Improving BCI: Neural Network on a Chip
Neuroprosthetics that are implanted into the tissue of the brain, remain the greatest fidelity system available to us. Slowly, year on year, the size of implanted electrodes is shrinking, and the complexity of the electrode arrays themselves, is growing.
However, one major hurdle, is fully utilising that electrode array. As it is, work is heading towards a variable height array. This is where individual electrode pins descend until they make strong contact with neuron dendrites, or another cell gets in the way.
The latter case is much more likely than the former, at least given current approaches. However, what if you could encourage dendrites to grow, right where you needed them? Where the electrode array could make full use of its connections, and achieve 90% or higher utilisation?
Guiding the axons of nerve cells, to grow where desired, is nothing new. There are various chemical and topographical modification techniques to do so, although none have successfully been tried in a living human brain. Also, controlling the axon does not directly equate to controlling the dendrites of a neuron.
In new work that was recently published in Langmuir (Rectifying and Sorting of Regenerating Axons by Free-Standing Nanowire Patterns: A Highway for Nerve Fibers), Christelle Prinz, a postdoc researcher in the Division of Solid State Physics at Lund University in Sweden, thinks they have achieved an order of magnitude better than this: A short- range chemical trigger. The trigger encourages cell growth to change direction, and does not propogate far into the brain.
Essentially, rather than have the electrodes descend to meet the neurons, the dendrites of all nearby neurons, reorient to meet the prosthetic. By using a prosthetic implant more similar to a biochip than an electrode grid, the researchers have shown proof of concept, of encouraging brain neurons to grow inside the interface scaffold, making the implant literally a part of the brain.
With interactions this deeply embedded; rather than a 2D array of pins you get a 3D direct interaction with the brain's circuitry, as axons grow and spread across the channels inside an implanted chip. Connection with the brain can occur anywhere inside or outside the chip, greatly increasing the interaction possibilities.
The first application for this research would be in neural network design. By using the rectifying pattern, sensory neurons can be guided to one electrode for detection and motor neurons to another electrode for activation.
Full Text of above paper (PDF)