Abstract: The authors present an idea of implementing genetic algorithm in neural network so as to break the time-consuming bottleneck of GA (genetic algorithm). They propose the architecture of the NN (neural network) based model for GA, and generalize both the crossover operation and mutation operation to the so-called all-crossover and multi-point mutation operations. Through a group famous test functions, they compare this algorithm and typical genetic algorithm in optimization of solution search. This study is significant for using hardware to implement GA and explicitly realizing the inherent parallelism of GA so as to improve the real time performance of GA and broaden the application range of GA.