IEEE Circuits and Systems Magazine - Q1 2018 - 42

enable its adoption in real-life applications, e.g., enhancing memristor-based computing precision, co-optimizing
algorithm and hardware for nonconvex optimization, and
determining the feasibility of other problems that can
benefit from memristor-based hardware implementation.
Some specific future directions are discussed below.
First, memristor-based computing systems have not yet
demonstrated a competitively high computation accuracy
for solving practical problems in the presence of hardware
variations. To enhance precision, extra hardware resources would be needed. It is thus essential to optimize a full
hardware system under given hardware resources. Problems
of interest include selection of device-level components in
hardware implementation, and design of energy-efficient
on-chip communication infrastructure.
Second, the convergence of ADMM for nonconvex
optimization is not guaranteed. Therefore, new optimization algorithms, appropriate for hardware design, are
desired to address nonconvex problems, e.g., artificial
neural network based applications. Traditional algorithms
to train neural networks, such as back-propagation or
other gradient-based approaches, require updating of
the gradient information at each iteration. This leads to
frequent writing/reading operations on memristor crossbars and thus an increasing amount of energy consumption. Motivated by that, innovation beyond the existing
algorithms is encouraged to co-optimize algorithm and
hardware for nonconvex optimization.
Third, in many scenarios, it is assumed that certain
solutions exist for the considered optimization and machine learning problems. However, it is possible that the
mapped problems on memristor crossbars are infeasible, e.g., no solution exists for an overdetermined linear
system. Therefore, a robust memristor crossbar-based
solver should be capable of identifying the feasibility of
problems. This identification procedure should be implemented by using device-level components subject to
limited hardware resources.
Fourth, there is much work to be done to expand the
applications of memristor crossbars from the end-user
perspective. Some potential lucrative applications include
memristor-based smart sensors, small footprint intelligent controllers in wearable devices, and on-chip training
platforms in autonomous vehicles and Internet of Things.
To sum up, memristor technology has the potential
to revolutionize computing, optimization and machine
learning research due to its orders-of-magnitude improvement in energy efficiency and computation speed.
Moving forward, engineers and scientists in different
fields, such as, machine learning, signal processing, circuits and systems, and materials should collaborate with
each other to make significant progress on this exciting
research topic.
42

ieee circuits and systems magazine

Sijia Liu (S'13-M'16) received the B.S. and
M.S. degrees in electrical engineering
from Xian Jiaotong University, Xian, China, in 2008 and 2011, respectively. He
received the Ph.D. degree (with All University Doctoral Prize) in electrical and
computer engineering from Syracuse University, Syracuse,
NY, USA, in 2016. He was a Postdoctoral Research Fellow at
the University of Michigan, before joining in IBM Research
AI. His research interests include resource management in
wireless sensor networks, optimization for machine learning, graph signal processing, and information fusion. He
received the Best Student Paper Award (third place) at the
42nd IEEE International Conference on Acoustics, Speech
and Signal Processing (ICASSP) in 2017. He was also among
the seven finalists of the Best Student Paper Award at the
Asilomar Conference on Signals, Systems, and Computers
in 2013. He was the winner of the Nunan research poster
competition at Syracuse University in 2012.
Yanzhi Wang is currently an assistant
professor at Syracuse University, starting from August 2015. He received B.S.
degree from Tsinghua University in 2009
and Ph.D. degree from University of
Southern California in 2014, under supervision of Prof. Massoud Pedram. His research interests
include neuromorphic computing, energy-efficient deep
learning systems, deep reinforcement learning, embedded systems and wearable devices, etc. He has received
best paper awards from International Symposium on Low
Power Electronics Design 2014, International Symposium
on VLSI Designs 2014, top paper award from IEEE Cloud
Computing Conference 2014, and best paper award and
best student presentation award from ICASSP 2017. He
has two popular papers in IEEE Trans. on CAD. He has received multiple best paper nominations from ACM Great
Lakes Symposium on VLSI, IEEE Trans. on CAD, and Asia
and South Pacific Design Automation Conference., and International Symposium on Low Power Electronics Design.
Makan Fardad (M'08) received the B.S.
degree from Sharif University of Technology, the M.S. degree in electrical
engineering from Iran University of Science and Technology, and the Ph.D.
degree in mechanical engineering from
the University of California, Santa Barbara. He was a
postdoctoral associate at the University of Minnesota
before joining the Department of Electrical Engineering
and Computer Science at Syracuse University. His research interests include modeling, analysis, and optimization of large-scale dynamical networks.
first quarter 2018

Table of Contents for the Digital Edition of IEEE Circuits and Systems Magazine - Q1 2018