Work

Brain-Inspired Hyperdimensional Learning for Resource-Constrained and Memristive Computing Hardware

Year: 2026
Type: article
Abstract: The proliferation of Internet of Things devices and the increasing demand for artificial intelligence at the edge have severely strained traditional computing paradigms, particularly the Von Neumann architecture. Memory bottlenecks and prohibitive power consumption necessitate a paradigm shift toward brain-inspired neuromorphic systems. Hyperdimensional computing has emerged as a promising computational framework that mimics human memory and cognition by computing with high-dimensional, holographic vectors. This paper presents a comprehensive investigation into the integration of hyperdimensional learning algorithms with emerging memristive hardware, addressing the critical challenges of executing intelligent workloads in resource-constrained environments. By leveraging the intrinsic error resilience of hyperdimensional spaces, we demonstrate that memristive crossbar arrays can execute complex vector-matrix multiplications efficiently in the analog domain, circumventing the non-idealities typical of nanoscale devices. The methodology encompasses novel vector encoding schemes tailored for spatial mapping onto non-volatile memory arrays, coupled with associative memory learning algorithms optimized for low-precision analog hardware. Detailed circuit-level simulations evaluate the proposed architecture across standard classification benchmarks, including image recognition and human activity detection. The findings indicate that memristive hyperdimensional computing achieves state-of-the-art energy efficiency and latency while maintaining classification accuracies comparable to conventional deep neural networks. Furthermore, the inherent robustness of high-dimensional representations successfully mitigates the impact of conductance drift and device variations inherent to memristive technologies. Ultimately, this research establishes a robust theoretical and practical foundation for deploying hyperdimensional intelligence on next-generation edge devices.
Language: English
Open Access status: diamond
OpenAlex ID: W7220307737
FWCI:
Cites: 27
Cited by:
Related to: 10