How Silicon Learns to Speak to the Living: Mark Hersam and the Printed Neurons
A black ink, dark as squid's ink, made of graphene flakes and molybdenum disulfide. An aerosol jet printer, the kind that might one day sit on a workbench beside a 3D printer for hobbyists. And, on a microscope stage, a thin slice of mouse cerebellum, still alive, still firing. The moment was almost banal in its setup — and astonishing in its outcome. For the first time, a neuron printed by a human and a neuron grown by nature began to speak to each other. The voltage spike traveled from the synthetic device into the living tissue. The biological neurons listened. They answered.
Illustration: Pierre Guité et Mid-Journey
That experiment, published in Nature Nanotechnology in April 2026 by a team at Northwestern University led by Mark C. Hersam and Vinod K. Sangwan, is not just a beautiful piece of engineering. It is a quiet announcement that one of the longest borders in science — the border between the electronic and the biological — has just become a little more porous. And the way it happened is as instructive as the result itself.
A Scientist Who Refuses to Choose a Department
Most academic biographies fit on a single line of letterhead. Mark Hersam's does not. He is the Walter P. Murphy Professor of Materials Science and Engineering at the McCormick School of Engineering. He is also a professor of medicine at the Feinberg School of Medicine. And, simultaneously, a professor of chemistry at Weinberg College of Arts and Sciences. Three appointments at Northwestern, three departments, three intellectual cultures that almost never share a coffee machine.
It is tempting to read this as accumulation — the trophy case of a prolific career. It is something else. It is a posture. Hersam has spent two decades insisting that the most interesting questions in science do not live inside disciplines; they live in the cracks between them. The printed neuron is what that posture produced when it finally found the right problem.
There is a long intellectual lineage behind this kind of stance. Norbert Wiener, founding cybernetics in 1948, proposed that the most important question of the twentieth century was not how machines differ from living things but in what language they could possibly converse. Information, he argued, was that language. Hersam's lab has, almost accidentally, built a small object that speaks it — fluently enough that a mouse neuron understands what it is saying.
The Strange Behavior of Two-Dimensional Materials
Hersam did not set out to imitate the brain. He set out to print semiconductors.
For years, his group had been working on flexible electronics built from two-dimensional materials — atomically thin sheets of substances like molybdenum disulfide (MoS₂) and graphene. MoS₂ behaves as a semiconductor; graphene as a near-perfect conductor. Suspend them as nanoscale flakes in a stabilizing polymer, load them into an aerosol jet printer, and you can spray patterns of working electronic devices onto bendy plastic. It is the kind of work that ends up in wearable sensors and foldable displays. Useful. Industrial. Far removed from neurobiology.
Then the devices started doing something odd. When the team passed current through them, the spikes did not look like the clean digital pulses of a transistor. They looked, eerily, like the action potentials of a neuron — fast, bursting, sometimes rhythmic, sometimes erratic.
The conventional engineering response would have been to stamp out the irregularity. Hersam's response was to lean into it.
The key innovation, Hersam explains, was to stop fighting the polymer that bound the inks together. "Instead of fully removing the polymer, we partially decompose it," he says. When current flows through the partially decomposed film, the decomposition continues, but unevenly. A narrow conductive filament forms inside the device, concentrating the current. The result is not a clean switch. It is something that fires — that integrates, holds, releases, and fires again. Something that behaves, in short, like a synapse.
Other groups had tried to build artificial neurons before, mostly using organic semiconductors. "Other labs have tried to make artificial neurons with organic materials," Hersam notes, "and they spiked too slowly." Speed matters. A neuron that fires on the wrong timescale is a neuron that speaks a language no biological circuit can hear.
The Northwestern team's MoS₂-and-graphene devices fired fast, with millisecond timing close enough to the real thing that the next experiment became irresistible.
The Moment a Mouse Cerebellum Answered Back
That next experiment required a different kind of scientist. Hersam called Indira M. Raman, the Bill and Gayle Cook Professor of Neurobiology at Northwestern, whose laboratory routinely keeps thin slices of mouse cerebellum alive for hours under a microscope, with their neural circuits intact and recordable.
The setup was deceptively simple. The artificial neurons, freshly printed on flexible plastic, were wired to deliver their voltage spikes into the living tissue. Microelectrodes recorded what the biological neurons did in response.
Not randomly. Not as a stress response. They fired the way they would have if another biological neuron had been talking to them. The artificial spikes — in amplitude, in duration, in rhythm — were realistic enough that the living circuitry treated them as conversation, not noise. The team could tune the printed devices to produce isolated spikes, sustained firing, or rhythmic bursts, mirroring the repertoire of real neurons. The cerebellum did not seem to notice the difference.
This is the moment that matters. Not because it solves any single application, but because it proves a particular kind of fluency. The printed neuron is not a metaphorical neuron, a "neural-network-inspired" computing element abstracted away from biology. It is a small piece of matter, made of inks and polymer, whose electrical behavior is biologically legible.
The Double Promise: A Limb That Feels, A Data Center That Sips
From this single demonstration, two enormous trajectories unfold, pointing in directions that almost never get discussed in the same sentence.
The first is medical. A neuroprosthetic device — a cochlear implant, a retinal array, a spinal-cord bypass — works only to the extent that it can persuade biological neurons to listen to it. Today's interfaces are crude: they push electrical pulses into nervous tissue and hope the brain learns to interpret them. The printed neurons hint at a different paradigm, in which the implant does not shout at the nervous system in an alien tongue but speaks to it in dialect. Flexible. Biocompatible. Cheap to print. Capable of being shaped to the exact anatomy of an inner ear or a peripheral nerve. "Brain-machine interfaces and neuroprosthetics for hearing, vision, and movement" is how the team describes it, with the restraint of researchers who know how long the road is from a mouse cerebellum to a human patient.
The second trajectory is, improbably, climate. "The world we live in today is dominated by artificial intelligence," Hersam says. "The way you make AI smarter is by training it on more and more data. To meet the energy demands of AI, tech companies are building gigawatt data centers powered by dedicated nuclear power plants."
The comparison he draws is brutal. "The brain is five orders of magnitude more energy efficient than a digital computer." A hundred thousand times. The reason is architectural. Digital computing is built on uniformity — billions of identical transistors switching in lockstep, moving data back and forth between memory and processor in ways that burn most of their energy on transport rather than thought.
"Silicon achieves complexity by having billions of identical devices," Hersam observes. "Everything is the same, rigid and fixed once it's fabricated. The brain is the opposite. It's heterogeneous, dynamic, and three-dimensional."
A printed neuron is none of those things in the silicon sense. It is heterogeneous — the conductive filament is a happy accident of polymer decomposition, slightly different in every device. It is dynamic — its history shapes its future firing. It is, in principle, stackable into three-dimensional architectures. Neuromorphic computing built from such elements would not just be a different kind of chip. It would be a different kind of physics for thinking. And if even part of the hundred-thousand-fold efficiency gap closes, the carbon mathematics of the next decade of AI changes shape.
The Bridge That Was Already Built
What makes this story unusual is not the discovery itself. Plenty of laboratories are racing toward neuromorphic computing — IBM's TrueNorth, Intel's Loihi, a half-dozen academic groups with their own variants. What makes it unusual is the path that led there.
Hersam did not migrate from materials science to neuroscience to find this result. He stayed exactly where he was — at the intersection of all three of his appointments — and waited for a problem to recognize itself. The triple title is not a credential. It is a search algorithm. By refusing to fully inhabit any single discipline, he kept himself in a position to notice when an ink for printing transistors started behaving like a synapse.
There is a quiet lesson in this for any organization that funds research. The standard pressure of modern science is toward specialization, because specialists are easier to evaluate, easier to publish, easier to compare to their peers. The kind of researcher who holds three departmental affiliations is, by most institutional metrics, harder to assess — and therefore harder to hire, harder to promote, harder to keep. The institutions that protect this kind of profile, against the gravitational pull of the disciplines, are the ones that get printed neurons.
Gaston Bachelard, the French philosopher of science, once wrote that "the scientific mind must form against nature." He meant that scientific insight comes from resisting the obvious categories that nature seems to hand us. The categories "biology" and "electronics," "wet" and "dry," "thinking" and "computing" are exactly such categories — true as far as they go, but not far enough. The printed neuron is one of those rare objects that reveals what a useful distinction also hides.
What This Says About the Convergence Ahead
It would be easy to read this story as a step toward a transhumanist horizon in which humans and machines fuse into a single hybrid system. On the bench at Northwestern, something more modest is happening. A slow, technical demonstration that the boundary between biological and electronic computation is not a wall but a translation problem. And that the translation, in at least some directions, is starting to work.
A child born today will live in a world in which a person who has lost their hearing may receive an implant whose neurons were printed, not micromachined. In which a data center the size of a small town may give way to one the size of a closet, because its chips compute the way brains do. In which the question of where "biology" ends and "engineering" begins becomes less a metaphysical puzzle and more a matter of which ink you ordered.
None of that is inevitable. But the demonstration that a printed device can hold an intelligible conversation with a living brain slice is the kind of small, concrete event that, in retrospect, tends to mark the moment a frontier moved.
The Conversations to Have
These are questions to ask while the work is still early — before the device leaves the laboratory.
Deaf communities have asked, for two generations, whether the cochlear implant is a medical solution or a cultural erasure. A printed device that speaks the dialect of the nervous system sharpens that question rather than soothing it. The more fluent the technology becomes, the easier it is to forget that fluency was never the only horizon. Some patients will reach for the implant. Some will refuse. Both choices have to stay possible.
What flows back. If a printed interface is biologically legible enough to send a signal a neuron will accept, it is also legible enough to record what the neuron is doing in return. The same property that makes the device a good speaker will make it a good witness. Witness for whom, with whose consent, under what authority.
The line between implant and organ will become irrelevant for many medical conditions — not by any grand program, but by the cumulative weight of a million small, reasonable clinical decisions. The question is not whether to accept or refuse that future. It is who sits at the table when its rules are written, and whose conceptions of personhood, body, and consent are allowed to count.
💡 Key Insights for Leaders
1. Cultivate trespassers, not just specialists: Mark Hersam holds three simultaneous professorships across materials science, medicine, and chemistry. Hire and promote people whose intellectual passport has many stamps.
2. Let the artifact surprise you: The printed neuron was not a goal. It was the consequence of noticing that an electronic ink, designed for flexible semiconductors, happened to fire like a synapse. Make sure your R&D culture has room for the "wait, that's weird" moment — and for the patience to chase what it means rather than engineer it away.
3. Resist the urge to fully remove the imperfection: Hersam's technical breakthrough was deciding not to fully decompose the polymer in the ink. Many organizations sanitize their processes so that nothing surprising can happen. Sometimes the conductive filament forms only inside the disorder you were going to clean up.
4. Pair technical promise with at least two distinct impact stories: This single innovation can credibly address neuroprosthetic medicine and the energy footprint of AI. Innovations with that kind of cross-domain reach attract capital, talent, and political support that single-purpose technologies cannot. Train your teams to articulate the second and third use case, not just the first.
5. Build collaborations across the wettest line: The pivotal moment in this work came when materials scientists put their device in contact with neurobiologist Indira Raman's mouse cerebellum. That is an institutional collaboration that is hard to make happen — different buildings, different journals, different vocabularies. The leaders who orchestrate those handshakes get experiments that no single department could have designed.
6. Hold the long view without sliding into hype: From a working printed neuron in a mouse slice to a clinical neuroprosthesis or a neuromorphic data center is a road measured in decades, not quarters. Communicate the trajectory honestly — the promise is enormous, but it is not next year. Leaders who can tell a true long story keep the funding and the trust that overnight-success narratives eventually destroy.
If a printed neuron can already hold a quiet conversation with a living mouse cerebellum, the relevant question for our institutions may no longer be whether the languages of matter and life can be bridged — but how many other bridges are quietly being built in laboratories where someone refused to choose a single department.
References:
Hadke, S. S., et al. Printed MoS₂ memristive nanosheet networks for spiking neurons with multi-order complexity. Nature Nanotechnology. DOI: 10.1038/s41565-026-02149-6. April 2026.
Morris, A. Printed neurons communicate with living brain cells. Northwestern Now, April 15, 2026.
Fellman, M. Printed Neurons Communicate with Living Brain Cells. Feinberg News Center, Northwestern University, April 27, 2026.
Printed Neurons That Mimic Brain Cells Could Slash AI's Energy Bill. April 20, 2026. Singularity Hub.
Bachelard, G. (1938). La Formation de l'esprit scientifique : Contribution à une psychanalyse de la connaissance objective. Paris : Vrin.
Wiener, N. (1948). Cybernetics: Or Control and Communication in the Animal and the Machine. Cambridge, MA : MIT Press.
