Highly energy-efficient information-handling dynamics of the brain

ABSTRACT Artificial intelligence can outperform humans in specific tasks but at extremely high energy cost. Here we show that the brain can agglomerate neurons into a sphere to achieve energy-efficient, ultra-long-period electrophysiological activities. Chaos dynamics and fractal theory demonstrated the principle of the brain to memorize and process information through the formation of electrophysiological strange attractors under the global cooperation of neurons in the spatiotemporal structure. The neural sphere-based information-handling dynamics of the brain can then be constructed, predicting a storage capacity of 7.48 × 1018 bytes (3 orders higher than the previous estimates) and a computational power of 6.24 × 1018 floating-point operations per second (FLOPS) (∼78 000 modern GPUs) for the human brain with a surprising energy efficiency up to 79% via Landauer’s principle, 8 orders higher than that of the latest computer chips.