Topologically Defined Neuronal Networks Controlled by Silicon Chips

Summary
Small neuronal networks with defined synaptic connection patterns could provide a unique tool for
studying fundamental concepts in neuroscience, because they are much simpler than in vivo systems
and give easy access to individual neurons. Field-effect transistors and capacitive stimulators implemented
on a semiconductor chip establish a non-invasive neuron-silicon interface, that enables the long
term supervision even of large neuronal assemblies.
The present thesis addresses the experimental and technological aspects of network design. Based on
existing transistor chips a new device was processed for the specific requirements here. Small chipcontrolled
networks were grown from neurons of the snail Lymnaea stagnalis and characterized.
First, a new method for directing neurite outgrowth was developed. It is based on topographic guidance
cues consisting of pits and narrow connecting grooves, which were processed from SU-8 polyester photoresist.
Neurons placed into the pits grew neurites that followed the grooves and established electrical
synapses upon contact with other neurites or somata. These networks were studied with standard electrophysiology
and the conductance of the synapses was determined. On average, it was larger than the
conductance of synapses between cell pairs grown on protein tracks.
Besides guiding their outgrowth, the topographic structures also keep neurites in the final geometry
and confine cell bodies to the pits. This is a major advantage compared to techniques using chemical
patterns; there, neurites and somata are frequently pulled away by forces exerted by the growth-cones.
Moreover, the structures are reusable many times, which makes them very efficient.
Building on established technologies for extracellular recording and stimulation of neural activity, a
transistor chip was processed next. Its layout, 16 bidirectional contacts arranged in a 4x4 array, was
especially designed for monitoring small, defined networks of snail neurons. Each contact comprised
a buried channel field-effect transistor for recording action potentials, surrounded by capacitive stimulation
spots. Chips were characterized electronically and tested with single neurons. The transistors
recorded a variety of signal types, similar to the results of previous works. Most of the extracellular
signals could be qualitatively explained with the point-contact model for the neuron-silicon interface
and the Hodgkin-Huxley model describing the voltage dynamics of the neuron. Extracellular stimulation
was very reliable, 5-10 square wave pulses with 1V-5V amplitude applied to the capacitive spots
evoked action potentials in the cell above.
The third step combined both technologies, SU-8 topographic structures and silicon chips. For the
first time, hybrid networks with defined geometry were realized. The most fundamental system, the
silicon-neuron-neuron-silicon loop, was studied in detail. Upon extracellular stimulation the respective
presynaptic neuron fired an action potential, which was detected by the transistor. The signal propagated
along the neurites, passed the synapse and depolarized the postsynaptic cell where it triggered
an action potential, if the input was strong enough. Again, neuronal activity was recorded from the
transistor underneath. Larger networks with three and four neurons were also grown and characterized.
The examples presented here constitute proof-of-principle experiments. They demonstrate that topologically
defined hybrid networks, combining biology and semiconductor technology, can be implemented
in functional systems, and pave the way for future applications such as a ‘living’ neurocomputer.