Experimental characterization and modeling of GaN-based power devices reliability

Power electronics plays a fundamental role in enhancing energy efficiency and promoting sustainable development, which is crucial for a more environmentally conscious future. Power electronics deals with the conversion and the control of electric power, using high-efficiency, reliable and even more compact electronic converters based on switching mode semiconductor power devices. In this scenario, GaN-on-Si devices showed capability to operate at high-voltage and -frequency with higher efficiency and comparable cost of the silicon counterparts. Although GaN transistors demonstrate impressive characteristics, there are degradation mechanisms affecting the device performance and reliability that need further investigation. This PhD dissertation is focused on the identification, characterization, and modeling of the root causes that limit the gate reliability of AlGaN/GaN HEMTs featuring pGaN gate technology, by accelerated life tests and electro-thermal simulations. The thesis main contributions are: Gate biases, temperatures, and device geometry dependencies of time-dependent gate breakdown (TDGB) in GaN-based power HEMTs with p-type gate under DC stress conditions are analyzed. Two failure mechanisms have been identified, hence, accurate field-acceleration fitting models are adopted to estimate gate lifetime. Combined experimental/simulation analysis was performed to study the TDGB under pulse stress conditions. Thanks to this investigation, reliability issues introduced by the switching operation have been highlighted. An extensive analysis on the role of both switching frequency and duty cycle on the TDGB of GaN-HEMTs with Schottky metal to p-GaN gate. Findings of this analysis are useful both for further technology improvement and for GaN-based power circuit designers. Additionally, storage/release mechanisms within the buffer layers responsible for Ron degradation were investigated by means of back-gating I-DLTS and TCAD simulations. A genetic algorithm has been employed to fit the experiments allowing to understand the work conditions dependence of DRon. Moreover, devices featuring different buffer layers composition are compared, providing useful information for the epi-stack optimization.