Northwestern Engineers Print Artificial Neurons That Successfully Communicate With Living Brain Cells

Engineers at Northwestern University have successfully printed flexible artificial neurons using nanoscale semiconductor materials that can generate electrical signals capable of activating living brain cells, a breakthrough with potential applications in brain-machine interfaces, neuroprosthetics, and more energy-efficient artificial intelligence hardware, the team reported in the journal Nature Nanotechnology.

The research, led by Professor Mark Hersam at Northwestern's McCormick School of Engineering, used a manufacturing technique called aerosol jet printing to deposit nanoscale flakes of molybdenum disulfide (MoS2) — a semiconductor — and graphene — an electrical conductor — onto flexible polymer substrates, creating artificial neurons that can be bent and shaped to match the contours of biological tissue.

When the printed neurons were tested against slices of living mouse brain tissue, they generated electrical signals realistic enough to trigger responses in actual biological neurons. The artificial neurons successfully communicated with the living cells, completing the first demonstration of printed electronics making meaningful electrochemical contact with real brain tissue — a result that opens a new frontier in the quest to develop electronics capable of speaking the electrical language of the nervous system.

Conventional brain implants — including cochlear implants, retinal prosthetics, and deep brain stimulators — rely on rigid metal electrodes that create a mechanical mismatch with the soft, constantly-moving tissue of the brain. This mismatch causes progressive inflammation and signal degradation over time, limiting the long-term efficacy of implanted devices and, in some cases, causing tissue damage at implantation sites. Flexible electronics that conform to neural tissue could dramatically extend implant lifetimes, improve signal quality, and reduce adverse biological reactions.

Hersam's group chose molybdenum disulfide because of its semiconductor properties at the nanoscale, where it behaves as a two-dimensional material similar to graphene — extraordinarily thin sheets of material with electronic properties tunable through quantum mechanical effects. The combination of MoS2 and graphene layers creates a composite that can replicate the threshold-based firing behavior of biological neurons, generating electrical spikes — action potentials — when triggered by incoming signals at sufficient intensity.

The aerosol jet printing technique allows the material to be deposited with fine spatial precision on flexible substrates that can conform to the contours of neural tissue. Unlike semiconductor lithography used in conventional chip manufacturing, aerosol jet printing works at much lower temperatures and can be applied to a wide range of surfaces — including, in principle, surfaces shaped to fit individual patients' neural anatomy. The printing process is also more easily scalable than many competing approaches to flexible bioelectronics.

The potential applications span several clinical domains. In neuroprosthetics, artificial neurons that can reliably interface with surviving neural circuits could restore function for patients with spinal cord injuries, hearing loss, or visual impairment with far greater precision than current electrode-based systems. In brain-machine interfaces for communication — devices enabling patients with ALS or locked-in syndrome to control computers or speech synthesizers directly from neural signals — higher-quality interfacing could increase bandwidth and reliability, potentially allowing for more nuanced and faster communication.

The energy efficiency implications for artificial intelligence computing are also significant. Biological neurons are extraordinarily efficient, processing information using electrochemical spikes with energy expenditures orders of magnitude below what silicon transistors consume per equivalent operation. Hardware that more closely mimics neuronal signaling — so-called neuromorphic computing — could reduce the power consumption of AI inference dramatically. This matters enormously given the escalating energy demands of large-scale AI infrastructure.

Hersam noted that the current demonstration used mouse brain tissue in laboratory conditions, and significant work remains to develop implantable devices suitable for human use. Regulatory pathways, long-term biocompatibility studies, and scalable manufacturing processes for clinical applications will require years of additional development before the technology reaches patients.