Tunable spike-frequency adaptation in organic artificial neurons

Neuromorphic computing is one of the most promising computing paradigms to build the next generation of energy-efficient computing systems. Many materials and devices are being investigated to develop these new architectures. Organic electrochemical transistors (OECTs) have emerged as candidate devices for the design of artificial synapses. Their ability to emulate a variety of neurological mechanisms, such as short-term and long-term plasticity, their operation relying on ionic transport and the possibility for global connectivity make them intrinsically similar to neurobiological membranes. Moreover, they present low switching energies and a wide range of tunability. Other neuromorphic components must be created with a similar technology to build an entirely organic neuromorphic system for spiking neural networks. Indeed, organic circuits that emulate biological neurons are crucial. Few examples of artificial neurons fabricated with OECT technology exist in literature. They emulate some spiking features of biological neurons with neurotransmitter or ion-based modulation. However, they all spike at a constant frequency for a constant applied stimulus and fixed operating conditions, thereby having limited neural encoding capability. Spike-frequency adaptation (SFA) is a fundamental neuronal mechanism that encodes information by modulating firing activities: adaptive neurons show an initial high frequency spiking activity at the onset of a constant stimulus, that gradually reduces with time to a steady-state response. In this thesis, a new circuit architecture to emulate spike-frequency adaptation in an all-organic artificial neuron is proposed. This circuit relies on the short-term plasticity property of OECTs and the possibility to fabricate OECTs with a secondary tuning gate, that acts on its transcharacteristic. The circuit is demonstrated first in a hybrid version, with combined OECT and standard CMOS technologies. Subsequently, an all-organic version of the circuit is proposed. The circuit operates at a supply voltage < 1V and shows tunable spike-frequency adaptation. All the device modelling and circuit simulations have been carried out on LTSpice.