The neuromorphic computing market will explode from 28.5 million to 1.325 billion dollars by 2030, representing growth of 89.7% per year. This projection finds an unexpected catalyst in the laboratories of Northwestern University: printed artificial neurons that do not merely imitate the brain—they communicate directly with it.
The experiment conducted by Professor Mark Hersam’s team crosses the barrier between the artificial and biological worlds. Tested on slices of mouse brain tissue, these artificial neurons successfully triggered real neural responses. This technical breakthrough redefines the possibilities of neural prosthetics and could reduce artificial intelligence’s energy consumption by a thousandfold.
The Barrier Between Artificial and Biological Collapses
The human brain is five orders of magnitude more energy-efficient than a digital computer, consuming only 12 watts of power to function continuously versus 250 watts to operate a simple GeForce Titan X graphics card. This fundamental difference explains why the race for artificial neurons is intensifying.
Hersam’s team developed artificial neurons using soft, printable materials, formulated from nanometric flakes of molybdenum disulfide (MoS2) serving as a semiconductor and graphene acting as an electrical conductor. Using a specialized printing technique called aerosol jet printing, the researchers deposited these inks on flexible polymer substrates.
The innovation lies in exploiting an apparent defect. While other researchers completely eliminated the stabilizing polymer from the inks after printing, Hersam partially decomposed it to create functionality similar to the brain. This inhomogeneous spatial decomposition forms a conductive filament that generates a sudden electrical response, resembling that of a neuron.
Energy Efficiency That Defies Classical Computing
Rather than generating simple single impulses, this new device produces complex signaling patterns, including individual spikes, continuous firing, and burst patterns that resemble actual neural communication. By capturing this diversity of signaling, each neuron can encode more information and execute more sophisticated functions, drastically reducing the number of components necessary.
This approach directly addresses current limitations in AI. As Hersam explains: “The way to make AI smarter is to train it on more and more data. This data-intensive training leads to a massive problem of electrical consumption. We therefore need to develop more efficient hardware.”
This technical transformation rests on a fundamental architectural difference. While computers address complex tasks by adding billions of identical components on rigid two-dimensional silicon chips, with each transistor behaving the same way and remaining fixed once manufactured, the brain operates differently, relying on diverse types of neurons accomplishing specialized roles, organized in flexible three-dimensional networks that constantly change.
Direct Dialogue Between Artificial and Biological Circuits
The Northwestern team crossed the decisive step: bidirectional communication. To test whether their artificial neurons could truly interface with biology, Hersam’s team collaborated with Indira M. Raman, who applied electrical signals from the artificial neurons to slices of mouse cerebellum.
The biological neurons responded, demonstrating that the synthetic signals were sufficiently convincing to activate real neural circuits. “You can see the living neurons responding to our artificial neuron,” Hersam confirms. “We have demonstrated signals that not only have the correct time scale but also the correct spike shape to interact directly with living neurons.”
This biocompatibility opens concrete perspectives for brain-machine interfaces. The global brain-computer interface market was valued at 1.488 billion dollars in 2020 and is expected to reach 5.463 billion dollars by 2030, with growth of 13.9%. In parallel, the neuroprosthetics market was valued at 12.66 billion dollars in 2023 and is expected to reach 26.12 billion dollars by 2030.
Energy Efficiency as a Civilizational Challenge
The urgent energy question of AI transforms this breakthrough into a civilizational necessity. Hersam observes that technology companies are building enormous data centers powered by nuclear power plants to meet AI’s energy needs, an approach that has limitations in terms of power supply and cooling, making essential new, more energy-efficient hardware.
The comparison of efficiencies reveals the scale of the challenge. According to an estimate based on detailed simulations, biological computation is approximately 900 million times more energy-efficient than artificial computing architecture. A human brain uses approximately 20 watts to operate, equivalent to the consumption of a computer monitor on standby, enabling 80-100 billion neurons to perform trillions of operations that would require the power of a small hydroelectric plant if performed artificially.
This remarkable efficiency is explained by evolution. The brain evolved under clear constraints of energy and space, this efficiency being explained by the systemic neurobiological use of analog computation, local wiring and computational strategies, as well as by adaptation and learning capacities.
An Industrial Transformation in the Making
With only two of these printable neurons and a few basic circuit components, researchers have produced sophisticated spike patterns, adjusting the length and frequency of spikes to match the timing of biological action potentials. This flexibility could transform AI’s energy efficiency, which today is already reinventing its approach to reduce consumption by a factor of 100.
The impact extends far beyond computing. This work paves the way for electronics capable of communicating directly with the nervous system, with potential applications in brain-machine interfaces and neuroprosthetics, including implants for hearing, vision, and movement.
The neuromorphic market confirms this trajectory. The market is shaped by emerging trends, namely the shift toward edge computing architectures, growing demand for brain-computer interfaces, and the convergence of quantum computing with neuromorphic systems. The globalization of neuromorphic computing will gain momentum thanks to growing demand for high-performance integrated circuits and the adoption of neuromorphic hardware by the health and automotive industries.
Northwestern has just demonstrated that imitating the brain is no longer enough. The next step consists in speaking directly to it. This conversation between the artificial and the biological could redefine the limits of intelligence itself, a dialogue that begins in test tubes and could end in our heads.