Efficient optimization approach for designing power device structure using machine learning. (1st April 2023)

• Record Type: Journal Article
• Authors: Yamano, Hayate; Kovacs, Alexander; Fischbacher, Johann; Danno, Katsunori; Umetani, Yusuke; Shoji, Tetsuya; Schrefl, Thomas
• Is Part Of: Japanese journal of applied physics. Volume 62:Number SC(2023)
• Publication Date: 2023-04-01
• Subjects: power device -- machine learning -- gallium oxide; Physics -- Periodicals; 621.05
• DOI: 10.35848/1347-4065/acb061
• Languages: English
• ISSNs: 0021-4922

Abstract

Abstract: Low power-loss semiconductor devices are necessary to achieve a carbon-neutral society. The optimization of device structures is known as a time-consuming process. In this work, we investigated an optimization approach with the help of machine learning. We applied an active learning scheme to optimize a gallium oxide Schottky barrier diode structure and demonstrated how this approach helps to reduce the number of time-consuming simulations for the optimization process. For the investigated work, the active learning strategy almost reduced the number of simulations by a factor of 2 in contrast to the conventional genetic optimization. In addition, we also demonstrated that machine learning models can be used to estimate the performance variations caused by process variations. This approach can also contribute to reducing the number of simulations and speeding up the structure design process.