Northwestern University engineers have printed artificial neurons that do more than mimic the brain on paper — they talk to it. In a study published in Nature Nanotechnology, researchers reported that their printed devices generated electrical signals realistic enough to trigger responses in living neurons from mouse brain tissue, a step that the field has chased for years as it tries to bridge rigid silicon electronics with the soft, three-dimensional wiring of biology.
The work was led by materials scientist Mark C. Hersam, with Vinod K. Sangwan as co-lead and neuroscientist Indira M. Raman contributing on the biological side. It marks one of the first clear demonstrations that flexible, printable electronics can function as drop-in partners to real neural circuits rather than only as passive sensors or stiff implants.
How the devices are built
Instead of fabricating neurons on conventional silicon wafers, the Northwestern team formulated a set of electronic inks from nanoscale flakes of molybdenum disulfide (MoS2), which acts as a semiconductor, and graphene, which serves as an electrical conductor. Using aerosol jet printing, the researchers deposited these inks onto flexible polymer substrates and then selectively decomposed a stabilizing polymer to create localized conductive pathways.
The result is a soft, low-cost device that produces complex, neuron-like firing patterns rather than simple on/off pulses. When the printed neurons were placed in contact with slices of mouse brain tissue, their signals were strong and natural enough to elicit responses from the living cells — the kind of two-way biocompatibility that previous rigid electrode arrays have struggled to achieve.
"The brain is the opposite. It's heterogeneous, dynamic and three-dimensional," Hersam said of conventional computing hardware in a statement accompanying the paper. "To move in that direction, we need new materials and new ways to build electronics."
Why it matters for AI
The near-term applications are medical. Flexible printed neurons could underpin a new class of neuroprosthetics — implants for hearing, vision and movement that conform to tissue rather than fighting it — and give brain-machine interface researchers a cheaper, more biologically compatible substrate than today's dense silicon probes.
The longer-term implications point squarely at AI infrastructure. Large language models are now bound almost as much by electricity and memory bandwidth as by algorithms, with data center power draw emerging as a constraint on further scaling. The human brain, by contrast, runs at roughly the power of a dim light bulb. Hardware that signals more like real neurons — asynchronous, event-driven, analog — is one of the few credible paths toward drastically cutting the energy cost of inference.
Neuromorphic computing has promised this shift for a decade, but most previous prototypes either lived on the same rigid CMOS stack as ordinary chips or failed to interact meaningfully with biology. A printable, flexible, biocompatible neuron is a far more plausible building block for systems that blur the line between AI accelerator and living tissue.
What to watch next
The study was done on brain slices, not in living animals, and scaling up from a handful of printed neurons to networks large enough to do useful computation is a serious engineering problem. Durability, noise, and the ability to manufacture consistent devices at volume will determine whether this line of work reaches clinics and data centers or stays confined to the lab bench. Even so, the demonstration that printed devices can hold a conversation with real neurons is the kind of quiet result that tends to look, in hindsight, like a turning point.
