Appendix 4: Memristor Applications
89
mainstream memristors-CMOS circuits can be used in applications that require
adaptability and nonvolatility. Memristors offer a compact solution for the future
nonvolatile memory which could be used in Internet-of-Things (IoT), robotics,
medical and memory applications. In particular, highly predictable memristors
with high yield and long retention would support an immense range of possible
applications exploiting highly compact analog memories, such as precision circuits
with low overhead trimming, precise crossbar computing applications (e.g., very
fast matrix multiplications, with high impact for deep neural networks and machine
learning at large), and adaptive neural (and non-neural) like circuits. As memristor
technology is constantly developing, more and more practical application concepts
based on memristor devices are being demonstrated in various research papers.
Great interest has been raised in the field of neuromorphic applications as various
memristive technologies/devices (such as oxide based memristors) can have their
resistance shifted in small steps which renders them ideal in emulating the operation
of physical synapses. Other novel applications, for example the memristor-based
humanoid robot control system presented in [80, 81], exploit the ability of memristive devices to store and process data in the same physical location. In all cases,
memristors are utilized to replace conventional designs (CMOS-based designs with
Von Neumann architectures) for reducing total power consumption and increasing
energy efficiency.
There is also a crucial need for accurate and reliable models for the nonlinear
dynamics of resistance switching memories, especially those composed of novel
unexplored materials, where the physical mechanisms underlying the inherently
complex dynamics emerging in the nanodevices are still under study. Provided a
reliable mathematical description of a memristor is available, a thorough analysis of
the model through the application of methodologies from nonlinear dynamic system
theory [82, 83] may allow to predict the set of input/initial condition combinations
under which it is safe to operate the nano-device. Furthermore, the adoption of
concepts from stability theory and nonlinear dynamics may provide insights into
the biasing circuit arrangement necessary to induce specific dynamical phenomena
in memristors, e.g., locally active behaviors, bifurcations, and complex attractors
[84, 85]. All in all, the development of accurate, reliable, and numerically stable
memristor circuit models, but also the analysis through concepts and methodologies
from the theory of dynamical systems, are essential tools to provide circuit designers
with a comprehensive picture of the nonlinear response of the nanostructures to
input/initial condition combinations expected in the application of interest, helping
them to take more conscious decisions on the most suitable circuit topology to meet
prescribed specifications.
Appendix 4: Memristor Applications
Though not currently offered as part of standard semiconductor processes, a number
of companies are working towards integrating memristors with standard CMOS
89
mainstream memristors-CMOS circuits can be used in applications that require
adaptability and nonvolatility. Memristors offer a compact solution for the future
nonvolatile memory which could be used in Internet-of-Things (IoT), robotics,
medical and memory applications. In particular, highly predictable memristors
with high yield and long retention would support an immense range of possible
applications exploiting highly compact analog memories, such as precision circuits
with low overhead trimming, precise crossbar computing applications (e.g., very
fast matrix multiplications, with high impact for deep neural networks and machine
learning at large), and adaptive neural (and non-neural) like circuits. As memristor
technology is constantly developing, more and more practical application concepts
based on memristor devices are being demonstrated in various research papers.
Great interest has been raised in the field of neuromorphic applications as various
memristive technologies/devices (such as oxide based memristors) can have their
resistance shifted in small steps which renders them ideal in emulating the operation
of physical synapses. Other novel applications, for example the memristor-based
humanoid robot control system presented in [80, 81], exploit the ability of memristive devices to store and process data in the same physical location. In all cases,
memristors are utilized to replace conventional designs (CMOS-based designs with
Von Neumann architectures) for reducing total power consumption and increasing
energy efficiency.
There is also a crucial need for accurate and reliable models for the nonlinear
dynamics of resistance switching memories, especially those composed of novel
unexplored materials, where the physical mechanisms underlying the inherently
complex dynamics emerging in the nanodevices are still under study. Provided a
reliable mathematical description of a memristor is available, a thorough analysis of
the model through the application of methodologies from nonlinear dynamic system
theory [82, 83] may allow to predict the set of input/initial condition combinations
under which it is safe to operate the nano-device. Furthermore, the adoption of
concepts from stability theory and nonlinear dynamics may provide insights into
the biasing circuit arrangement necessary to induce specific dynamical phenomena
in memristors, e.g., locally active behaviors, bifurcations, and complex attractors
[84, 85]. All in all, the development of accurate, reliable, and numerically stable
memristor circuit models, but also the analysis through concepts and methodologies
from the theory of dynamical systems, are essential tools to provide circuit designers
with a comprehensive picture of the nonlinear response of the nanostructures to
input/initial condition combinations expected in the application of interest, helping
them to take more conscious decisions on the most suitable circuit topology to meet
prescribed specifications.
Appendix 4: Memristor Applications
Though not currently offered as part of standard semiconductor processes, a number
of companies are working towards integrating memristors with standard CMOS