90
2 Fundamental Properties of Mem-Elements
processes. Memristor devices have been so far proven useful to multiple application
areas ranging from neuromorphic computing as artificial synapses to biosensors.
This section provides an overview of memristor applications.
Artificial Synapses
Brain-inspired computing is an emerging field, which can extend the capabilities of
information technology beyond the Von Neumann paradigm. Biologically inspired
systems, aiming to emulate the nervous system of living beings, are the best solution
to solve ill-posed problems, such as real time interaction with the external environment or pattern recognition. In biological systems, neurons are interconnected
and they interact among them through synapses, which can change their strength
in order to inhibit (synaptic depression) or facilitate (synaptic potentiation) the
connection between two neurons. This ability is called synaptic plasticity and it
is a key mechanism used by the brain in learning processes. In neuromorphic
architectures, whilst CMOS technology is typically used to design artificial neurons,
the implementation of artificial synapses poses a serious challenge. In fact, synapses
outnumber neurons by 3–4 orders of magnitude, therefore calling for high-density
and low-power devices. Moreover, they should be CMOS-compatible in order to
be easily integrated with CMOS-based neurons [86]. In this scenario, Resistive
Switching (RS) devices, or memristors, are considered good candidates to be used
as artificial synapses due to their high scalability and low power consumption.
Memristors are two-terminal devices able to change their conductance from a High
Conductance State (HCS) to a Low Conductance State (LCS), and vice versa. The
change is induced upon application of proper electrical stimuli. This behavior is
used to emulate synaptic plasticity. More specifically, the transition from HCS to
LCS emulates a depression operation whereas the reverse transition emulates a
potentiation operation [86]. Several other applications of memristive devices as
artificial synapses include: circuits for encryption/decryption of medical data [87],
computer arithmetic systems [88], spiking networks implementing unsupervised
learning [89], real-time encoding and compression of neuronal spikes [90], attractor
networks [91], oscillatory networks for unconventional computing [92], unbiased
generation of random numbers [93], emulation of short-term synaptic dynamics
[94], modeling of biochemical reactions [95], maze-solving computations [96],
and circular buffer for real-time signal processing [97]. Furthermore, memristive
devices can have other functions in the emulation of biological computation, e.g.,
the implementation of neural dynamics.
2 Fundamental Properties of Mem-Elements
processes. Memristor devices have been so far proven useful to multiple application
areas ranging from neuromorphic computing as artificial synapses to biosensors.
This section provides an overview of memristor applications.
Artificial Synapses
Brain-inspired computing is an emerging field, which can extend the capabilities of
information technology beyond the Von Neumann paradigm. Biologically inspired
systems, aiming to emulate the nervous system of living beings, are the best solution
to solve ill-posed problems, such as real time interaction with the external environment or pattern recognition. In biological systems, neurons are interconnected
and they interact among them through synapses, which can change their strength
in order to inhibit (synaptic depression) or facilitate (synaptic potentiation) the
connection between two neurons. This ability is called synaptic plasticity and it
is a key mechanism used by the brain in learning processes. In neuromorphic
architectures, whilst CMOS technology is typically used to design artificial neurons,
the implementation of artificial synapses poses a serious challenge. In fact, synapses
outnumber neurons by 3–4 orders of magnitude, therefore calling for high-density
and low-power devices. Moreover, they should be CMOS-compatible in order to
be easily integrated with CMOS-based neurons [86]. In this scenario, Resistive
Switching (RS) devices, or memristors, are considered good candidates to be used
as artificial synapses due to their high scalability and low power consumption.
Memristors are two-terminal devices able to change their conductance from a High
Conductance State (HCS) to a Low Conductance State (LCS), and vice versa. The
change is induced upon application of proper electrical stimuli. This behavior is
used to emulate synaptic plasticity. More specifically, the transition from HCS to
LCS emulates a depression operation whereas the reverse transition emulates a
potentiation operation [86]. Several other applications of memristive devices as
artificial synapses include: circuits for encryption/decryption of medical data [87],
computer arithmetic systems [88], spiking networks implementing unsupervised
learning [89], real-time encoding and compression of neuronal spikes [90], attractor
networks [91], oscillatory networks for unconventional computing [92], unbiased
generation of random numbers [93], emulation of short-term synaptic dynamics
[94], modeling of biochemical reactions [95], maze-solving computations [96],
and circular buffer for real-time signal processing [97]. Furthermore, memristive
devices can have other functions in the emulation of biological computation, e.g.,
the implementation of neural dynamics.