Neural network based modeling of diffusion process for high-speed avalanche photodiodes fabrication

This paper presents the modeling methodology of Zn diffusion process utilized for high-speed avalanche photodiode fabrication using neural networks. The modeling scheme can characterize the effects of diffusion process conditions on the performance metrics of diffusion process. Three different InP-based test structures with different doping concentrations in diffused layer are explored. Three input factors (sealing pressure, amount of Zn ₃P₂ source per volume, and doping concentration of diffused layer) are examined with respect to the two performance metrics (diffusion-rate and Zn doping concentration) by means of D-optimal design experiment. Diffusion rate and Zn doping concentration in diffused layer are characterized by a response model generated by training feed-forward error back-propagation neural networks. It is observed that the neural network based process models developed here exhibit good agreement with experimental results.