Ageing of GaN HEMT devices: which ... - Journal of Semiconductors

8 downloads 21 Views 641KB Size Report
conditions. Finally, regarding the mass of data attainable with ... a voltage-temperature stressed device. ... During the recovery measurements, the devices are.

Vol. 37, No. 1

Journal of Semiconductors

January 2016

Ageing of GaN HEMT devices: which degradation indicators? A. Divay1 , O. Latry1; Ž , C. Duperrier2 , and F. Temcamani3 1 Groupe

de Physique des Matériaux, Université et INSA de Rouen - UMR CNRS 6634 - Normandie Université, Saint Etienne du Rouvray, France 2 University of Cergy, ETIS UMR 8051 CNRS, ENSEA, 95000 Cergy-Pontoise, France 3 ECS-Lab, ENSEA, 95000 Cergy-Pontoise, France

Abstract: A following of diverse degradation indicators during the ageing in operational conditions of AlGaN/GaN HEMTs (high electron mobility transistors) is proposed. Measurements of pulsed I –V , Schottky barrier height, RF output power and gate current versus output power during the early phase of the ageing test (2000 h on a 6000 h total) are presented. These preliminary results give insight on some of the principal degradation indicators that are interesting to follow during an ageing test close to operational conditions on such components. Key words: GaN; HEMT; ageing tests; reliability DOI: 10.1088/1674-4926/37/1/014001 PACS: 85.30.Tv

1. Introduction The AlGaN/GaN HEMT (high electron mobility transistors) technology is very promising for radio frequency power and high-voltage switching applicationsŒ1 . However, the lack of reliability feedback is, at this time, one of its major drawbacks. A variety of studies were conducted in order to address those reliability questionsŒ2 4 . DC and step stress may be interesting in order to know exactly which parameter is important on a degradation mechanismŒ5 7 or with normalized tests like MILHDBK in order to qualify a technology for foundriesŒ8 . Yet, the best way to characterize the ageing of a component in its system (as an RADAR for example) is through operating conditions. Finally, regarding the mass of data attainable with this kind of test, the question of which degradation parameter to follow is important. After the ageing, the components will be investigated physically until the atomic scale using the laboratory instruments (EMMI, FIB, TEM, etc.) in order to link the electrical degradations to physical ones. This paper presents the preliminary results of an ageing test close to operating conditions (2000 h on 6000 h), and different ageing indicators are shown (pulsed I –V measurements, Schottky characterization, outpout power, Ig D f .Pin /, etc.).

2. Method The power amplifiers used for our stress have been designed using 0.5 m gate length and 32  250 m gate width GaN-on-SiC AlGaN/GaN HEMTs. The amplifiers were intended to permit in-situ measurements: two switches allow bypassing the decoupling capacitors (Figure 1). It is then possible to stop the ageing and to perform pulsed and diode measurements without unplugging the amplifiers and losing the bench RF calibration. Six components are aged in parallel branches, close to operational conditions but stressed in temperature and voltage.

EEACC: 2560

The input power is pulsed (duty cycle at 10%): Pin D 35 dBm (around 4 dB of compression) for a class B amplification (Idq D 200 mA). Half of the components are stressed at Vds D 50 V (nominal) and Tc D 160 ıC (stress 1), the other half at Vds D 60 V and Tc D 150 ıC (stress 2). These two conditions are chosen in order to discern if there is a different degradation mechanism between a temperature-only stressed device and a voltage-temperature stressed device. The junction temperature is approximately the same between the two conditions, at 185 ıC (the difference in Vds is compensated with the Tc shift). During the ageing, the input, output and reflected power are monitored. The average voltages and currents on the gate and the drain, as the case temperature, are also recorded in order to follow the degradation in the components. At logarithmic intervals, the ageing in stopped for recovery measurements, which would not be possible without the in-situ setting. These measurements include diode characterization and pulsed I –V and Ig D f .Pin /. As of now, these measurements were done at t0 , 43, 132, 500, 1000 and 2000 h.

3. Preliminary results of the ageing test The first element visible during the ageing is the lowering of the RF output power (Figure 2). The noise visible on the curve is due to the room temperature shift from the day/night cycle, changing slightly the input power. This output power decrease may be a consequence of an evolution of the GaN structure; however, using this data only, it is not sufficient to conclude. A set of complementary measurements during the ageing test is then necessary in order to gain more information. This task is fulfilled with recovery measurements. Regarding all the aged components, a 0.4 to 1 dBm diminution of output power is observed. This degradation is generally accompanied in our measurements with a decrease in mean Ids (Vth shift). The law of Pout diminution is of the form Pout D at b with a D Pout (t D t0 / and b is between –0.002 and –0.005, depending

† Corresponding author. Email: [email protected] Received 13 October 2015

© 2016 Chinese Institute of Electronics


J. Semicond. 2016, 37(1)

A. Divay et al.

Figure 1. Diagram of the amplifiers on which the components are mounted. The switches permit the in-situ measurements. The impedance adaptation and the RF filters are not represented.

Figure 2. Lowering of the output power during the early phase of the ageing for one of the components (stress 2), time in logarithmic scale.

on the components. It seems that the output power diminishes quicker with stress 2 than with stress 1, to be confirmed with the end of the ageing tests. The actual drop of output power is between 1% and 2%. The mean drain current is also followed during the ageing and is found to diminish with time in the same way as Pout . However, this effect is reversible after recovery measurements. This tendency indicates that the threshold voltage may shift due to traps with a virtual gate effect. These trapped electrons are then released when the RF signal is shut down and the recovery measurements are made. Regarding the gate current during the ageing test, its mean value is slowly rising (Figure 3). As this data is extracted during the ageing test and not in recovery measurements, the day/night cycle can also be observed. A peak in input power is present between 800 and 1200 h but then returns to normal values and is due to a shift in the room’s temperature (correlated with Pin /. This slow rise may be due to an increase of compression, hence the little augmentation in forward current. This hypothesis will be verified with Pin –Pout measurements at the end of the ageing tests. During the recovery measurements, the devices are cooled down to a case temperature of 35 ıC. They are biased at Vds D 50 V (nominal) and Idq D 200 mA. The mean gate current is then measured as a function of input RF power (Figure 4). It increases with input power, as the compression settles. However, this Ig evolution starts sooner and increases higher

Figure 3. Mean value of Igs during the ageing. The noise is due to the precision of the ageing bench and the day/night cycle. Precise gate characterization is done with diode measurements.

as the ageing goes on. The compression seems to come faster on an aged component. This result will be validated with Pin – Pout measurements in the end of the ageing test. The Schottky junction being the most sensitive part on HEMTs, the following of its characteristic is crucial during an ageing test. This contact is then precisely characterized in forward and reverse diode during recovery measurements for the purpose of observing its physical state. These measurements indicate a surprising diminution of reverse gate leakage current (Figure 5). As seen on other workŒ9; 10 , the barrier height has a tendency to rise during the ageing test. In our recovery measurements, it levels off in an exponential saturated way (Figure 6), as the ideality factor diminishes in the complementary way. This evolution suggests a change in the Schottky interface state induced by the joined effects of temperature and voltage. Several studies have shown the stability of the Schottky contacts (using transition metals on a GaN or AlGaN layer) regarding interdiffusion effects at high temperaturesŒ7; 11 . This change is then more likely to stem from a more subtle change in interface state (local bonds) as the Schottky barrier height is clearly dependent on these interfacial states and not only on the work function of the metal forming the junctionŒ12 . Finally, pulsed I –V measurements were conducted during recovery measurements at three bias points: (Vgs , Vds / D (0


J. Semicond. 2016, 37(1)

A. Divay et al.

Figure 4. (Color online) Mean gate current versus input power (Idq D 200 mA, Vds D 50 V). The gate current increases sooner and higher with the ageing.

Figure 7. (Color online) Variation of gm during the ageing test with the quiescent bias (Vgs0 , Vds0 / D (–7 V, 0 V), stimulating traps close to the gate.

Figure 5. (Color online) Reverse diode characteristic during the ageing test. Surprisingly, the reverse leakage current decreases.

Figure 8. (Color online) Variation of the maximum of gm (taken at Vgs D 1 V) during the ageing test with all quiescent biases. A detrapping effect is visible on the third quiescent bias with an odd point at  1000 h.

Figure 6. Forward Schottky characteristics during recovery measurements. The barrier height is increasing with a tendency to saturate at a higher value.

V, 0 V), (–7 V, 0 V) and (–7 V, 50 V). These bias points are chosen respectfully to have the less possible trapping effects, a gate lag and a drain lag characterization. From these measurements, we observe a decrease in the slope of input–output char-

Figure 9. (Color online) Measurement of a current transient for trapping effects analysis.

acteristics (Id –Vg /. We then extracted the transconductance for each component (Figure 7). It appears that the maximum of gm decreases during the ageing, for each of the bias points. No


J. Semicond. 2016, 37(1)

A. Divay et al. end of the ageing tests. Furthermore, we have shown that a single evolution of 'b is not self-sufficient to determine the evolution of the Schottky barrier. Pulsed I –V and trapping measurements may help to characterize a probable subtle change in interface states at this junction, before a physical analysis. Finally, we have shown that in order to make good assumptions on an ageing test close to operational conditions, the need for a complete characterization of the component at each step is crucial.

Acknowledgement The authors would like to thank the French Department of Defense (DGA) for its financial support to this work. Figure 10. Arrhenius plot of the three defects observed in traps measurement. Three activation energies are extracted: 0.45, 0.86 and 0.65 eV.

real Vp shift is observed. This indicates that the hypothesis of Au–Ni diffusion is less probable and has yet to be proved with physical analysis in the end of the ageing tests. By taking only the gm maximum of each curve, we note that the devices show little gate lag and more drain lag (Figure 8). The decrease in gm is much higher with the (–7 V, 50 V) bias point than the others. An odd point is visible on the (–7 V, 50 V) characterization and is also visible on some other devices. This effect may be due to detrapping effects during the characterization. Finally, trapping measurements are done at t0 and at the end of the ageing, due to their relative high length compared to other recovery measurements (one day per component). Drain current transients are measured in a range of temperature from 10 to 90 ıC (range of the heating elements on the ageing bench), without disconnecting any RF connector thanks to in-situ measurements. An example of a transient measurement is presented in Figure 9. Three defects are observed at the initial state t0 (Figure 10): two detrapping constants (DP1 at 0.45 eV and DP2 at 0.65 eV) and a trapping constant TP1 at 0.86 eV. These defects are attributed in the literature to various physical causes: DP1 is related to C/O/H impuritiesŒ13 , DP2 to VGa C oxygen complexes (Ga vacancies)Œ14 and TP1 also to gallium vacanciesŒ15 . These values will be compared with measurements on aged devices in order to characterize a possible evolution in trapping effects. A possible evolution of traps close to the gate may be partially linked to the evolution of the barrier height with time.

4. Conclusion These results indicate that the more information we can obtain on the different ageing indicators, the better hypotheses we can produce in order to explain these changes. Measurements ranging from pulsed I –V , diode and Ig D f .Pin / coupled with the following of Pout , Idmean and Igmean are a minimum for ageing close to operating conditions. A comparison between trapping measurements before and after the ageing tests may also reveal interesting information. However, due to the long time a complete trapping study will need, it may be difficult to follow the dynamic of trapping effects at each recovery measurement for each component, hence their use at the beginning and the

References [1] Wu Y W, Moore M, Saxler A, et al. 40-W/mm double fieldplated GaN HEMTs. IEEE 64th Device Research Conference, 2006: 151 [2] Fonder J B, Latry O, Duperrier C, et al. Compared deep class-AB and class-B ageing on AlGaN/GaN HEMT in S-band pulsed-RF operating life. Microelectron Reliab, 2012, 52(11): 2561 [3] Del Alamo J A, Joh J. GaN HEMT reliability. Microelectron Reliab, 2009, 49(9–11): 1200 [4] Meneghesso G, Verzellesi G, Danesin F, et al. Reliability of GaN high-electron-mobility transistors: state of the art and perspectives. IEEE Trans Device Mater Reliab, 2008, 8(2): 332 [5] Marcon D, Viaene J, Favia P, et al. Reliability of AlGaN/GaN HEMTs: permanent leakage current increase and output current drop. Microelectron Reliab, 2012, 52(9/10): 2188 [6] Di Lecce V, Esposto M, Bonaiuti M, et al. Experimental and simulated DC degradation of GaN HEMTs by means of gate–drain and gate–source reverse bias stress. Microelectron Reliab, 2010, 50(9–11): 1523 [7] Chou Y C, Leung D, Smorchkova I, et al. Degradation of AlGaN/GaN HEMTs under elevated temperature life testing. Microelectron Reliab, 2004, 44(7): 1033 [8] Lambert B, Thorpe J, Behtash R, et al. Reliability data’s of 0.5 m AlGaN/GaN on SiC technology qualification. Microelectron Reliab, 2012, 52(9/10): 2200 [9] Fonder J B, Chevalier L, Genevois C, et al. Physical analysis of Schottky contact on power AlGaN/GaN HEMT after pulsed-RF life test. Microelectron Reliab, 2012, 52(9/10): 2205 [10] Singhal S, Roberts J C, Rajagopal P, et al. GaN-on-Si failure mechanisms and reliability improvements. IEEE 44th International Annual Reliability Physics Symposium Proceedings, 2006: 95 [11] Li H D, Wong T L, Wang N, et al. Effect of the starting surfaces of GaN on defect formation in epitaxial Co thin films. J Appl Phys, 2011, 110(9): 093501 [12] Tung R T. Recent advances in Schottky barrier concepts. Mater Sci Eng R: Reports, 2001, 35(1–3): 1 [13] Tapajna M, Simms R J T, Pei Y, et al. Integrated optical and electrical analysis: identifying location and properties of traps in AlGaN/GaN HEMTs during electrical stress. IEEE Electron Device Lett, 2010, 31(7): 662 [14] Stuchlíková L, Šebok J, Rybár J, et al. Investigation of deep energy levels in heterostructures based on GaN by DLTS. IEEE 8th International Conference on Advanced Semiconductor Devices & Microsystems (ASDAM), 2010: 135 [15] Polyakov A Y, Smirnov N B, Govorkov A V, et al. Influence of high-temperature annealing on the properties of Fe doped semiinsulating GaN structures. J Appl Phys, 2004, 95(10): 5591