A D-Band InP Power Amplifier Featuring Fully AI-Generated Passive Networks

Abstract

This work presents a high-efficiency D-band power amplifier (PA) implemented in 250-nm indium phosphide (InP) technology. A key innovation is the integration of artificial intelligence (AI) into the radio frequency power amplifier (RFPA) design flow for the automated generation of passive networks—a traditionally labor- and computation-intensive process. By utilizing a machine learning (ML) model as a surrogate for time-consuming electromagnetic (EM) solvers, this approach enables rapid exploration of a broader design space, improving productivity and uncovering potentially nonintuitive structures. The physics augmentation in the ML model mitigates its dependence on big training datasets, achieving an S-parameter prediction error of less than 0.5 dB at the target frequency of 125 GHz. Using fully AI-generated passive networks, the PA delivers a saturated output power ( ${P} _{\text {sat}}$ ) of 15.3 dBm and a peak power-added efficiency (PAE) of 26%. Additionally, it achieves a peak gain of 12.7 dB and a 3-dB bandwidth spanning 110–140 GHz and demonstrates its capability to support high-order quadrature amplitude modulation (QAM) signals at tens of Gb/s data rates.