15.3 EC-2.0eV Trap-Related Dynamic RON in GaN/Si MISHEMTs

Large dynamic on-resistance due to a trap at EC-2.0 eV is observed in industry-supplied AlGaN/GaN MISHEMTs. This trap is likely in the GaN buffer and possibly related to carbon. AlGaN/GaN metal-insulator-semiconductor high electron mobility transistors (MISHEMTs) for high voltage switching applications are promising due to the ability to simultaneously achieve high breakdown voltage and low on-resistance beyond the theoretical limits of Si technology [1]. However, electrically active defects in these structures can negatively impact the device operation leading to increasing static on-resistance (RON), dynamic RON, and threshold voltage instabilities [2]. In this work, we correlated degradations to a specific trap, identified its physical location and possible physical source using MISHEMT-based defect spectroscopy. The AlGaN/GaN MISHEMTs in this study were fabricated on a Si (111) substrate with a dielectric layer in the access regions and under the gate. Multiple field plates were optimized to achieve a breakdown voltage in excess of 600V. On the same wafer, 300 たm x 300 たm AlGaN/GaN Schottky diodes were fabricated without the dielectric layer to investigate the location and physical source of the trap(s). At large off-state VDS, the MISHEMTs exhibited large dynamic RON with a ~400 s time constant determined by fitting with an exponential decay function (Fig. 1a). An onset to the dynamic RON is evident in Fig. 1b with a threshold at 200 V, and the dynamic RON reaches ~15X higher than the static RON. If the dynamic RON recovers due to thermal emission from a defect state, the dynamic RON would exhibit strong temperature dependence, but Fig. 2 shows the dynamic RON transients are temperature independent. To further explore the trap(s) responsible for the dynamic RON, constant drain current deep level optical spectroscopy (CID-DLOS) measurements were used [3]. The sample was subjected to off-state VDS biasing of 250 V then the dynamic RON transient was recorded with the sample subjected to monochromatic light. In Fig. 3 the steady-state RON is plotted after 100 s with monochromatic light shining. This time is significantly shorter than the dynamic RON time constant so if the photon energy is lower than the trap energy the trap does not emit and the RON remains high. If the trap emission is stimulated optically (i.e. hv>EC-ET), the dynamic RON will quickly return to the static RON. Fig. 3 shows a single onset in the dynamic RON where it recovers with optical energies above 2.0 eV indicating a single trap at EC-2.0 eV that is also responsible for the temperature-independent recovery (~300 s in Fig. 2). From the recovery and using the model to estimate trap concentrations in HEMTs [4], the trap density of this level was estimated to be 9.6x1012 cm-2, which is very close to the 2DEG concentration and explains the large RON change. To identify where the defect is located in the MISHEMT device and possible physical source, the Schottky diodes were used. To determine the defect location, AlGaN/GaN Schottky diodes were used. Fig. 4 shows the capacitance-based deep level optical spectroscopy measurements with a depletion depth of 0.7 たm, which was almost entirely in the GaN buffer layer. Figure 4 shows the same EC-2.0 eV trap as the MISHEMT measurements. The fact that the depletion is almost entirely in the GaN buffer means the measurement is mostly sensitive to traps in the buffer, so the presence of the EC-2.0 eV trap suggests this level is a GaN buffer trap. To confirm the location of the EC-2.0 eV trap, nanometerscale deep level optical spectroscopy (nano-DLOS) was performed on the AlGaN/GaN Schottky device. Nano-DLOS is a scanning probe microscopy technique where the surface potential is monitored at a fixed position. This allows lateral mapping of defect concentrations and the presence of the 2DEG at the AlGaN/GaN interface acts to screen any trap response in the GaN from the nano-DLOS tip. The nano-DLOS results in Fig. 5 show little evidence of an EC-2.0 eV level nor evidence of other typical GaN peaks such as EC-3.28 eV. The only feature in the AlGaN layer is the EC-3.76 eV trap previously observed in similar AlGaN material [5]. Additional evidence to the location and source of this level is available from previous studies of carbon incorporation in n-type GaN. It was found that carbon contributed to several deep levels in ntype GaN co-doped with Si and C, but C led to a level forming at EC-1.94 eV, which is very similar to the EC-2.0 eV level reported here indicating these levels may be the same trap [6]. These electron traps are negatively charged, which leads to a 2DEG reduction and increased dynamic RON. It is demonstrated that large dynamic RON is present in these MISHEMTs due to a high concentration of EC-2.0 eV traps likely in the GaN buffer that are possibly related to the presence of carbon. This suggests that careful buffer design and limiting carbon incorporation is necessary to reduce RON. ACKNOWLEDGEMENTS The was partially supported by the Semiconductor Research Corporation under Task #2298.001, Office of Naval Research under contracts N00014-13-1-0815, N00014-13-1-0796, and N00014-09-1-0242 (monitored by Dr. P. Maki), and National Reconnaissance Office under contract NRO000-13-C-0308. REFERENCES [1] Y. Wu, et al, IEEE Electron Device Lett. 25, 117 (2004). [2] U. K. Mishra et al., Proc. IEEE 96, 287 (2008). [3] A. Arehart et al., Phys. Stat. Sol. C 8, 2242 (2011). [4] A. Arehart et al., Proc. 2010 IEEE International Electron Devices Meeting (IEDM), 20.1.1 (2010). [5] A. Arehart et al., J. Appl. Phys. 109, 114506 (2011). [6] A. Armstrong et al., J. Appl. Phys. 98, 053704 (2005). [7] R. Chu et al., IEEE Electron Device Lett. 32, 632 (2011). 15 309 CS MANTECH Conference, May 18th 21st, 2015, Scottsdale, Arizona, USA 10 -4 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 10 4 10 5 0 2 4 6 8 10 12 14