Verification Technology Driven By Coverage Based On Genetic Algorithm

With the development of integrated circuit industry, consistence between RTL description and definition of function has become a critical issue in verification. Verification driven by coverage is widely accepted for higher reliability and effectiveness; however, the vector-generating technology will stimulate the DUV by heavily repeated vectors, which leads to the slow convergence of coverage and low effectiveness in the whole process of IC designing.In this thesis, firstly a method is proposed to apply Simple Genetic Algorithm (SGA) to verification driven by coverage. This method makes a definition of the SGA encoding hierarchy in verification, involves the functional coverage of different generations in the fitness calculation, and gets the inheritance strategy including proportion of selection operator, uniform operator and an improved mutation operator. Based on the results experimented on NCC accelerator in Matlab and VMM platform, the proposed method is proved to achieve more efficient than total-random method and other SGA using other parameters and strategies.This thesis further presents a novel Elite Genetic Algorithm (EGA), which will be applied to verification driven by coverage. The method upgrades the SGA using elites, constructs a vector album with a specific structure and content, and fuels the algorithm by combining the elite-keeping and elite-crossing strategy. Besides an adaptive crossing function is proposed to achieve crossing operation. The experiments reveal the significant superiority of the EGA comparing to not only total-random method but SGA.