Scientists just built a living neural network they can wire up, train, and use as a computer

by

Researchers at Princeton have taken a network of living neurons, wired it with flexible electronics, and used electrical pulses to train it to compute. This is not a simulation. The neurons are real, the wiring is real, and the device that made it possible kept the network stable and functional for six months.

The work, published in Nature Electronics on 23 April 2026, describes a device the researchers call 3D-MIND. It is a three-dimensional flexible electronic array that wraps directly around a living neural network grown in a laboratory dish. Electrodes distributed across multiple planes record action potentials, the electrical pulses neurons use to communicate, from different depths simultaneously. The same array can deliver targeted electrical stimulation to specific regions to reshape how the network is wired.

A flexible 3D electronic array wraps around a living neural network

The neural network is real, living tissue. Neurons are cultured to grow in three dimensions rather than spread flat across a surface. The electronics conform to the network’s irregular shape without crushing or damaging it.

The key result that separates this from earlier attempts is time. Previous devices that tried to interface electronics with three-dimensional neural tissue degraded within days or weeks. The chemistry at the boundary between living cells and electronic materials is hostile, and the mechanical mismatch between soft tissue and rigid components causes failures. The Princeton device maintained stable recordings for six months, long enough to watch a neural network’s connection map change over time, something no lab had done in three dimensions before.

The researchers then used the stimulation capability to do something more deliberate. They applied electrical pulses to specific connections within the network, strengthening some and weakening others.

By repeatedly stimulating selected neural pathways, researchers strengthened some connections and weakened others

This mimics how the brain learns: connections between neurons that fire together repeatedly grow stronger. By repeating this in a controlled way, the team converted the living network into a reservoir computing system, a type of computing architecture that uses a complex, dynamic physical system to process information. The network was not just being observed. It was being used as hardware.

Why does this matter beyond the lab?

The first reason is drug development. Brain disease drugs have an unusually high failure rate in clinical trials. A large part of the reason is that the standard test model, a flat layer of neurons on a dish, behaves differently from three-dimensional tissue in a living brain. The geometry changes how neurons connect, how signals travel, and how cells respond to drugs. A three-dimensional network that can be monitored in full volume over six months, with drug responses tracked across all depths in real time, is a substantially more accurate model. It does not replicate a human brain. But it removes one of the most significant sources of distortion between the lab and the clinic.

The technology matters beyond the lab

The second reason is neuromorphic computing. Silicon chips process information by moving electrons through transistors. The brain processes information by moving signals through neurons, and it does certain tasks, particularly pattern recognition and learning, at energy costs no silicon chip has matched. This work is the first demonstration that a living three-dimensional neural network can be stably wired, trained, and used as a computing substrate over an extended period. The gap between this and anything deployable is large. But the proof did not exist before this paper.

The limitations are straightforward. These are cultured neurons in a laboratory, not tissue from a living organism. Scaling to a network with meaningful computational capacity would require fabrication advances not described here. The six-month stability figure applies to a controlled research device.

None of that makes this incremental. The combination of three-dimensional architecture, long-term electronic stability, and trainable living computation has not been demonstrated in a single device before. It changes what neuroscientists can measure, what pharmacologists can test, and what engineers can build with biology as a component.

Sources

More Like This