Researchers at Northwestern University’s McCormick School of Engineering have demonstrated printed neuromorphic devices that generate action-potential-shaped voltage spikes and trigger responses in live mouse cerebellum tissue, according to a paper published in Nature Nanotechnology on 15 April. The work, led by Mark C. Hersam, Walter P. Murphy Professor of Materials Science and Engineering, with research associate professor Vinod K. Sangwan and neurobiology collaborator Indira M. Raman, is the first reported demonstration of a printed artificial neuron communicating directly with living neural tissue at biologically realistic voltages and timescales.
The device is built from aerosol-jet-printed inks containing nanoscale flakes of molybdenum disulfide, which behaves as a semiconductor, and graphene, which acts as a conductor. The inks are deposited onto flexible polymer substrates. The group’s key methodological choice was to partially decompose the stabilising polymer during processing, rather than burning it off entirely, as previous printed-electronics work has done. The residual polymer creates narrow conductive filaments that give the device memristive switching and produce spiking dynamics that earlier printed devices have not achieved.
Raman’s laboratory in Northwestern’s Weinberg College of Arts and Sciences applied the printed neurons’ output to slices of mouse cerebellum. The synthetic voltage spikes matched biological action potentials in timing and duration closely enough to reliably activate real neurons in the tissue. Neuromorphic hardware has produced spike-shaped signals in silicon for more than a decade, but most of those devices cannot drive wet tissue without an intermediate conditioning stage. Printed, flexible devices that speak the correct electrochemical language directly to neurons are rare in the literature.
Fabrication economics for BCI hardware
The dominant clinical-stage BCI electrode programmes (Neuralink, Synchron, Precision Neuroscience, Paradromics, INBRAIN) use fabrication processes that resemble semiconductor manufacturing: wafer-based deposition, photolithographic patterning, chemical etching, and cleanroom handling. These processes are capital-intensive, constrain device geometry to what cleanroom tools can produce, and keep per-channel cost high. Aerosol-jet printing can deposit functional materials onto arbitrary flexible substrates at roll-to-roll speeds and at a fraction of the per-unit cost of lithographic processes. If printed neuromorphic devices eventually augment or replace lithographically fabricated electrodes for some classes of neural interface, the cost structure of BCI hardware changes.
Hersam framed the broader motivation in energy-efficiency terms, noting that the brain is “five orders of magnitude more energy efficient than a digital computer” and pointing to potential applications spanning AI compute, hearing restoration, vision restoration, and motor prosthetics. The team lists brain-machine interfaces and neuroprosthetics as long-term targets for the technology alongside lower-power neuromorphic AI computing.
What the study does not answer
The work is preclinical and distant from any human indication. The team tested on brain slices, not in vivo. Biocompatibility of the printed MoS₂-graphene composite under chronic implant conditions, long-term electrochemical stability of the ink in cerebrospinal fluid, and scaling from a single device to a multi-channel array are all open questions the paper does not address. The demonstration also uses a cerebellar slice, not cortex, and targeting cortical tissue at clinically relevant densities is a separate engineering problem.
The paper also does not provide a head-to-head energy-per-spike comparison between the printed devices and conventional silicon neuromorphic hardware, so the energy-efficiency argument, while implicit in the materials choice, is not yet quantified at the device level.
Lab background
Hersam’s laboratory has published 2D-material printed-electronics work since 2012, including earlier demonstrations of printed synapses, inverters, and memristive logic elements using similar ink chemistries. This Nature Nanotechnology paper is the first from the group to demonstrate coupling to living tissue, which moves the programme out of pure materials science and into the application space that the authors have been signalling for several years. The cerebellum-slice validation was done jointly with Raman’s laboratory in Northwestern Neurobiology.
Funding for the printed-electronics research line has come from the US Department of Energy, the National Science Foundation, and DARPA across multiple materials programmes. The group’s next milestones, based on the paper’s discussion, are multi-channel printed arrays and in vivo validation, though no timeline has been disclosed.
Clinical-stage BCI hardware is on a multi-year trajectory toward higher channel counts, lower per-unit cost, and more flexible form factors. Every company in the category is pushing against the same fabrication ceiling. The Nature Nanotechnology paper is not a product. It is a data point suggesting printed approaches, which have been absent from that conversation, may need to be in it.