Automated Design of CMOS Operational Amplifier Using a Neural Network

Abstract

Automated design tools for digital systems have found widespread use. A reliable automated tool for the design of analogue circuits is highly desirable. Automating the design of analogue circuits presents a challenge due to their highly nonlinear nature, with multiple trade-offs among different performance metrics. In this paper the sizing of the components of a two-stage differential amplifier to achieve desired characteristics for performance metrics will be demonstrated. The performance metrics used include DC Gain, Phase Margin, Slew Rate, Unity-Gain Bandwidth and Area. In order to achieve this, a neural network was trained to select component sizes that achieve the desired performance characteristics. The calculated sizes are then transferred to LTspice in order to perform SPICE simulation for design verification.