Every day, the human brain carries out trillions of calculations while consuming roughly the same amount of electricity as a dim light bulb. Despite decades of research, scientists have yet to fully understand how it stores such vast amounts of information and processes it with such remarkable efficiency.
A new study, grounded in physical principles, offers an unexpected answer. Instead of viewing neurons as independent processors, researchers argue that groups of neurons work together as collective systems. The findings were published in the journal National Science Review.
The model developed by the team suggests that the brain operates near the theoretical limit of energy efficiency. It also indicates that the brain can store about a thousand times more information than previously thought, and could serve as a blueprint for creating far more efficient artificial intelligence.
For decades, neuroscience has explained thinking through individual parts of the brain. Researchers have traced how a single neuron activates, how a synapse transmits a signal, and how brain regions interact with each other. This work laid the foundations of modern neuroscience. However, it never fully explained why the brain is so economical with energy.
Wanglin Guo, a professor at Nanjing University of Aeronautics and Astronautics (NUAA), has spent his career studying the physics of materials and energy, not neurons. Most of his best-known work focuses on nanomaterials and generating electricity from moving water. His team, led by Jinxuan Ma, brought this physical approach to the study of the brain. The key idea was to stop looking at neurons individually. Instead, they considered the entire group as a single entity and asked what they could do together.
To study the group as a whole, the team created a simplified model of a neuron. Each one replicated the basic shape and electrical behavior of a real cell. They grouped many of these digital neurons into a sphere and allowed it to settle into a position with the lowest energy. They call this cluster a neural sphere.
The researchers then ran the model. They varied how the neurons were connected, their number, and their initial state. During these experiments, the clusters neither quieted down nor activated randomly. Instead, they generated stable electrical rhythms that repeated over long cycles.
The team tested this idea across multiple nervous systems. The scale of the study ranged from a roundworm with a few hundred neurons to the human brain with tens of billions of neurons.
Here, the rhythms do the main work. In this model, memories are distributed rather than localized. They form according to a pattern that emerges in the entire cluster after a small push. Different inputs nudge the cluster toward different starting points. From there, it establishes its own unique rhythm.
This rhythm not only stores information but also moves it to the next stage of processing. To describe how simple groups give rise to such complex and stable patterns, the researchers borrowed tools from chaos theory and fractal geometry. Both fields study how order arises from tangled, repetitive motion.
This part differs most from standard views. Until now, most estimates of brain capacity boiled down to summing up the states of individual synapses. Viewing information as a collective, constantly changing flow opens up far more possibilities.
This extra space shows up in the numbers. The model predicts a human brain memory capacity of about 7.5 billion gigabytes. That is roughly a thousand times more than estimates based on counting synaptic states.
The synaptic counting method has already yielded unexpected results. In one widely cited study, researchers mapped the fine neural connections in a fragment of brain tissue. They found that each synapse contains much more information than expected.
The new framework significantly increases the total memory capacity by treating it as a dynamic rather than a fixed resource. The model also estimates overall computing power. It calculates the brain’s computational capacity at about 78,000 units, equivalent to the power of high-performance graphics cards. All of this operates within a living human body consuming about 20 watts.
The claim about efficiency is based on a law from physics. Erasing one bit of information incurs a minimum energy cost, known as the Landauer limit. This is a lower bound set by thermodynamics, and no computer can surpass it.
According to the model’s calculations, the brain’s information processing costs are about 1.26 times this minimum threshold. This translates to an efficiency of nearly 79 percent. The minimum level is not just theory. In a 2012 experiment that captured a single microscopic particle, researchers measured the heat released when erasing one bit. The result matched the predicted minimum.
Today’s best AI chips far exceed this limit. The study estimates that they consume about a billion times more energy than the permissible limit for moving and deleting information. This gap highlights just how efficient the brain already appears to be.
What is new here is the method of counting. The brain’s functioning is described not as signals in individual cells, but as patterns spreading across interacting groups. This perspective predicts both a significantly larger memory capacity and efficiency close to the physical limit.
The most obvious application lies in hardware. Engineers developing neuromorphic chips—processors modeled after the brain—now have a concrete goal and a possible strategy. Instead of powering energy-hungry clocks with high voltage, future designs could rely on the collective rhythms described by this model.
These estimates are bold and derived from a model rather than direct recordings, so other research groups will test them. Still, this work unites neuroscience and chip design into a single common principle. The brain stores information through the collective motion of many of its parts, and this storage method is remarkably energy-efficient.
About the Author
