Neural network based MOSFET channel length and width decision method for analogue integrated circuits

Abstract

The channel length and width of a MOSFET are two important parameters selected by the experience of the integrated circuit designer. Since drain current of a transistor is directly adjusted by the aspect ratio, the wrong selection of these parameters changes the circuit characteristics. In this work neural networks are used to decide the most suitable selection of channel length and width of MOSFET. Both p-channel and n-channel transistors are modelled by multi layer perceptron (MLP) neural network and the channel length and width are predicted by MLP. MOSFET level 3 is modelled by MLP, training and test data are obtained from HSPICE design environment with YITAL 1.5 mu parameters. Developed network is tested with the current mirror and the diffential amplifier circuits. Estimated aspect ratios for each transistor are compared with the HSPICE simulation results.