AI-Assisted Modeling and Optimization of RF Circuits Using HSPICE and Neural Metaheuristics

Abstract

This paper suggests an AI-aided modeling and optimization approach to RF circuits, that combines the HSPICE simulation with the neural metaheuristic approaches, thus allowing amplifying the automation and accuracy of performance matching in circuits. Its first goal will be to overcome some of the drawbacks of manually sweeping parameters and using trial and error procedures in the common RF design process by adding a smart optimization loop that will improve important parameters including gain, return loss and power efficiency. The approach uses transistor-level modeling and waveform-accurate simulation by HSPICE and involves search of design space by a neural metaheuristic engine, i.e., the use of a hybrid neuralgenetic algorithm that explores the design space efficiently and computes no gradients. Iterative tuning is also automated so that fitness is judged using the result of the simulation and optimal parameters written back into HSPICE completing the loop. An example to prove the followed approach would be designing a Class-A RF power amplifier, at 2.4 GHz based on 0.18 micrometer CMOS technology. The optimization resulted in a 23% gain in small-signal gain, a 6 dB increase in return loss and a 30% decrease in weighted total harmonic distortion (THD) of the manually tuned designs. The difference in time of performing optimization exceeded 80 per cent, which illustrates the efficiency of the framework. The findings indicate that intensive applications of AI to conventional circuit simulators can tremendously shorten the RF design cycles and enhance the outcomes of performance. This project forms a starting point towards next generation automation of RF design across wireless and communication systems.