artificial neural networks and hybrid evolutionary algorithms for multi-objective optimization of analog integrated circuit design

The design of analog integrated circuits demands the careful optimization of multiple interdependent parameters, including transistor sizes, bias currents, and passive components, to meet stringent performance targets such as gain, bandwidth, phase margin, and power efficiency. to address this challenge, this work introduces a computational intelligence framework that combines artificial neural networks (anns) with a hybrid genetic algorithm–particle swarm optimization (ga–pso) strategy. the framework was validated on two representative circuits: a two-stage cmos operational amplifier with miller compensation and a differential lc voltage-controlled oscillator (lc-vco) operating at 2.8 ghz in 0.18-µm cmos technology. extensive hspice simulations generated datasets that enabled the ann to capture the complex nonlinear relationships between design variables and performance metrics. the method successfully predicted optimal device dimensions and biasing conditions, achieving a 160% improvement in figure of merit (fom) for the amplifier and a fom of 118.1 dbc/hz for the lc-vco, comparable to state-of-the-art designs. these results demonstrate the framework’s versatility and scalability, providing a flexible soft-computing tool for multi-objective optimization across diverse analog circuit topologies.