An ANN-Based Approach to the Modelling and Simulation of Analogue Circuits

Abstract

This paper explores the automatic creation of behavioural models of analogue circuits from simulation data. Artificial neural networks (ANNs) are used to model the circuit, speeding up the simulations and providing residual error results when compared with an off-the-shelf simulator. Since the pair input-output is generated through SPICE simulation, the model is trained in a supervised manner. This work proposes to model the circuit behaviour using a multilayer perceptron with delay lines. As a novelty, it introduces an approach that can model the circuit behaviour for different circuit sizes. The proposed method was applied to a set of amplifiers and the results obtained show the effectiveness of the model in behavioural modelling of analogue circuits. A generator capable of converting the Python ANN to Verilog-A was also developed and used to convert the model from Python to this hardware description language, so that the models are ready to be integrated with the circuit simulator. Simulating the circuit using the developed models was 5 times faster than simulating it at the transistor level.