A flexible strip of printed circuitry, thinner than a human hair and built from atom-thick sheets of carbon and molybdenum disulfide, just did something no synthetic device has done before: it fired electrical pulses that living brain cells recognized and responded to as if they came from a neighboring neuron.
The achievement, published in June 2026 in Nature Nanotechnology, comes from a team at Northwestern University led by materials scientist Mark C. Hersam and neurophysiologist Indira M. Raman. Their printed artificial neurons generated spike trains that activated Purkinje cells in mouse cerebellum tissue slices, bridging a gap between synthetic hardware and biological neural networks that researchers have been trying to close for more than a decade.
Printing a neuron from nanomaterial inks
The devices start as liquid inks containing graphene and molybdenum disulfide (MoS2), two materials that can be suspended in solution and deposited with an aerosol-jet printer onto flexible plastic substrates. The printer lays down a sandwich structure: graphene on the bottom, MoS2 in the middle, graphene on top. That stack functions as a memristor, a component whose electrical resistance shifts depending on the voltage history applied to it.
When voltage is applied, a polymer layer within the device partially breaks down, forming a tiny conductive filament. As that filament grows and ruptures under cycling voltage, the device snaps between high-resistance and low-resistance states. Physicists call this behavior snap-back negative differential resistance: past a certain voltage, pushing harder actually produces less current, creating a built-in instability. Pair that instability with simple resistors and capacitors, and the circuit begins to oscillate on its own, producing discrete voltage spikes much like the action potentials that neurons use to communicate.
By adjusting device geometry and circuit parameters, the Northwestern team tuned the spiking frequency across several orders of magnitude, reaching up to 20 kHz. That range overlaps with the firing rates of biological neurons, which was the whole point: to build hardware that speaks the brain’s electrical language.
Making contact with living tissue
The critical experiment moved from physics into biology. Thin slices of mouse cerebellum were kept alive in a nutrient bath while the printed devices were placed against the tissue surface. When the artificial neurons fired, nearby Purkinje cells, the large, elaborately branched neurons responsible for coordinating motor output, responded with their own characteristic firing patterns.
Purkinje cells are among the most distinctive neurons in the brain. Each one receives roughly 200,000 synaptic inputs and serves as the sole output channel of the cerebellar cortex. Getting these cells to respond to a synthetic signal is a meaningful test because their firing behavior is well characterized and easy to distinguish from noise or artifact.
The experiment was conducted ex vivo, meaning the brain tissue was maintained outside a living animal in controlled laboratory conditions. That is standard practice for electrophysiology studies and allows precise measurement, but it is not the same as operating inside a living brain, a distinction that matters for anyone thinking about future medical applications.
Where this fits in the field
The Northwestern team is not the only group working on artificial devices that interface with biological neurons. A separate study published in Nature Electronics previously demonstrated an organic artificial spiking neuron designed for biointerfacing, using organic and ionic materials rather than inorganic two-dimensional nanosheets. Other research groups have explored silicon-based neuromorphic chips and conducting-polymer electrodes for similar goals.
What distinguishes the Northwestern work is the specific combination: printed fabrication (which could eventually enable low-cost, scalable manufacturing), inorganic two-dimensional nanomaterials (which offer chemical stability and tunable electronic properties), and direct functional communication with living neurons confirmed in tissue. The “first” in the headline refers to that particular intersection, not to the broader concept of artificial spiking devices aimed at biological systems.
The result also builds on years of groundwork within Hersam’s lab. A 2018 preprint by overlapping authors described multi-terminal memtransistors fabricated from polycrystalline monolayer MoS2, demonstrating that two-dimensional semiconductors could mimic synapse-like and neuron-like behaviors in rigid, lithographically patterned hardware. That work, which remains an unreviewed preprint as of June 2026, established gate tunability and high switching ratios but did not involve printed devices or biological interfacing. A 2022 perspective article by Sangwan and Hersam in Matter laid out a roadmap for “bio-realistic” neuronal computing, arguing that neuromorphic devices needed to match biological timescales, adopt soft and flexible form factors, and physically interface with tissue rather than chase abstract computational benchmarks. The new paper checks several of those boxes.
What the technology cannot do yet
The gap between a successful slice experiment and a working brain implant remains wide. No data on long-term biocompatibility, immune response, or device degradation in living animals has been published. Inside the body, these devices would face challenges that a nutrient bath does not present: micro-movements from pulsating blood vessels, temperature fluctuations, inflammatory responses, and the slow encapsulation process by which the brain walls off foreign objects in scar tissue.
Mechanical durability is another open question. The flexible substrates can conform to curved tissue surfaces, but whether they survive weeks or months of chronic implantation under real physiological conditions has not been tested.
Manufacturing scalability is similarly unresolved. Aerosol-jet printing is well established in laboratory settings, but the available sources describe fabrication only in qualitative, proof-of-concept terms. No published data addresses yield rates, device-to-device variability at scale, or per-unit cost. A practical brain-machine interface would require many synchronized artificial neurons, not just a handful of devices on a bench, so those manufacturing details will eventually matter as much as the electrical performance.
Existing brain-machine interfaces, such as the Utah microelectrode array used in clinical trials or the flexible electrode threads developed by Neuralink, take a fundamentally different approach: they record and stimulate with passive electrodes rather than generating neuron-like spikes from active circuitry. Whether spike-generating artificial neurons offer a practical advantage over conventional stimulation electrodes in real clinical scenarios is a question that only future comparative studies can answer.
What comes next for printed artificial neurons
The milestones to watch are straightforward: in vivo animal trials, longer-duration stability and biocompatibility studies, and independent replication by laboratories outside Northwestern. Each step will test whether the promising physics demonstrated in a dish translates into reliable performance inside a living brain.
If the technology does scale, the potential applications extend beyond prosthetics. Artificial neurons that can be printed onto flexible substrates and tuned to specific frequencies could serve as research tools for probing neural circuits, as components in closed-loop systems that respond to brain activity in real time, or as building blocks for hybrid biological-electronic networks that do not yet exist outside theoretical proposals.
For now, the Northwestern result stands as a proof of concept with a clear message: synthetic devices built from two-dimensional nanomaterials can produce electrical signals that living neurons treat as real. The hard engineering work of turning that proof into something a surgeon could implant is only beginning.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
