Artifical Intelligence Parameter Extraction Methods

In this chapter, we introduce a pioneering approach employing three distinct machine learning (ML) models: Artificial Neural Network (ANN), Logistic Regression (LR), and Decision Tree (DT). These models are employed for the comprehensive analysis and extraction of internal parameters of Schottky photodiodes (SPDs), all while operating in the absence of any prior knowledge of the intricate and nonlinear thermionic emission (TE) expression governing the device current. Our methodology is validated through extensive training, evaluation, and demonstration of the ML models across a dozen proprietary datasets. These datasets represent the current responses of graphene oxide (GO) doped p-Si Schottky barrier diodes (SBDs) under varying conditions, encompassing both different levels of GO doping (0, 1, and 10%) and illumination levels spanning from complete darkness (0 mW/cm2) to 30 mW/cm2. Notably, the predictions made by these models are consistent, even at an intensity of 60 mW/cm2. The training dataset incorporates independently calculated values of barrier height ( \(\phi \) ), ideality factor (n), and series resistance ( \(R_s\) ), all derived using the Cheung-Cheung method, for each diode. Subsequently, the models accurately predict these parameters at unspecified intensities, including 80 and 100 mW/cm2 on the model development data and 5% and 20% GO doping levels, which were not part of the development dataset. Remarkably, the ANN achieves an outstanding training and test accuracy of 100% and 99.99%, respectively, across most datasets, while the LR and DT models attain a validation and test accuracy of 100%. This demonstrates the ML models’ remarkable capacity to efficiently grasp the photodiodes’ photo responses and predict the internal SBD parameters with exceptional accuracy, all without relying on any inherent comprehension of the thermionic emission (TE) equation for SBDs. Our proposed ML models hold the potential to significantly streamline the analysis phase in device development cycles and can be readily extended for application to diverse datasets across various domains.