MACHINE LEARNING BASED ALGORITHM FOR SINGLE EVENT TRANSIENT CURRENT PREDICTION IN 14NM FINFETS

Abstract

In this thesis a novel machine learning regression-based model for circuit-level simulation of Single Event Upset (SEU) transient current. The phenomenal success of FinFET technology in terms of the level of integration and performance over planar MOSFETs have paved their way for usage in aerospace integrated circuits and defence applications. But their sensitivity to radiation hazards in such applications remains the primary concern. With the recent advancement in technology, the semiconductor industry has shifted its focus on device analysis before the actual fabrication so that radiation effects may be mitigated by the circuit designers before actual fabrication. The Technology Computer-Aided Design (TCAD) tools are being used to design the structure and analyse the device parameters. These tools are computation exhaustive and time consuming. This work tries to explore the feasibility of using machine learning for predicting device parameters and Single Event Transient current (SET) using unsupervised learning technique. A 14nm 3D FinFET device was designed using TCAD tool and a dataset with various parameters was generated. This dataset was used to train a SET Model using a Random Forest Regressor. A feedforward neural network-based SET pulse current model was also developed and compared with the previous model. The 10% dataset was randomly chosen as a subset to test this algorithm and predict SET current. The comparison between actual and predicted data shows a high possibility to supplement TCAD for accelerating device performance analysis for large-scale circuits. This study may be further expanded to other digital & analog circuits to help designers mitigate SET effects in design phase.