Aerosol Jet Printing, Molybdenum Disulfide And A Response From Living Mouse Neurons: Northwestern University Just Changed What Brain-Computer Interfaces Can Become
This article was compiled and made possible with the help of the following source(s): Nature Nanotechnology, Northwestern Now, EurekAlert!, ScienceDaily and SciTechDaily.
For decades, the boundary between biological and artificial intelligence has been defined by one fundamental barrier: silicon chips cannot meaningfully communicate with living neurons. They can receive signals from them, they can send pulses to them, but genuine bidirectional neural communication, the kind that mimics how brain cells actually coordinate, has remained firmly out of reach.
A study published in Nature Nanotechnology on April 15, 2026 from Northwestern University has now produced results that begin to dissolve that barrier in a way that no previous research has managed.
The Northwestern team successfully printed artificial neurons using a technique called aerosol jet printing, a high-precision fabrication process that uses a focused stream of aerosolized ink to deposit electronic materials onto flexible substrates with extraordinary resolution.
The artificial neurons were constructed from molybdenum disulfide, known in materials science as MoS₂, a two-dimensional semiconductor that carries a remarkable combination of properties: extreme thinness at the atomic scale, mechanical flexibility, semiconducting behaviour and high sensitivity to electrical signals.
When placed in contact with real mouse brain cells in a laboratory setting, these printed neurons successfully triggered measurable neural responses in the living tissue. The artificial cells were not merely passive receivers. They communicated.
To understand why this is significant, it helps to first understand what makes this problem so hard. Biological neurons operate through electrochemical gradients, ion channels and synaptic connections that transmit signals at energy levels measured in femtojoules, which are quadrillionths of a joule.
Conventional silicon electronics work at energy levels orders of magnitude higher and communicate through binary voltage states that share nothing structurally with the analog, continuous-wave signalling of the brain.
Every previous attempt to bridge this gap has faced the same problem: how do you create an artificial device that is simultaneously soft enough not to damage brain tissue, chemically compatible enough not to trigger immune responses, electrically precise enough to detect and generate signals at the scale of individual neurons, and robust enough to function reliably over time?
MoS₂ addresses several of these challenges simultaneously. Research from MIT’s Microsystems Technology Laboratories, published in Advanced Materials in 2024, demonstrated that two-dimensional transition metal dichalcogenides, the material class that includes MoS₂, can be tuned to exhibit synaptic plasticity, meaning they can strengthen or weaken their signal transmission in response to repeated stimulation, directly mimicking Hebbian learning, the foundational mechanism by which biological synapses change through experience.
The Northwestern team built on this body of work by demonstrating not just that MoS₂ could simulate neural behaviour in isolation but that it could do so when integrated directly with living biological neural networks.
Parallel research from the University of Bath, published in Bioelectronics in 2023, demonstrated that silicon-based artificial synapses could replicate the basic firing behaviour of biological neurons under controlled laboratory conditions. But silicon’s rigidity and its incompatibility with biological tissue remained significant limitations that prevented in-vivo application. MoS₂, being mechanically flexible at atomic thicknesses, removes the rigidity problem entirely.
The energy dimension of this research deserves close attention. The human brain operates at approximately 20 watts, consuming roughly the same power as a dim light bulb, while achieving computational performance that no silicon system has come close to replicating in terms of efficiency per operation. IBM Research has estimated that the human brain is approximately 100,000 times more energy-efficient than today’s best digital computers when performing equivalent cognitive tasks.
This efficiency gap is the central challenge facing AI hardware development, and it is precisely this gap that neuromorphic computing, using hardware that mimics the architecture and operating principles of biological neural networks, aims to close.
The Northwestern result therefore carries two distinct sets of implications. The first concerns medical neurotechnology. Cochlear implants, visual prosthetics and deep brain stimulators all rely on electrodes that interface with neural tissue, and all face the same core limitation: the signal handshake between electronics and biology is crude, one-directional and rapidly degraded by the body’s immune response to foreign materials.
If artificial neurons made from MoS₂ can genuinely integrate with living tissue and establish bidirectional communication, the door opens to implantable devices that do not merely stimulate neurons but participate in neural circuits, adapting and responding in real time rather than delivering preprogrammed pulses. The clinical implications for conditions including spinal cord injury, deafness, blindness and neurodegenerative disease are substantial.
Researchers at the BrainGate consortium at Brown University have spent two decades refining cortical implants for people with paralysis, and their work consistently identifies the degradation of the electrode-tissue interface over time as the primary technical barrier to clinical deployment. Materials that can form stable, responsive interfaces with living neurons without triggering fibrosis or immune rejection would represent a fundamental advance in translating laboratory results into usable therapies.
The second set of implications concerns computing hardware. Neuromorphic chips, including Intel’s Loihi 2 and IBM’s NorthPole processor, attempt to replicate the brain’s parallel, event-driven, low-power architecture using conventional semiconductor fabrication techniques. The results have been promising but constrained by the physical properties of silicon and the energy overhead of conventional transistor design.
MoS₂-based neuromorphic hardware could potentially achieve lower switching energies and higher integration densities than silicon, enabling neuromorphic chips that more closely approximate the brain’s actual operating efficiency rather than simply borrowing its architectural principles.
The evaluation of the Northwestern result must also be honest about what remains unknown. The demonstration involved mouse neurons in laboratory conditions, not an intact brain, and the artificial neurons triggered responses without full characterisation of what those responses represented functionally.
Triggering a neural response is not the same as forming a functional synapse, and the pathway from a laboratory result to a clinically deployable brain-computer interface requires solving problems in biocompatibility, long-term stability, scalability of fabrication and immune response that no single study can address.
What the Northwestern study establishes, however, is that printed artificial neurons made from two-dimensional semiconducting materials can produce meaningful interactions with living neural tissue. That is a new fact about the world, and it opens research directions that were not credibly accessible before.
The brain evolved over hundreds of millions of years to become the most energy-efficient information processing system on Earth. Science has spent decades trying to reverse-engineer it in silicon. The Northwestern result suggests that a different approach, using materials that share the flexibility, sensitivity and scale of biological systems, may bring us closer to building devices that do not merely interface with the brain but speak its language.
Like this science-focused extensive analysis driven article? Read our previous article here.
