Fig. 1: Transistors and their use in neuro-synaptic-mimicking devices.
From: Synaptic and neural behaviours in a standard silicon transistor
a, General structure of an ANN comprising neurons and synapses (left). Some of the typical I–V or resistance behaviours of these devices are schematically represented: thresholding behaviour of neurons, update of the synaptic strength (weight) of synapses between neurons, and short-term plasticity displayed by synapses in spiking neural networks. b, Examples of electronic devices capable of mimicking some neuro-synaptic behaviours. Left, a single neuron core can comprise more than 20 CMOS transistors. Middle, the most basic digital cell used as a binary synapse is a six-transistor static random access memory cell. Right, an alternative implementation of a multilevel synapse using a floating-gate device. c, Representative examples of emerging memory devices, such as memristors based on phase-change materials or metals on oxides, which are being explored for their use as neurons and synapses in neuromorphic circuits. sn, synaptic update input event (typically a voltage or current pulse); Vsyn, post-synaptic signal as input to the neuron block; Ibias, neuron circuit reference bias current; Vth, threshold voltage setting; Vadap, input voltage for adaptive thresholding control; Ifb, feedback current; Vrf, output refractory period setting voltage; Vspike, spiking neuron output voltage; DEP, depression; POT, potentiation; VB, bias voltage (read/update control); SRAM, static random access memory.
