An Open-Source ANN-based Analog/RF Intellectual Property for SkyWater130nm Technology

Abstract

The demand for reusable design has become essential in today's fast-paced world, driving the integration of Intellectual Property (IP) into integrated circuit design. While digital IPs are favored for their ease of development, reconfiguration, and reuse, analog/RF IPs often lack flexibility, offering only a single general solution and requiring design iterations to meet specific targets. Automatic synthesis tools promise relief for this inflexibility, yet balancing accuracy and efficiency remains crucial for analog/RF IP reuse and resizing. Circuit modeling via artificial neural networks (ANNs) has emerged as a solution to this challenge. In this paper, we present an open-source analog/IP for SkyWater130nm technology developed through an ANN-supported analog/RF multi-objective automation tool. ANN circuit models are tested as performance estimators in the optimization loop, eliminating the need for a circuit simulator. Further validation is conducted by comparing obtained Pareto-optimal fronts (POF) with SPICE results. We demonstrate results for four distinct analog/RF building blocks, with the generated IP offering multiple solution points for each circuit and ANN-based models for immediate fine-tuning or new designs.