The plasma cleaning process cleaning the residual containments in the assembly processes is one of the major processes in the plastic ball grid arrays (PBGA) assembly processes. Especially, when there is any micron particle existed before wire bonding and molding, parts of nonconforming will increase. With the rapid growth in production, yield improvement to reduce the production cost is essential for PBGA industry. In fact, the plasma cleaning process before wire bonding and molding both enhances the PBGA yield and reduces the production cost effectively. This study combines neural network with Taguchi method to structure a well-trained prediction model, further searches for the optimal plasma cleaning parameter design through genetic algorithm, and finally enhances the process yield and production quality in a central Taiwan PBGA company.