Microprocessor design verification by two-phase evolution of variable length tests
James E. Smith, Mike G. Bartley, Terence C. Fogarty · 2002
This paper discusses the use of a genetic algorithm to generate test programs for the verification of the design of a modern microprocessor. The algorithm directly learns sequences of assembly-code instructions which satisfy a coverage metric for one specific part of a design. The complexity of the design is such that it is not simple to predict in advance the length of the program needed to achieve coverage, and there is a severe time penalty for evaluating long tests. This has led to the development of a genetic algorithm which uses a two phase mechanism for variation in string length, through maintenance of a diverse population with varying lengths coupled with a "meta-algorithm" for periodic larger increases.