For AlGaN/GaN high-electron-mobility transistors (HEMTs), a gate dielectric or surface-passivation layer must do more than cover the semiconductor. Interface traps can capture and release charge as operating conditions change, contributing to threshold-voltage instability and reliability concerns. A study of annealed, approximately 2.6 nm ALD Al2O3 on AlGaN/GaN MIS HEMT capacitors illustrates an important design trade-off: Al2O3 GaN passivation can be useful when minimizing added parasitic capacitance is a priority, but the thin dielectric does not necessarily provide strong suppression of gate leakage.
Why assess both interface traps and leakage?
AlGaN/GaN HEMTs use a two-dimensional electron gas (2DEG) channel, and dielectric layers may serve as a gate insulator, a surface-passivation layer, or both. Their electrical behavior depends in part on the quality of the dielectric/(Al)GaN interface and on the dielectric itself. Traps at the interface can respond to changing bias and signal frequency, while defects within the dielectric can provide paths for current through the gate stack.
These effects call for different measurements. Capacitance-voltage (C–V) characterization can reveal charge-trapping behavior and help estimate interface-trap densities. Temperature-dependent current-voltage (I–V) measurements can help identify the mechanism responsible for leakage. Considering the results together gives engineers a more complete view than treating capacitance or gate current as a stand-alone measure of dielectric quality.
What C–V measurements revealed about Al2O3 GaN passivation
The capacitors were fabricated on a GaN-on-Si(111) structure. A 10 nm ALD Al2O3 layer was deposited initially, then thinned in the gate region after the ohmic-contact annealing process. C–V analysis estimated the remaining gate-region dielectric thickness at approximately 2.6 nm.
Low slow-trap response, but measurable fast traps
In dual-sweep C–V measurements, the voltage hysteresis around threshold was approximately 1 mV. The study used this hysteresis to estimate a slow interface-trap density of about 3.33 × 109 cm−2 for the trap states sampled by that measurement. This small hysteresis indicates limited slow charge trapping under the measurement conditions; it should not be interpreted as an inventory of every trap state at the interface.
Frequency-dependent conductance measurements probed faster interface states. For the measured energy range of 0.275 to 0.408 eV below the GaN conduction-band edge, the extracted density ranged from 8 × 1012 to 5 × 1011 eV−1 cm−2. The difference between these results and the slow-trap estimate reflects the fact that the methods respond to different trap time constants and energy ranges.
The study also notes a measurement consideration for MIS HEMTs: the barrier layer between the dielectric interface and the 2DEG has resistance and capacitance that can affect conductance analysis. The reported assessment considered this potential artifact, but the results still apply to the specific structure and measurement approach used. For process development, comparing multiple characterization methods can help avoid treating one extracted trap density as a complete description of interface behavior.
Trap-assisted tunneling governs reverse-bias leakage
Temperature-dependent I–V measurements were used to examine electron transport through the thin Al2O3 under reverse bias. In the measured voltage range, the estimated electric field across the dielectric was 3.69 to 4.34 MV cm−1. Analysis identified trap-assisted tunneling (TAT) as the dominant reverse-bias transport mechanism in that range.
In TAT, electrons move through the insulator by tunneling via defect-related energy levels rather than crossing the dielectric in a single step. The trap energy associated with the measured conduction was estimated at 2.13 ± 0.02 eV below the Al2O3 conduction-band edge. The current increased with temperature, consistent with phonon-assisted TAT as described in the study.
Poole–Frenkel emission and Fowler–Nordheim tunneling were also considered. The analysis did not identify either as the dominant mechanism in the selected reverse-bias region at temperatures above 30 °C. This distinction matters because reducing interface traps and reducing leakage through the dielectric are related but separate engineering objectives: favorable C–V hysteresis does not, by itself, establish that gate current will be low.
Annealing changes the dielectric as well as the interface
The Al2O3 experienced a rapid thermal anneal with a maximum temperature of 900 °C for 40 seconds during processing. Separate measurements on Al2O3 films showed an increase in dielectric constant from about 8 before annealing to about 9 afterward, while ellipsometry indicated an approximately 8.4% reduction in film thickness.
X-ray photoelectron spectroscopy showed a change in the fitted Al 2p signal after annealing, including an increased component associated with Al-O-H. The authors suggest that annealing may enable a reaction between Al-O bonds and hydrogen residues or defects present in the ALD film. They propose that this chemistry could contribute to the observed TAT, but it is an interpretation rather than a definitive identification of the cause of leakage.
Practical implications for HEMT process decisions
For engineers and process teams, the results highlight the need to match dielectric selection and thickness to the intended device function. A thin Al2O3 layer can serve as AlGaN/GaN surface passivation where limiting passivation-associated parasitic capacitance is important. In the studied structure, adding the annealed dielectric also reduced reverse-bias current density by approximately two orders of magnitude compared with the Schottky capacitor reference. However, TAT remained the dominant leakage mechanism through the MIS dielectric, so the layer was not considered ideal when the goal is significant suppression of gate leakage.
When evaluating a similar process, teams should therefore assess both slow and fast interface-trap response, characterize leakage over relevant bias and temperature conditions, and account for the effects of post-deposition annealing. These measurements help distinguish a passivation layer that supports the desired capacitance behavior from a gate dielectric that also meets leakage requirements.
Conclusion
Annealed, approximately 2.6 nm ALD Al2O3 presents a useful but application-dependent option for AlGaN/GaN structures. The measured low slow-trap density supports its potential for surface passivation, while frequency-dependent conductance reveals a separate population of faster interface traps. Most importantly, reverse-bias transport through the dielectric was governed by trap-assisted tunneling. The practical lesson is to evaluate interface quality and gate leakage as distinct design criteria before adopting thin Al2O3 in a MIS HEMT process.