desirable for millimeter wave applications. With the development of alternative substrates such as free-standing chemically vapor deposited (CVD) diamond [4-9] device degradation due to thermal effects can be reduced improving the overall power performance and reliability of AlGaN/GaN devices [10]. Recent GaN-on-diamond work has demonstrated power output of 4 W/mm and fT ~ 85GHz [11].

The electrical insulating and thermal dissipation properties of CVD diamond are attractive features for integration into electronic devices. The theoretical thermal conductivity of widely available polycrystalline CVD diamond (800-1500 W/m-K) is superior to that of silicon and 6-H silicon carbide, 141 W/m-K and 490 W/m-K respectively. Existing work has demonstrated comparisons between AlGaN/GaN epitaxial layers on CVD diamond and silicon carbide (SiC) substrates highlighting device performance based on thermal dissipation of the substrates [12]. The heat generated in the AlGaN/GaN active layers is dissipated away by the diamond. The closer the CVD diamond layer is to the active layer, greater heat

dissipation will result. However, limitations due to stress-induced bowing of the wafer, interface quality, and atomic bonding still exist. The innate fragility of a 100-micron thick CVD diamond wafer typically requires that a silicon carrier wafer for processing. In addition, single crystal diamond growth (~2000 W/m-K) has yet to be achieved in low-cost wafer form which would provide better heat dissipation qualities.

II. FABRICATION AND MEASUREMENTS

The epitaxial structure grown at Nitronex Corp on high-resistivity silicon (111) wafers consists of a 175Å Al0.26GaN0.74 barrier with a 20Å GaN cap. A ~1.9 µm GaN isolating buffer/nucleation layer is grown to gradually reduce the dislocations from the wafer to the AlGaN layer. Group4 Labs transferred the AlGaN/GaN layer by atomic bonding onto a 100mm 100µm-thick CVD diamond wafer (~800 W/m-K) using their proprietary transfer process, coring out a 25mm-round wafer for device processing. No silicon carrier wafer was attached to the diamond wafer afterwards, demonstrating the ability to handle the 100µm-thick wafer using our standard device fabrication process.

Dual-gate HEMT devices were fabricated using the same process for both the silicon and the CVD diamond wafers. First, a two-step BCl3/Cl2/Ar and Cl2/Ar dry etch in a PlasmaTherm 770 ICP system was used for mesa isolation. Ohmic contacts were then patterned via optical lithography and metal deposition was done by evaporation using a Ti/Al/Ni/Au metal stack. A 30-second 850˚C anneal in N2 yielded ohmic contacts with ~4.5µm source-drain spacing. A PMMA/MMA bilayer photoresist stack was used to pattern T-gates by electron beam lithography on a JEOL JBX 6300 system using a 100kV accelerating voltage. Gates were formed by metal evaporation of 200Å/3800Å Ni/Au. Gate lengths were 180nm and 200nm for the Si and CVD diamond samples respectively. Finally, 1000Å of Si3N4 passivation layer was deposited at 300˚C by PECVD.

U.S. Government work not protected by U.S. copyright

Full-wafer DC and small signal characterization was performed from 1 to 26GHz on an HP8510 and HP4142 network and parameter analyzer with a Cascade Microtek 12K prober for both wafers. The AlGaN/GaN on silicon (279 sites tested) demonstrated IDS,max= 676.4 mA/mm ±27.2 mA/mm (4%), gm,peak= 343 mS/mm ±18.6 mS/mm (5.4%), ft = 52.6 GHz ±2.1 GHz (3.9%), and fmax = 99.3 GHz ±7.8 GHz (7.8%). The CVD diamond wafer (30 sites tested) demonstrated IDS,max= 498.4 mA/mm ±63.7 mA/mm (12.8%), gm,peak= 229.3 mS/mm ±18.6 mS/mm (8.1%), ft = 47.9 GHz ±3.2 GHz (6.7%), and fmax = 83.9 GHz ±1.9 GHz (2.3%). Process control monitor (PCM) structures were measured on a on a Keithley 450 system. Sheet resistance and contact resistance measurements yielded RSH, Si = 409.5 Ω/sq ±10.8 Ω/sq (2.6%) and RSH, Diamond = 637.9 Ω/sq ±25.0 Ω/sq (3.9%). Contact resistance of RC,Si = 0.12 Ω·mm ±0.02 Ω·mm (17.7%) and RC,

Diamond = 0.34 Ω·mm ±0.11 Ω·mm (31.4%) for the silicon and CVD diamond substrates respectively.

III. EXPERIMENT AND DISCUSSION

Initial small-signal and DC measurements on the CVD diamond and silicon wafer yielded similar results for certain sites which enabled further evaluation of the selected HEMT devices. For this extended study, devices from the Si and CVD diamond wafers with similar contact and sheet resistances to were selected and paired to establish a baseline for performance. Given that both wafers had the same Nitronex AlGaN/GaN epitaxial layers, DC and small signal performance would result with great similarity. Three pairs of HEMT devices were selected for comparison as presented in Table 1. Additional small signal, static and pulsed-IV (DC) measurements at varying chuck temperatures up to 200˚C up at VDS =20V were obtained to identify performance degradation due to heating. During wafer-level measurements, the devices were intentionally not optimally heat-sunk. As a result, heat-sinking was substantially provided by stationary air convection.

Among the most notable differences between devices on CVD diamond and silicon substrates were their transfer characteristics (Fig. 1). As expected, the thermal effects

decreased the devices’ maximum current and extrinsic transconductance, ~33% and ~50% respectively for both substrate types. Thermionic effects of the sub-threshold

TABLE I. AlGaN/GaN HEMT DEVICE PARAMETERS ON SILICON AND CVD DIAMOND SUBSTRATES

Wafer Reticle ID

RC (Ω·mm)

RSH (Ω/sq)

gm,peak (mS/mm)

IDSS (mA/mm)

Imax (mA/mm)

ft (GHz)

fmax (GHz)

VBKa

(V) Vknee

(V) VT (V)

NC11 (S) 03,20 0.227 637.97 235.73 365.24 491.20 49.61 102.72 50.00 1.42 -1.72

G425 (D) 12,19 0.221 636.41 243.53 351.78 505.87 47.97 80.38 49.99 1.48 -1.56

NC11 (S) 04,09 0.366 631.04 237.84 368.62 504.27 49.19 104.83 22.99 1.34 -1.66

G425 (D) 14,14 0.364 636.01 237.60 354.97 505.07 49.59 84.41 49.99 1.57 -1.55

NC11 (S) 09,20 0.370 602.71 262.04 387.13 513.27 50.04 108.28 50.00 1.41 -1.75

G425 (D) 13,18 0.346 603.65 234.81 360.48 518.33 47.18 77.80 50.00 1.55 -1.56 (S) = silicon substrate, (D) = CVD diamond substrate a Breakdown voltage measured at IDS = 1mA/mm and VGS < VT

(a)

(b)

Figure 1. Transfer characteristic comparison of 2x150 µm AlGaN/GaN HEMT devices on silicon (a) and CVD diamond substrates (b). Measurements were obtained at VDS = 20V and VGS = +1V to -3V by increasing the plate temperature from 25˚C to 200˚C.

region are also evident in these plots, depicting greater threshold shift for devices on the silicon substrate.

In addition, power measurements at 10 GHz X-band were obtained on a Maury Microwave load pull system at 25˚C. All devices were pre-matched and tested under continuous wave (CW) for maximum power added efficiency (PAE) at 10% Idss, VDS = 15V-20V (Fig. 2). As depicted, the output power (POUT) and gain (GT) are traced almost identical as the input power (Pin) increases. However, the power-added efficiency (PAE) of the CVD diamond substrate is greater. Measurements indicate ≥10% higher PAE on AlGaN/GaN HEMTs on CVD diamond. The difference could be attributed to lower junction temperature and thermal resistance of the CVD diamond substrate. Further studies are needed to provide a definitive relationship between PAE and thermal efficiency of CVD diamond.

Pulsed I-V measurements were taken at 25˚C and 100˚C to observe the thermal effects under different biasing conditions (Fig. 3). Performance degradation increases at higher drain bias due to self-heating of the device at 25˚C and 100˚C. However, devices on CVD diamond demonstrate slightly better performance at VDS >10V. Under pulsed conditions, the devices on a silicon substrate have more noticeable advantage at 100˚C.

Another observation was the variation in threshold voltage as a function of temperature. The devices on CVD diamond demonstrated better VTH stability at higher temperatures (Fig. 4). Threshold voltage increased slightly compared to devices on the silicon wafer, 2.6% vs. 14.7% respectively. Gate leakage was observed to increase more for devices on CVD diamond as a function of temperature (Fig. 5). This could be attributed to barrier surface imperfections. An etch process used to remove the protective silicon nitride layer after the atomic bonding process of the AlGaN/GaN layer onto the

(a) (b) (c)

Figure 2. X-band 10 GHz CW power charactersistic comparison of 2x150 µm AlGaN/GaN HEMT devices on silicon and CVD diamond substrates. HEMT devices on CVD diamond demonstrate higher PAE while Pout and Gt values are almost identical. (a) VDS = 20V, 10% IDSS (b) VDS = 20V, 10% IDSS (c) VDS =15V, 10% IDSS

(a) (b)

Figure 3. Pulsed (200ns) I-V (dotted line) and static dc (solid line) measurement comparison of 2x150 µm AlGaN/GaN HEMT devices on silicon and CVD diamond substrates. Measurements were obtained at VDS = 20V and VGS = +1V to -3V by increasing the plate temperature from (a) 25˚C and (b) 100˚C.

CVD diamond may have altered the surface. Closer examination of the barrier surface would have to be carried out to better understand the source of increased gate leakage.

IV. CONCLUSION

The electronic devices group at the Air Force Research Lab’s Sensors Directorate, in cooperation with Group4 Labs, has demonstrated the successful application of free-standing CVD diamond substrates for AlGaN/GaN HEMT fabrication. This new alternative substrate can improve device cooling on wafer due to the proximity of the active layer to CVD diamond substrate which has innate high thermal conductivity and electrical isolating properties. With the production of 100µm-thick free standing CVD diamond wafers, the device fabrication process has improved significantly. The results of this study demonstrate that GaN-on-diamond atomic

attachment does not negatively impact the performance of the AlGaN/GaN HEMTs while stressed under similar conditions. In addition, this study has provided a preliminary observation of thermal effects of HEMTs on CVD diamond in comparison to those fabricated on a silicon wafer with nominally the same epitaxial structure. Due to the controlled nature of the experiment, the performance advantage can be attributed to the substrates and their thermal properties. The differences in using different substrates for AlGaN/GaN HEMTs are evident; however, there is still more work to be done to accurately measure advantages of this material with respect to its thermal efficiency primarily due to the inconsistency of device mounting and packaging for thermal characterization. AFRL is in the process of a comprehensive comparative study of AlGaN/GaN on silicon, silicon carbide and CVD diamond wafers provided by Group4 Labs. A broad range of DC, small signal, power, and thermal analysis will be done. Greater emphasis will be placed on standardizing the bonding and packaging techniques to ensure the thermal characterization of the devices is as accurate as possible to quantify any device performance differences among the different substrate materials being studied.

ACKNOWLEDGMENT The authors would like to thank J. Breedlove and P. Cassity

for their assistance with the metal deposition and etching and Group4 Labs for supplying the CVD diamond substrates.

REFERENCES [1] Y. F. Wu, M. Moore, A. Saxler, T. Wisleder, and P. Parikh, Proc. 64th

DRC Tech. Dig., 2006, pp. 151–152. [2] J. Chung, W.E. Hoke, E.M. Chumbes, T. Palacios, IEEE Electron

Device Letters., vol. 31, no. 3, Mar. 2010. [3] D.I. Babić, Q Diduck, P. Yenigalla, D. Francis, F., Falli, F. Ejeckam,

J.G. Felbinger, L.F. Eastman, IEEE Transactions on Electronic Devices (unpublished)

[4] G. H. Jessen, J. K. Gillespie, G. D. Via, A. Crespo, D. Langley, J. Wasserbauer, F. Faili, D. Francis, D. Babic, F. Ejeckam, S. Guo, I. Eliashevich, 2006 IEEE Compound Semiconductor Integrated Circuit Symposium, pp. 271-274, Nov. 2006.

[5] D.C. Dumka, P. Saunier, "AlGaN/GaN HEMTs on Diamond Substrate," Proc. 65th DRC Tech. Dig, 2007, pp.31-32.

[6] J.W. Zimmer, G. Chandler, 2008 Int’l Conference on Compound Semiconductor Manufacturing Technology.

[7] K.D. Chabak, J.K. Gillespie, V. Miller, A. Crespo, J. Roussos, M. Trejo, D.E. Walker, G.D. Via, G.H. Jessen, J. Wasserbauer, F. Faili, D.I. Babic, D. Francis, F. Ejeckam, IEEE Electron Device Lett., vol. 31, no. 2, Feb. 2010.

[8] J.G. Felbinger, L.F. Eastman, J. Wasserbauer, F. Faili, D.I. Babić, D. Francis, and F. Ejeckam, 2009 Workshop on Compound Semiconductor Devices and Integrated Circuits.

[9] D. Francis, F. Faili, D. Babić, F. Ejeckam, A Nurmikko, H. Maris, Material Research Society 2009

[10] S. Tirelli, D. Marti, H. Sun, A.R. Alt, H. Benedickter, E.L. Piner, C.R. Bolognesi, IEEE Electron Device Lett., vol. 31, no. 4, Apr. 2010.

[11] Q. Diduck, J. Felbinger, L.F. Eastman, D. Francis, J. Wasserbauer, F. Faili, D.I. Babić, F. Ejeckam, IEEE Electron Device Lett., vol. 45, no. 14, Jul. 2009.

[12] J.G. Felbinger, M.V.S. Chandra, Y. Sun, L.F. Lester, J. Wasserbauer, F. Faili, D. Babić, D. Francis, F. Ejeckman, IEEE Electron Device Lett., vol. 28, no.11, Nov. 2007.

Figure 4. Threshold voltage (VT) measurement comparisons of 2x150 µm AlGaN/GaN HEMT devices on silicon (NC11) and CVD diamond (G425) substrates. Measurements were obtained at VDS = 20V and VGS = +1V to -3V at different plate temperatures from 25°C to 200°C.

Figure 5. Gate current leakage (Igl) measurement comparisons of 2x150 µm AlGaN/GaN HEMT devices on silicon (NC11) and CVD diamond (G425) substrates. Measurements were obtained at VDS = 20V and VGS = +1V to -3V at different plate temperatures from 25°C to 200°C.