Parametric yield estimation of GaAs/AlGaAs multiple quantum well avalanche photodiodes using neural networks

GaAs/AlGaAs multiple quantum well avalanche photodiodes can be used for image capture in high definition systems. Since millions of devices are required for imaging arrays, it is critical to evaluate performance variations of individual devices. Even in a defect-free manufacturing environment, random variations in the fabrication process lead to varying device performance. Accurate performance prediction requires characterization of these variations. This paper presents a methodology for modeling the parametric performance of MQW APDs. This approach requires a model of the probability distribution of each process variable, as well as a model to account for correlation between measured process data and device performance metrics. These models enable the computation of the joint probability density function required for predicting performance using the Jacobian transformation. This density function can be numerically integrated to determine parametric yield. Neural networks are used as a tool for generating these models. It is shown that using a small number of test devices with varying active diameters, barrier and well widths, and doping concentrations enables prediction of APD gain and noise in larger populations of devices. This approach will allow device yield prediction prior to manufacturing to evaluate the impact of design decisions.