Applsci 11 11254
Applsci 11 11254
Abstract: This paper presents an overview of emerging memory technologies. It begins with the
presentation of stand-alone and embedded memory technology evolution, since the appearance of
Flash memory in the 1980s. Then, the progress of emerging memory technologies (based on fila-
mentary, phase change, magnetic, and ferroelectric mechanisms) is presented with a review of the
major demonstrations in the literature. The potential of these technologies for storage applications
addressing various markets and products is discussed. Finally, we discuss how the rise of artificial
intelligence and bio-inspired circuits offers an opportunity for emerging memory technology and
shifts the application from pure data storage to storage and computing tasks, and also enlarges the
range of required specifications at the device level due to the exponential number of new systems
and architectures.
Keywords: memory; nonvolatile memory; reliability; data storage; artificial intelligence; computing;
von Neumann; emerging memory; neuromorphic circuit
complexity. Then, pressure was exerted on emerging memory concepts in order to pursue
scaling to more aggressive technology nodes, opening the path to new class of embedded
technologies.
In 2015, Intel and Micron developed the 3D XPoint Technology based on a phase
change memory (PCM). This new class of memory enabled filling the latency gap between
DRAM and Flash, often referred to as storage class memory. This made possible new sys-
tem memory architectures with improved performances, bringing innovation to the non-
volatile memory (NVM) arena. The first announcement was made in 2015, and the tech-
nology has been available on the open market under the brand name Optane (Intel) since
2017 [9].
We are now in the more than Moore era, and we work toward new systems (includ-
ing in-memory computing and non von Neumann architectures) that emulate the human
brain to achieve high energy efficiency, parallelism, and ability in cognitive tasks, such as
object recognition, association, adaptation, and learning [10]. This offers a strong oppor-
tunity for new memory technologies.
This article presents the evolution of memory technologies since the invention of
Flash memory, and describes how the exponential increase in new systems (essentially
coming from in-memory computing architectures, artificial intelligence, and neuromor-
phic circuits) offers opportunities to back-end emerging memory technologies. Section
two presents the memory technologies’ market and trends for both stand-alone and em-
bedded sectors. Section three presents an overview of emerging memories with key de-
monstrators presented in the literature. Finally, Section four analyzes how emerging
memories can be used in new computing systems, for both von Neumann and non von
Neumann architectures.
PDAs, MP3 players, and digital cameras were the drivers in the 1990s [11]. With the emer-
gence of smartphones, tablets, USB drives, and SSD (solid state drives), a new driving
force led to a NAND market increase in the digital era of the 2000s [12]. More recently,
artificial intelligence and other new applications offer a strong opportunity for emerging
memories using new switching mechanisms. The next 10 years are expected to provide
many growth opportunities for the semiconductor industry, with a continuous increase
in technical and business challenges. In 2026, the NAND Flash market is predicted to rep-
resent the current GDP of medium-sized countries. Competitive pressures within many
segments of the semiconductor industry will increase significantly in the future, but sem-
iconductor companies that have innovative businesses as well as product strategies are
expected to achieve financial success [13].
The memory market can be divided into two categories, high capacity standalone
memories and embedded memories where the memory device is integrated into a core
CMOS process flow. These two technologies are presented and described in the following
sections.
Three sectors can be distinguished for embedded memories, which are key enablers
of today’s wide variety of microcontroller products [21]:
- Automotive microcontrollers: these represent the most important market. The main
applications include powertrain, body and convenience, safety, connectivity, secu-
rity, etc.
- Smartcard microcontrollers: these represent a wide range of applications, including
ticketing, authentication, government ID, payment, security, etc. This market has re-
cently increased with the rise of the Internet of Things (IoT) and connected objects.
- Consumer and industrial: these also represent a wide range of applications, and in-
clude any type of machine or equipment that can use memory.
Smaller than the standalone memories market, the embedded market still represents
tens of billion dollars every year with huge volume growth. Embedded specifications de-
pend on the market and applications [22]. The most stringent market is the automotive,
with aggressive specifications especially for grade 0 (−40 °C to +150 °C ambient operating
temperature range).
In terms of technology, each competitor develops and uses its own memory cell con-
cept. They all rely on charge trapping mechanisms (in a poly-Si floating gate or in a nitride
charge trapping layer) and differ in cell architecture, from 1 T (one transistor) and 1.5 T
(self-aligned control gate, select gate, and split gate, which is discussed further on) to 2 T
options (separated select gate and memory gate). Production is currently at the 40 nm
node, while 28 nm technologies are increasing in volume production. The current main
embedded NVM charge storage-based cell concepts in production are described in [20].
traffic was expected to grow by a factor of 1455 from 2018 to 2020, following an ex-
ponential evolution, as reported by IBS [26]. In particular, video is expected to main-
tain a growth rate of 70–80% every year for the next decade [26].
According to CISCO, data traffic is forecast to increase 2.5 times over the period 2016
to 2021 [27]. The majority of the increase in data traffic is due to the increase in video
content delivery, projected to account for 82% of traffic by 2021 [27]. The devices that will
thus be the major contributors are smartphones and TVs with respective data traffic per-
centages of 33% and 30% [28].
We are also in the era of IoT, with the exponential growth of connected objects in our
everyday life. We have smartphones, smart watches, smart clothes, etc. The number of
network-connected devices is increasing, from an estimated 18 billion in 2020 to 46 billion
in 2030, which represents an almost 300% increase in that period, with an average com-
pound annual growth rate (CAGR) of 9.8% [28]. The majority of this growth is related to
automation network connected devices and, in particular, to IoT, which will represent
about half of connected device stocks in 2030 [28].
it may be too difficult for one memory to fill the gap, two types of storage class memories
can be distinguished: memory mapped and storage mapped [11]. Memory mapped SCM
should combine read/write and endurance performances close to DRAM, with improved
retention time (without being a true nonvolatile memory) and lower cost. Storage mapped
SCM should be nonvolatile, cost much less than DRAM, and have improved perfor-
mances (speed and endurance) with respect to Flash memories. In particular, the
3DXpoint using PCRAM fits well between NAND and DRAM in the computing memory
hierarchy [31].
Moreover, memory performance and speed growth have not kept pace with that of
processors (memory latency remains almost constant), leading to a performance gap
known as the “memory wall” [35,36]. This performance gap between processor and
memory is widening with technology scaling [37]. In von Neumann systems, separation
of computing and memory becomes critical for performance and efficiency when data
movement becomes prominent, as is the case in current system. Memory access energy
costs 1–3 orders of magnitude higher than computing [38], and consumes approximately
1000 times the energy of a complex addition [36]. Currently, data movement between the
main memory and conventional computation units is a major contributor to the total sys-
tem energy consumption in consumer devices. For instance, it was reported that data
movement accounts for 62.7% of the total energy consumed by Google consumer work-
loads [39]. This is why new non von Neumann architectures were proposed and devel-
oped to exploit locality and near-memory computing in order to reduce the cost and en-
ergy of data movement.
art in a variety of video, image, audio, and text processing tasks. They are already de-
ployed in many real-world applications such as Google’s image and voice search, Apple’s
Siri, Facebook’s DeepText and DeepFace, Microsoft’s Skype Translator, and many others
[40]. They should also enter a wider range of applications in the future, including auton-
omous vehicles, education, and healthcare. The rise of neuromorphic circuits offers
unique opportunities for emerging memories. Among them, two terminal back-end mem-
ories integrated in crossbar arrays could significantly improve the efficiency of deep neu-
ral networks as will be discussed in the next section.
In conclusion, these new systems, offering significant computing energy, are also
very demanding in terms of high reliability memories and require new memory technol-
ogies with excellent capacity, bandwidth, and performance. Research perspectives on
emerging memories for new non von Neumann systems are discussed in the next sections.
Figure 2. TEM cross sections of an HfO2 based OxRAM integrated in the BEOL of a 28 nm FDSOI technology, from [44]
(top) and an OxRAM/OTS device integrated in the BEOL of a 130 nm CMOS process, from [45] (bottom). The OTS (Ovonic
Threshold Switch) backend selector is composed of a C/GeSeSbN/C stack.
time of 100 ns. Less than 10 ns programming times were also reported in the literature
[46].
Resistive RAM underwent a strong improvement in its performance and maturity.
Demonstrations of macros are now reported down to the 22 nm node. Bitcell area of less
than 0.05µm² [47] was achieved by Intel, and 20 nm cell size functionality was demon-
strated by Panasonic with high performances [48]. The endurance of macros of more than
one megabyte now reach 10–100 kcycles, while the intrinsic endurance was demonstrated
to be more than 1010 cycles for single cells [49], showing the technology’s reliability can be
further improved. In particular, Panasonic demonstrated a limited resistance distribution
shift up to 100 k with 10-ppm bitcell resolution [50], and Leti showed no fail after 105 cycles
on 16 kb arrays on a 28 nm node [44].
High-temperature operation is affected by oxygen vacancies (or metal) temperature
assisted diffusion, limiting RRAM retention. In particular, device stability becomes critical
when low programming currents are used [51]. Renesas demonstrated that it was possible
to reach a −6δ margin with a one bit ECC) after 40 min at 200 °C (equivalent to 10 years
at 85 °C), which corresponds to a 0.1% chip failure of two-megabyte cells [52].
Finally, good scalability was shown: scaled memory devices include: 10 nm crossbar
OxRAM [53] and 5 nm liner CBRAM [54]; further density improvements could be ob-
tained with vertical RRAM architecture [43].
As filamentary switching can be observed in most integrated layers, many elements
were investigated until some consensus finally appeared on transition metal oxides
(Ta2O5, HfO2, Al2O3, etc). Depending on the integrated materials, various device perfor-
mances are possible, enabling targeting of various applications.
Today the main application of RRAM is in embedded products. For classical embed-
ded technologies, chip cost reduction will be very difficult beyond 2× nm nodes, due to
their limited scalability and increasing complexity for integration in sub−28 nm CMOS
nodes [20]. For that reason, RRAM is a strong candidate for future nodes due to its simple
bitcell structure and low process complexity. RRAM is highly studied for IoT for its good
density and low power consumption. Panasonic proposed the first commercially available
implementation of RRAM. They offered a microcontroller for portable healthcare, security
equipment, or sensor processing applications. Renesas also investigated RRAM for low-
power microcontroller units (MCUs) for IoT applications [52]. Intel [47,55] and TSMC [56]
both demonstrated RRAM based macros in 22 nm in a 1T1R configuration for embedded
applications, with comparable features (85 °C 10-year retention, 10 kcycles endurance).
Intel studies mobile and RF applications and TSMC targets eFlash, IoT, and smartcards.
Adesto offers RRAM based EEPROM, as a compatible serial memory for discrete and em-
bedded memory applications. They target IoT and other energy-conscious applications.
The startup Crossbar is also very active and provides RRAM for IoT system-on-chips but
also persistent memory solutions.
In the case of embedded applications, the memory is integrated in the BEOL above
the logic. In particular, the memory cell is integrated above the select transistor in the 1T1R
configuration. The bitcell area is limited by the transistor more than by the memory itself
[57]. Thus, it is important to have low operating voltages. Ideally, the RRAM would use
logic transistors where the voltage was ~1 V. Due to the short RRAM programming time
(~100 ns), the transistor could operate in overdrive mode [57]. This leads a targeted RRAM
programming voltage of ~1.5 V to be compatible with logic CMOS to reach the best bitcell
density. For more advanced nodes (1× nm nodes), the memory could be integrated in a
1S1R configuration with a backend selector [58,59] to reduce the bitcell size and target
higher capacities than the 1T1R architecture allows [60].
The main issue for RRAM is the variability and related resistance distribution spread
for high and low resistive states. Indeed, the conductive filament is composed with a finite
number of atoms, and the memory operation is governed by stochastic phenomena
[61,62]. Thus, from cycle to cycle, the conductive filament can have various shapes and is
Appl. Sci. 2021, 11, 11254 9 of 26
Figure 3. 3D schematics and SEM cross section of a GeSbTe (GST) phase change memory integrated
with GeSeSbN OTS selector from [71].
Phase change memories offer low voltage operation (<3 V), fast behavior (~100 ns
switching), and nonvolatility. Very high endurance (1012 cycles at the single cell level) has
Appl. Sci. 2021, 11, 11254 10 of 26
been demonstrated. It is a two terminal device and can be integrated into the BEOL. Phase
change memories do not require any initialization step (unlike RRAM, for example) and
work in unipolar mode. Moreover, phase change memories can have an analog behavior,
which is suitable for some neuromorphic applications. The device characteristics can be
tuned through material engineering (doping, etc.). Thus, they have the capability to ad-
dress the high temperature retention required in embedded applications and the high
speed required, for example, in storage class memory applications depending on the
elected stack [72,73].
In terms of limitations, phase change memory technology suffers from resistance
drift that can affect its high temperature retention [74,75]. Optimized programming
schemes (better than detection threshold adjustment over time) were proposed to improve
immunity to drift for multilevel operations [76]. Moreover, they generally need elevated
(hundreds of µA) currents to operate (linked to the melting temperature of the material
and to the current density needed to achieve such a temperature), which can be a limita-
tion for consumption and crossbar integration. This is balanced, nevertheless, by the fact
that the programming current decreases as the cell size is called: a 20 nm confined cell can
be reset at <100 µA [77,78]. However, at high density, thermal disturbance among neigh-
boring cells can become critical and may require additional layers. In terms of speed,
quenching time can limit the programming speed. In terms of process and integration,
PCRAM may also require complex alloys (ternary or quaternary materials) with an accu-
rate control of the layer composition. They can also need specific device structures such
as wall architecture to improve programming efficiency with more technological steps
and lithography levels than other emerging memories.
PCRAM has a high maturity and products already use this technology. The most
well-known is the 3D Crosspoint technology developed by Intel and Micron that inte-
grates the PCRAM with a backend selector in crosspoint arrays. The technology is used
as a storage class memory to fill the latency gap between DRAM and NAND in the
memory hierarchy. Two types were initially envisaged with two different locations in the
system [1]: “Storage mapped” is part of the memory hierarchy with typical 128 Gb
memory capacities. The memory is faster than NAND, has higher endurance, and is 10X
more dense than conventional memories. “Memory mapped” is a shadow of DRAM; data
in DRAM are copied to the 3D XPoint in order to expand the size of the main memory.
Intel has proposed Optane. Intel Optane DC Persistent and SSD/Caches memories can
achieve 100s ns and 1s µs, respectively, allowing significant improvement of computer
architecture [79]. Micron also offered the X100 NVMe™ SSD cache memory based on 3D
XPoint™ Technology. STMicroelectronics also provides PCRAM technology for automo-
tive embedded applications for 28 nm node and beyond [80,81]. The memory uses a chal-
cogenide ternary material and is integrated in the BEOL of 28 nm FDSOI technology, with
a cell size of 0.036 µm². A bit error rate of <10−8 was achieved after multiple bakes at 150
°C and 10 k cycling of code storage memory was shown.
So far, phase change memory technology has followed the Gartner Hype Cycle, with
R&D in the 1960s, followed by the first product generation (Samsung for mobile phones),
negative press in the 2000s (reporting drift issue during retention), and a second product
generation (3D-Xpoint) now in the market.
In summary, PCRAM is a mature technology showing high reliability, fast speed,
and high endurance, which makes it a strong candidate for both automotive grade em-
bedded applications and storage class memories.
Figure 4. TEM cross-sections of perpendicular STT-MRAM integrated in a four kilobyte array in the
BEOL of a 130 nm CMOS process, from [82].
Toggle MRAM uses a one transistor, one MTJ (magnetic tunnel junction) cell to pro-
vide a simple high-density memory. During a read, the pass transistor is activated and
data are read by comparing the resistance of the cell to a reference device. During writes,
the magnetic field from Write Line 1 and Write Line 2 writes the cell at the intersection of
the two lines but does not disturb other cells on either line. Another MRAM technology
uses a spin torque transfer property, which is the manipulation of the spin of electrons
with a polarizing current, to establish the desired magnetic state of the free layer to pro-
gram, or write, the bits in the memory array. Spin transfer torque MRAM (STT-MRAM)
provides a significant reduction in switching energy compared to toggle MRAM and is
highly scalable, enabling higher density memory products.
The advantage of MRAM is fast switching speed compared to other nonvolatile
memories, with ~1–10 ns read and write erase times and very good endurance (up to 1015
cycles).
One major concern with MRAM is the scalability that is difficult due to the complex-
ity of etching many layers with good conformity. Moreover, MRAM etching generally
uses ion beam etching, which is not suitable for extensive scaling. The other generally
reported issue is MRAM data retention. However, recent material and stack development
allowed significantly improved retention performances [83], which can be attributed to
the adoption of a dual MgO interface instead of the standard single layer.
MRAM also suffers from small ON/OFF current ratio compared to the other emerg-
ing memories: STT-MRAM has a small window and TMR reduces with temperature
[83,84]. Multilevel is difficult to achieve and MRAM requires a good sense amplifier. One
decade ago, magnetic memories were envisaged for SRAM replacement (last level cache)
due to their high speed and endurance. However, the retention and reliability of MRAM
has been highly improved, opening the range of applications to eDRAM (Samsung [85]),
embedded (Samsung [85], GlobalFoundries [86], Intel [87]), Industrial (Everspin [88]), and
even automotive (TSMC) applications.
Intel recently presented significant improvements in MRAM technology [89,90] for
embedded applications. STT-MRAM is also proposed by GF for Embedded, MCU, and
IoT applications [86]. Avalanche is shipping perpendicular MRAM for SRAM (manufac-
tured at partner foundries) but also announced the production in 2020 of a 22 nm MRAM
for Flash replacement for nonvolatile embedded applications [84]. The limitation of spin
memories was retention, but recent achievements by TSMC have shown that this technol-
ogy can be envisaged for automotive applications in the near future [82].
Appl. Sci. 2021, 11, 11254 12 of 26
Several products already exist in the market, such as the spin transfer torque MRAM
for DDR3 and the DDR4 (DRAM) product of Everspin using STT-MRAM, serial periph-
eral interface (SPI) and parallel interface MRAM (using toggle MRAM technology) from
Everspin, and the SPSRAM (a persistent SRAM using STT-MRAM technology with serial
peripheral interface) and SPNOR (perpendicular STT-MRAM for embedded Flash and
embedded SRAM used in system-on-chips) memories from Avalanche technology.
In summary, MRAM offers excellent endurance suitable for DRAM and SRAM ap-
plications, but has also shown recent increased stability for embedded applications. Stack
complexity is generally invoked as a device limitation. The next challenge will be to in-
crease its scalability and capacity.
Figure 5. SEM and TEM cross sections of ferroelectric TiN/10 nm Si(1%)-implanted HfO2/TiN ca-
pacitors integrated between M4 and M5 of a 130 nm CMOS, from [91].
Ferroelectrics are theoretically an ideal solution for low write power nonvolatile
memories. However, the complexity of ferroelectric perovskites has hindered the scaling
of such devices to competitive feature sizes. The discovery of ferroelectricity in hafnium
oxide solved this issue, and led to renewed interest by the scientific community in this
concept for various applications, due to its CMOS compatibility [92–94].
Three types of ferroelectric-based memories are:
- Ferroelectric FET (FeFET): the ferroelectric material is embedded in the gate stack of
a transistor. The nonvolatile polarization of the material acts as a remnant control
gate and leads to a threshold voltage shift of the characteristics. This concept offers
ultra-low power but is a three terminal device and can thus hardly be envisaged for
high-density applications. Moreover, the effect vanishes for thin layers, making the
concept hardly scalable. Finally, the degradation of the interface layer between the
ferroelectric and the semiconductor channel limits endurance, in particular, due to
trapped charge that affects the conduction of the FET below the ferroelectric [92]. For
all of these reasons FeFET are targeting Flash or EEPROM rather than DRAM replace-
ment. In particular, it is now seen as an alternative to Flash for ultra-low power ap-
plications [91], due to its 10 fJ/bit consumption and five-nanosecond programming
Appl. Sci. 2021, 11, 11254 13 of 26
speed. FeFET based eNVM solutions were integrated into leading edge technologies:
GlobalFoundries FeFET technology was embedded into the 28 nm gate first HKMG
low power CMOS platform, showing 6δ distribution, reasonable endurance, and sta-
ble data retention [95]. GF also demonstrated a 22 nm node on FDSOI CMOS tech-
nology [96].
- Capacitor based ferroelectric RAM (FeRAM): in this case, the cell resembles a DRAM
with the capacitor dielectric replaced by the ferroelectric. Recent reports verified an-
tiferroelectric properties for pure ZrO2 dielectrics used in DRAM stacks. By employ-
ing electrodes with different work function values, a built-in bias is introduced
within the anti-ferroelectric stack, thus creating two stable nonvolatile states [97]. It
demonstrated 1010 endurance and 10 ns speed combined with 100 °C retention, mak-
ing this concept very promising for a dense (6 F²) and new class of nonvolatile
DRAM. In FeRAM, reading is destructive as it is performed by switching the ferroe-
lectric into a specific direction and measuring the contrast between a switching and
a nonswitching event. Thus, programming is required after each reading operation.
In terms of FeRAM reliability challenges, trapped charges at the ferroelectric-elec-
trode interface have to be controlled to improve retention (requiring careful interface
engineering), while dielectric breakdown induced by high coercive film has to be
prevented to insure high endurance [92].
- Ferroelectric tunneling junctions (FTJ): in this case, the memory is a two terminal de-
vice. In the FTJ, a very thin ferroelectric film is used that allows tunneling and the
tunneling current is modulated by the polarization of the ferroelectric. A critical issue
for this concept (which is more prospective than the previous ones) is the low read
current.
In summary, FeRAM is a simple and low cost memory offering very low consump-
tion (~10 fJ/bit), suitable for low power applications (IoT, etc). Its high endurance and non-
volatility also make it a promising technology for the future, including neuromorphic cir-
cuits.
STT MRAM MRAM SOT PCM stand PCM RRAM stand RRAM FeRAM FeFET
SCM/ embedded Cache alone embedded alone embedded
DRAM
Capacity >1Gb 10-100Mb >1Mb Gb 10-100Mb ~Gb 1-10Mb Poor Small
targetted
Scalability Medium Medium Poor Good Good Medium Good Medium Poor
Retention >1 yr Automotive 85-100°C 85-100°C Automotive 10ys 85°C 10ys >85°C 85-100°C SMT
110°C 150°C 10ys compliant
Latency 10ns 10ns <ns 100ns 100ns 100ns 100ns <20ns 5ns
Power pJ/bit pJ/bit fJ/bit 10pj/bit 10pj/bit 1-10pj/bit 1-10pj/bit 10fj/bit 10fj/bit
>200µA ~100µA
Endurance 1010 >106 >1010 107 106 107 106 >1011 104-105
(destructive
read)
Variability NA NA NA Issue (drift) Issue (drift) Issue Issue Variability Variability
(variability, (variability, @small size @small size
noise) noise)
Space DRAM NVM Cache SCM MPU, MCU SCM MPU, MCU DRAM Flash
(storage, (storage,
memory) memory)
Maturity, Products: Product: No product Products: Product No product Products: Products Good
example of Everspin, Avalanche, Intel/ sampling: ST Panasonic, (PZT): Texas
products Avalanche TSMC Micron, Microelectr Dialog, Instruments,
(persistent (offers STT- Intel onics TSMC Fujitsu,
SRAM) MRAM) Cypress
Figure 6. Tentative benchmark of various emerging memory technologies for various applications.
Finally, Figure 7 focuses on macros for embedded applications. Most macros are in-
tegrated in 28 nm or 22 nm nodes. Cell size, limited by the selected transistor, is always in
the range of 0.04–0.05 µm². The best endurance has been demonstrated for MRAM with
106 cycles (10–100 kc for Resistive RAM, 100 kc for Phase Change RAM). Various applica-
tions are targeted, from smartcard and IoT to automotive grade MCU.
Figure 7. Table summarizing the emerging memory macros for embedded applications.
could be managed by the OS (operating system) of the chip as it is today for storage
memory. The storage memory market being essentially cost-driven, it is likely that
RRAM will only replace the now well-established flash technology when it is cost-
competitive; as flash density continues to increase due to monolithic 3D integration,
this will take time. In the meantime, the emerging memory could be used as an inter-
mediate step between flash and the main memory, as for instance an ultra-fast SSD
used to store data with frequent access. The large difference in latency is likely to
make acceptable a higher cost. Currently, the gap in latency between the main
memory, which is in the order of 30–50 ns and the one of flash-based storage, which
is about 100 µs (much more in writing) is vast, and having a technology enabling
microsecond scale latencies would undoubtedly be an improvement for data centric
applications. For these reasons, 1 µs latency and 106 cycle high density RRAM and
PCRAM (possibly in vertical architectures) could be two strong candidates for this
role.
- It can be used in the main memory area, either as a DRAM companion chip, located
on the same memory bus, or as a replacement of the DRAM. Inserting the emerging
memory on the main memory bus, alongside DRAM components, will present sig-
nificant system benefits. It will improve data integrity management; data can be
quickly secured locally, on a word-by-word basis, instead of using current cumber-
some journaling or check-pointing schemes to protect data from events such as a loss
of power supply. This can significantly reduce data traffic in a data center, as it is
estimated that data integrity management can represent up to 80% of the file system
usage in PetaFlop data centers. As the emerging memory should become more dense
and less expensive than DRAM, it will enable much greater capacity than main mem-
ories; this is especially important for big data applications, where storing large mul-
tidimensional tables in the main memory enables a tremendous performance ad-
vantage, because these tables often need to be accessed in a different order than the
way it has been stored in the storage memory, which can only be accessed sequen-
tially. It will also simplify atomic operations in transactional databases, as RRAM are
byte addressable and nonvolatile. They propose a much more efficient solution than
current NVDIMMs composed of a mix of DRAM and flashes powered by a bulky
supercapacitor during the data transfer from DRAM to Flash when the main power
supply is lost. For these reasons, according to the specifications, fast (~100 ns) and
high endurance (~109 cycles) RRAM and PCRAM could succeed.
- It can also be used as a last-level cache replacement or complement; it is unlikely that
memory technology could become fast enough to be used as a first-level cache. In
this case, the cache capacity could be made much higher, diminishing the external
bandwidth requirements. Here, the most stringent requirements would be speed (<30
ns) and endurance (>1016). We cannot rely on wear-leveling in this case as the cache
capacity would not be sufficient in regard to its bandwidth, each bit being written
frequently. A high endurance of 1016 cycles is likely to reserve this application for
STT-MRAM. FeRAM could be placed between an SCM memory (memory type) and
the DRAM due to its high endurance. Finally, the only technology that exhibits per-
formances close to SRAM is the SOT-MRAM. As far as cost and power consumption
are concerned, the reference there is embedded DRAM (eDRAM); the RRAM has to
be cheaper and less consuming than eDRAM to be competitive.
A summary of the possible implementation of emerging memories in the memory
hierarchy is presented in Figure 8. In this figure, storage class memory refers to a class of
memory that stands between DRAM and disk storage in the data storage hierarchy. In
other work, this class of memory has been defined as persistent memory [98]. The differ-
ence at the system level between persistent storage, nonpersistent DRAM extension, and
persistent memory at the architecture level is not discussed in this paper.
Appl. Sci. 2021, 11, 11254 16 of 26
Figure 8. Possible implementation of emerging memories in the memory hierarchy based on the
benchmark presented in Figure 6.
this context, several techniques have been proposed in the literature to solve the imper-
fections of current technologies: in [113], a mixed hardware–software neural-network im-
plementation combines long-term storage in PCM (for weight data), near-linear updates
of volatile capacitors, and weight-data transfer with ‘polarity inversion’ to cancel out in-
herent device-to-device variations. In [114,115], RRAM are used in low precision binarized
neural networks (BNN), to cope with device non-idealities. In this type of neural network,
both synaptic weights and neuron activations are implemented by binary values (+1 or
−1) after a network training process.
Panasonic developed neuromorphic computing based on analog RRAM, resistive an-
alog neuromorphic device (RAND), as a low power solution for edge application [116].
The authored demonstrated MNIST recognition and sensor application in which several
networks could be configured at the same time.
Finally, multiple bits-per-cell was demonstrated in OxRAM in 1T4R configuration
due to gradual SET/RESET [117]. The authors claim that this makes the structure suitable
for multiple deep learning applications and showed high degrees of inference accuracy
within 0.01% of ideal values. However, statistical cell-to-cell variability limits the maxi-
mum number of levels because of the overlap between adjacent resistance distributions
[118], and OxRAM resistance relaxation after programming leads to overlap of the
memory states, which can alter the circuit reliability [119].
- Spiking brain inspired neural network
In spiking neural networks, the approach is to mimic human brain behavior and to
represent data with spikes. Spiking neural networks (SNN) with spike-timing dependent
plasticity (STDP) are then capable of replicating bio-realistic online/unsupervised learning
[120,121], which is not the case for deep learning architectures relying on supervised back-
propagation. In SNN processes, the simultaneous spiking activity at two neurons can lead
to a potentiation of the synapse connecting them, meaning that two neurons which are
active in response to the same event, should be linked by a relatively strong synaptic con-
nection.
In this field, various demonstrations were proposed in the literature, aiming at im-
plementing STDP in hardware synapses adopting various classes of emerging memories
(including RRAM, CBRAM, STT-MRAM, and PCM). In [122], and [123] OxRAM synapses
based on HfO2 and SiOx resistive memory technologies were combined with analog neu-
rons in spiking neural networks for MNIST digit classification. The OxRAM, coding the
synaptic weights of the network, was thus used for inference in the DNN, which required
high reading operations but only a very limited number of cycles.
As for an analog neural network, no emerging technology can afford the high num-
ber of cycles that can offer continuous learning of the system. Thus, today, learning is
performed offline, and synaptic weights are then coded in the nonvolatile memories for
inference.
In conclusion, the emergence of high capacity memories, with good endurance, ide-
ally with analogic behavior (or at least multilevel) can significantly improve the efficiency
of data transfer and allow the emergence of new computing and non von Neumann sys-
tems.
5. Conclusions
The general context of nonvolatile memories is characterized by several key points.
First, the era of big data in which we live implies a constant and tremendous increase in
data volume generation associated with the increase in the number of connected objects.
The introduction into the market of 3D-NAND allowed pursuing a density increase. At
the system level, the memory hierarchy suffers from two limitations, memory wall (be-
tween SRAM and DRAM) and latency gap (between DRAM and Flash), offering oppor-
tunities for new technologies. Data deluge also changed the paradigm of computing sys-
tem; limited today by data transfer more than computing. This leads to the necessity for
Appl. Sci. 2021, 11, 11254 20 of 26
more efficient and specialized architectures, such as in-memory computing and neuro-
morphic circuits.
At the component level, there is currently a renewed interest in emerging memory
technologies (RRAM, PCRAM, MRAM, FeRAM, etc), based on “old” concepts due to the
combination they offer in terms of fast speed, high endurance, and nonvolatility. Today,
no universal memory has been discovered so far, and we move towards the co-existence
of various concepts, more and more specialized to a specific application.
Currently, 3D-NAND is dominant in the world of standalone memories and there is
no clear need for a new concept to replace it. However, there is an opportunity for emerg-
ing memories to enter the memory hierarchy in new (von Neumann and non von Neu-
mann) computing systems to improve efficiency and performances. Innovation will thus
consist in new architectures made possible by the advent of new memory technologies
showing more and more maturity. This will require strong collaboration and mutual un-
derstanding between device engineers and system architects. On the other hand, it is more
difficult and expensive to maintain current embedded charge based technologies for new-
technology nodes. Various emerging technologies are thus called upon to enter the em-
bedded memory market. The wide range of existing applications in this domain should
result in the appearance of various technologies depending on the applications. To fill
these requirements, there is a place for a dedicated research to improve emerging memory
performances, based on evolving concepts, new materials, and also optimized program-
ming schemes, which should be adapted to the physics of emerging devices.
References
1. Ishimaru, K. Future of Non-Volatile Memory -From Storage to Computing-. In Proceedings of the 2019 IEEE International Elec-
tron Devices Meeting (IEDM), 7-11 Dec. 2019, San Francisco, CA, USA; pp. 1.3.1–1.3.6.
https://doi.org/10.1109/IEDM19573.2019.8993609.
2. “Data centres of the world will consume 1/5 of Earth’s power by 2025”, Available online: https://data-economy.com/data-cen-
tres-world-will-consume-1-5-earths-power-2025/ (accessed on 10/12/2021).
3. Masuoka, F.; Asano, M.; Iwahashi, H.; Komuro, T.; Tanaka, S. A new flash E2PROM cell using triple polysilicon technology. In
Proceedings of the 1984 International Electron Devices Meeting, San Francisco, CA, USA, 9-12 Dec. 1984; pp. 464–467.
https://doi.org/10.1109/IEDM.1984.190752.
4. Available online: https://spectrum.ieee.org/chip-hall-of-fame-toshiba-nand-flash-memory (accessed on 10/12/2021).
5. Ogura, T. Flash Memory. In CMOS Processors and Memories. Analog Circuits and Signal Processing; Iniewski, K., Eds.; Springer:
Dordrecht, The Netherlands, 2010. https://doi.org/10.1007/978-90-481-9216-8_7.
6. Taehee Cho; Yeong-Taek Lee; Eun-Cheol Kim; Jin-Wook Lee; Sunmi Choi; Seungjae Lee; Dong-Hwan Kim; Wook-Ghee Han;
Young-Ho Lim; Jae-Duk Lee et al..; et al. A dual-mode NAND flash memory: 1-Gb multilevel and high-performance 512-Mb
single-level modes. IEEE J. Solid-State Circuits 2001, 36, 1700–1706. https://doi.org/10.1109/4.962291.
Appl. Sci. 2021, 11, 11254 21 of 26
7. H. Tanaka; M. Kido; K. Yahashi; M. Oomura; R. Katsumata; M. Kito; Y. Fukuzumi; M. Sato; Y. Nagata; Y. Matsuoka et al., . Bit
Cost Scalable Technology with Punch and Plug Process for Ultra High Density Flash Memory. In Proceedings of the 2007 IEEE
Symposium on VLSI Technology, Kyoto, Japan, 12-14 June 2007; pp. 14–15. https://doi.org/10.1109/VLSIT.2007.4339708.
8. Available online: https://www.renesas.com/us/en/about/press-room/renesas-electronics-develops-industrys-first-28nm-em-
bedded-flash-memory-technology-microcontrollers (accessed on 10/12/2021).
9. Available online: https://www.intel.com/content/www/us/en/products/details/memory-storage/optane-memory.html (ac-
cessed on 10/12/2021).
10. Available online: https://www.research.ibm.com/articles/brain-chip.shtml (accessed on 10/12/2021).
11. Cappelletti, P. Non volatile memory evolution and revolution. In Proceedings of the 2015 IEEE International Electron Devices
Meeting (IEDM), Washington, DC, USA, 7-9 Dec. 2015; pp. 10.1.1–10.1.4. https://doi.org/10.1109/IEDM.2015.7409666.
12. Aritome, S. Nand Flash Memory Devices. in NAND Flash Memory Technologies. IEEE 2015, 37–92, Online ISBN:
9781119132639, https://doi.org/10.1002/9781119132639.ch3.
13. International Business Strategies; Inc. (IBS): October 2019, Los Gatos, Ca, USA.
14. Hwang, C. Nanotechnology enables a new memory growth model. IEEE 2003, 91, 1765–1771.
https://doi.org/10.1109/JPROC.2003.818323.
15. Aritome, S. NAND Flash Memory Revolution. In Proceedings of the 2016 IEEE 8th International Memory Workshop (IMW),
Paris, France, 15-18 May 2016; pp. 1–4. https://doi.org/10.1109/IMW.2016.7495285.
16. Yoshiaki Fukuzumi; Ryota Katsumata; Masaru Kito; Masaru Kido; Mitsuru Sato; Hiroyasu Tanaka; Yuzo Nagata; Yasuyuki
Matsuoka; Yoshihisa Iwata; Hideaki Aochi et al. Optimal Integration and Characteristics of Vertical Array Devices for Ultra-
High Density, Bit-Cost Scalable Flash Memory. In Proceedings of the 2007 IEEE International Electron Devices Meeting, 10-12
Dec. 2007; pp. 449–452. https://doi.org/10.1109/IEDM.2007.4418970.
17. N. Shibata; K. Kanda; T. Shimizu; J. Nakai; O. Nagao; N. Kobayashi; M. Miakashi; Y. Nagadomi; T. Nakano; T. Kawabe et al.
13.1 A 1.33Tb 4-bit/Cell 3D-Flash Memory on a 96-Word-Line-Layer Technology. In Proceedings of the 2019 IEEE International
Solid- State Circuits Conference—(ISSCC), San Francisco, CA, USA, 17-21 Feb. 2019; pp. 210–212.
https://doi.org/10.1109/ISSCC.2019.8662443.
18. Kalavade, P. 4 bits/cell 96 Layer Floating Gate 3D NAND with CMOS under Array Technology and SSDs. In Proceedings of the
2020 IEEE International Memory Workshop (IMW), Dresden, Germany, 17-20 May 2020; pp. 1–4.
https://doi.org/10.1109/IMW48823.2020.9108135.
19. Inaba, S. 3D Flash Memory for Data-Intensive Applications. In Proceedings of the 2018 IEEE International Memory Workshop
(IMW), Kyoto, Japan, 13-16 May 2018; pp. 1–4. https://doi.org/10.1109/IMW.2018.8388775.
20. Yoshio Nishi, ed. 2014: Advances in Non-volatile Memory and Storage Technology, Chapter 4: Improving embedded Flash
memory technology p. 139.
21. Strenz, R. Review and Outlook on Embedded NVM Technologies—From Evolution to Revolution. In Proceedings of the 2020
IEEE International Memory Workshop (IMW), Dresden, Germany, 17-20 May 2020; pp. 1–4.
https://doi.org/10.1109/IMW48823.2020.9108121.
22. Wang, H. Challenges in Automotive Memory Solutions. In Proceedings of the 2018 IEEE International Memory Workshop
(IMW), Kyoto, Japan, 13-16 May 2018; pp. 1–7. https://doi.org/10.1109/IMW.2018.8388782.
23. David Reinsel, John Gantz, John Rydning IDC White Paper, Data Age 2025; Seagate: April 2017. Available online:
https://www.import.io/wp-content/uploads/2017/04/Seagate-WP-DataAge2025-March-2017.pdf (accessed on 10/12/2021).
24. David Reinsel, John Gantz, John Rydning. IDC White Paper, Data Age 2025. In The Digitalization of the World, From Edge to Core;
#US44413318; November 2018, Available online: https://www.seagate.com/files/www-content/our-story/trends/files/idc-
seagate-dataage-whitepaper.pdf (accessed on 10/12/2021).
25. Available online: https://www.whishworks.com/blog/big-data/understanding-the-3-vs-of-big-data-volume-velocity-and-vari-
ety (accessed on 10/12/2021).
26. Impact of Ai on Electronics and Semiconductor Industries; International Business Strategies, Inc. (IBS): April 2020.
27. The Zettabyte Era: Trends and Analysis. Cisco. Available online http://www.hit.bme.hu/~jakab/edu/HTI18/Litr/Cisco_The_Zet-
tabyte_Era_2017June__vni-hyperconnectivity-wp.pdf (accessed on 10/12/2021).
28. 4E Electronics Devices and Network Annex (EDNA). Total Energy Model for Connected Devices. June 2019. Available online:
https://edna.iea-4e.org/library (accessed on 10/12/2021).
29. Andrae, A. Total Consumer Power Consumption Forecast. Nord. Digit. Bus. Summit 2017, 5 Oct. 2017. Available online:
https://www.researchgate.net/publication/320225452_Total_Consumer_Power_Consumption_Forecast (10/12/2021).
30. Available online: www.iea-4e.org (accessed on 10/12/2021).
31. DerChang Kau; Stephen Tang; Ilya V. Karpov; Rick Dodge; Brett Klehn; Johannes A. Kalb; Jonathan Strand; Aleshandre Diaz;
Nelson Leung; Jack Wu et al. A stackable cross point Phase Change Memory. In Proceedings of the 2009 IEEE International
Electron Devices Meeting (IEDM), Baltimore, MD, USA, 7-9 Dec. 2009; pp. 1–4. https://doi.org/10.1109/IEDM.2009.5424263.
32. Hady, F.; Converging Memory and Storage. In Proceedings of the 2019 SNIA’s Persistent Memory Summit, Santa Clara, CA,
USA, 24 Jan. 2019. Available online: https://www.snia.org/sites/default/files/PM-Summit/2019/presentations/01-PMSummit19-
Hady-Keynote.pdf (accessed on 10/12/2021).
Appl. Sci. 2021, 11, 11254 22 of 26
33. Handy, J.; Coughlin, T. The Long and Winding Road to Persistent Memories. In Proceedings of the 2018 SNIA’s Storage Devel-
oper Conference, 24-27 Sept. 2018. Available online: https://www.snia.org/sites/default/files/SDC/2018/presenta-
tions/PM/Handy_J_Coughlin_T_The_Long_&_Winding_Road_to_Persistent_Memories.pdf (accessed on 10/12/2021).
34. Foong, A.; Hady, F. Storage as Fast as Rest of the System. In Proceedings of the 2016 IEEE 8th International Memory Workshop
(IMW), Paris, France, 15-18 May 2016; pp. 1–4. https://doi.org/10.1109/IMW.2016.7495289.
35. Wulf, W.A.; McKee, S. Hitting the memory wall: Implications of the obvious. ACM SIGARCH Comput. Archit. News 1995.
https://doi.org/10.1145/216585.216588.
36. Mutlu, O. Processing Data Where It Makes Sense in Modern Computing Systems: Enabling In-Memory Computation. In Pro-
ceedings of the Shortcourse of 2019 International Memory Workshop, Monterey, CA, USA, 12-15 May 2019.
37. Kulkarni, J.P. Circuit Technologies for Efficient In-Memory Computing. In Proceedings of the Shortcourse of 2019 International
Memory Workshop, Monterey, CA, USA.
38. Raychowdhury, A. Towards Memory-centric Autonomous Systems: A Technology and Device Perspective. In Proceedings of
the IEDM 2019 Short Course 2: Technologies for Memory-Centric Computing, 7-11 Dec. 2019, San Francisco, CA, USA.
39. Boroumand, A.; Ghose, S.; Kim, Y.; Ausavarungnirun, R.; Shiu, E.; Thakur, R.; Kim, D.; Kuusela, A.; Knies, A.; Ranganathan, P.;
et al. Google Workloads for Consumer Devices: Mitigating Data Movement Bottlenecks. In Proceedings of the 23rd International
Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS), Williamsburg, VA, USA,
24-28 March 2018.
40. Venkataramani, S.; Ranjan, A. ; Banerjee, S. ; Das, D. ; Avancha, S. ; Jagannathan, Durg, A. ; A. ;Nagaraj, D. ; Kaul, B. ; Dubay,
P. ; et al. SCALEDEEP: A scalable compute architecture for learning and evaluating deep networks. In Proceedings of the2017
ACM/IEEE 44th Annual International Symposium on Computer Architecture (ISCA), 24-28 June 2017; pp. 13–26.
https://doi.org/10.1145/3079856.3080244.
41. Nielsen, P.H.; Bashara, N.M. The reversible voltage-induced initial resistance in the negative resistance sandwich structure.
IEEE Trans. Electron Devices 1964, 11, 243–244. https://doi.org/10.1109/T-ED.1964.15319.
42. Hickmott, T.W. Low-Frequency Negative Resistance in Thin Anodic Oxide Films. J. Appl. Phys. 1962, 33, 2669–2682. [Online]
https://doi.org/10.1063/1.1702530.
43. I.G. Baek; M.S. Lee; S. Seo; M.J. Lee; D.H. Seo; D.-S. Suh; J.C. Park; S.O. Park; H.S. Kim; I.K. Yoo et al. Highly scalable nonvol-
atile resistive memory using simple binary oxide driven by asymmetric unipolar voltage pulses. IEDM Technical Digest. In
Proceedings of the IEEE International Electron Devices Meeting 2004, 13-15 Dec. 2004; pp. 587–590.
https://doi.org/10.1109/IEDM.2004.1419228.
44. L. Grenouillet; N. Castellani; A. Persico; V. Meli; S. Martin; O. Billoint; R. Segaud; S. Bernasconi; C. Pellissier; C. Jahan et al.
16kbit 1T1R OxRAM arrays embedded in 28nm FDSOI technology demonstrating low BER, high endurance, and compatibility
with core logic transistors. In Proceedings of the 2021 IEEE International Memory Workshop (IMW), Dresden, Germany, 16-19
May 2021; pp. 1–4. https://doi.org/10.1109/IMW51353.2021.9439607.
45. J. Minguet Lopez; N. Castellani; L. Grenouillet; L. Reganaz; G. Navarro; M. Bernard; C. Carabasse; T. Magis; D. Deleruyelle; M.
Bocquet et al. Ge-Se-Sb-N-based OTS scaling perspectives for high-density 1 S1R crossbar arrays. In Proceedings of the 2021
IEEE International Memory Workshop (IMW), Dresden, Germany, 16-19 May 2021; pp. 1–4.
https://doi.org/10.1109/IMW51353.2021.9439606.
46. Chen Wang, Huaqiang Wu, Bin Gao, Wei Wu, Lingjun Dai, Xinyi Li, He Qian, “Ultrafast RESET Analysis of HfOx-Based RRAM
by Sub-Nanosecond Pulses”, Adv. Electron. Mater. 2017, 3, 1700263, DOI: 10.1002/aelm.201700263.
47. O. Golonzka; U. Arslan; P. Bai; M. Bohr; O. Baykan; Y. Chang; A. Chaudhari; A. Chen; J. Clarke; C. Connor et al., Non-Volatile
RRAM Embedded into 22FFL FinFET Technology. In Proceedings of the 2019 Symposium on VLSI Technology, Kyoto, Japan,
9-14 June 2019; pp. T230–T231. https://doi.org/10.23919/VLSIT.2019.8776570.
48. Y. Hayakawa; A. Himeno; R. Yasuhara; W. Boullart; E. Vecchio; T. Vandeweyer; T. Witters; D. Crotti; M. Jurczak; S. Fujii et al..
Highly reliable TaOx ReRAM with centralized filament for 28-nm embedded application. In Proceedings of the2015 Symposium
on VLSI Technology (VLSI Technology), Kyoto, Japan, 16-18 June 2015; pp. T14–T15.
https://doi.org/10.1109/VLSIT.2015.7223684.
49. T. Hirtzlin; M. Bocquet; M. Ernoult; J. - O. Klein; E. Nowak; E. Vianello; J. - M. Portal; D. Querlioz. Hybrid Analog-Digital
Learning with Differential RRAM Synapses. In Proceedings of the 2019 IEEE International Electron Devices Meeting (IEDM),
San Francisco, CA, USA, 7-11 Dec. 2019; pp. 22.6.1–22.6.4. https://doi.org/10.1109/IEDM19573.2019.8993555.
50. Satoru Ito; Yukio Hayakawa; Zhiqiang Wei; Shunsaku Muraoka; Koichi Kawashima; Hideto Kotani; Kazuyuki Kouno; Ma-
sayoshi Nakamura; Guo An Du; Jiann Fu Chen et al.. ReRAM Technologies for Embedded Memory and Further Applications.
In Proceedings of the 2018 IEEE International Memory Workshop (IMW), Kyoto, Japan, 13-16 May 2018; pp. 1–4.
https://doi.org/10.1109/IMW.2018.8388846.
51. J. Guy; G. Molas; C. Cagli; M. Bernard; A. Roule; C. Carabasse; A. Toffoli; F. Clermidy; B. De Salvo; L. Perniola Guidance to
reliability improvement in CBRAM using advanced KMC modelling. In Proceedings of the 2017 IEEE International Reliability
Physics Symposium (IRPS), Monterey, CA, USA, 2-6 April 2017; pp. PM-2.1–PM-2.5. https://doi.org/10.1109/IRPS.2017.7936384.
52. M. Ueki; K. Takeuchi; T. Yamamoto; A. Tanabe; N. Ikarashi; M. Saitoh; T. Nagumo; H. Sunamura; M. Narihiro; K. Uejima et al.
Low-power embedded ReRAM technology for IoT applications. In Proceedings of the 2015 Symposium on VLSI Technology
(VLSI Technology), Kyoto, Japan, 16-18 June 2015; pp. T108–T109. https://doi.org/10.1109/VLSIT.2015.7223640.
Appl. Sci. 2021, 11, 11254 23 of 26
53. B. Govoreanu; G.S. Kar; Y-Y. Chen; V. Paraschiv; S. Kubicek; A. Fantini; I.P. Radu; L. Goux; S. Clima; R. Degraeve et al.. 10 × 10
nm2 Hf/HfOx crossbar resistive RAM with excellent performance, reliability and low-energy operation. In Proceedings of the
2011 International Electron Devices Meeting, Washington, DC, USA, 5-7 Dec. 2011; pp. 31.6.1–31.6.4.
https://doi.org/10.1109/IEDM.2011.6131652.
54. J. Guy; G. Molas; P. Blaise; C. Carabasse; M. Bernard; A. Roule; G. Le Carval; V. Sousa; H. Grampeix; V. Delaye et al. Experi-
mental and theoretical understanding of Forming, SET and RESET operations in Conductive Bridge RAM (CBRAM) for
memory stack optimization. In Proceedings of the 2014 IEEE International Electron Devices Meeting, San Francisco, CA, USA,
15-17 Dec. 2014; pp. 6.5.1–6.5.4. https://doi.org/10.1109/IEDM.2014.7046997.
55. Pulkit Jain; Umut Arslan; Meenakshi Sekhar; Blake C. Lin; Liqiong Wei; Tanaya Sahu; Juan Alzate-vinasco; Ajay Vangapaty;
Mesut Meterelliyoz; Nathan Strutt et al. 13.2 A 3.6 Mb 10.1 Mb/mm2 Embedded Non-Volatile ReRAM Macro in 22 nm FinFET
Technology with Adaptive Forming/Set/Reset Schemes Yielding Down to 0.5 V with Sensing Time of 5 ns at 0.7 V. In Proceed-
ings of the 2019 IEEE International Solid- State Circuits Conference—(ISSCC), San Francisco, CA, USA, 17-21 Feb. 2019; pp. 212–
214. https://doi.org/10.1109/ISSCC.2019.8662393.
56. Chung-Cheng Chou; Zheng-Jun Lin; Chien-An Lai; Chin-I Su; Pei-Ling Tseng; Wei-Chi Chen; Wu-Chin Tsai; Wen-Ting Chu;
Tong-Chern Ong; Harry Chuang et al.. A 22nm 96KX144 RRAM Macro with a Self-Tracking Reference and a Low Ripple Charge
Pump to Achieve a Configurable Read Window and a Wide Operating Voltage Range. In Proceedings of the 2020 IEEE Sympo-
sium on VLSI Circuits, Honolulu, HI, USA, 16-19 June 2020; pp. 1–2. https://doi.org/10.1109/VLSICircuits18222.2020.9163014.
57. J. Sandrini; L. Grenouillet; V. Meli; N. Castellani; I. Hammad; S. Bernasconi; F. Aussenac; S. Van Duijn; G. Audoit; M. Barlas et
al. OxRAM for embedded solutions on advanced node: Scaling perspectives considering statistical reliability and design con-
straints. In Proceedings of the 2019 IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec.
2019; pp. 30.5.1–30.5.4. https://doi.org/10.1109/IEDM19573.2019.8993484.
58. J. Minguet Lopez; D. Alfaro Robayo; L. Grenouillet; C. Carabasse; G. Navarro; R. Fournel; C. Sabbione; M. Bernard; O. Billoint;
C. Cagli et al. Optimization of RRAM and OTS selector for advanced low voltage CMOS compatibility. In Proceedings of the
2020 IEEE International Memory Workshop (IMW), Dresden, Germany, 17-20 May 2020; pp. 1–4.
https://doi.org/10.1109/IMW48823.2020.9108126.
59. D. Alfaro Robayo; G. Sassine; J. Minguet Lopez; L. Grenouillet; A. Verdy; G. Navarro; M. Bernard; E. Esmanhotto; C. Carabasse;
D. Deleruyelle et al. Reliability and Variability of 1S1R OxRAM-OTS for High Density Crossbar Integration. 2019 IEEE Interna-
tional Electron Devices Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp. 35.3.1–35.3.4.
https://doi.org/10.1109/IEDM19573.2019.8993439.
60. G. Molas; D. Alfaro Robayo; J. Minguet Lopez; L. Grenouillet; C. Carabasse; G. Navarro; C. Sabbione; M. Bernard; C. Cagli; N.
Castellani et al. Crosspoint Memory Arrays: Principle, Strengths and Challenges. In Proceedings of the 2020 IEEE International
Memory Workshop (IMW), Dresden, Germany, 17-20 May 2020; pp. 1–4. https://doi.org/10.1109/IMW48823.2020.9108143.
61. Padovani, A.; Larcher, L.; Pirrotta, O.; Vandelli, L.; Bersuker, G. Microscopic Modeling of HfOx RRAM Operations: From Form-
ing to Switching. IEEE Trans. Electron Devices 2015, 62, 1998–2006. https://doi.org/10.1109/TED.2015.2418114.
62. Jérémy Guy; Gabriel Molas; Philippe Blaise; Mathieu Bernard; Anne Roule; Gilles Le Carval; Vincent Delaye; Alain Toffoli;
Gérard Ghibaudo; Fabien Clermidy et al. Investigation of Forming, SET, and Data Retention of Conductive-Bridge Random-
Access Memory for Stack Optimization. IEEE Trans. Electron Devices 2015, 62, 3482–3489.
https://doi.org/10.1109/TED.2015.2476825.
63. Gilbert Sassine, Cécile Nail, Philippe Blaise, Benoit Sklenard, Mathieu Bernard, Rémy Gassilloud, Aurélie Marty, Marc Veillerot,
Christophe Vallée, Etienne Nowak, Gabriel Molas “Hybrid-RRAM Towards Next Generation of Non-volatile Memory: Cou-
pling of Oxygen Vacancies and Metal Ions. Adv. Electron. Mater. 2018, Vol. 5, Issue 2, 1800658.
https://doi.org/10.1002/aelm.201800658.
64. Bricalli, A.; Ambrosi, E.; Laudato, M.; Maestro, M.; Rodriguez, R.; Ielmini, D. SiOx-based resistive switching memory (RRAM)
for crossbar storage/select elements with high on/off ratio. In Proceedings of the 2016 IEEE International Electron Devices Meet-
ing (IEDM), San Francisco, CA, USA, 3-7 Dec. 2016; pp. 4.3.1–4.3.4. https://doi.org/10.1109/IEDM.2016.7838344.
65. Nathan Gonzales; John Dinh; Derric Lewis; Nad Gilbert; Bard Pedersen; Deepak Kamalanathan; John R. Jameson; Shane
Hollmer An Ultra Low-Power Non-Volatile Memory Design Enabled by Subquantum Conductive-Bridge RAM. In Proceedings
of the 2016 IEEE 8th International Memory Workshop (IMW), Paris, France, 15-18 May 2016; pp. 1–4.
https://doi.org/10.1109/IMW.2016.7493566.
66. Sassine, G.; Cagli, C.; Nodin, J.; Molas, G.; Nowak, E. Novel Computing Method for Short Programming Time and Low Energy
Consumption in HfO2 Based RRAM Arrays. IEEE J. Electron Devices Soc. 2018, 6, 696–702.
https://doi.org/10.1109/JEDS.2018.2830999.
67. Gabriel Molas, Gilbert Sassine, Cecile Nail, Diego Alfaro Robayo, Jean-François Nodin, Carlo Cagli, Jean Coignus, Philippe
Blaise and Etienne Nowak “(Invited) Resistive Memories (RRAM) Variability: Challenges and Solutions. ECS Trans. 2018, 86,
35. https://doi.org/10.1149/08603.0035ecst.
68. Scott Sills; Shuichiro Yasuda; Alessandro Calderoni; Christopher Cardon; Jonathan Strand; Katsuhisa Aratani; Nirmal
Ramaswamy Challenges for high-density 16Gb ReRAM with 27nm technology. In Proceedings of the 2015 Symposium on VLSI
Technology (VLSI Technology), Kyoto, Japan, 16-18 June 2015; pp. T106–T107. https://doi.org/10.1109/VLSIT.2015.7223639.
69. Kazuhiro Ohba; Shuichiro Yasuda; Tetsuya Mizuguchi; Hiroaki Sei; Tomohito Tsushima; Masayuki Shimuta; Tsunenori Shi-
imoto; Tetsuya Yamamoto; Takeyuki Sone; Seiji Nonoguchi Cross Point Cu-ReRAM with BC-Doped Selector. In Proceedings
Appl. Sci. 2021, 11, 11254 24 of 26
of the 2018 IEEE International Memory Workshop (IMW), Kyoto, Japan, 13-16 May 2018; pp. 1–3.
https://doi.org/10.1109/IMW.2018.8388824.
70. Available online: https://www.flashmemorysummit.com/English/Collaterals/Proceedings/2016/20160809_Keynote4_WD_Si-
varam.pdf (accessed on 10/12/2021).
71. G. Navarro; A. Verdy; N. Castellani; G. Bourgeois; V. Sousa; G. Molas; M. Bernard; C. Sabbione; P. Noé; J. Garrione Innovative
PCM+OTS device with high sub-threshold non-linearity for non-switching reading operations and higher endurance perfor-
mance. In Proceedings of the 2017 Symposium on VLSI Technology, Kyoto, Japan, 5-8 June 2017; pp. T94–T95.
https://doi.org/10.23919/VLSIT.2017.7998208.
72. H. Y. Cheng; J. Y. Wu; R. Cheek; S. Raoux; M. BrightSky; D. Garbin; S. Kim; T. H. Hsu; Y. Zhu; E. K. Lai et al. A thermally robust
phase change memory by engineering the Ge/N concentration in (Ge, N)xSbyTe z phase change material. In Proceedings of the
2012 International Electron Devices Meeting, San Francisco, CA, USA, 10-13 Dec. 2012; pp. 31.1.1–31.1.4.
https://doi.org/10.1109/IEDM.2012.6479141.
73. Gabriele Navarro; Guillaume Bourgeois; Julia Kluge; Anna Lisa Serra; Anthonin Verdy; Julien Garrione; Marie-Claire Cyrille;
Nicolas Bernier; Audrey Jannaud; Chiara Sabbione Phase-Change Memory: Performance, Roles and Challenges. In Proceedings
of the 2018 IEEE International Memory Workshop (IMW), Kyoto Japan, 13-16 May 2018; pp. 1–4.
https://doi.org/10.1109/IMW.2018.8388845.
74. Li, J.; Luan, B.; Lam, C. Resistance drift in phase change memory. In Proceedings of the 2012 IEEE International Reliability
Physics Symposium (IRPS), Anaheim, CA, USA, 15-19 April 2012; pp. 6C.1.1–6C.1.6. https://doi.org/10.1109/IRPS.2012.6241871.
75. Elliott, S.R. Electronic mechanism for resistance drift in phase-change memory materials: Link to persistent photoconductivity.
J. Phys. D: Appl. Phys. 2020, 53, 214002. https://doi.org/10.1088/1361-6463/ab795e.
76. Stanisavljevic, M.; Pozidis, H.; Athmanathan, A.; Papandreou, N.; Mittelholzer, T.; Eleftheriou, E. Demonstration of Reliable
Triple-Level-Cell (TLC) Phase-Change Memory. In Proceedings of the 2016 IEEE 8th International Memory Workshop (IMW),
Paris, France, 15-18 May 2016; pp. 1–4. https://doi.org/10.1109/IMW.2016.7495263.
77. M. J. Kang; T. J. Park; Y. W. Kwon; D. H. Ahn; Y. S. Kang; H. Jeong; S. J. Ahn; Y. J. Song; B. C. Kim; S.W. Nam et al. PRAM cell
technology and characterization in 20nm node size. In Proceedings of the 2011 International Electron Devices Meeting, Wash-
ington, DC, USA, 5-7 Dec. 2011; pp. 3.1.1–3.1.4, doi: 10.1109/IEDM.2011.6131478.
78. Gael F. Close; Urs Frey; Jack Morrish; Richard Jordan; Scott C. Lewis; Tom Maffitt; Matthew J. BrightSky; Christoph Hagleitner;
Chung H. Lam; Evangelos Eleftheriou A 256-Mcell Phase-Change Memory Chip Operating at $2{+}$ Bit/Cell. IEEE Trans. Cir-
cuits Syst. I: Regul. Pap. 2013, 60, 1521–1533. https://doi.org/10.1109/TCSI.2012.2220459.
79. Pangal, K. Emerging Memories including Cross-Point: Opportunities and Challenges. In Proceedings of the IEEE IEDM 2017
Shortcourse, 2 Dec. 2018, San Francisco, CA, USA.
80. F. Arnaud; P. Zuliani; J.P. Reynard; A. Gandolfo; F. Disegni; P. Mattavelli; E. Gomiero; G. Samanni; C. Jahan; R. Berthelon et al.
Truly Innovative 28nm FDSOI Technology for Automotive Micro-Controller Applications embedding 16MB Phase Change
Memory. In Proceedings of the 2018 IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 1-5 Dec.
2018; pp. 18.4.1–18.4.4. https://doi.org/10.1109/IEDM.2018.8614595.
81. Arnaud, F.; Haendler, S.; Clerc, S.; Ranica, R.; Gandolfo, A.; Weber, O. 28 nm FDSOI Platform with Embedded PCM for IoT,
ULP, Digital, Analog, Automotive and others Applications. In Proceedings of the ESSCIRC 2019—IEEE 45th European Solid
State Circuits Conference (ESSCIRC), Cracow, Poland, 23-26 Sept. 2019; pp. 7–10.
https://doi.org/10.1109/ESSCIRC.2019.8902913.
82. L. Tillie; E. Nowak; R. C. Sousa; M.-C. Cyrille; B. Delaet; T. Magis; A. Persico; J. Langer; B. Ocker; I-L Prejbeanu et al. Data
retention extraction methodology for perpendicular STT-MRAM. In Proceedings of the 2016 IEEE International Electron De-
vices Meeting (IEDM), San Francisco, CA, USA, 3-7 Dec. 2016; pp. 27.3.1–27.3.4. https://doi.org/10.1109/IEDM.2016.7838492.
83. W.J. Gallagher; Eric Chien; Tien-Wei Chiang; Jian-Cheng Huang; Meng-Chun Shih; C.Y. Wang; Chih-Hui Weng; Sean Chen;
Christine Bair; George Lee et al. 22 nm STT-MRAM for Reflow and Automotive Uses with High Yield, Reliability, and Magnetic
Immunity and with Performance and Shielding Options. In Proceedings of the 2019 IEEE International Electron Devices Meet-
ing (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp. 2.7.1–2.7.4. https://doi.org/10.1109/IEDM19573.2019.8993469.
84. Zihui Wang; Xiaojie Hao; Pengfa Xu; Longqian Hu; Dongha Jung; Woojin Kim; Kimihiro Satoh; Bing Yen; Zhiqiang Wei; Lien-
chang Wang et al. STT-MRAM for Embedded Memory Applications. In Proceedings of the 2020 IEEE International Memory
Workshop (IMW), Dresden, Germany, 17-20 May 2020; pp. 1–3. https://doi.org/10.1109/IMW48823.2020.9108144.
85. K. Lee; J. H. Bak; Y. J. Kim; C. K. Kim; A. Antonyan; D. H. Chang; S. H. Hwang; G. W. Lee; N. Y. Ji; W. J. Kim et al. 1Gbit High
Density Embedded STT-MRAM in 28nm FDSOI Technology. In Proceedings of the 2019 IEEE International Electron Devices
Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp. 2.2.1–2.2.4. https://doi.org/10.1109/IEDM19573.2019.8993551.
86. V. B. Naik; K. Lee; K. Yamane; R. Chao; J. Kwon; N. Thiyagarajah; N. L. Chung; S. H. Jang; B. Behin-Aein; J. H. Lim et al.
Manufacturable 22 nm FD-SOI Embedded MRAM Technology for Industrial-grade MCU and IOT Applications. In Proceedings
of the 2019 IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp. 2.3.1–2.3.4.
https://doi.org/10.1109/IEDM19573.2019.8993454.
87. J.G. Alzate; U. Arslan; P. Bai; J. Brockman; Y. J. Chen; N. Das; K. Fischer; T. Ghani; P. Heil; P. Hentges et al. 2 MB Array-Level
Demonstration of STT-MRAM Process and Performance Towards L4 Cache Applications. In Proceedings of the 2019 IEEE In-
ternational Electron Devices Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp. 2.4.1–2.4.4.
https://doi.org/10.1109/IEDM19573.2019.8993474.
Appl. Sci. 2021, 11, 11254 25 of 26
88. S. Aggarwal; H. Almasi; M. DeHerrera; B. Hughes; S. Ikegawa; J. Janesky; H. K. Lee; H. Lu; F. B. Mancoff; K. Nage et al. Demon-
stration of a Reliable 1 Gb Standalone Spin-Transfer Torque MRAM For Industrial Applications. In Proceedings of the 2019
IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp. 2.1.1–2.1.4.
https://doi.org/10.1109/IEDM19573.2019.8993516.
89. O. Golonzka; J. -G. Alzate; U. Arslan; M. Bohr; P. Bai; J. Brockman; B. Buford; C. Connor; N. Das; B. Doyle et al.. MRAM as
Embedded Non-Volatile Memory Solution for 22FFL FinFET Technology. In Proceedings of the 2018 IEEE International Electron
Devices Meeting (IEDM), San Francisco, CA,1-5 Dec. 2018; pp. 18.1.1–18.1.4. https://doi.org/10.1109/IEDM.2018.8614620.
90. Liqiong Wei; Juan G. Alzate; Umut Arslan; Justin Brockman; Nilanjan Das; Kevin Fischer; Tahir Ghani; Oleg Golonzka; P.
Hentges; R. Jahan et al.. 13.3 A 7Mb STT-MRAM in 22FFL FinFET Technology with 4ns Read Sensing Time at 0.9V Using Write-
Verify-Write Scheme and Offset-Cancellation Sensing Technique. In Proceedings of the 2019 IEEE International Solid- State
Circuits Conference—(ISSCC), San Francisco, CA, USA, 17-21 Feb. 2019; pp. 214–216.
https://doi.org/10.1109/ISSCC.2019.8662444.
91. L. Grenouillet; T. Francois; J. Coignus; S. Kerdilès; N. Vaxelaire; C. Carabasse; F. Mehmood; S. Chevalliez; C. Pellissier ; F. Trio-
zon et al. Nanosecond Laser Anneal (NLA) for Si-Implanted HfO2 Ferroelectric Memories Integrated in Back-End of Line
(BEOL). In Proceedings of the 2020 IEEE Symposium on VLSI Technology, 16-19 June 2020; pp. 1–2.
https://doi.org/10.1109/VLSITechnology18217.2020.9265061.
92. Mulaosmanovic, H.; Lomenzo, P.D.; Schroeder, U.; Slesazeck, S.; Mikolajick, T.; Max, B. Reliability aspects of ferroelectric haf-
nium oxide for application in non-volatile memories. In Proceedings of the 2021 IEEE International Reliability Physics Sympo-
sium (IRPS), Monterey, CA, USA, 21-25 March 2021; pp. 1–6. https://doi.org/10.1109/IRPS46558.2021.9405215.
93. T. Mikolajick; U. Schroeder; P. D. Lomenzo; E. T. Breyer; H. Mulaosmanovic; M. Hoffmann; T. Mittmann; F. Mehmood; B. Max;
S. Slesazeck Next Generation Ferroelectric Memories enabled by Hafnium Oxide. In Proceedings of the 2019 IEEE International
Electron Devices Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp. 15.5.1–15.5.4.
https://doi.org/10.1109/IEDM19573.2019.8993447.
94. L. Grenouillet; T. Francois; J. Coignus; N. Vaxelaire; C. Carabasse; F. Triozon; C. Richter; U. Schroeder; E. Nowak Performance
assessment of BEOL-integrated HfO2-based ferroelectric capacitors for FeRAM memory arrays. In Proceedings of the 2020 IEEE
Silicon Nanoelectronics Workshop (SNW), Honolulu, HI, USA, 13-14 June 2020; pp. 5–6.
https://doi.org/10.1109/SNW50361.2020.9131648.
95. Sven Beyer; Stefan Dünkel; Martin Trentzsch; Johannes Müller; Andreas Hellmich; Dirk Utess; Jan Paul; Dominik Kleimeier;
John Pellerin; Stefan Müller et al. FeFET: A versatile CMOS compatible device with game-changing potential. In Proceedings of
the 2020 IEEE International Memory Workshop (IMW), Dresden, Germany, 17-20 May 2020; pp. 1–4.
https://doi.org/10.1109/IMW48823.2020.9108150.
96. S. Dünkel; M. Trentzsch; R. Richter; P. Moll; C. Fuchs; O. Gehring; M. Majer; S. Wittek; B. Müller; T. Melde et al. A FeFET based
super-low-power ultra-fast embedded NVM technology for 22nm FDSOI and beyond. 2017 IEEE International Electron Devices
Meeting (IEDM), San Francisco, CA, USA, 2-6 Dec. 2017; pp. 19.7.1–19.7.4. https://doi.org/10.1109/IEDM.2017.8268425.
97. Pesic, M.; Knebel, S.; Hoffmann, M.; Richter, C.; Mikolajick, T.; Schroeder, U. How to make DRAM non-volatile? Anti-ferroelec-
trics: A new paradigm for universal memories. In Proceedings of the 2016 IEEE International Electron Devices Meeting (IEDM),
San Francisco, CA, USA, 3-7 Dec. 2016; pp. 11.6.1–11.6.4. https://doi.org/10.1109/IEDM.2016.7838398.
98. Hady, F.T.; Foong, A.; Veal, B.; Williams, D. Platform Storage Performance With 3D XPoint Technology. IEEE 2017, 105, 1822–
1833. https://doi.org/10.1109/JPROC.2017.2731776.
99. Carboni, R.; Ielmini, D. Stochastic Memory Devices for Security and Computing. Adv. Electron. Mater. 2019, 5, 1900198.
https://doi.org/10.1002/aelm.201900198.
100. Sahay, S.; Kumar, A.; Parmar, V.; Suri, M. OxRAM RNG Circuits Exploiting Multiple Undesirable Nanoscale Phenomena. IEEE
Transactions on Nanotechnology 2017, 16, 560–566. https://doi.org/10.1109/TNANO.2016.2647623.
101. Giuseppe Piccolboni; Gabriel Molas; Daniele Garbin; Elisa Vianello; Olga Cueto; Carlo Cagli; Boubacar Traore; Barbara De
Salvo; Gérard Ghibaudo; Luca Perniola Investigation of Cycle-to-Cycle Variability in HfO2-Based OxRAM. IEEE Electron Device
Lett. 2016, 37, 721–723. https://doi.org/10.1109/LED.2016.2553370.
102. Chen, A. Comprehensive assessment of RRAM-based PUF for hardware security applications. In Proceedings of the 2015 IEEE
International Electron Devices Meeting (IEDM), Washington, DC, USA, 7-9 Dec. 2015; pp. 10.7.1–10.7.4.
https://doi.org/10.1109/IEDM.2015.7409672.
103. Perrissin, N.; Gregoire, G.; Lequeux, S.; Tillie, L.; Strelkov, N.; Auffret, S.; Prejbeanu, L.D.B.; Sousa, R.C.; Vila, L.; Dieny, B.; et
al. Perpendicular shape anisotropy spin transfer torque magnetic random-access memory: Towards sub-10 nm devices. J. Phys.
D: Appl. Phys. 2019, Vol. 52, Number 23, 234001.
104. D. R. B. Ly; J-P. Noel; B. Giraud; P. Royer; E. Esmanhotto; N. Castellani; T. Dalgaty; J -F. Nodin; C. Fenouillet-Beranger; E.
Nowak Novel 1T2R1T RRAM-based Ternary Content Addressable Memory for Large Scale Pattern Recognition. In Proceedings
of the 2019 IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp. 35.5.1–35.5.4.
https://doi.org/10.1109/IEDM19573.2019.8993621.
105. Tsai, H.; Ambrogio, S.; Narayanan, P.; Shelby, R.M.; Burr, G.W. Recent progress in analog memory-based accelerators for deep
learning. J. Phys. D Appl. Phys. 2018, 51, 283001. https://doi.org/10.1088/1361-6463/aac8a5.
106. Available online: https://www.ibm.com/blogs/research/2016/12/the-brains-architecture-efficiency-on-a-chip/ (accessed on).
Appl. Sci. 2021, 11, 11254 26 of 26
107. Ielmini, D.; Wong, H.-S.P. In-memory computing with resistive switching devices. Nat. Electron. 2018, 1, 333–343,
DOI:10.1038/S41928-018-0092-2.
108. Wu, T.F.; et al. 14.3 A 43pJ/Cycle Non-Volatile Microcontroller with 4.7µs Shutdown/Wake-up Integrating 2.3-bit/Cell Resistive
RAM and Resilience Techniques. In Proceedings of the 2019 IEEE International Solid- State Circuits Conference—(ISSCC), San
Francisco, CA, USA, 2019; pp. 226–228. https://doi.org/10.1109/ISSCC.2019.8662402.
109. Sheng, X.; Graves, C.E.; Kumar, S.; Li, X.; Buchanan, B.; Zheng, L.; Lam, S.; Li, C.; Strachan, J.P. “Low-Conductance and Multi-
level CMOS-Integrated Nanoscale Oxide Memristors. Adv. Electron. Mater. 2019, 5, 1800876.
https://doi.org/10.1002/aelm.201800876.
110. Strukov, D.; Indiveri, G.; Grollier, J.; Fusi, S. “Building brain-inspired computing. Nat. Commun. 2019, 10, 4838.
https://doi.org/10.1038/s41467-019-12521-x.
111. Geoffrey W. Burr, Robert M. Shelby, Abu Sebastian, Sangbum Kim, Seyoung Kim, Severin Sidler, Kumar Virwani, Masatoshi
Ishii, Pritish Narayanan, Alessandro Fumarola et al. . Adv. Phys. X 2017, Vol. 2, No. 1, pp. 89–124,
https://doi.org/10.1080/23746149.2016.1259585.
112. Vinay Joshi, Manuel Le Gallo, Simon Haefeli, Irem Boybat, S. R. Nandakumar, Christophe Piveteau, Martino Dazzi, Bipin Ra-
jendran, Abu Sebastian, Evangelos Eleftheriou, Accurate deep neural network inference using computational phase-change
memory. Nat. Commun. 2020, 11, 2473. https://doi.org/10.1038/s41467-020-16108-9.
113. Stefano Ambrogio, Pritish Narayanan, Hsinyu Tsai, Robert M. Shelby, Irem Boybat, Carmelo di Nolfo, Severin Sidler, Massimo
Giordano, Martina Bodini, Nathan C. P. Farinha et al. Equivalent-accuracy accelerated neural-network training using analogue
memory. Nature 2018, 558, 60–67. https://doi.org/10.1038/s41586-018-0180-5.
114. Tifenn Hirtzlin, Marc Bocquet, Bogdan Penkovsky, Jacques-Olivier Klein, Etienne Nowak, Elisa Vianello, Jean-Michel Portal
and Damien Querlioz “Digital Biologically Plausible Implementation of Binarized Neural Networks with Differential Halfnium
Oxide Resistive Memory Arrays. Front. Neurosci. 13, 1383, 9 January 2020 | https://doi.org/10.3389/fnins.2019.01383.
115. M. Bocquet; T. Hirztlin; J.-O. Klein; E. Nowak; E. Vianello; J.-M. Portal; D. Querlioz “In-Memory and Error-Immune Differential
RRAM Implementation of Binarized Deep Neural Networks. Tech. Dig. IEDM 2018, San Francisco, CA, USA, 1-5 Dec. 2018, DOI:
10.1109/IEDM.2018.8614639.
116. Takumi Mikawa; Ryutaro Yasuhara; Koji Katayama; Kazuyuki Kouno; Takashi Ono; Reiji Mochida; Yuriko Hayata; Masayoshi
Nakayama; Hitoshi Suwa; Yasushi Gohou et al. Neuromorphic computing based on Analog ReRAM as low power solution for
edge application. In Proceedings of the 2019 IEEE 11th International Memory Workshop (IMW), Monterey, CA, USA, 12-15
May 2019; pp. 1–4. https://doi.org/10.1109/IMW.2019.8739720.
117. E.R. Hsieh; M. Giordano; B. Hodson; A. Levy; S.K. Osekowsky; R.M. Radway; Y.C. Shih; W. Wan; T. F. Wu; X. Zheng et al.
High-Density Multiple Bits-per-Cell 1T4R RRAM Array with Gradual SET/RESET and its Effectiveness for Deep Learning. In
Proceedings of the 2019 IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp.
35.6.1–35.6.4. https://doi.org/10.1109/IEDM19573.2019.8993514.
118. Milo, V.; Zambelli, C.; Olivo, P.; Pérez, E.; Mahadevaiah, M.K.; Ossorio, O.G.; Wenger, C.; Ielmini, D. Multilevel HfO2-based
RRAM devices for low-power neuromorphic networks. APL Mater. 2019, 7, 081120. https://doi.org/10.1063/1.5108650.
119. E. Esmanhotto; L. Brunet; N. Castellani; D. Bonnet; T. Dalgaty; L. Grenouillet; D. R. B. Ly; C. Cagli; C. Vizioz; N. Allouti et al.
High-Density 3D Monolithically Integrated Multiple 1T1R Multi-Level-Cell for Neural Networks. In Proceedings of the 2020
IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 12-18 Dec. 2020; pp. 36.5.1–36.5.4.
https://doi.org/10.1109/IEDM13553.2020.9372019.
120. Maass, W. Networks of Spiking Neurons: The Third Generation of Neural Network Models. NeuralNetworks 1997, 10, 1659–
1671.
121. Ielmini, D.; Ambrogio, S. «Emerging neuromorphic devices». Nanotechnology 2020, 31,, 9, 092001. https://doi.org/10.1088/1361-
6528/ab554b.
122. A. Valentian; F. Rummens; E. Vianello; T. Mesquida; C. Lecat-Mathieu de Boissac; O. Bichler; C. Reita Fully Integrated Spiking
Neural Network with Analog Neurons and RRAM Synapses. In Proceedings of the 2019 IEEE International Electron Devices
Meeting (IEDM), San Francisco, CA, USA, 7-11 Dec. 2019; pp. 14.3.1–14.3.4. https://doi.org/10.1109/IEDM19573.2019.8993431.
123. Amir REGEV; Alessandro BRICALLI; Giuseppe PICCOLBONI; Alexandre VALENTIAN; Thomas MESQUIDA; Gabriel MO-
LAS; Jean-François NODIN Fully-Integrated Spiking Neural Network Using SiOx-Based RRAM as Synaptic Device. In Pro-
ceedings of the 2020 2nd IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS), Genova, Italy,
31 Aug.-2 Sept. 2020; pp. 145–148. https://doi.org/10.1109/AICAS48895.2020.9073840.