Summary: Moving beyond traditional views that treat memory and computation as static synaptic weights or isolated neuronal spikes, researchers modeled local functional circuits as parameterized digital neural ensembles embedded within a multi-scale “neural sphere.”
Their analysis shows that information in the brain is carried by collective spatio-temporal electrical patterns governed by nonlinear dynamics, chaos, and fractal behavior. The framework indicates the brain’s information processing operates extremely close to thermodynamic limits—requiring only about 1.26 times the Landauer limit and achieving up to 79% energy efficiency—while projecting a theoretical maximum memory capacity of 7.48 × 1018 bytes.
Key Facts
- Near-Landauer Computational Efficiency: The model estimates energy consumed for neural computation and communication at roughly 1.26 times the Landauer limit (the theoretical minimum energy to erase one bit), corresponding to an operational energy efficiency of about 79%.
- Sharp Contrast with Current AI Silicon: Contemporary semiconductor AI hardware exceeds the Landauer limit by roughly 109 (one billion) times, suggesting substantial opportunities for energy-efficiency improvements in neuromorphic chip design.
- Revised Memory Bound: The theoretical maximum storage capacity is estimated at 7.48 × 1018 bytes—approximately three orders of magnitude greater than estimates based on linear synaptic summation.
- Equivalent Computational Throughput: The brain’s modeled computational throughput reaches up to 6.24 × 1018 floating-point operations per second (FLOPS), about the equivalent of 78,000 modern high-performance GPUs.
- Spatio-Temporal Pattern Encoding: Instead of encoding information solely in single synapses or neurons, the brain appears to use initial conditions across neural ensembles to generate periodic potential waveforms that function as dynamic physical carriers of information.
Source: Science China Press
The human brain is one of nature’s most energy-efficient intelligent systems. With roughly 20 watts of metabolic power it supports perception, memory, learning, reasoning and motor control. How the brain handles such complex information processing with so little energy remains a central question in neuroscience and information science.
In a recent paper published in National Science Review, Jinxuan Ma and Wanlin Guo from Nanjing University of Aeronautics and Astronautics propose a theoretical framework the authors call highly energy-efficient information-handling dynamics of the brain. The work reframes how we think about memory, computation and energy use in biological neural networks.
The study offers a physical perspective on how large amounts of information may be stored and processed through dynamic electrophysiological activity, emphasizing the spatio-temporal organization of ensembles rather than isolated spikes or static synaptic weights.
Traditional models often focus on single-cell action potentials, synaptic transmission, or region-to-region connectivity to explain brain function. While those approaches have driven enormous progress, they do not fully account for the brain’s remarkable energy efficiency. Ma and Guo shift the emphasis to the coordinated, multi-dimensional dynamics of neural ensembles and how these dynamics interact with metabolic constraints.
In their simulations, the authors abstract local functional circuits into neural ensembles and represent biological neurons as parameterized digital units that capture morphology and electrophysiological properties. Using an energy-minimization principle, they construct a neural sphere model that unifies structure and function and permits simulation of local brain activity across scales and under metabolic limits. The team systematically explored ensembles with varied topologies, initial membrane potential states, and neuronal population sizes.
Their simulations demonstrate that neuronal populations can generate rich, periodic electrophysiological activity. Rather than storing information in single synapses or neurons, the brain may encode information in dynamic electrical patterns that arise from the coordinated behavior of many cells. These spatio-temporal patterns—describable using nonlinear dynamics, chaos theory and fractal mathematics—serve as dynamical carriers of information that can preserve initial-condition-dependent signals and participate in ongoing computation.
Interpreting these dynamical waveforms as information carriers leads to striking quantitative implications. The model predicts a maximum storage capacity of 7.48 × 1018 bytes for the human brain, roughly 1,000 times larger than earlier synapse-count-based estimates. It also predicts peak computational power around 6.24 × 1018 FLOPS, comparable to tens of thousands of modern GPUs. Perhaps most notable, the estimated energy cost for brain computation and communication is only about 1.26 times the fundamental Landauer limit, implying energy efficiency of up to 79%—orders of magnitude better than current silicon-based AI processors.
These results suggest that the brain’s efficiency arises not just from biophysical components but from how information is organized across space and time in collective dynamics. Encoding data in periodic, fractal-like electrophysiological attractors allows a high-density state space for memory and low-cost transitions for computation.
Key Questions Answered
Q: What is the Landauer limit and why does it matter here?
A: The Landauer limit is a thermodynamic lower bound on the energy required to erase one bit of information. Demonstrating that brain computation operates at about 1.26 times this limit indicates neural networks can approach fundamental physical efficiency bounds for information processing.
Q: Why does this model yield a storage estimate three orders of magnitude larger than older models?
A: Earlier estimates add up static synaptic states linearly. This framework treats neural ensembles as nonlinear, high-dimensional dynamical systems in which information is encoded by spatio-temporal and fractal waveform patterns, greatly expanding the available representational state space.
Q: How could these findings influence future AI hardware?
A: Current AI chips consume far more energy relative to physical limits. By emulating the brain’s sparse, collective spatio-temporal dynamics rather than relying on brute-force clocked computation, engineers may be able to develop neuromorphic systems that dramatically reduce power use while maintaining or increasing computational capacity.
Editorial Notes
- This article was edited by a Neuroscience News editor.
- The underlying journal paper was reviewed in full.
- Additional context was provided by staff editors.
About this research summary
Author: Bei Yan
Source: Science China Press
Contact: Bei Yan – Science China Press
Image credit: Neuroscience News
Original Research: Open access. “Highly energy efficient information handling dynamics of brain” by Jinxuan Ma (马 锦 暄) and Wanlin Guo (郭 万 林), National Science Review. DOI: 10.1093/nsr/nwag373
Abstract
Highly energy-efficient information handling dynamics of the brain
Artificial intelligence can outperform humans on specific tasks but typically at extremely high energy cost. This study shows how the brain can group neurons into coherent spheres to produce energy-efficient, ultra-long-period electrophysiological activity. Chaos dynamics and fractal theory explain how the brain forms electrophysiological strange attractors through coordinated spatio-temporal activity, enabling memory and computation. The neural-sphere information-handling framework predicts a potential storage capacity of 7.48 × 1018 bytes and computational power of 6.24 × 1018 FLOPS for the human brain, with energy efficiency up to 79% relative to Landauer’s principle—orders of magnitude higher than modern computer chips.