GAUSSIAN PROCESS REGRESSION WITH BAYESIAN METHOD FOR DOPING OPTIMIZATION FOR ADAPTIVE RADIATION HARDENING OF FINFETS IN SPACE GRADE PROCESSORS

This is solely based to optimize FINFETs doping profiles for space grade processors using Gaussian Process Regression model trained on the TCAD simulations dataset. A cleaned dataset of FINFET devices which was created with the values of source/drain (SD) concentration, halo dose, body doping, gate work-function, and oxide charge as input features, and extracted electrical outputs - threshold voltage Vth, on-current Ion, off-current Ioff, and Ion/Ioff ratio along with a radiation-hardening label. Separate GPR models for Vth, log10 (Ion), log10 (Ioff), and log10(Ion/Ioff) achieve test R2 scores of approximately 0.96-0.98 with Root Mean Square Error (RMSE) around 7mv for Vth and sub-decade errors for current-related quantities.