A tiny strip of printed nanomaterials fired an electrical spike, and a living mouse brain cell answered back. That exchange, reported in April 2025 by researchers at Northwestern University, marks the first time an artificial neuron built from two-dimensional materials has carried on a two-way electrical conversation with biological neurons in brain tissue. The result, published in Nature Nanotechnology, does not mean printable brain implants are around the corner. But it does clear a barrier that has stalled the field for years: getting synthetic hardware to speak the brain’s own electrical dialect.
What the team actually built
The device is a memristor, a type of electronic switch whose resistance changes depending on the voltage history it has experienced. Think of it as a component that can “remember” whether it was recently on or off, much like a biological neuron retains a trace of its last firing. The Northwestern team, led by materials scientist Mark C. Hersam, stacked three atomically thin layers: graphene on the top and bottom, with molybdenum disulfide (MoS2) sandwiched in between. They deposited this stack onto a flexible plastic substrate using aerosol-jet printing, a technique that sprays nanomaterial inks through a fine nozzle.
When voltage is applied, the MoS2 layer forms and breaks tiny conductive filaments through a process driven by localized heating. Each filament snap creates a sharp voltage spike that mimics the all-or-nothing firing of a real nerve cell. By tuning the thickness of the layers and the geometry of the device, the researchers controlled how fast the spikes occur, how long each spike lasts, and how long the device rests before it can fire again. That rest period, called the refractory period, is one of the defining features of biological neurons, and matching it has been a persistent challenge for artificial alternatives.
The printed neurons produced spiking patterns across a wide frequency range, up to 20 kHz, and maintained stable performance over more than one million switching cycles on the bench. That range matters because neurons in different brain regions fire at very different rates. A device locked to a single frequency would be like a radio stuck on one station: useless for tracking the brain’s shifting rhythms.
How it talked to living tissue
The critical experiment moved beyond electronics benchmarks. Working with neuroscientists at Northwestern, Hersam’s team placed the printed devices next to ex vivo slices of mouse hippocampus, a brain region central to memory and one of the best-studied circuits in neuroscience. When the artificial neurons fired, they reliably triggered postsynaptic responses in nearby living cells, meaning the biological neurons recognized the synthetic spikes as genuine input and passed the signal along through local circuits.
The communication ran in both directions. When the researchers pharmacologically altered the excitability of the brain slice, making neurons more or less responsive, those changes showed up in the electrical recordings from the artificial devices. The printed neurons were not just broadcasting into the tissue; they were picking up the tissue’s replies.
According to Northwestern’s announcement, Hersam framed the design choice in terms of speed: prior attempts using organic materials produced neurons that spiked too slowly, while metal-oxide versions switched too fast. The graphene/MoS2 combination lands in a middle range that overlaps with the firing rates of real hippocampal and cortical neurons, enabling the kind of back-and-forth that earlier neuromorphic chips could not sustain with living tissue.
Why previous approaches fell short
Artificial spiking circuits are not new. Silicon-based neuromorphic chips have been generating spike-like signals for decades, and a foundational review in Frontiers in Neuroscience laid out the requirements for real-time dialogue with living neurons: not just digital pulses, but analog dynamics including membrane-like integration and realistic refractory behavior. Most silicon designs meet some of those criteria on rigid chips but struggle to interface directly with soft, wet brain tissue.
Meanwhile, existing clinical implants like the Utah array, a bed-of-nails electrode grid used in brain-computer interface research, can record from and stimulate neurons but do not themselves behave like neurons. They deliver current pulses on command from an external processor rather than generating autonomous, brain-like spiking. Newer flexible electrode systems, including those developed by companies like Neuralink, improve mechanical compatibility with brain tissue but still rely on conventional stimulation rather than neuromorphic firing patterns.
The Northwestern approach is distinct because the device itself produces spikes that are biologically realistic in shape, timing, and adaptability. A commentary in Nature Electronics has outlined how electrical stimulation fits alongside chemical modalities in the broader landscape of biohybrid neural interfaces, noting that the field is still debating whether purely electrical approaches can capture the full complexity of synaptic communication or whether hybrid strategies combining voltage and neurotransmitter delivery will prove necessary.
What has not been proven yet
The distance between a successful brain-slice experiment and a working implant remains substantial. Several key unknowns stand between this proof of concept and clinical relevance.
No in vivo testing. Every result so far comes from ex vivo slices, which preserve local circuit wiring but lack blood flow, immune surveillance, and the mechanical forces of a living brain. The devices have not been tested in a living animal, and chronic implant performance, how the materials hold up over weeks or months inside warm, wet tissue, is entirely uncharacterized.
Biocompatibility is unresolved. Graphene and MoS2 are generally regarded as low-toxicity materials, but long-term tissue reactions depend on particle size, surface chemistry, and degradation byproducts. The Nature Nanotechnology paper demonstrates device performance, not tissue safety over extended implantation.
Scaling to useful arrays. The current experiments involve small numbers of devices interacting with slices of hippocampal tissue. A clinically meaningful interface for a condition like epilepsy or Parkinson’s disease would likely require hundreds or thousands of independently controllable channels. How densely the printed neurons can be packed without generating problematic heat or electrical crosstalk has not been studied.
Chemical signaling is missing. Real synapses do not communicate through voltage alone. They release neurotransmitters into a narrow cleft, triggering receptor-specific responses on the receiving cell. Whether purely electrical stimulation from memristive devices can replicate the richness of that chemical dialogue, or whether future implants will need to combine electrical and chemical delivery, remains an open and actively debated question.
No independent replication. As of May 2026, no outside laboratory has published a replication of the full device stack and brain-interfacing protocol. The underlying physics of threshold switching in similar memristive systems is well established in the literature, but the specific combination of aerosol-jet-printed graphene/MoS2 neurons communicating with living tissue has been demonstrated only by the Northwestern group.
Where the work fits in the larger race for better brain interfaces
Neural implant research is moving on multiple fronts simultaneously. Electrode-based systems are getting thinner and more flexible. Optogenetic approaches use light to control genetically modified neurons with millisecond precision. Ultrasound-based stimulation is being explored as a way to reach deep brain structures without surgery. Each strategy has tradeoffs in invasiveness, precision, longevity, and the type of information it can exchange with the brain.
The Northwestern printed neurons occupy a specific niche: they aim to be the first synthetic devices that do not just stimulate or record but actually participate in neural circuits as functional peers, firing and responding in real time. If that capability holds up in living animals and eventually in humans, it could enable a new class of implants that adapt dynamically to the brain’s activity rather than delivering pre-programmed stimulation patterns. For patients with epilepsy, that might mean a device that detects the earliest signs of a seizure and responds with precisely timed counter-spikes to halt it. For Parkinson’s disease, it could mean stimulation that adjusts moment to moment as the brain’s dopamine-related circuits fluctuate.
None of that is imminent. Animal trials, biocompatibility studies, regulatory review, and manufacturing scale-up each represent years of work. Engineers will still need to solve practical problems: packaging printed arrays onto substrates that conform to the brain’s curved surfaces, routing signals through minimally invasive connectors, and writing algorithms that translate between the brain’s patterns and the artificial network’s logic. But the fundamental question, whether a printed device can generate spikes that living neurons treat as real, now has a clear answer. The brain listened, and it talked back.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
