Machine Learning-Based Analogue Circuit Design for Stage Categorization and Evolutionary Optimization

The goal of this study is to provide a two-stage bottom-up approach that uses machine learning to simplify the development of analogue devices. Analogue CMOS (complementary metal-oxide-semiconductor) circuit designers use their expertise to address complex problems, involving multiple factors and goals. The research delves into the possibilities of libraries housing open-source machine-learning models to aid designers and provides a framework outlining the specifics of ML-generated analogue circuits. Creating neural network designs has traditionally relied on commercialized CMOS or software simulations; however, these approaches don't guarantee optimal performance. The proposed approach is validated using a three-stage device design. In the first stage, the type is accurately predicted with a precision of 89.75% utilising a machine learning approach called the decision tree. Both of these rule induction strategies are used to generate prediction logic. The second stage involves forecasting the usual parameters for each stage type using four learning techniques: decision trees, gradient-boosted trees, random forests, and support vector machines. When compared to other approaches, the support vector machine outperforms them all while exhibiting the lowest error rates.