Engineers at Northwestern University have created printed artificial neurons that can go beyond imitation and interact directly with real brain cells. These flexible, low-cost devices generate electrical signals that closely resemble those produced by living neurons and can activate biological brain tissue.
In experiments using mouse brain slices, artificial neurons were able to trigger responses in real neurons. This result demonstrates a new level of compatibility between electronic devices and living neural systems.
Towards brain interfaces and energy-efficient AI
This advance brings researchers closer to electronics that can connect directly to the nervous system. Potential applications include neuroprosthetics such as brain-machine interfaces and implants that help restore hearing, vision, and movement.
The technology also represents a new generation of brain-inspired computing systems. By replicating the way neurons communicate, future hardware will be able to perform complex tasks using far less energy. The brain remains the most energy-efficient computing system known, and scientists hope to apply those principles to modern technology.
The study is scheduled to be published in the journal April 15th natural nanotechnology.
“The world we live in today is dominated by artificial intelligence (AI),” said study leader Mark C. Hersam of Northwestern University. “The way to make AI smarter is to train it with more data. This data-intensive training leads to huge power consumption problems, so we need to come up with more efficient hardware to process big data and AI. The brain is five orders of magnitude more energy efficient than digital computers, so it makes sense to look to the brain for inspiration for the next generation of computing.”
Hartham is an expert in brain-inspired computing and holds multiple positions at Northwestern University, including Walter P. Murphy Professor of Materials Science and Engineering in the McCormick School of Engineering. He is also Professor of Medicine at Northwestern University Feinberg School of Medicine and Professor of Chemistry at Weinberg College of Arts and Sciences. Additionally, he serves as Chair of the Department of Materials Science and Engineering, Director of the Center for Materials Science and Engineering, and a member of the International Institute for Nanotechnology. He co-led the research with Vinod K. Sangwan, an associate professor at McCormick College.
Why the brain performs better than traditional silicon
Modern computers handle increasing workloads by packing billions of identical transistors onto rigid, two-dimensional silicon chips. Each component operates in the same way, and once manufactured, the system remains fixed.
Brains work very differently. It consists of many types of neurons, each with a specialized role, arranged in a soft three-dimensional network. These networks constantly change, form connections, and adjust as learning occurs.
“Silicon achieves complexity by having billions of identical devices,” Hersam said. “Everything is the same and rigid and fixed once manufactured. The brain is the opposite. It is heterogeneous, dynamic, and three-dimensional. Moving in that direction requires new materials and new ways to build electronics.”
Artificial neurons have been developed before, but most generate very simple signals. To achieve more complex operations, engineers typically require larger networks of devices, which increases energy usage.
Printable materials enable brain-like behavior
To better mimic real neural activity, Hersam’s team constructed artificial neurons using soft, printable materials that better match the structure of the brain. Their approach relies on an electronic ink made from nanoscale flakes of molybdenum disulfide (MoS2), which acts as a semiconductor, and graphene, which acts as an electrical conductor. These materials were deposited onto flexible polymer surfaces using aerosol jet printing.
Previously, researchers treated these inks as a defect because the polymers in them interfere with their electrical performance. As a result, it was deleted after printing. In this work, the team used the same features to enhance the device.
“Rather than completely removing the polymer, it partially degrades it,” he said. “Then, when we apply an electric current to the device, we further degrade the polymer. This degradation occurs spatially non-uniformly, leading to the formation of conductive filaments, which confine all the current to a small area in space.”
That narrow conductive path causes a sudden electrical response similar to the firing of a neuron. The resulting device can generate a variety of signals that closely resemble real neural communications, including single spikes, continuous firing, and burst patterns.
Each artificial neuron can generate more complex signals, so fewer components are needed to perform advanced tasks. This can significantly improve computing efficiency.
Testing artificial neurons on real brain tissue
To assess whether artificial neurons could indeed interact with biological systems, the researchers partnered with Indira M. Raman, the Bill and Gail Cook Professor of Neurobiology at Weinberg College. Her team applied artificial signals to slices of the mouse cerebellum.
The results showed that electrical spikes match key biological properties such as timing and duration. These signals reliably activated real neurons and triggered neural circuits in a manner similar to natural brain activity.
“Other labs have tried to make artificial neurons using organic materials, but their spikes were too slow,” Hersam said. “Alternatively, they used metal oxides, but this is too fast. We’re in a time range that hasn’t been demonstrated before with artificial neurons. We see living neurons responding to artificial neurons. So we’ve demonstrated not only the correct timescale, but also the correct spike-shaped signal that interacts directly with living neurons.”
Impact on low-cost, sustainable manufacturing and AI
Beyond performance, this new approach offers environmental and practical benefits. The manufacturing process is simple and inexpensive, and the layered printing method places material only where it is needed, reducing waste.
Improving energy efficiency is especially important as artificial intelligence systems become increasingly demanding. Large data centers already consume large amounts of electricity and require large amounts of water for cooling.
“To meet the energy needs of AI, tech companies are building gigawatt-scale data centers powered by dedicated nuclear power plants,” Hersam said. “It is hard to imagine that the next generation of data centers will require 100 nuclear power plants, so it is clear that this high power consumption will limit further expansion of computing. Another problem is that consuming gigawatts of power generates a lot of heat. Data centers are cooled with water, so AI is putting a lot of stress on the water supply. No matter how you look at it, AI We need to develop more energy-efficient hardware for
The study, “Multi-order complexity spiking neurons enabled by printed MoS2 memristive nanosheet networks,” was supported by the National Science Foundation.
