Learning to Design Analog Circuits to Meet Threshold Specifications

Automated design of analog and radio-frequency circuits using supervised or reinforcement learning from simulation data has recently been studied as an alternative to manual expert design. This research aims to address the common challenge faced by users who have threshold performance criteria rather than an exact target vector. We propose a novel method that generates a dataset from simulation data, enabling a system to be trained via supervised learning to design circuits that meet these threshold specifications. Our work represents an extensive evaluation of automated analog circuit design, experimenting with a significantly diverse range of circuits including linear, nonlinear, and autonomous configurations. The results demonstrate that our method consistently achieves a success rate exceeding 90% within a 5% error margin and enhances data efficiency by more than an order of magnitude. A demonstration of the developed system is available at circuits.streamlit.app.