Machine Learning based Single Photon Avalanche Diode Device Modeling

This paper presents machine learning (ML) based novel single photon avalanche diode (SPAD) device modeling. We investigate three popular ML models, Extreme Gradient Boosting (XGBoost), Long-Short Term Memory (LSTM) and Feed Forward Neural Network (FFNN) for predicting the I-V characteristics of SPAD devices. Due to the data-driven nature, all of these models are viable alternatives to Technology Computer-Aided Design (TCAD) and can significantly reduce the computational cost. We trained and evaluated each of the models using a large amount of real-world I-V data produced through TCAD simulations and compared their performances. The FFNN model demonstrated a smooth convergence through the learning period with a coefficient of determination <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(R^{2})$</tex> of 0.9779 which means that it can generalize the model well. In comparision, LSTM and XGBoost outperformed FFNN with a higher <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> of 0.9835 and 0.9929 respectively. Therefore, XGBoost captures the overall trend of the data better among these three models. Based on these findings, the XGBoost model proves to be more efficient in predicting the I-V characteristics of SPADs.