Abstract: In this paper, we present a neural-based manufacturing process control system to improve the lot yield of wafer lab. The process is as follows: 1. A model based on feedforward neural networks is proposed to simulate the wafer manufacturing process. Learning from the historical technological records with a special dynamic learning method, the neural-based model can approximate the function relationship between the technological parameters and the wafer yield precisely. 2. We also use a gradient-descent method to search a set of optimal technological parameters that lead to the maximum yield by simulation. 3. We optimize the specifications of the wafer fab according to the optimal parameters. The wafer yield increases by 11.2 % after the optimized specifications are applied to the wafer fabrication assembly.