MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202621080176 A) filed by Indian Institute Of Technology Gandhinagar on June 29, 2026, for Synthetic Antiferromagnetic Free-Layer Mram Based Compute-In-Memory Accelerator System For Binarized Neural Networks.
Inventors include Verma, Gaurav; Das, Subham; and Sisodia, Naveen.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: ABSTRACT SYNTHETIC ANTIFERROMAGNETIC FREE-LAYER MRAM BASED COMPUTE-IN-MEMORY ACCELERATOR SYSTEM FOR BINARIZED NEURAL NETWORKS A synthetic antiferromagnetic free-layer MRAM based compute-in-memory accelerator system for binarized neural networks (100) is disclosed. The system comprises a plurality of magnetic random access memory (MRAM) synaptic devices arranged in one or more crossbar arrays, at least one processing unit configured to perform in-memory neural network computations, at least one peripheral circuit comprising sensing, read, write, analog-to-digital conversion, and control circuitry, and a memory control circuitry configured to execute multiply-accumulate, XNOR, and population-count operations within the crossbar arrays. Conductance states of the MRAM synaptic devices represent neural network weights for the execution of in-memory computations. Each MRAM synaptic device comprises a magnetic tunnel junction having a synthetic antiferromagnetic composite free layer configured to undergo magnetic switching based on combined spin-transfer torque and spin-orbit torque mechanisms. The disclosed architecture enables energy-efficient, scalable, and high-throughput neuromorphic and artificial intelligence processing.
Disclaimer: Curated by HT Syndication.
