High-performance neuromorphic computing architecture of brain

TL;DR

This paper proposes a neuromorphic computing architecture inspired by the brain's neural spheres, demonstrating high storage capacity and energy efficiency surpassing current computer chips, based on chaos dynamics and fractal theory.

Contribution

It introduces a novel brain-inspired architecture based on neural spheres, integrating chaos and fractal principles for energy-efficient computation.

Findings

Predicts a storage capacity of 7.48×10^18 Bytes.

Achieves a computational power of 6.24×10^18 FLOPS.

Energy efficiency up to 79%, surpassing current chips.

Abstract

Artificial intelligence can outperform humans in specific tasks but consumes substantial energy. How the human brain can work at just 20 watts with complex cognitive intelligence? Here we decode the fundamental information strategy unit of brain, neural sphere, which agglomerates neurons into sphere to achieve energy-efficient and exhibits many ultra-long period or random electrophysiological activities. Chaos dynamics and fractal theory demonstrated the mathematical principle of neural spheres to memorize and process through different electrophysiological activities which depend on strange attractors. A high-performance neuromorphic computing architecture of brain was then constructed which predicts a storage capacity of $7.48\times 10^{18}$ Bytes and a computational power of $6.24\times 10^{18}$ FLOPS for human brain. At this capacity, the energy efficiency of the human brain after long-term evolution can be up to 79% via Landauer's principle, 8-order higher than that of the latest computer chips, supporting the rationality of the proposed architecture.