Design of Terahertz InP pHEMT Using Machine Learning Assisted Global Optimization Techniques

Abstract

This paper presents an optimal design of terahertz InP-based pseudomorphic high electron mobility transistors (pHEMT) powered by an artificial intelligence (AI) technique. Unlike the traditional physics-based design optimization methods, the new technique employs a machine learning-assisted global optimization algorithm. A state-of-theart commercial pHEMT operating at millimeter-wave frequencies was used to calibrate the physics-based model. Based on the pHEMT, the proposed machine learning-assisted optimization method was implemented with the constraint of gate length, i.e., 100 nm. The simulation results show significant improvement in terms of cut-off frequency, i.e., 57%, and maximum oscillation frequency, i.e., 30%, compared to the commercial design. To the best of our knowledge, this is the first time to employ machine learning-assisted global optimization techniques to pHEMT design, showing high potential in terms of numerical simulation and device design for ultrafast semiconductor devices.