IEEE Circuits and Systems Magazine - Q4 2019 - 8

Among the top articles in the memristive circuit and
system literature, less than 15% are verified by
implementation and measurement.
states are represented as the memory states stored on
the memristor. Therefore, no memory read or write is
necessary before and after the logical operations. This
is a substantial change compared to how CMOS circuits
operate. Operating in memory domain is where CMOS
has a significant disadvantage with respect to memristive technologies, since it requires a significantly larger
area as well as power consumption to do a similar operation. Consequently, in-memory operation is the most
propitious path for memristors as an emerging technology, especially considering the von-Neumann bottleneck which regards the constraints in data transmission
between memory and processing units [57]. Few recent
studies [42], [65]-[69] performed on in-memory computations have already shown promise.
The rest of this paper is organized as follows; In
Section II, we first present our findings regarding the
proportion (or from our point of view, in this case, disproportion) of physical implementation and simulation
among memristor-based systems and circuits in the literature. Then, in Section III, we argue as to why this disproportion is very important in this field and should be
ameliorated in the future. Next, in Section IV, we delve
deeper into the shortcomings of models and some designs by presenting three examples. These experiments
show how proper operation of memristive circuits cannot be taken for granted as it is often done in many memristive system and circuit designs. In Section V, we provide some suggestions as potential directions to take in
order to tackle existing challenges. Finally, we draw our
conclusions in Section VI.

15%
49%
36%

Theory/Device
Circuit/System Sim.-Only
Circuit/System Impl./Meas.

Figure 2. Distribution of "Top 30" literature works on "memristors" showing the share of circuit/system implementation/
measurements (15%) vs. circuit/system simulations or theory/device level works (85%). We note that 52% of the search
results were indeed circuit/system articles (considering both
practical and simulation-only works).
8

IEEE CIRCUITS AND SYSTEMS MAGAZINE

II. Physical Implementation & Verification
in the Literature
Studying the literature, to our surprise, there is a considerable shortage of physical implementation and
verification among the myriads of circuits and systems
proposed for memristive devices. To understand the dimensions of this shortage better, we conducted a survey
in the literature. In this survey we searched five keywords, namely 'memristor', 'memristor circuit', 'memristor system', 'memristor device', 'memristive circuit', and
'memristive system', and picked the first thirty results
(considering papers appearing in more than one search
only once and dismissing it in subsequent appearances)
and checked whether they were verified based on physical implementation or not. To make sure that our sample
is not biased or limited to a certain community we chose
"Google Scholar" as our search engine. "Google Scholar" searches a wide range of available materials including world-wide patents databases, research databases
(such as IEEE, Elsevier, Springer, ResearchGate, ArXive,
and others), university databases (for theses), and other
published papers on the web (white papers published by
companies or papers published by the authors on their
personal website). Our results, visualized in Fig. 2, show
that from the 142 works (38 of the 180 results were repetitive appearances), 51% were indeed circuit and system designs, however, only 30% of them (15.5% of all the
search results) were based on physical implementation
or measurements.
Knowing that some researchers prefer to use specific
device names, we ran the same experiments with new
keywords. Given the better reception of ReRAM among
various types of memristive device, we chose the following keywords; 'RRAM', 'ReRAM', 'resistive RAM', 'resistive memory', 'resistance switch', 'resistance switching',
'resistive switch', 'resistive switching.' The result for this
set of search shows that the majority of the papers using this keyword are at a device and model level, and
not circuit or system design. From the 211 works, only
11% were circuit and system designs, of which 78% were
physically implemented. This means that only 9% of the
total number of works found using those additional keywords were implementations.
We also did a brief search with PCM, where out of
the 30 results, 9 papers (30%) were circuit or system designs, 89% of those containing actual implementations,
yielding 27% implementations in total. Surprisingly, none
FOURTH QUARTER 2019

IEEE Circuits and Systems Magazine - Q4 2019

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