Northwestern University printed artificial neurons that mouse brains accepted as real. The technique, published in Nature Nanotechnology this week, uses a specialized semiconductor material and a carbon-based conductor — molybdenum disulfide and graphene — printed onto a flexible polymer sheet, and the key was a stabilizing polymer that should have been removed before testing. Hersam's team left it in by accident. The mouse neurons treated the artificial ones as one of their own, according to Northwestern's news office.
That cross from artificial to living tissue is the part nobody has figured out. Hersam's result demonstrates the crossing is physically possible. Whether we understand what we are crossing into — and whether anyone has asked — is a different question.
"Other labs have tried to make artificial neurons with organic materials, and they spiked too slowly," Hersam told Northwestern's news office. "Or they used metal oxides, which are too fast. We are within a temporal range that was not previously demonstrated for artificial neurons."
The reason Hersam landed in that range was the accident. The electronic ink contains a stabilizing polymer that interferes with electrical current — conventional practice burns it off after printing. Hersam's team partially decomposed it instead. When current passes through the remaining polymer, it breaks down unevenly, forming a narrow conductive filament. That constriction produces the sharp voltage spikes that look like biological signals. Other labs tried to eliminate this imperfection. Hersam found it was essential.
Whether that biological realism matters depends on what you are building toward. Tech companies are building data centers powered by dedicated nuclear power plants. Some are negotiating water rights to keep the cooling systems running. If neuromorphic computing — chips designed to mimic the brain's architecture — can close even part of the gap between current AI hardware and biological efficiency, the implications are significant. A recent perspective in PNAS noted that neuromorphic approaches are among the few proposed paths to breaking through the efficiency wall AI is running into. Intel's Loihi, IBM's TrueNorth, and Europe's BrainScaleS project have all demonstrated three to five orders of magnitude better energy efficiency than conventional chips for specific tasks, according to a review in MDPI Engineering Proceedings. One recent neuromorphic system was reported to achieve 5,600 times the energy efficiency of GPU-based edge AI on continual learning benchmarks, though that figure traces to a secondary aggregator rather than a primary study.
Northwestern's paper is more fundamental than those systems in one respect: it is a materials insight that the imperfection created by partial polymer decomposition is not something to engineer away but the actual mechanism producing biological realism. Conventional neuromorphic chips have largely tried to solve the energy problem by building many identical spiking units and letting scale compensate. Hersam's work suggests that may be the wrong bet — that neuromorphic computing will need to embrace heterogeneity, not eliminate it.
The paper is not a product. It is not a chip. There is no efficiency measurement comparing these neurons to conventional circuits, no manufacturing process, no industry partner. Hersam himself acknowledges the distance between what he has shown and what would be needed to move the needle on data center power consumption. "The brain is five orders of magnitude more energy efficient than a digital computer," his press release states. "The way you make AI smarter is by training it on more and more data. This data-intensive training leads to a massive power-consumption problem."
Northwestern's paper offers a direction, not a destination. The neurons fired in a dish. What comes next is the harder question: Hersam's team is planning to test whether the printed neurons can interface with living neural tissue, not just a dish of cells. Whether that step has been funded, reviewed, or green-lit by any ethics body is not in the paper. Whether anyone asked what happens if it works is a question neither Hersam nor anyone else has answered yet.
