This case study explains how Analog Devices improved compact-model extraction for GaAs pHEMT devices used in high-frequency RF and millimeter-wave circuits. Traditional workflows required deep specialist knowledge, repetitive manual tuning, and several days of work, while model accuracy often degraded across wide bias conditions. Keysight and ADI combined a derivative-free machine-learning optimizer in IC-CAP with a hybrid ANN-enhanced ASM-HEMT model. The optimizer simultaneously fitted multiple DC and S-parameter targets, while neural-network capacitance and resistance elements corrected bias-dependent limitations without abandoning physical behavior. Extraction time fell from days to hours, agreement improved from 250 MHz to 30 GHz, and large-signal PAE, output power, IMD3, and load-line

Join for free to read