Open-Source AI for Analog Correction: From RF Power Amplifiers to Energy-Efficient Silicon

Abstract

Slides for the talk F5.4, "Open-Source AI for Analog Correction: From RF Power Amplifiers to Energy-Efficient Silicon", presented by Chang Gao at ISSCC 2026 Forum 5, "Analog for AI and AI for Analog: What the Analog/RF People Can Do and Leverage in the AI Era", on February 19, 2026.

AI offers a new paradigm for correcting analog/RF non-idealities, but challenges in design, benchmarking, and deployment hinder its adoption. This talk argues for an open-source approach to bridge these gaps. It introduces OpenDPD, a framework for PA linearization that has enabled co-design of optimized mixed-precision and sparse AI models and AI-DPD hardware accelerators, and discusses the path from algorithm to silicon with extensions to circuits such as ADCs and PLLs.