Fully-Binarized, Parallel, RRAM-based Computing Primitive for In-Memory Similarity Search

TL;DR

This paper introduces a fully-binarized, RRAM-based in-memory similarity search primitive that performs parallel XOR operations for efficient Hamming distance computation, validated through experiments and simulations.

Contribution

It presents a novel fully-binarized XOR-based in-memory similarity search scheme using RRAM arrays, enabling parallel match operations and energy-efficient computation.

Findings

Experimental validation on fabricated RRAM arrays confirms functionality.

Achieves 17 fJ energy per XOR operation in simulations.

Projected 1.5x energy savings at 28 nm nodes compared to state-of-the-art.

Abstract

In this work, we propose a fully-binarized XOR-based IMSS (In-Memory Similarity Search) using RRAM (Resistive Random Access Memory) arrays. XOR (Exclusive OR) operation is realized using 2T-2R bitcells arranged along the column in an array. This enables simultaneous match operation across multiple stored data vectors by performing analog column-wise XOR operation and summation to compute HD (Hamming Distance). The proposed scheme is experimentally validated on fabricated RRAM arrays. Full-system validation is performed through SPICE simulations using open source Skywater 130 nm CMOS PDK demonstrating energy of 17 fJ per XOR operation using the proposed bitcell with a full-system power dissipation of 145 $\mu$W. Using projected estimations at advanced nodes (28 nm) energy savings of $\approx$1.5$\times$ compared to the state-of-the-art can be observed for a fixed workload. Application-level validation is performed on HSI (Hyper-Spectral Image) pixel classification task using the Salinas dataset demonstrating an accuracy of 90%.