Automatic Regression Framework for MRAM Compact Models Calibration including Stochasticity

Abstract

This paper introduces a comprehensive optimization framework designed to enhance the regression of physical parameters for MRAM technologies, including STT-MRAM, VCMA, and SOT devices. As MRAM emerges as a promising candidate for next-generation memory technologies, the accuracy of physical models becomes crucial for the development and optimization of these devices. However, regressing precise physical parameters from device measurements poses significant challenges due to the complex nature of the phenomena involved and the noise inherent in experimental data. Our framework addresses these challenges by implementing advanced data processing techniques to clean and preprocess measurement data, facilitating the accurate calibration of both electrical and magnetic switching physical models. Furthermore, it incorporates statistical models to account for device variations and intrinsic stochastic behavior, offering a robust solution for the optimization of MRAM technologies. The efficacy of our framework is demonstrated through comprehensive simulations and experimental validations against in-house STT, SOT, and VCMA-MRAM devices. The framework fills the gap between test structures and circuit compact models, required for the development of future MRAM-based applications.