XOR-CiM: An Efficient Computing-in-SOT-MRAM Design for Binary Neural Network Acceleration

In this work, we leverage the uni-polar switching behavior of Spin-Orbit Torque Magnetic Random Access Memory (SOT-MRAM) to develop an efficient digital Computing-in-Memory (CiM) platform named XOR-CiM. XOR-CiM converts typical MRAM sub-arrays to massively parallel computational cores with ultra-high bandwidth, greatly reducing energy consumption dealing with convolutional layers and accelerating X(N)OR-intensive Binary Neural Networks (BNNs) inference. With a similar inference accuracy to digital CiMs, XOR-CiM achieves similar to 4.5x and 1.8x higher energy-efficiency and speed-up compared to the recent MRAM-based CiM platforms.

Automated summary

In this work, a digital Computing-in-Memory (CiM) platform called XOR-CiM is developed by leveraging the uni-polar switching behavior of Spin-Orbit Torque Magnetic Random Access Memory (SOT-MRAM). XOR-CiM converts MRAM sub-arrays to parallel computational cores with high bandwidth, reducing energy consumption and accelerating inference for X(N)OR-intensive Binary Neural Networks (BNNs). Compared to recent MRAM-based CiM platforms, XOR-CiM achieves similar inference accuracy but shows approximately 4.5x and 1.8x higher energy-efficiency and speed-up.