Artificial Neural Network Driven Optimization for Analog Circuit Performance

This paper introduces a novel Deep Learning driven approach for analog circuit optimization, predicting key design variables (e.g., effective channel width, load resistance, bias voltage, etc.) based on given performance metrics (voltage gain, bandwidth, slew rate, noise, and harmonic distortion). We have implemented instant training of the dataset employing initial parameter estimation which outperforms the Multi Objective Optimization (MOO) technique based on Genetic Algorithm (GA). The presented framework is successfully validated with extensive numerical simulations on various circuits such as common-source amplifier with resistive load, active load, differential amplifier with active load. For validation, UMC 180nm technology node PDK has been utilized in SPICE simulations.