Enhanced Photon Detection Probability Model for Single-Photon Avalanche Diodes in TCAD with Machine Learning

Abstract

Accurate photon detection probability (PDP) modeling is important for the optimized design of single-photon avalanche diodes (SPADs) using modern standard CMOS technologies. To ensure a planar active region of a SPAD, the edge of the depletion region must have a lower electric field, so a lower doping concentration is needed. However, this edge effect may have a negative impact on the total PDP, especially for small-sized SPADs. In this paper, we proposed an enhanced PDP modeling process by combining the Technology Computer-Aided Design (TCAD) simulations with machine learning (ML) techniques. Using this ML-TCAD PDP model, we investigated the influence of the edge effect on the PDP of SPADs by varying the diameter of the SPADs from 1.75 μm to 8.75 μm. After generating the sample simulation data, Gaussian process regression (GPR) and deep neuron network (DNN) are applied to train the model. With the application of principal component analysis (PCA), the accuracy of the trained models was significantly improved. Overall, this ML-TCAD PDP model provides an optimized and accelerated design process for SPADs, thus saving simulation time and reducing the design iterations required in the traditional design process of SPADs.