SOT-MRAM Memories for Energy Efficient Embedded and AI Applications

Modern data intensive and intelligent applications are slowly gearing to be a hardware deployable technology with the advancements in non-conventional computing architectures. With this comes the major challenge of holistic efficiency particularly low power and high-performance implementations in the light of memory system architectures. In this context, modern spintronics based non-volatile memory technologies (NVM) such as spin–orbit torque has significant advantages over existing SRAM with advantages like non-volatility, ultra-low power consumption, higher density, and scalability. In this article we present and demonstrate the simulation framework for the evaluation of energy-time performance of computing architectures running embedded and AI applications. The suitability of spintronic based SOT-MRAM for modern energy-efficient computations and relevant integration architectures in comparison to existing technology is established. Our investigations reveal that SOT-MRAM is ~7% faster, >30% energy efficient and ~40% better in EDP than SRAM technology and therefore can be better choice for modern computing architectures.