Variation analysis of spintronic device using machine learning algorithm

In this article, the focus is on using machine learning methods to analyse non-volatile memory devices. This is important because the production of integrated circuits in the sub-micrometre range depends on advancements in manufacturing process technology, and it is crucial to evaluate how manufacturing tolerances affect the functionality of contemporary integrated circuits. Traditionally, Monte Carlo-based techniques have been used to accurately evaluate the impact of manufacturing tolerances on the functionality of integrated circuits. However, these techniques are computationally time-consuming. We will propose a scheme to "learn" the variation of the read margin (parallel and anti-parallel resistance) performance of spintronics devices. The machine learning approach, artificial neural network, is focused on this study (Read margin of spin transfer torque (STT)) spintronics devices. The accuracy for STT by Artificial Neural Network (ANN) is calculated with the help of the MATLAB deep learning toolbox. Regression models using machine learning (ML) are fast and precise over a variety of input ranges, making them ideal for device modelling. The ML algorithm has emerged as a potential substitute for Monte Carlo-based techniques. It can reduce the computational load needed in a Monte Carlo simulation covering all process corners, design parameters, and operating conditions. The article demonstrates the effectiveness of the ML algorithm by performing various simulations on spin transfer torque (STT) non-volatile memory. The proposed scheme involves "learning" the variation of the read margin performance of spintronic devices as a function of its material and geometric parameters. In conclusion, the use of machine learning techniques based on the different regression methods is a promising approach to increase the prediction time of result analysis as compared to SPICE simulation time.