BP Neural Network Modeling and Solving Acceleration of Analog ICs

A novel accuracy-aware modeling and solving approach of adopting a back-propagation neural network (BP-NN) targeting to a classical analog five-transistor operational transconductance amplifier (OTA) is proposed in this paper. Three complex BP-NN algorithms are described amply in performance model training between multiple design factors and performance metrics, where genetic algorithm (GA), particle swarm optimization (PSO) and mind evolutionary algorithm (MEA) are introduced to further speed up the searching process for global solutions. Effectiveness comparison is performed by running simulations using SMIC 180 nm/1.8 V CMOS technology, three complex algorithms with longer time-consuming of 10 h, 30 m and 28 m demonstrate significantly superior features in solving accuracy, which show lower mean square errors (MSEs) by improvements of 82.7%, 99.7% and 99.9%, respectively. The results show that the proposed BP-NN modeling approach can be effectively aided in accelerating the global design solution of a long-cycle, nonlinear and multivariable-solved analog circuit.

Automated summary

This paper proposes a novel accuracy-aware modeling and solving approach using a back-propagation neural network (BP-NN) for a classical analog five-transistor operational transconductance amplifier (OTA). Three complex BP-NN algorithms are described in detail for performance model training, utilizing genetic algorithm (GA), particle swarm optimization (PSO), and mind evolutionary algorithm (MEA) to speed up the search for global solutions. Simulations using SMIC 180 nm/1.8 V CMOS technology show that the algorithms have significantly superior features in solving accuracy, with improvements in mean square errors (MSEs) of 82.7%, 99.7%, and 99.9% respectively. The results demonstrate that the proposed BP-NN modeling approach effectively aids in accelerating the global design solution of long-cycle, nonlinear, and multivariable-solved analog circuits.