MIMHD: Accurate and Efficient Hyperdimensional Inference Using Multi-Bit In-Memory Computing

TL;DR

This paper introduces MIMHD, a multi-bit in-memory hyperdimensional computing platform that significantly improves accuracy and energy efficiency over binary approaches by utilizing multi-bit HVs and specialized hardware.

Contribution

The paper presents a novel multi-bit in-memory HDC inference platform with hardware-aware retraining, achieving higher accuracy and efficiency than binary HDC implementations.

Findings

MIMHD with 3-bit HVs achieves 92.6% accuracy, 8.5% higher than binary.

MIMHD offers 84.1x energy improvement over GPU.

MIMHD with 3-bit HVs is 4.3x faster and more energy-efficient than binary HDC accelerators.

Abstract

Hyperdimensional Computing (HDC) is an emerging computational framework that mimics important brain functions by operating over high-dimensional vectors, called hypervectors (HVs). In-memory computing implementations of HDC are desirable since they can significantly reduce data transfer overheads. All existing in-memory HDC platforms consider binary HVs where each dimension is represented with a single bit. However, utilizing multi-bit HVs allows HDC to achieve acceptable accuracies in lower dimensions which in turn leads to higher energy efficiencies. Thus, we propose a highly accurate and efficient multi-bit in-memory HDC inference platform called MIMHD. MIMHD supports multi-bit operations using ferroelectric field-effect transistor (FeFET) crossbar arrays for multiply-and-add and FeFET multi-bit content-addressable memories for associative search. We also introduce a novel hardware-aware retraining framework (HWART) that trains the HDC model to learn to work with MIMHD. For six popular datasets and 4000 dimension HVs, MIMHD using 3-bit (2-bit) precision HVs achieves (i) average accuracies of 92.6% (88.9%) which is 8.5% (4.8%) higher than binary implementations; (ii) 84.1x (78.6x) energy improvement over a GPU, and (iii) 38.4x (34.3x) speedup over a GPU, respectively. The 3-bit $\times$ is 4.3x and 13x faster and more energy-efficient than binary HDC accelerators while achieving similar accuracies.