IEEE Circuits and Systems Magazine - Q4 2019 - 7

d to Me
mristor
Theory
Recon
[8]
figurab
le
Logic C
STDP
ir
c
u
D
it
e
ry
[32]
Statefu monstratio
Complel Logic Proto n [58]
type [3
mentary
3]
RRAM
[30]
Hybrid
C
Embed rossbar/CM
ded RR
O
AM (28 S Chip [59]
nm) [17
Comm
]
Percepercial RRAM
Memris tron Pattern Microcontroll
C
e
tive Ne
uristor lassification [6r [18]
[54]
0]
16 Gb
RRAM
Prototy
pe [61]
12 × 12
Percep
tron Sy
stem [6
Memri
2]
Spikingstor Dot-Prod
uct Eng
Neural
ine
Network
Prototy[47]
Sparse
pe [55]
Coding
Chip [6
3]

s Linke

Hybrid

R

Semiconductor Manufacturing Company (TSMC) in 2019
[27]. More importantly, they can be used for purposes
other than secondary memory, code-storage, or similar
conventional ways non-volatile memories are often used
[24]. There have been several efforts in using memristors for implementing various logical functions [28]-
[44], calculations [38], [45]-[49], and other applications
[50], [51] such as learning [51]-[55] and even cancer detection [56]. Fig. 1 summarizes some of the major events
in the memristive community since 2008 [57].
Logic circuit design is the key to the development of
memristor-based computing systems. A noteworthy observation in this regard is that the majority of these logics are inspired by CMOS and in a certain way mimic behaviors of a CMOS circuit or replace parts of it. Among
these logics CMOS/Memristor Threshold Logic (based
on the Logic Threshold Gate (LTG)) [28], [37], Ratioed
Logic [28], [30], and CMOS-like Logic [28] are among the
most prominent ones. One of the major common properties of all of these logics is that they operate in the
voltage domain (information and logical states are represented in the voltage of a node). This includes some
more recent ones like Scouting Logic [44] too. Due to the
voltage representation of the values, they often compete with their traditional and far more mature CMOS
counterparts. However, memristive technology can be
more successful if its native properties such as having memory are exploited. This idea has been used in
IMPLY Logic [31], [35], [64] and Memristor-Aided Logic
(MAGIC) [39], [43] which fundamentally operate in the
memory domain. That is, the information and logical

Device

I. Introduction
educing energy consumption is a crucial goal in
the current circumstances of rapidly growing computational load. Mobile systems such as smartphones, embedded systems, wearable electronics, and
Internet of Things (IoT) devices, which are often powered
by batteries or rely on energy harvesting, require optimal
utilization of the available energy [1]. A key factor in the
energy budget of modern computing devices is memory
[2]-[4]. In a modern chip, the number of transistors required to store data has a significant and increasing impact on the total transistor count [5], and consequently
on the production cost. A promising solution for these
problems is using memristors.
Although memristive behavior has been observed
before [6], [7], a turning point for this type of basic
circuit elements was when Hewlett Packard (HP) presented some of the (circuit level) applications of their
passive solid-state Resistive Random Access Memory
(ReRAM) devices with memristive characteristics in
2008 [8]. They advocated the memristors and their applications in the scientific community, especially that
of circuits and systems. Thanks to their non-volatility,
memristors could decrease the overall power consumption of the system dramatically [9]. Moreover, the relatively simple structure of memristors, allows compact
implementations (device size of sub 10 nm # 10 nm [10]
and 3 nm # 3 nm has been already reported [11], [12])
which can reduce their size up to one tenth of their
Random Access Memory (RAM) counterparts [13]. It is
worth noting that memristive behavior is not limited to
ReRAMs. Other devices such as Phase Change Memory
(PCM) and Spin Transfer Torque (STT) also show similar behaviors as described by Leon Chua in 1971 [14].
Since in many cases these devices face similar challenges, in this paper, we refer to them under the umbrella
term of "memristors." However, one should keep in mind
that each of them has a different mechanism of operation which needs to be taken into account while working
with them.
A natural candidate application for memristors has
been in memory systems [9], [13], [15]-[22]. The possibility of integrating 1 TB of storage on a single chip [23]
makes this technology a very attractive candidate for
memory-intensive big-data applications [24]. Especially
given that they can be integrated with Complementary
Metal-Oxide Semiconductor (CMOS) technology with
minimum changes. For examples, see MOSIS C5 CMOS
[25], or CMOS Back End Of Line (BEOL) Memristor service [26], or the news on the planned offering of Taiwan

2008 2009 2010

2012

2013

2014 2015 2016 2017

Figure 1. A summary of major events and milestones in
memristive community according to [57].

Nima TaheriNejad and David Radakovits are with the Institute of Computer Technology, TU Wien, 1040 Vienna, Austria (email: nima.taherinejad@
tuwien.ac.at, david.radakovits@tuwien.ac.at).
FOURTH QUARTER 2019

IEEE CIRCUITS AND SYSTEMS MAGAZINE

7
IEEE Circuits and Systems Magazine - Q4 2019

Table of Contents for the Digital Edition of IEEE Circuits and Systems Magazine - Q4 2019
