Abstract: To improve the photolithography precision of Si support system, some main process parameters in soft bake, exposure, development were analyzed with orthogonal experiment. The influence degree of the parameters, including soft bake time, exposure time, development time and temperature, and their best combination were found. On this basis, a proper back propagation (BP) neural network was designed and trained and the experimental data were analyzed and optimized by the BP neural network. Verification experiments supported the correctness of the BP neural network. When the photoresist is 1.55 m thick, the optimum process parameters are: soft bake temperature 100 ℃, time 90 s; exposure time 5 s; development temperature 15 ℃, time 90 s. On this condition, the linewidth error can be less than 0.3 m.