A Low-temperature Approach to Spiking Neural Circuits

Neuromorphic circuits seek to emulate biological neurons and synapsesfor use in applications that involve processing large quantities of impreciseinformation. Implementations of the soma (cell-body) using silicon CMOS have been proposed, but these cannot be fabricated over large areas on flexible substrates that are often indispensable. We have fabricated spiking soma circuits that operate at biological time scales and are compatible with flexible substrates using ambipolar nanocrystalline-silicon thin-film transistors with a maximum process temperature of 250°C. These circuits can easily be scaled down to realize high circuit densities. Depending on the complexity of the soma circuit, increased biological functionality such as spike frequency adaptation, and setting specific refractory periods can be achieved. The frequency-current (discharge) curve of the circuit also resembles that of biological neurons. In addition, we have designed a simple synapse circuit that operates using rectangular voltage pulses to produce a Hebbian learning rule that depends on both spike frequency and timing. Finally, SPICE simulations show that these synapse designs can perform useful computation including associative learning, coincidence detection, and extraction of a fundamental frequency.