Neuromorphic computers have architectures that can mimic a human brain with billions of neurons and
their synapses to store, retrieve and process information using future chip technology. One of the driving
forces of neuromorphic computing is the requirement for low energy consumption, as present day chips
still consume tens if not hundreds of watts for a relative small fraction of tasks capable in a human
brain. Neurons are either digital or analogue and can work asynchronously. However, the power
performance and density of synapses is of critical importance as they far exceed neurons in number.
Two terminal memristor devices such as (Resistive Ram RERAM, Conductive Bridge Memory (CB
RAM), Phase Change Memory (PCM) and ferroelectric Memory (FeRAM) have been shown to have
excellent properties for non-volatile RAM applications that mimic the properties of synapses. [1-4]. In
biological systems, signal transmission and learning process occurs concurrently and the learning
process is via a change in synaptic weight. The present two terminal synaptic devices that function as a
connecting element between a pre and post neuron terminals, have the limitation that signal
transmission and learning functions cannot be carried out simultaneously. The learning function in two
terminal devices are carried out by feeding the signal from post neuron to the synaptic device terminal
to modulate the synaptic weight while the synaptic transmission is inhibited. However, three terminal
synaptic devices are able to realise both signal transmission and learning functions, where signal
transmission is carried via the channel medium and synaptic weights are modulated independently via
the gate terminal. Of the three terminal based synaptic devices, Ferroelectric FETs (FeFETs) and ion
gated FETs/TFTs are the most discussed [4-9]. Oxide based memristors in recent years have reported
properties of signal transmission and self-learning: at the heart of their mechanism lies a motion of
positively charged oxygen vacancies that have been widely examined in the context of stability and
performance of Transparent oxide thin film transistors (TFTs). In this talk, we will present the latest
developments that summarise and compare the performance of a ZnO/Ta2O5 based memristor “system”
with other materials reported to date. The synaptic functions are emulated at a low programming voltage
of 200 mV, an order of magnitude smaller than in conventional RERAM. The synaptic behaviour of
our transistors is found to be independent of environmental factors. By the application of relatively
higher pre-synaptic spike signal at the gate terminal, a long term memory retention lasting up to several
(>3) hours is observed. We will present data analysis that explains the relaxation kinetics of the oxygen
vacancies in these samples and draw conclusions on possible mechanisms related to observed synaptic
and transistor characteristics.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)