In this paper, using the realistic market data of NAND flash memory in the past decade, we provide theoretical arguments and empirical evidence for how genetic algorithms (GA) can he used for efficient estimation of macro-level diffusion models. Under the comparison of two methods, we perform two parts; one is the estimation of growth of single generation of NAND flash memory and the other is the estimation of growth of multi-generation NAND flash memory. In the first part, we find the estimated ability of GA is as good as nonlinear least square (NLS), but GA can overcome initial guess problems that NLS couldn't. In the second part, we find the result of estimation by NLS method cannot converge. Thus, GA is superior to NLS when we perform the estimation of multi-generation flash memory. According to the preliminary results above, we could conclude that GA is suited for estimation of diffusion model than that of the NLS; in particular, for multi-generation NAND flash memory.