Artificial Neural Network Based model for single event upset current prediction in 14 nm FinFET

A surrogate model of Single Event Upset (SEU) transient current for circuit-level modeling is presented in this work using machine learning regression. The use of FinFET technology in defense and aerospace-integrated circuits has been made possible by its extraordinary success in terms of integration and performance compared to planar MOSFETs. Their susceptibility to radiation risks in these kinds of applications is still the main worry, though. In order for circuit designers to minimize radiation impacts prior to real manufacture, the semiconductor industry has refocused its efforts on device analysis prior to production due to recent technological advancements. The structure and device parameters are being designed using Technology Computer-Aided Design (TCAD) tools. Nevertheless, these techniques require a lot of effort and processing. The viability of employing machine learning to forecast device characteristics and Single Event Transient (SET) current through an unsupervised learning technique is investigated in this work. The TCAD tool is used to create a 14nm 3D FinFET device, and a dataset with multiple parameters is produced. A feedforward neural network and a Random Forest Regressor model are trained using this dataset in order to forecast the SET pulse current. To test this technique and forecast SET current, a section of the 10% dataset was selected at random. The accuracy of the comparison between the actual and anticipated data is good. This demonstrates that TCAD can be replaced by machine learning models to speed up device performance assessment for large-scale circuits. In comparison, our suggested approach predicts the SET current in just 8–10 seconds. SET impacts can be lessened at the design process with the use of this study.