A team of engineers at Northwestern University recently made a major breakthrough by developing printed artificial neurons that can directly interact with real brain cells. These devices are not only soft, flexible and cheap to manufacture, but can also produce electrical signals that are highly similar to living neurons. In laboratory tests using slices of mouse brain tissue, the artificial neurons successfully stimulated real neurons and elicited measurable responses. This achievement demonstrates an unprecedented level of compatibility between electronic systems and biological neural networks.
This research opens up important avenues for electronic devices capable of communicating with the nervous system. The technology is expected to support the development of brain-computer interfaces and neuroprosthetic devices, including implants designed to restore hearing, vision or movement. At the same time, the findings also point to a future of more efficient computing. By replicating the way neurons send signals—a core feature of the brain as the most energy-efficient computing system known—next-generation hardware will be able to handle complex tasks using far less energy than current systems.
The research was published in the journal Nature Nanotechnology on April 15. "The world we live in today is dominated by artificial intelligence. To make artificial intelligence smarter, you need to train it with more and more data. This data-intensive training leads to huge energy consumption problems. Therefore, we must develop more efficient hardware to process big data and artificial intelligence. Since the brain is five orders of magnitude more energy efficient than digital computers, it is reasonable to look to the brain for inspiration for the next generation of computing."
As computing needs grow, traditional systems respond to these challenges by adding more of the same components. Modern chips contain billions of transistors arranged on a rigid, flat piece of silicon, with each element performing the same function. Once manufactured, these systems cannot be changed. The brain works completely differently. It is composed of multiple types of neurons, each with specialized roles, organized in a soft three-dimensional network. These networks continuously adapt, forming new connections and reshaping existing connections as learning occurs. "Silicon achieves complexity by having billions of identical devices. Everything is identical, rigid, and fixed once it's made. The brain is the opposite. It's heterogeneous, dynamic, and three-dimensional. To move in that direction, we need new materials and new ways of building electronics," Hessam explained.
Although artificial neurons have been created before, most produced signals that were too simple. To generate more complex behaviors, engineers often rely on large networks, which increases energy consumption. To better match the behavior of real neurons, the researchers designed their device using soft, printable materials. They created specialized e-inks from sheets of molybdenum disulfide, a material that functions as a semiconductor while graphene acts as a conductor. These inks are deposited onto flexible polymer surfaces using a method called aerosol jet printing.
Previously, the polymer component in these inks was considered a drawback because it interferes with electrical flow, so it was usually removed after printing. In this case, the research team instead took advantage of it. "Instead of completely removing the polymer, we partially break it down," the researchers said. "Then, when we run an electrical current through the device, we further drive the polymer's breakdown. This breakdown occurs in a spatially non-uniform manner, leading to the formation of conductive filaments so that all current flow is confined to a narrow area in space." This narrow conductive path creates a sudden electrical response similar to a neuron's firing. As a result, artificial neurons can produce a wide variety of signals, including single spikes, steady firing, and burst patterns, closely mimicking real neural activity. Because each device can handle more complex signaling, fewer components are required overall, which could significantly improve the efficiency of future computing systems.
To determine whether these artificial neurons could interact with real biological systems, the research team collaborated with Indira Raman, a professor of neurobiology at the Weinberg School. Her team applied artificial signals to slices of mouse cerebellum. The results showed that these electrical spikes matched key features of natural neuronal activity, including timing and duration. These signals reliably activate real neurons and trigger neural circuits in a manner similar to natural brain signals. "Other labs have tried to make artificial neurons out of organic materials, but they fire too slowly," Hessam said. "Or they use metal oxides, which are too fast. We're on a time scale that hasn't been demonstrated in artificial neurons before. You can see living neurons responding to our artificial neurons. So we've shown signals that not only have the right time scale, but also the right spike shape, can interact directly with living neurons."
This new approach also offers environmental and practical advantages. The manufacturing process is simple and cost-effective, and the additive printing method uses materials efficiently and reduces waste by placing them only where they are needed. As artificial intelligence systems continue to expand, improving energy efficiency is particularly important. 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," Hessam noted. "It's clear that this huge power consumption will limit the further expansion of computing, because it is difficult to imagine that the next generation of data centers will require 100 nuclear power plants. Another problem is that when you dissipate gigawatts of power, you generate a lot of heat. Because data centers are cooled with water, AI is putting a severe strain on the water supply. No matter how you look at it, we need to develop more energy-efficient hardware for AI."
This research was supported by the National Science Foundation.