A Hybrid Numerical–Deep Learning Approach for Feature Importance Analysis in Junctionless Infrared Phototransistors

Abstract

This article presents a physics‐informed deep learning (DL) framework for the modeling and optimization of submicron junctionless infrared phototransistors (JLPs). By integrating TCAD‐based simulations with DL classification and feature importance analysis, the study identifies key parameters that govern device figures of merit (FoMs) such as I on , I off , I on /I off ratio, and responsivity. Results highlight the critical influence of doping and structural variations on phototransistor performance and demonstrate the utility of DL in guiding low‐cost, high‐efficiency device design for optical communication systems. Beyond performance prediction, the proposed approach provides valuable design guidelines that capture complex parameter interactions and improve device reliability. This work highlights the potential of physics‐informed DL as a strategic tool to accelerate the development of scalable, energy‐efficient phototransistors for next‐generation low cost CMOS‐based optoelectronic applications.