Design of an on-chip Hilbert fractal inductor using an improved feed forward neural network for Si RFICs

Akhendra Kumar Padavala,Bheema Rao Nistala

doi:10.3906/elk-1705-362

Abstract

Design of an on-chip Hilbert fractal inductor using an improved feed forward neural network for Si RFICs

Highlights

The rapid growth in wireless communication systems demands low cost, miniature radio frequency integrated circuits (RFICs)
The conventional neural network is a two hidden layer neural network each with 25 hidden neurons having the same input and output data parameters used in FNNPSOGSA
The results showed that the particle swarm optimization and gravitational search algorithm (PSOGSA) algorithm converges after 20 epochs with a mean square error of 0.0006781, whereas the LM algorithm converges after 50 epochs with a mean square error of 0.05

Summary

Introduction

The rapid growth in wireless communication systems demands low cost, miniature radio frequency integrated circuits (RFICs). Several heuristic optimization algorithms have been proposed to train the neural network such as genetic algorithm (GA) and simulated annealing (SA) in data mining [13] and microwave filter [14] applications Both these algorithms suffer from slow converge rates and tapering at local minima. The modelling of fractal inductors using ANN was not reported earlier.In this paper, an efficient feed forward neural network trained by hybrid particle swarm optimization and the gravitational search algorithm (PSOGSA) is proposed for designing complex Hilbert fractal inductors based on design specifications. Value 11.9 4 20 mΩ/2 2 μm 10 Ω cm effective fractal length, the width of metal traces, and frequency are considered as inputs to design the neural model. PSOGSA has an advantage of high convergence rate and it is does not tend to be tapered at local minima

PSOGSA learning algorithm

Results and discussion

Conclusion