Optimization Of Digital Circuits Based On Improved Genetic Algorithm

In recent years, genetic algorithm is adopted in circuit design, which developed Evolvable Hardware (EHW). As a result, circuit can be designed without large quantity of knowledge, thus effectively reduce the design cost. However, when classic genetic algorithm is utilized in the optimization of digital circuits, the diversity of its later population is reduced, which could easily cause problems such as premature convergence, slow convergence, etc.The thesis introduces the research status of EHW and genetic algorithm, and describes the operating principle of EHW. Based on these, the utilization of genetic algorithm in EHW is studied. The main contents are listed as followings:(1)When classic genetic algorithm is utilized in the optimization of digital circuits, as the low probability of low fitness individual being selected, the population’s diversity is reduced, which could easily lead to premature convergence. In order to solve this problem, the paper proposes a new selection mode, the modified roulette mode. In this mode, the population is firstly numbered by its fitness, and then grouped by its number. By artificially set the selection probability of each group, the selection probability of low fitness individuals is enhanced.(2)In order to shorten the optimization time enlarged by the modified roulette mode, the best individuals are reserved in the selection process. The relatively high fitness individuals in a population, the number of which is βN (where N is the number of the population, m is a parameter and0.02<β<0.05), are directly multiplied to the next generation without intersection or mutation. By utilizing this selection method, the optimization time could be shortened.(3)After evaluating the fitness function, the classic generic algorithm directly goes to intersection and mutation, thus the population’s distribution information cannot be made full use. In the later phase of circuit optimization, premature convergence could be developed as a result of the concentration of high fitness individuals. Based on statistical principles, genetic programming is introduced in circuit optimization process. By counting each population’s fitness function values, genetic programming evaluates population’s fitness distribution, then programs the population based on the evaluation result, thus increase population’s diversity.