Chapter 9
Memristor Cellular Neural Networks
Computing in the Flux-charge Domain
9.1 Introduction
A memristor is a nonlinear device obeying Ohm’s law but, unlike a resistor, the
memristor resistance, also called memristance, depends upon the history of the
voltage applied or the current flowing through it. A memristor is then both a
nonlinear and a memory element in the (v, i)-domain. Another unique property is
nonvolatility, namely, when current (or voltage) is turned off, the memristor can
keep in memory the final value of charge, flux, or memristance, thereafter (see
Chap. 2).
It is widely believed that memristors are potentially useful and will play a
major role in the design of neural networks (NNs), smart computers, and future
brain-like machines for efficient applications in edge computing and the Internet
of Things (IoT) era [1–7]. Memristors are indeed expected to provide various
advantages, such as scalability, small on-chip area, low power dissipation in the
synapse implementation, efficiency, and better adaptation capability with respect to
their CMOS counterparts.
So far the literature has mainly investigated the use of memristors to implement
adaptive synapses in neuromorphic architectures [2, 3]. In these applications, use
is made of the fine-resolution programming of the memristance, tuned by the
input amplitude, pulsewidth, and frequency, in memristor acting as a nonvolatile
memory. Memristors are subject to low voltages during their operation as analog
circuit elements and (relatively) high voltages to program their memristance, i.e.,
memristors are exploited as pulse-programmable resistances.
Goal of the chapter is to take a different viewpoint and exploit the nonlinear
dynamic features in the (v, i)-domain and nonvolatility of memristors to implement
a class of NNs that features some potential advantages over the standard cellular
neural networks (SCNNs) proposed by Chua and Yang [8]. We refer to the networks
as memristors SCNNs (M-SCNNs). In a M-SCNN, nonlinear memristors are used
within the cells in place of linear (memoryless) resistors of SCNNs. Via a suitable
© Springer Nature Switzerland AG 2021
F. Corinto et al., Nonlinear Circuits and Systems with Memristors,
https://doi.org/10.1007/978-3-030-55651-8_9
343
Memristor Cellular Neural Networks
Computing in the Flux-charge Domain
9.1 Introduction
A memristor is a nonlinear device obeying Ohm’s law but, unlike a resistor, the
memristor resistance, also called memristance, depends upon the history of the
voltage applied or the current flowing through it. A memristor is then both a
nonlinear and a memory element in the (v, i)-domain. Another unique property is
nonvolatility, namely, when current (or voltage) is turned off, the memristor can
keep in memory the final value of charge, flux, or memristance, thereafter (see
Chap. 2).
It is widely believed that memristors are potentially useful and will play a
major role in the design of neural networks (NNs), smart computers, and future
brain-like machines for efficient applications in edge computing and the Internet
of Things (IoT) era [1–7]. Memristors are indeed expected to provide various
advantages, such as scalability, small on-chip area, low power dissipation in the
synapse implementation, efficiency, and better adaptation capability with respect to
their CMOS counterparts.
So far the literature has mainly investigated the use of memristors to implement
adaptive synapses in neuromorphic architectures [2, 3]. In these applications, use
is made of the fine-resolution programming of the memristance, tuned by the
input amplitude, pulsewidth, and frequency, in memristor acting as a nonvolatile
memory. Memristors are subject to low voltages during their operation as analog
circuit elements and (relatively) high voltages to program their memristance, i.e.,
memristors are exploited as pulse-programmable resistances.
Goal of the chapter is to take a different viewpoint and exploit the nonlinear
dynamic features in the (v, i)-domain and nonvolatility of memristors to implement
a class of NNs that features some potential advantages over the standard cellular
neural networks (SCNNs) proposed by Chua and Yang [8]. We refer to the networks
as memristors SCNNs (M-SCNNs). In a M-SCNN, nonlinear memristors are used
within the cells in place of linear (memoryless) resistors of SCNNs. Via a suitable
© Springer Nature Switzerland AG 2021
F. Corinto et al., Nonlinear Circuits and Systems with Memristors,
https://doi.org/10.1007/978-3-030-55651-8_9
343