Dielectric reliability in high-voltage GaN metal-insulator-semiconductor high electron mobility transistors

Abstract

As the demand for more energy-efficient electronics increases, GaN has emerged as a promising transistor material candidate for high-voltage power management applications. The AlGaN/GaN Metal-Insulator-Semiconductor High Electron Mobility Transistor (MIS-HEMT) constitutes the most suitable device structure for this application as it offers lower gate leakage than its HEMT counterpart. GaN has excellent material properties, but there are still many challenges to overcome before its widespread commercial deployment. Time-dependent dielectric breakdown (TDDB), a catastrophic condition arising after prolonged high-voltage gate stress, is a particularly important concern. This thesis investigates this crucial reliability issue in depth. Using a robust characterization strategy, we have studied not only the dielectric breakdown behavior in GaN MIS-HEMTs but also the evolution of the device subthreshold characteristics in the face of high bias stress. This allows us to work towards understanding on a more physical level the underlying degradation behind dielectric breakdown in order to inform future device lifetime models. We begin by looking at positive gate stress TDDB, a classic condition studied in the silicon CMOS community for many years. In order to understand the impact of TDDB, we must also understand how transient degradation effects such as threshold voltage (VT) shift may impact our results and ensure we can disentangle the permanent degradation associated with TDDB. With the foundational understanding of TDDB we establish under these positive gate stress conditions, we turn our attention to OFF-state stress which is a more relevant stress condition that mimics the most common state of these GaN power switching transistors in power management circuits. In order to develop accurate lifetime models for GaN MIS-HEMTs, we show that much care must be taken to ensure that device lifetime does not become distorted by transient trapping-related degradation effects. It is also crucial to have a physics-based lifetime model that gives confidence in making lifetime projections from data collected in the span of hours to lifetime estimations on the order of many years.