Waveform learning based on a genetic algorithm and its application to signal integrity improvement

Abstract

A novel waveform generator and its design methodology are proposed in this paper. In the idea, no complicated circuits but only a trace in a printed circuit board is used as the generator: the trace is divided into multiple segments of different widths and lengths. Any desired output waveforms can be generated by adjusting each segment's width and length because the multiple reflection waves occur in the trace and they are superposed onto the input waveform. The adjusting, or the width-length design however comes to a combinatorial explosion problem. In order to overcome the design difficulty we use a genetic algorithm (GA) by mapping the segmentally divided transmission lines onto chromosomes in the GA, and make them learn the desired output waveform. We apply the proposed waveform generator to signal integrity improvement in high-speed interconnections, and demonstrate its remarkable efficiency using a prototype board.