Clo-HDnn: A 4.66 TFLOPS/W and 3.78 TOPS/W Continual On-Device Learning Accelerator with Energy-efficient Hyperdimensional Computing via Progressive Search

2025-07-23

AI Summary

To address the low energy efficiency and high computational overhead in edge-device continual learning, this work proposes a domain-specific accelerator architecture based on hyperdimensional computing (HDC). Methodologically: (i) it introduces a gradient-free continual learning mechanism that eliminates backpropagation; (ii) it incorporates a low-cost Kronecker-product-based hypervector encoder coupled with weight re-clustering for feature extraction; and (iii) it supports dual-mode operation and progressive search, encoding only a subset of query hypervectors. Experimental results demonstrate that the accelerator achieves 4.66 TFLOPS/W and 3.78 TOPS/W energy efficiency on feature extraction and classification tasks, respectively—outperforming state-of-the-art accelerators by 7.77× and 4.85×. Computational complexity is reduced by up to 61%. This is the first systematic application of HDC to hardware acceleration for edge continual learning, delivering significant improvements in energy efficiency and real-time performance while preserving accuracy.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Systems and Infrastructure for Web, Mobile and WoT: Cloud, edge and content delivery systems for the Web

Abstract

Clo-HDnn is an on-device learning (ODL) accelerator designed for emerging continual learning (CL) tasks. Clo-HDnn integrates hyperdimensional computing (HDC) along with low-cost Kronecker HD Encoder and weight clustering feature extraction (WCFE) to optimize accuracy and efficiency. Clo-HDnn adopts gradient-free CL to efficiently update and store the learned knowledge in the form of class hypervectors. Its dual-mode operation enables bypassing costly feature extraction for simpler datasets, while progressive search reduces complexity by up to 61% by encoding and comparing only partial query hypervectors. Achieving 4.66 TFLOPS/W (FE) and 3.78 TOPS/W (classifier), Clo-HDnn delivers 7.77x and 4.85x higher energy efficiency compared to SOTA ODL accelerators.

Problem

• Designs an on-device continual learning accelerator
• Optimizes accuracy and efficiency with hyperdimensional computing
• Achieves high energy efficiency for learning tasks

Innovation

• Hyperdimensional computing for efficient learning
• Gradient-free continual learning for knowledge storage
• Progressive search reduces complexity by 61%