SCIENCE AND TECHNOLOGY OF ADVANCED MATERIALS

2023, VOL. 24, NO. 1, 2162323

https://doi.org/10.1080/14686996.2022.2162323

FOCUS ISSUE REVIEW

A review of memristor: material and structure design, device performance,

applications and prospects
Yongyue Xiaoa, Bei Jianga, Zihao Zhanga, Shanwu Kea, Yaoyao Jina, Xin Wenb and Cong Yea
a
Hubei Key Laboratory of Ferro-& Piezoelectric Materials and Devices, Faculty of Physics and Electronic Science, Hubei University, Wuhan,
China;
b
Faculty of Chemical Technology and Engineering, West Pomeranian University of Technology in Szczecin, Szczecin, Poland

ABSTRACT ARTICLE HISTORY

With the booming growth of artificial intelligence (AI), the traditional von Neumann computing Received 16 October 2022
architecture based on complementary metal oxide semiconductor devices are facing memory Revised 9 December 2022
wall and power wall. Memristor based in-memory computing can potentially overcome the Accepted 21 December 2022
current bottleneck of computer and achieve hardware breakthrough. In this review, the recent KEYWORDS
progress of memory devices in material and structure design, device performance and applica­ Artificial intelligence;
tions are summarized. Various resistive switching materials, including electrodes, binary oxides, in-memory computing;
perovskites, organics, and two-dimensional materials, are presented and their role in the memristor; material and
memristor are discussed. Subsequently, the construction of shaped electrodes, the design of structure design; device
functional layer and other factors influencing the device performance are analyzed. We focus performance
on the modulation of the resistances and the effective methods to enhance the performance.
Furthermore, synaptic plasticity, optical-electrical properties, the fashionable applications in
logic operation and analog calculation are introduced. Finally, some critical issues such as the
resistive switching mechanism, multi-sensory fusion, system-level optimization are discussed.

The recent progress of memory devices in material and structure design, device performance
and applications were summarized. Various resistive switching materials are presented and the
role in the memristor were systematically discussed. The construction of the shaped electrode,
the design of functional layer and other factors influencing the device performance are analyzed.
The excellent and advanced performances of the memristor from the frontier researches are
illustrated. The fashionable applications in logic operation and analog calculation are
introduced. Memristor: RS materials, structure design, performances, and applications

CONTACT Cong Ye [email protected] Hubei Key Laboratory of Ferro-& Piezoelectric Materials and Devices, Faculty of Physics and Electronic
Science, Hubei University, Wuhan 430062, China
© 2023 The Author(s). Published by National Institute for Materials Science in partnership with Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits
unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Sci. Technol. Adv. Mater. 24 (2023) 2 Y. XIAO et al.

1. Introduction have simple structure, fast reading and writing speed,

good scalability, high density, low cost, and are compa­
Human society is facing the contradiction between the
tible with complementary metal oxide semiconductor
growing demand for artificial intelligence versus insuffi­
(CMOS) process.
cient computing power in the era of big data. The current
In the future, memristor is one of the promising
von Neumann architecture has two serious problems for
candidates for high-density, high-energy efficiency,
the separation of storage and computing in intelligent ultra-fast, low-latency, low power, large-capacity non-
computing [1]: 1. The access speed of memory is far volatile memory. Therefore, many companies
behind the calculation speed of the central processing (Samsung, Panasonic, HP, Micron, Sony, Yangtze
unit (CPU), and greatly reduces the utilization of the Memory Technologies Co., Ltd. (YMTC), Crossbar
central processing unit, which is called ‘memory wall’ etc.) are engaged in research and development of mem­
[2,3]; 2. The loss of power consumption caused by data ristors. The traditional method to simulate a neuron or
transmission has seriously undermined the development a synapse requires dozens of traditional electronic
of the chip, which is known as the ‘power wall’ [4]. As devices such as transistors and capacitors [7].
predicted by Moore’s law, computing power has been Although breakthroughs have been made in brain-
enhanced by continuously reducing device size, alleviat­ inspired computing systems based on traditional
ing the von Neumann bottleneck [5,6], but these methods CMOS technology in recent years [13], there are still
still belong to the von Neumann architecture, and the many challenges in terms of the integration and power
storage and computing of data are still separated from consumption. With a continuously adjustable resistance
each other, which cannot fundamentally solve the pro­ under an applied electric field, memristors are ideal
blems. Hence, the researchers investigate new computing neurosynaptic bionic components, providing
systems with non-von Neumann architectures [7,8]. a potential solution for high-density, low-power brain-
In-memory computing architecture, represented by inspired computing chips. Furthermore, the memristor
brain-inspired computing, is one of the non-von arrays can build more integrated neural network struc­
Neumann architectures [9,10]. However, its develop­ tures, including artificial neural networks (ANN) con­
ment has been slowed in the past decades, which is volutional neural networks (CNNs), deep neural
mainly due to the lack of a physical device that can realize networks (DNNs), recurrent neural networks (RNNs),
the architecture of integrated storage and computing and spike neural networks (SNNs). The memristor
system. The problem was not solved until the advent of arrays directly use Ohm’s law for addition and
the memristor. The theoretical memristor was proposed Kirchhoff’s law for multiplication, thus enabling the
by Leon O. Chua as the fourth basic circuit element in parallel multiplication-accumulation (MAC) operations
1971 [11], and the first physical implementation of the required by neural networks, resulting in a significant
memristor was experimentally verified by Strukov et al. increase in the speed and power efficiency of neural
at Hewlett-Packard (HP) Labs in 2008 [12]. Memristors networks [14].

Scheme 1. Memristor: RS materials, structure design, performances, and applications. Reproduced with permission from [15,16]
copyright 2021, Wiley, [17] copyright 2021, American Chemical Society, [18] copyright 2021, Springer Nature, [19] copyright 2021,
American Chemical Society, [20] copyright 2018, Royal Society of Chemistry, [21] copyright 2022, Elsevier, [22] copyright 2020,
Springer Nature, [23] copyright 2020, American Chemical Society, [24] copyright 2019, Wiley, [25] copyright 2022, American
Chemical Society, [26] copyright 2022, Wiley.
Sci. Technol. Adv. Mater. 24 (2023) 3 Y. XIAO et al.

In this review, as shown in Scheme 1, various resistive Cu-Te [36], Hf-Ta [34], Ag-Cu [35], etc.) intentionally
switching (RS) materials of the memristor are intro­ designed to stabilize the RS behavior.
duced, including common electrodes as well as particular
electrode materials, binary oxides, perovskites, organics,
2.2 RS materials
and two-dimensional (2D) materials. Next, the latest
structural designs, such as electrode engineering, resistive The RS materials are the most important part of the
layers, and three-dimensional (3D) integration, are intro­ memristors, and different RS materials have different
duced, and their role in the memristor is systematically resistive characteristics. RS materials can be generally
addressed. Subsequently, the construction of shaped elec­ classified as inorganic and organic materials.
trodes, the design of functional layers, and other factors Inorganic materials (binary oxide, perovskite, 2D
affecting device performance are analyzed. Excellent per­ materials etc.) usually exhibit a more stable, faster,
formance of the memristor is illustrated by selected and robust RS behavior, while organic materials have
research results. Finally, applications of memristors in the advantages of high flexibility, simple preparation
logic and simulation are summarized, and some critical method and low cost.
issues such as RS mechanisms, multi-sensory fusion, and
system-level optimization are looked ahead, providing 2.2.1 Binary oxides
further developments and new insights for memristor- Binary oxides have a simple composition, high stabi­
based in-memory computing. lity, low cost, simple preparation process, and are
compatible with the traditional CMOS process. They
include TiOx [46–50], SiOx [51–53], AlOx [54], NiOx
2. Materials of memristors
[55,56], CuOx [57,58], ZnOx [59,60], HfOx [61,62],
The memristor mainly has a metal-insulator-metal TaOx [63,64], WOx [65,66], ZrOx [67,68], SnOx [69]
(MIM) sandwich structure, including two layers of elec­ etc. In particular, HfOx and TaOx are the most promis­
trodes and a middle resistive layer formed by ing binary oxides due to their sub-ns operation speed
a semiconductor or insulator. The properties and resis­ and ultimate endurance of more than 1010 cycles [70].
tance mechanism of memristors are closely related to Among the many resistive materials, binary oxides are
the component materials of the devices, which include the most abundant and have excellent resistive proper­
SET/RESET voltage, on/off ratio, switching speed, ties, such as ultra-high on/off ratio >1010 in HfOx, sub-
endurance, retention, power consumption, etc. ns switching speeds in HfOx [71], and TaOx [70], and
extreme endurance >1012 cycles in TaOx [70].
However, conventional binary oxide memories gener­
2.1 Electrode materials
ally have high power consumption or low uniformity.
Memristor electrodes not only carry electric current, In order to solve these problems, our group prepared
but may also participate in the resistive reaction. They a bismuth-doped tin oxide (Bi: SnO2) memristor with
are commonly made from the following materials: con­ an ITO/Bi: SnO2/TiN structure by magnetron sputter­
ventional metals [27–31], noble metals [32,33], alloys ing. As shown in Figure 1(a), the self-compliance
[34–36], carbon-based materials such as graphene [37] current, switching voltage and operating current of
and carbon nanotubes [38], nitrides such as TiN [16] the Bi: SnO2 memristor were significantly smaller
and TaN [39], transparent conductive flexible oxides than those of the ITO/SnO2/TiN device. As shown in
such as indium tin oxide (ITO) [29,39,40] and doped Figure 1(b), the high resistance state (HRS) and low
ITO [41], F-doped tin oxide (FTO) [42–46], etc. The resistance state (LRS) of Bi: SnO2 memristors had
common electrode materials can be classified into four higher resistance values, and the operating current of
types according to their different role in RS behavior. Bi: SnO2 devices was reduced by more than an order of
First, the electrodes are mainly used as current conduct­ magnitude compared to that of SnO2 devices. With the
ing media and have almost no effect on the RS behavior. content of 4.8% Bi doping, the SET operating power of
Such electrode materials mainly include inert metals, doped device was 16 µW for ITO/Bi: SnO2/TiN mem­
such as Au [32] and Pt [33]. Second, the electrodes are ory cell of 0.4 × 0.4 µm2, which was cut down by two
responsible for the formation of conductive filaments orders of magnitude. Transmission electron micro­
(CFs), which is exactly the case of cation migration- scopy (TEM) observation of the Bi: SnO2 devices
based memristors. In these memristors, CFs are formed revealed that the bismuth atoms surround the surface
by electrochemical dissolution and deposition of elec­ of SnO2 crystals to form the coaxial Bi CFs [69].
trochemically active metal electrodes including mainly On the other hand, with the rapid development of
Cu [30] and Ag [27]. Finally, there are some novel portable and wearable devices, it is urgent to develop
electrodes, such as ITO [39], FTO [42], the graphene flexible memristors for data storage and integration with
[37], which are generally used to prepare flexible and other electronic devices in flexible electronic systems. We
transparent memristors, TiN [16] electrodes prepared have fabricated a resistance random access memory
using CMOS processes, and alloy electrodes (such as (RRAM) with bilayer TiO2/HfO2 structure based on
Sci. Technol. Adv. Mater. 24 (2023) 4 Y. XIAO et al.

Figure 1. (a) I–V characteristic for ITO/SnO2/TiN and ITO/Bi: SnO2/TiN memristors. (b) The on/off state resistance ratio for ITO/SnO2
/TiN and ITO/Bi: SnO2/TiN memristors. Reproduced with permission from [69], copyright 2020, Wiley. (c) Typical I–V curves of the
ITO/TiO2/HfO2/Pt resistive RRAM. The inset is a schematic diagram of the device. (d) LRS/HRS resistances for the ITO/TiO2/HfO2/Pt
RRAM under tensile strain with different bending radius. Inset: Photograph of the ITO/TiO2/HfO2/Pt RRAM on polyvinyl
naphthalate substrate. Reproduced with permission from [49], copyright 2019, Wiley. (e) 100 sequential I–uV curves for the
Pt/HfO2/BFO/HfO2/TiN and the Pt/HfO2/TiN devices. The inset is schematic diagram of RS mechanism of the Pt/HfO2/BFO/HfO2/TiN
during the set process. (f) High-resolution TEM image of a Hf nanofilament, inserts: images of FFT patterns for hexagonal crystal Hf
in the functional layer of Pt/HfO2/BFO/HfO2/TiN device. Reproduced with permission from [16], copyright 2021, Wiley.

polyethylene naphthalate substrate, as illustrated in 2013, Panasonic produced a microcontroller named

Figure 1(c), which had 500 times of mechanical bending. MN101 L, which adopted TaOx-based memristor using
The coefficients of variation for the high and low resis­ 0.18 μm process as an embedded memory. In 2014, Sony
tance states were 3.2% and 3%, respectively. In fabricated a 16 Gb test chip based on oxide memristors
Figure 1(d), no degradation of performance was using a 27 nm process. In 2016, the Institute of
observed under mechanical stresses with bending radii Microelectronics of the Chinese Academy of Sciences
ranging from 70 to 10 mm. The asymmetric hourglass- fabricated a 4-layer self-selective oxide memristor array
like oxygen vacancy (Vo) distribution at the HfO2/TiO2 with 3-dimensional vertical structure as a test chip.
interface played a key role in the performance of this Taiwan Semiconductor Manufacturing Company
flexible RRAM device [49]. Limited (TSMC) has been able to produce embedded
In order to improve the performance of the memris­ oxide memristors modules using a 22 nm process [73].
tors, optimize simulate synaptic function and perform Samsung reported a Pt/Ta2O5/TaOx/Pt memristor with
neuromorphic calculations, our research group designed a switching speed of 10 ns and an endurance of 1012
a three-layer HfO2/BiFeO3(BFO)/HfO2 memristor by cycles [74]. Liu’s group in the Institute of
inserting a 4 nm conventional ferroelectric BFO layer. Microelectronics of the Chinese Academy of Sciences
As demonstrated in Figure 1(e), the memristor has designed a Ta/TaOx/W memristor with a 2-transistor
improved RS performance, such as with the storage 2-resistor (2T2 R) PUF scheme with assisting circuit
window of 104 and the multi-level storage capability. techniques for dense and reliable cryptographic key gen­
The pattern recognition simulation based on neuro­ eration [75].
morphic network was conducted with recognition accu­
racy of 91.2%. The CFs consisting of single crystals of 2.2.2 Perovskites
hafnium (Hf) with a hexagonal lattice structure were Perovskites are materials with ABX3 structure. The
observed using high-resolution TEM (Figure 1(f)). doping of the A-site with metal ions of low valence
Attributed to the BFO inserting layer, more oxygen causes oxygen vacancies as the material [76]. By par­
vacancies originate from the BFO layer, which makes tially doping the A- and B-site with cations of different
the formation of Hf CF become easier [16]. valence or radius, a wide variety of physical and che­
Not only academics, but also the industry has engaged mical properties can be achieved. Where A is NH2
in the research on binary oxides. In 2010, Unity reported CH=NH21+ (FA1+), CH3NH31+ (MA1+), or Cs+, Ag+;
a 64 Mb test chip based on oxide memristor [72]. In B is Pb2+, or Sn2+; and X is Cl-, or Br-, or I- or
Sci. Technol. Adv. Mater. 24 (2023) 5 Y. XIAO et al.

Figure 2. (a) (i) Schematic showing the MAPbI3 quantum wires (QWs)/nanowires (NWs) in a porous alumina membrane stacked
between crisscrossed Ag/ITO and Au finger electrodes, supported by a polyethylene terephthalate substrate. (ii) An enlarged view of
an individual QW comprising the Ag filament sandwiched between ITO/Ag and Au contacts. (iii) Crystal structure of MAPbI3. (b)
Schematic diagram of MAPbI3 QWs memory device. Insert: Highly magnified cross-sectional view of 153 nm2 device after Ag
evaporation showing the distinct SiO2 layer between the Ag electrode and MAPbI3QWs. Reproduced with permission from [17],
copyright 2021, American Chemical Society. (c) Schematic diagram of the Cs2AgBiBr6-based device. I–V characteristics of Cs2AgBiBr6-
based device in different harsh environments: (d) Burnt by luminous cone of alcohol lamp for 10 s. (e) Heated from 303 to 453 K. (f)
Exposed under 60Co γ-ray irradiation with a total dose of 5 × 105 rad (SI). Reproduced with permission from [94], copyright 2019, Wiley.

a combination of them [77]. Common perovskites 10 s in alcohol burner flames, and withstand
include MAPbI3 [78–81], (C4H9NH3)2PbBr4 [82], a temperature of 180°C or 60Co γ-ray irradiation at
LaFeO3 [83], MAPbBr3 [84], CH3NH3PbClXI3 [85– a dose of 5 × 105 rad (SI) (Figure 2(f)). These parameters
87], etc. Due to the superior optoelectronic properties are superior to those of commercial flash memory
of perovskite materials, such as suitable tunable band devices [94]. Although memristors based on perovskite
gap, high optical absorption coefficient, high photo­ have been hotly investigated, the perovskite memristor
luminescence quantum yield, narrow-band emission, still has some disadvantages like the incompatibility
relatively short absorption wavelength, long diffusion with CMOS processes, which limits their practical
length and low cost, perovskites are ideal materials for applications.
solar cells [88,89], photodetectors [90,91], light-
emitting diodes [92,93], etc. Because of novel optoe­ 2.2.3 Organic materials
lectronic properties such as mixed ion and electron With the rapid development of portable and wearable
conductivity, switchable majority carrier concentra­ devices, it is urgent to develop flexible memristors for
tion and slow photocurrent decay, perovskites are data storage in flexible electronic systems. Due to the low
promising materials for the design of next-generation cost, solution processability, flexibility and the ability of
neuromorphic memristors. As illustrated in large scale preparation, the organic materials (organic
Figure 2(a), Poddar et al. fabricated an RRAM with small molecules SU-8 [95], monochloro copper phtha­
a perovskite quantum well/nanowire (NW) array as locyanine (ClCuPc) [96], Ru-complex of 2-(phenylazso)
the switch matrix, with an on/off ratio of ~107, multi- pyridine ([RuII(L)3] (PF6)2) [97], and organic polymers
bit memory capability, ultrafast programming speed PEDOT: PSS [98,99], fluoropolymer [100], etc.) are pro­
of ~ 100 ps (Figure 2(b)) [17]. mising candidates for flexible memristors. Liu et al. pre­
A serious deficiency of perovskite materials is lack of pared an Ag/2DPBTA+PDA/ITO memristor, which
environmental stability. To solve this problem, Cheng exhibited good switching performance with high relia­
et al. firstly utilized lead-free double perovskite Cs2 bility and reproducibility, with switching ratios ranging
AgBiBr6 for environmentally robust memristors [94]. from 102 to 105, depending on the thickness of the film
A memristor with ITO/Cs2AgBiBr6/Au structure exhib­ [101]. However, the 2DP-based non-volatile memristor
ited performance of 105 seconds of retention and 104 had an endurance of only 200 cycles at 0.1 V [101]. To
times of mechanical bending (Figure 2(c)). Most impor­ overcome the problem of low endurance in organic
tantly, the performance of the memristor remains memristors, Xu et al. reported a new, flexible, and robust
robust in harsh environments. The device could last diffusion memristor based on a copolymer of
Sci. Technol. Adv. Mater. 24 (2023) 6 Y. XIAO et al.

Figure 3. (a) Mechanical and electrical performances of the developed organic memristor. Reproduced with permission from [104],
copyright 2021, Wiley. (b) Schematic illustrations of the array consisting of pV3D3/Al2O3 memristors and their multiply-accumulate
operations. Reproduced with permission from [105], copyright 2022, Wiley. (c) Textile chip made from DNA-bridged memristor
cross-bar arrays. Each contact point represents a memristor. (d) Photograph and the operation mechanism of a proof-of-concept
all-fabric data-processing system. Scale bars in (e) is 0.5 cm. Reproduced with permission from [106], copyright 2020, Wiley.

trifluorochloroethylene and vinylidene fluoride (FK- biodegradable, bioabsorbable, environmentally

800). The device with the structure of Ag/FK-800/Pt friendly, biocompatible, implantable, and cheap
had an switching endurance of over 106 cycles [102]. [109,110]. Xu et al. prepared a robust DNA-bridged
Although the endurance of the organic memristors memristor based on electrophoretically DNA active
with the structure of Ag/FK-800/Pt has been greatly layers on fiber electrodes (Figure 3(c)). The unique
enhanced, their manufacturing yield is still low. Zhang structure and orientation of the DNA molecule com­
et al. achieved a record yield of 90% for polymer bined with silver nanoparticles provided best-in-class
memristors with miniaturization and low power con­ performance, such as a low operating voltage of 0.3 V,
sumption by implementing a 2D conjugation strategy low power consumption of 100 pW, and high switch­
[103]. This organic memristor had a low power con­ ing speed of 20 ns. As demonstrated in Figure 3(d),
sumption of 10−15 J/bit by constructing coplanar basic logic calculations, such as implication and
macromolecules with 2D conjugated thiophene deri­ NAND, were demonstrated as a function of the textile
vatives [103]. Park et al. fabricated a solution- chip, and a full-fabric information processing system
processed flexible organic memristor array with self- was demonstrated by integrating the memristor with
selectivity by systematically designing the diffusion of the power and light-emitting modules [106].
carbon fibers in a polymer medium, as shown in
Figure 3(a). In this array, the cellular self-selectivity 2.2.4 2D materials
effectively suppressed the sneaking current paths and 2D materials have attracted widespread attention due to
the maximum size of the memristor arrays is found to their atomic-scale thickness as well as their excellent
be more than 1 M bits. Meanwhile the neural networks electrical properties and novel characteristics, such as
based on these devices achieved a high recognition flexibility and transparency [56]. A single layer or several
accuracy of over 90% [104]. layers can be exfoliated from the bulk 2D crystal due to
To improve the reliability of the memristors based on weak van der Waals (vdW) forces between adjacent layers
organics, Cha etal. prepared Cu/poly (1,3,5-trivinyl- [111]. These characteristics make 2D materials as an
1,3,5-trimethylcyclotrisiloxane) (pV3D3)/Al2O3/Pt alternative candidate for the atomic-scale thickness,
memristors, using pV3D3 and Al2O3 films for organic- excellent electrical properties, and novel characteristics,
inorganic bilayer stacking of synaptic unit cells [105], as such as flexibility, transparency. For these reasons,
illustrated in Figure 3(b). The 5-bit multilevel retention researchers investigated 2D memristors with vertical
of 104 s was firstly achieved. The added layer of Al2O3 and planar structures. As illustrated in Figure 4(a), our
played a key role in mitigating Joule heating of RESET group proposed flexible memristors with vertical struc­
process and conductivity fluctuations, resulting in reli­ tures based on polyvinyl alcohol-graphene oxide hybrid
able retention characteristics and stable switching beha­ nanocomposites [112]. These devices excellent optical
vior [105]. and electrical properties (a narrow band gap (0.2 − 2
Some novel organic materials (deoxyribonucleic eV) and respond to light spanning from the visible to
acid (DNA) [106,107] and egg proteins [108], etc.) infrared regimes) with high stability against mechanical
have received widespread attention because they are stress at voltages below 0.5 V. Tang et al. prepared an Ag/
Sci. Technol. Adv. Mater. 24 (2023) 7 Y. XIAO et al.

Figure 4. (a) Schematic of an ITO/PVA-GO/ITO memristor, where PVA stands for poly(vinyl alcohol) and GO for graphene oxide.
Reproduced with permission from [112], copyright 2020, Wiley. (b) Schematic of an Ag/InSe/Ag memristor. Inset is the optical image
of the device with a channel length of 8 μm. Reproduced with permission from [113], copyright 2021, AIP Publishing LLC. (c)
Sandwich-like structure of the TiS3-based memristor. Reproduced with permission from [17], copyright 2021, American Chemical
Society. (d) Schematic diagram of the WS2/MoS2 heterojunction memristor. Reproduced with permission from [111], copyright 2021,
Royal Society of Chemistry. (e) The typical structure of Al/Ti3C2: Ag/Pt device. Reproduced with permission from [114], copyright 2021,
Elsevier. (f) Schematics of the multilayer Gr/MoS2(MGM) cellulose paper memory. Reproduced with permission from [115], copyright
2019, American Chemical Society.

InSe/Ag memristor using 2D InSe nanosheets functional layer materials are damaged during this pro­
(Figure 4(b)) [113]. The device had a low SET voltage of cess, the final prepared devices have good performance
0.3 V and an on/off ratio of 4.5 × 103 at a read voltage of and endurance and can operate stably for a long time.
0.1 V. Furthermore, external strain, light, electric and The heterostructure of 2D materials in memristors are
magnetic fields can be applied to boost up the character­ stacked with graphene [118,119], transition metal
istics of 2D based memristors. Our research group dichalcogenide (TMD, including MoS2 [116,117],
designed a 2D transition metal trihalide (TMTC)-based WSe2 [120,121], MoTe2 [122,123], etc.), black phos­
optoelectronic memristor, as shown in Figure 4(c). The phorus (BP) [124–126] and hexagonal boron nitride
memristor exhibited stable bipolar RS property due to the (h-BN) [127,128] to form vdW heterojunctions with
excellent optical and electrical properties of titanium atomically flat interfaces. Zhang et al. used a low-cost
trisulfide (TiS3) functional layer. Multi-level storage was transfer-free 2D material synthesis method to fabricate
achieved by applying the wavelengths between 400 and a novel memristor based on WS2/MoS2 2D semicon­
808 nm, and the synaptic properties such as Long-term ductor heterojunctions (Figure 4(d)) [111]. The metal/
Depression (LTD), Long-term Plasticity (LTP) and heterojunction/metal (MHM) structure utilized
Spiking-timing-dependent plasticity (STDP) were rea­ a reliable band modulation mechanism rather than
lized. Besides, Pavlovian associative learning was success­ a random ion filament formation mechanism to achieve
fully mimicked on the TiS3-based artificial synapses [40]. good memristive properties. The RS mechanism based
Owing to the small thickness of 2D material, the 2D on band modulation instead of CFs could be imple­
materials based memristors normally have low endur­ mented to eliminate material degradation during SET/
ance (tens of cycles), poor performance (on/off ratio RESET process. A prototype MHM memristor based on
<10), and unstable operation. Hence, vdW heterojunc­ WS2/MoS2 heterojunction was successfully developed
tions have attracted attention due to their controllable, with a switching ratio of 104 and an endurance over 120
scalable, and programmed synthesis techniques switching cycles, showing that the performances of
[116,117]. The vdW heterojunctions with high quality memristor based on 2D WS2/MoS2 heterojunction are
heterojunction interfaces are formed by stacking differ­ superior to single MoS2 or WS2 layer memristors [111].
ent 2D materials together by vdW forces. The hetero­ In order to improve the electrical, magnetic and
junctions can be constructed by selecting different 2D optical properties of 2D materials, some researchers
materials, adjusting the energy band structure, control­ chased doping to modify 2D materials. Yan et al. fabri­
ling the transport of the electrons, and mildly modulat­ cated a 2D MXene Ti3C2-based memristor (Figure 4(e))
ing the resistive properties by voltage. Since no by silver nanoparticle doping [114]. Compared with the
Sci. Technol. Adv. Mater. 24 (2023) 8 Y. XIAO et al.

pure Ti3C2 device, the silver nanoparticle-doped Ti3C2 doping. Conventional doping methods such as ther­
memristor exhibited bidirectional continuous current mal diffusion and ion implantation cause damage to
transition behavior. This method not only overcome atomically thin 2D materials. The type of charge car­
the issue of poor electrical performance of conventional riers in 2D semiconductors strongly depends on the
memristors due to the dissolution of silver electrodes contacts and surrounding dielectrics, so the introduc­
under electric fields, but also solved the problem of tion of suitable metal contacts and dielectrics with
abrupt behavior of pure Ti3C2 devices. Meanwhile, the fixed charges may be an effective strategy to achieve
device realized basic decimal arithmetic operations, stable doping of 2D semiconductors. Furthermore, it
such as addition and multiplication [114]. is technically possible to tune the energy band struc­
Generally, 2D material-based memristors are pre­ ture and electronic properties of 2D materials by ion
pared on silicon wafers or polymer substrates, but intercalation and surface modification [129]. 4. Direct
these substrates are neither cheap nor biodegradable. deposition of metal electrodes on 2D materials usually
In contrast, cellulose paper is a good alternative due to breaks covalent bonds in the atomic lattice and intro­
its good biodegradability, low cost, recyclability, light duces defects, resulting in Fermi level pinning, form­
weight and mechanical flexibility. As demonstrated in ing Schottky contacts instead of the desired ohmic
Figure 4(f), Yalagala et al. prepared a paper-based contacts, and generating large contact resistance [129].
flexible memristor device for the first time, which
used a nanohybrid material consisting of a multilayer
3. Structure design
graphite/MoS2 with silver and copper as the top and
bottom electrodes, respectively. The memristor exhib­ In order to enhance the performance of memristors,
ited an endurance of 500 cycles, a switching ratio of some researchers adopted electrode engineering
~104, and outstanding flexibility [115]. method to modulate the performance [20,130–134].
However, 2D materials still face the following chal­ Ahn et al. prepared Nb/NiO/Nb memristors by pre­
lenges: 1. Large area preparation; 2. Instability in the paring well-aligned Nb nanopin array bottom electro­
ambient atmosphere, which leads to performance des, as shown in Figure 5(a). Due to the enhanced
degradation. Usually, h-BN and organic molecular electric field induced by the nanopin electrode,
films are used as vdW packaging layers, which signifi­ a lower SET/RESET voltage was observed. The nio­
cantly improve the device performance and its stability bium nanopin electrode minimized the dispersion of
in the ambient atmosphere. Oxide dielectric passiva­ the LRS and HRS currents and the variation of the
tion is feasible. Due to the suspended bondless surface SET/RESET voltages [20]. As illustrated in Figure 5(b),
of the 2D materials, additional seed layer growth and Huang et al. prepared a W/AlOx/Al2O3/Pt via-hole
oxidation processes are required, but these processes structured bilayer memristor. The devices consisted
may introduce impurities or defects to disrupt the of an oxygen-deficient AlOx layer and a near stoichio­
inherent material structure [129]. 3. Controllable metric ratio Al2O3 layer which were deposited by radio

Figure 5. (a) Schematic drawing of a NiO RRAM cell device composed of an Nb nanopin bottom electrode, a NiO thin film, and an
Nb top electrode. The thickness (d) of the NiO film is about 80 nm. Inset: A scanning electron microscopy image of the Nb nanopin
array. Reproduced with permission from [20], copyright 2018, Royal Society of Chemistry. (b) Schematics of the W/AlOx (RF)/Al2O3
(ALD)/Pt memristor with 250 nm via-hole structure. Reproduced with permission from [135], copyright 2020, IEEE. (c) The structure
of the W/MgO/SiO2/Mo memristor. Reproduced with permission from [15], copyright 2021, Wiley. (d) A schematic of a 1D1R
memory device. (e) IV characteristics of a 1D1R memory device. Reproduced with permission from [136], copyright 2010, Wiley. (f)
3D schematic of 1S1R integration crossbar arrays. The enlarged view is a vertical stacking cell of a selector and a memristor with
top, middle, and bottom electrodes. (g) Typical I–V curves of the integrated 1S1R cell, where the forming process and normal
operation cycles are included. Reproduced with permission from [137], copyright 2019, Wiley.
Sci. Technol. Adv. Mater. 24 (2023) 9 Y. XIAO et al.

frequency magnetron sputtering and atomic layer cross-array node has a self-rectifying effect, that is,
deposition, respectively. The devices exhibited high one resistor (1 R) structure [154]. All four solutions
speed (28 ns), high endurance (108 cycles at 100 K, are designed to achieve rectification characteristics,
1010 cycles at 298 K, and 107 cycles at 400 K), uniform which can suppress the interference of leakage cur­
resistance distribution, and large on/off ratio and good rents. Cho et al. connected Schottky-type Al/p-Si
retention [135]. Moreover, the multilayer structure diodes in series with an unipolar organic memories
can improve the performance of the memristor to form a 1D1 R crossover array (Figure 5(d)) [136].
[21,137–143]. Wu et al. fabricated a memristor of Reversible switching characteristics of the unipolar
W/MgO/SiO2/Mo structure with nonvolatile analog memristor were observed under forward bias condi­
switching characteristics, as demonstrated in tions, while under reverse bias conditions the switch­
Figure 5(c). The weight conductance of the memristor ing was significantly suppressed due to the
could be precisely adjusted and it was used to fabricate rectification characteristics of the diode (Figure 5(e))
a single-layer SNN for speech recognition with a high [136]. As shown in Figure 5(f,g), Sun et al. developed
recognition of 94%, which was equivalent to the accu­ a 1S1 R cell consisting of a Pt/Ta2O5−α/TaOβ/Pt
racy in software [15]. (PTTP) memristor and a novel Ag/TaOx/TaOy/TaOx
One of the advantages of memristors for next gen­ /Ag (ATTA) (x<y) selector. The all-TaOx-based inte­
eration non-volatile memories is their good scalability. grated 1S1 R cells were fabricated by magnetron sput­
For 3D integration of memristors, the crossbar array tering, and exhibits nonlinear characteristics avoiding
suffers from a crosstalk current issue, which can lead undesired crosstalk current [137]. Adam et al.
to misreading of information [144]. Normally, there reported a monolithic integrated 3D crossbar circuit
are four main solutions: 1. Each memory cell is con­ based on metal oxide memristors, which was suitable
nected to a diode to form a one diode one resistor for analog neuromorphic computing applications
(1D1 R) structure [145–147]; 2. Each memory cell at [155]. This circuit was based on Pt/Al2O3/TiO2-x
the cross-array node is connected to a transistor to /TiN/Pt memristors consisting of two 10 × 10 cross­
form a one transistor one resistor (1T1 R) structure bars, as demonstrated in Figure 6(a,b). The integrated
[141,148–151]; 3. Each memory cell is connected to crosspoint memristors were optimized for analog
a selector to form a one selector one resistor (1S1 R) computing applications, allowing the successful
structure [137,152,153]; 4. The memory cell at the switching of 200 devices in the crossbar circuits, and

Figure 6. (a) Scanning electron microscopy top-view image of the fabricated circuit with a zoom on a stacked Pt/Al2O3/TiO2–x/TiN/
Pt memristors to highlight the clean electrode edges. (b) Equivalent circuit for two Pt/Al2O3/TiO2–x/TiN/Pt memristors in the
stacked configuration, in particular, highlighting that the middle electrode (gray) is shared between bottom (red) and top (blue)
devices. The cross-section photograph showing the material layers and their corresponding thicknesses. Reproduced with
permission from [155], copyright 2017, IEEE. (c) (i) 3D monolithic integrated memristor circuits. Schematic of the 3D circuits
composed of high-density staircase output electrodes (blue) and pillar input electrodes (red). (ii) Sideview of 3D row banks. Each
row bank in the 3D array operates independently. (iii) Sideview from column side showing unique staircase electrodes.
Reproduced with permission from [22], copyright 2020, Springer Nature. (d) An optical image of a wafer with transistor arrays.
(e) Close-up of chip image showing arrays of various sizes. (f) Microscope image showing the 1T1 R structureof the cell. Scale bar,
10 µm. (g) Cross-sectional scanning electron microscopic image of an individual 1T1R cell, which is cut in a focused ion beam
microscope from the dashed line in (g). Scale bar, 2 µm. (h) Cross-sectional transmission electron microscopic image of the
integrated Ta/HfO2/Pt memristor. Scale bar, 2 nm. Reproduced with permission from [141], copyright 2018, Springer Nature.
Sci. Technol. Adv. Mater. 24 (2023) 10 Y. XIAO et al.

most importantly, the conductivity of the device could this review, the performances of memristors are intro­
be precisely adjusted within the dynamic operating duced from the following aspects: RS performance,
range [155]. As shown in Figure 6(c), Lin et al. synaptic plasticity, and optical-electrical performance.
reported a 3D circuit consisting of eight layers of
monolithic integrated memristor devices, which
implemented a CNN function by programming paral­ 4.1 RS performance
lel operation kernels into a 3D array and achieved
With a decade of research, the RS performances of the
software-comparable accuracy (98.1%) in recognizing
memristors have been significantly improved. The on-off
handwritten digits in a MNIST [22]. Li et al. integrated
a Ta/HfO2/Pt memristor with a foundry-made tran­ ratio is defined as the ratio between the currents in the
sistor array on a 6-inch wafer. Each memristor was high and low resistance states of the memristor. As shown
connected to a series transistor in a ‘1T1 R’ configura­ in Figure 7(a), Lu et al. fabricated a Ag/TiN/HfOx/HfOy
tion (Figure 6(d–h) show the integrated memristor /HfOx/Pt memristor, and it had an ultra-high switching
array from the wafer scale to the nanometer scale) ratio of 1010 [162]. We have fabricated a Pt/BFO/HfO2
[141]. The device of 1T1 R structure showed robust, /TiN device by inserting a 2 nm BiFeO3 layer, achieving
linear, and symmetric synaptic weight updates. As a pulse endurance of 108 cycles (Figure 7(b)) [61]. Lee et al.
a result, a multilayer neural network implemented in demonstrated an TaOx-based memristor with an endur­
a 128 × 64 memristor array was trained on the MNIST ance of more than 1012 cycles [70].
training set of 80,000 images, after which an accuracy Power consumption is also a key parameter in
of 91.71% was achieved on the complete 10,000 images memristors, especially in order to solve the power
test set [141]. wall problem [166]. Wang et al. fabricated 2D MXene
Ti3C2-based memristors and the power consumption
was 0.35 pJ/bit, as illustrated in Figure 7(c) [114]. Fu
4. Device performance et al. fabricated a forming-free V/VOx/HfWOx/Pt
Resistive materials and device structures can decide the memristor and power consumptions for writing and
performances of the memristors. For non-volatile mem­ erasing are 0.06 pJ/bit and 0.9 pJ/bit, respectively
ory, we often focus on the on/off ratio [156,157], endur­ [26]. Wang et al. proposed an ultra-low power Ta2
ance [70,158], retention [159] and switching speed O5/Al2O3 bilayer-gate-dielectric synaptic transistor
[160,161]. For neuromorphic applications of memristor, and it showed a low power consumption of 19.9 aJ/
the conductance linearity of the device is important. In bit [167].

Figure 7. (a) Repeatable 100 cycles of Ag/TiN/HfOx/HfOy/HfOx/TiN/Ag device under an ICC of 3 mA, indicating an extremely high
ON/OFF ratio of over 1010. Reproduced with permission from [162], copyright 2022, Wiley. (b) Endurances of HfO2-based
memristors before and after inserting the BFO layer. Reproduced with permission from [61], copyright 2020, Wiley. (c) The
power consumption variation of Ti3C2 and Ti3C2: Ag devices. Reproduced with permission from [114], copyright 2021, Elsevier. (d)
Cumulative probability distribution of 1,024 cells with respect to 32 independent conductance states. Reproduced with permission
from [163], copyright 2020, Springer Nature. (e) Switching characteristics of TiN/HfO2/ITO device for 10 switching cycles. The inset
shows a schematic diagram of the device. Reproduced with permission from [164], copyright 2014, IOP. (f) The DC I-V characteristic
of the flexible Pt/Hf17.66Ti13.79O68.55/ITO selector device in 100 voltage sweep cycles. Reproduced with permission from [165],
copyright 2019, Royal Society of Chemistry.
Sci. Technol. Adv. Mater. 24 (2023) 11 Y. XIAO et al.

Table 1. Comparison of different types of memristors.

Set voltage Reset voltage Retention Switching Endurance Operation
Configuration Structure (V) (V) (s) ratio (cycles) speed Ref.
Oxide-RRAM Pt/Ta2O5-x/TaO2-x/Pt −1 2 – >10 1012 10 ns [70]
Ag/TiN/HfOx/HfOy/HfOx/Pt – – – 1010 106 60 ns [71]
Pt/SiOx:Pt/Ta ~-0.6 1–1.2 107 103 3 × 107 <100 ps [53]
Perovskite- ITO/Ag/MAPbI3/Au 2.4 −2.2 4.2 × 107 ~107 ~106 100 ps [17]
RRAM
ITO/Cs2AgBiBr6/Au 1.53 −3.4 105 >102 103 - [94]
Organic-RRAM Ag/SU-8Ag/Pt <0.3 <0.7 2 × 103 ~106 102 - [95]
Ag/2DPBTA+PDA/ITO 0.90 −1 8 × 104 ~105 2 × 102 - [101]
Ag/FK-800/Pt – – – >102 >106 340 μs [102]
Ag/DNA/AgNP/Pt 0.3 −0.2 105 106 1000 20 ns [106]
2D-RRAM ITO/PVA-GO/ITO −0.2 0.2 104 >10 5 × 102 - [112]
Ag/InSe/Ag 0.3 −0.7 3.5 × 104 4.5 × 103 3 × 102 - [113]
Pt/h-BN/Ag 0.3 −0.1 – 108 107 50 ns [127]

Multi-level property may be concerned for mem­ fluctuation over time if no bias is applied (preferably
ristors. Yao et al. implemented CNNs based on high- less than 10% over >10 years) [176]. As shown in Table
yield, high-performance and uniform memristor 1, which presents a comparison of the Set/Reset voltage,
crossbar arrays. The CNNs integrated eight 2,048-cell retention, switching ratio, endurance, operation speed
memristor arrays to improve parallel-computing effi­ for different types of memristors. For the oxide based
ciency [163]. Figure 7(d) shows the distribution of device, the memristor with the structure of
1024 memristors with 32 different conductance states, Ag/TiN/HfOx/HfOy/HfOx/Pt has the ultra-high on/off
where all curves are separated without any overlap. ratio >1010 and sub-ns switching speeds [71], but it has
This system with a five-layer memristor-based CNN some disadvantages, such as high power consumption
achieved a high accuracy of over 96% to perform or low uniformity. In order to solve the problem, dop­
MNIST image recognition [163]. ing in the switching layer [177], design double layer
The switching speed reflects the ability of the mem­ structure [178,179] have been taken by some academics.
ory to achieve fast operation, and the high switching For Perovskites based device, it can be seen that the
speed shows the great potential of the memristors as memristor with the structure of ITO/Ag/MAPbI3/Au
high-speed memory. Poddar et al. developed has the operation speed of 100 ps and the on/off ratio of
a memristor with a high density of monocrystalline 107 [17]. But the environmental stability needs to be
perovskite quantum wires as the switching matrix, improved. Some scholars have improved the environ­
showing a switching ratio of ~107 and an ultrafast mental stability of the memristors based on perovskites
switching speed of ~ 100 ps [17]. Besides the above by new materials of functional layers, such as lead-free
properties, self-compliance of current has been bicalcitonite Cs2AgBiBr6 [94]. For organic memristors,
reported, which is used to simplify the peripheral the memristor with the structure of Ag/SU-8Ag/Pt has
circuits. As demonstrated in Figure 7(e), our research low power consumption of 1.5 fW and the switch ratio
group fabricated a memristor with TiN/ITO structure, of 106 [95]. To improve the endurance of the memris­
which demonstrated a self-compliance current effect tors based on organics, organic-inorganic bilayer stack­
for the ITO electrode [164]. Another novel character­ ing has been taken [105]. For 2D based memristors, low
istic of memristor is self-rectifying effect, which can endurance (tens of cycles), poor performance (on/off
effectively suppress the crosstalk current and signifi­ ratio <10) are obstacles to the future applications. The
cantly simplify the complexity of circuit [168–175]. preparation technology of the functional layer is impor­
Besides, selector is often combined with memristor tant. It can be seen that the memristor with the structure
for 3D integration. As illustrated in Figure 7(f), we of Pt/h-BN/Ag has the low Set/Reset voltage of 0.3/−0.1
fabricated a Pt/HfTiO/ITO selector on a flexible sub­ V, switch ratio of 108, endurance of 107 cycles and
strate polyvinyl naphthalate (PEN), which exhibited operation speed of 50 ns [127].
excellent bending reliability in the bending radius
range of 50 mm to 30 mm without degradation of the
operating performance. This selector device could be
4.2 Synaptic plasticity
switched on within 60 ns at 4.4 V and turned off
within 50 ns under a 0.5 V bias [165]. Memristors have become an ideal candidate for artificial
Commercial non-volatile memories must meet synaptic devices in neuromorphic systems. To simulate
requirements for integration with current ICs. Among synapses, the memristor exhibit a simulated switching
these requirements are integration density up to 1 giga­ behavior that requires a gradual SET and RESET pro­
byte (GB)/mm2, writing voltage <3 V, power consump­ cesses rather than an abrupt one [180]. Recently, some
tion <10 pJ, switching speed <10 ns, endurance >1010 research groups have investigated synaptic properties
cycles, HRS/LRS resistance ratio >10, and low resistance using materials based on ion migration or
Sci. Technol. Adv. Mater. 24 (2023) 12 Y. XIAO et al.

Figure 8. (a) EPSC triggered by positive gate voltage pulses with different widths. Insert: schematic of organic field-effect transistor
[191]. (b) The learning experience behaviors of Cu/SiC/W memristor for the learning/forgetting process [192]. (c) The repeated LTP
and LTD characteristics of the device demonstrating good stability with minimal cycle-to cycle variations. (d) PPF index obtained
by applying paired pulses [193]. (e) Asymmetric Hebbian STDP rule obtained in HfO2/BFO/HfO2 memristor [16].

electrochemical reduction reactions, such as TiOx [181], operating voltage, and long retention time. As demon­
HfOx [182], TaOx [183,184], ZnOx [57,60], organic [185– strated in Figure 9(b), during illumination ranging from
187] et al. Qiao et al prepared a memristor base onC12- ultraviolet (380 nm) to near infrared (785 nm), the
BTBT. As shown in Figure 8(a), a positive voltage pulse Schottky barrier height is elevated further so that the
with the assistance of UV light irradiation (1.35 resetting voltages and power consumption decrease
mW cm−2) was applied to the memristor to trigger the [23]. The conductance of most optoelectronic synaptic
excitatory postsynaptic currents (EPSC) [188]. Kapur devices could only be reversibly tuned by a combination
et al. fabricated a memristor based on silicon carbide of optical and electrical signals [193–195]. For an ideal
(SiC) and the conductance could be modulated gradually. optoelectronic neuromorphic device, its weight is repre­
Moreover, the synaptic function of learning-forgetting- sented by the conductance, which should be all optically
relearning processe was successfully emulated and tunable. Hu et al. implemented an all-optical control
demonstrated using a 3 × 3 array of SiC-based memris­ analog memristor based on InGaZnO switching layer
tors (Figure 8(b)) [189]. Ali et al. fabricated memristors (Figure 9(c)) [191]. As illustrated in Figure 9(d), the
with Ag/GeSe/Pt/Ti/SiO2 structure and simulated synap­ membrane conductance of this memristor was reversibly
tic functions, such as long time range enhancement tunable in a continuous range only by changing the
(LTP), long time range inhibition (LTD) (Figure 8(c)), wavelength of the incident light. The photoinduced
paired pulse facilitation (PPF) (Figure 8(d)) and others membrane conductance state was non-volatile. This
[190]. Our group prepared a Pt/HfO2/BFO/HfO2/TiN device simulated STDP by all-optical modulation [191].
artificial synapse, realizing the physiological synapse Moreover, some scholars have applied memristors with
STDP characteristics (Figure 8(e)) [16]. Furthermore, good opto-electronic properties for the neural network of
the memristors simulated other synaptic functions, such the next generation. Seo et al. demonstrated an optical
as short-term potentiation, post-tetanic potentiation, neurosynaptic device by implementing both synaptic
nonassociative learning, associative learning, synaptic and optical sensing functions on h-BN/WSe2 heteroge­
scaling and spike-rate-dependent plasticity [189]. neous structures (Figure 9(e)) [192]. The device simu­
lated the human visual system to recognize color and
color mixed patterns in an optical neural network
4.3 Optical-electrical performance
(ONN). The system achieved a recognition rate of over
Nowadays, photoelectric memristors attracted much 90% for color pattern recognition tasks by this ONN
attention for the excellent optical-electrical performance. (Figure 9(f)) [192]. Some academics have focused on
As shown in Figure 9(a), Zhou et al. from our group piezoelectrical and thermoelectrical performance of
prepared a optoelectronic memristor based on BP memristors. Liu et al. proposed a multilevel memristor
nanosheets (BP@PS NSs) coated with polystyrene (PS) based on bamboo-like GaN microwires. The piezotronic
[23]. With the aid of PS, the BP@PS-based memristor has effect was introduced to modulate the SET voltage and
good RS characteristics such as no initial preforming, low the general performance of the memristor [177].
Sci. Technol. Adv. Mater. 24 (2023) 13 Y. XIAO et al.

Figure 9. (a) Schematic showing light modulation of the BP@PS memristor. (b) I-V curves of the BP@PS memristor modulated by
different wavelengths. Reproduced with permission from [23], copyright 2020, American Chemical Society. (c) Schematic of the
realization of an AOC memristor. MC and NIR denote the memconductance and near-infrared, respectively. (d) Reversible
regulation of the memconductance by means of 100 blue light pulses (D = 1 s, I = 1 s, and P = 20 µw cm−2) and 100 NIR light
pulses (D = 1 s, I = 1 s, and P = 24 µw cm−2). Reproduced with permission from [191], copyright 2021, Wiley. (e) Integration of the
h-BN/WSe2 optic-neural synaptic device. A Schematic of the human optic nerve system, the h-BN/WSe2 synaptic device integrated
with h-BN/WSe2 photodetector, and the simplified electrical circuit for the ONS device. Here, the light sources were dot lasers with
wavelengths of 655 nm (red), 532 nm (green), and 405 nm (blue) with a fixed power density (P) of 6 mW cm−2 for all wavelengths.
(f) Left: Developed ONN for recognition of 28 × 28 RGB-colored images. Right: Recognition rate as a function of number of training
epochs for ONN and conventional NN. Reproduced with permission from [192], copyright 2018, copyright 2021, Springer Nature.

Lappalainen et al. fabricated a thermal-electric VO2 after power failure. Digital logic operations based on
memristor. A basic thermoelectric logic circuit of the non-volatile memristors can solve this problem, because
memristor was demonstrated to be used to simulate the its logic output is the resistance of the device. The mem­
operation of synapses between neurons [178]. ristors array in the chip can perform distributed parallel
computing, which will greatly improve the computa­
tional efficiency [201]. Mao et al. reported a damage-
5. Applications
free Au/h-BN/Au memristor using a clean, water assisted
In-memory computing based memristors are normally metal transfer approach by physically assembling Au
divided into logic operations using abrupt memristors, electrodes onto the layered h-BN. As shown in
and analog calculations using gradual memristors. The Figure 10(a,b), the memory arrays fabricated by vdW
abrupt memristor has an abrupt transition between metal integration technology and direct metal evapora­
high and low resistance states, which is used as the tion. By using the memristors as logic gates and latches,
logic input and output to realize non-volatile informa­ simultaneous data storage and ‘stateful’ implication logic
tion processing. Gradual memristor can continuously (IMP) is achieved in a longitudinal array with suppressed
change its resistance during SET and RESET processes, potential path currents (Figure 10(c–e)) [202].
which is suitable for building efficient ANN [14]. The Krishnaprasad et al. demonstrated low variability
cycle-to-cycle variation of the memristors [179] and synapses using chemical vapor deposited 2D MoS2 as
linearity [196] are critical factors to build high compu­ a switching medium with Ti/Au electrodes [204]. Logical
tational accuracy for ANN. In order to obtain the gra­ operations of AND, OR, and NOT were implemented by
dual resistances of memristors, the following methods monolithic integration of MoS2 synapses with MoS2
can be used to fabricate the memristor such as doping in leaky integrate-and-fire neurons [204]. Yuan et al.
the switching layer [197], design double layer structure demonstrated an efficient logic method based on
[198,199], electrode engineering [200], etc. Pt/Ta/Ta2O5/Pt/Ti memristors, where the two input
variables were represented by different physical quanti­
ties, implementing all 16 Boolean logic functions in less
5.1 Logic operation
than 3 steps using single device [205].
The conventional CMOS digital logic operations are With the advent of the era of big data, the security
usually based on complementary transistors. This tran­ of personal information and hardware is becoming
sistor-based logic circuit cannot maintain the logic state more and more important. The physical unclonable
Sci. Technol. Adv. Mater. 24 (2023) 14 Y. XIAO et al.

Figure 10. (a) Au/h-BN/Au memristors array and IMP logic implementation. The color map plots of the resistance distribution of the 8
× 8 memory array fabricated via (a) Direct metal evaporation and (b) Metal transfer technique. (c) Schematic of the IMP logic circuit.
VSET and VCOND are simultaneously applied to memristors Q and P, respectively. RG is chosen to be 500 Ω. (d) Experimental results of
IMP operations for four input conditions. p and q are the resistance states of device P and Q, respectively. “1” refers to low-resistance
state and “0” refers to high-resistance state. p′ and q′ are the states of device P and Q, respectively after the logic operations. Variable
p remains unchanged because VCOND is lower than VSET, so p and p′ should have the same value. The readout voltage is 0.1 V. (e) True
table of the logic operation. Reproduced with permission from [202], copyright 2022, Wiley. (f) Photograph of the PUF system. ZC706
FPGA evaluation board from Xilinx is used. The chip contains an 8-kb memristor array and supports parallel operation for up to eight
memristor devices. (g) The change in CC between binary PUF data (top left inset) and analog resistance values measured in secure
mode (top right inset). (h) The change of prediction accuracy with incremental training epoch for traditional NVM PUF, whose data
are steadily stored by distinct high and low conductance states, and the developed concealable PUF. Reproduced with permission
from [203], copyright 2022, American Association for the Advancement of Science.

function (PUF) is a promising security primitive that subthreshold slope variation of a transistor could be
uses the random variations inherent in electronic used as the entropy source of a PUF to generate
hardware to generate digital keys [206]. As demon­ a physical key [207]. By combining this subthreshold
strated in Figure 10(f), Gao et al. certified a hideable slope PUF with a memristor-based XOR logic func­
PUF at the chip level by integrating an array of mem­ tion, an in situ encryption/decryption scheme in
ristors [203]. The switching characteristics of HfOx- a compact 1T1 R structure was proposed.
based memristors were used to efficiently implement Experiments demonstrated that the subthreshold
PUF hiding/recovery via SET/RESET operations. slope PUF had good reproducibility, uniqueness and
A PUF recovery with zero-bit error rate and significant uniformity. Encryption and decryption of three 16-bit
attack resistance was achieved. This hideable feature, binary sequences were successfully implemented in
coupled with the inherent noise in the memristor a 1T1 R device using a PUF key of subthreshold
array, allowed the PUF to effectively resist both inva­ slope [207].
sive and noninvasive attacks which were the main Encryption and decryption of information is another
threats to modern hardware security. As illustrated in way to ensure information security. Song et al. proposed
Figure 10(g), an attacker executes an attack in an optimized XOR logic gate based on TiN/Ti/HfOx
a simulated manner by analyzing the correlation /TiN memristor [208]. The encryption and decryption
between valid PUF data and conductance distribution based on the XOR circuit consisting of two memristors
in secure mode. In contrast, the conductance obtained was verified by successfully performing parallel electri­
from the secure mode via microprobes was disordered cal tests [208]. Furthermore, the true random number
and had a correlation coefficient of less than 0.4 over generator (TRNG) is critical for cryptographic applica­
250 cycles. Multilayer fully connected perceptron tions. The advantages of TRNG based on memristors
could be used to perform such attacks on the conceal­ are rich and easily extractable randomness, low cost,
able PUFs, as shown in Figure 10(h), where the trained high integration density, low operational power con­
neural network predicted the accuracy of 70% from the sumption, and fast response time [209,210]. Predictable
recovered PUF data, which is insufficient for breaking random number generators (RNGs) open the door to
the PUF [203]. Yang et al. demonstrated the the attacks that can hack the devices and compromise
Sci. Technol. Adv. Mater. 24 (2023) 15 Y. XIAO et al.

Figure 11. (a) (i) Artificial somatosensory system consisting of an MFSN array and an SNN classifier to emulate tactile perception.
(ii) the MFSN unit is composed of a pressure sensor and a NbOx-based memristor. The output frequency of the MFSN unit is
affected by the pressure intensity and temperature, while the output amplitude of spikes depends only on the temperature. (b)
Schematic of an SNN comprising 400 input neurons, 50 hidden neurons, and 8 output neurons for the simulation. (c) Evolution of
the training accuracy with training epoch at three different modes (pressure, temperature, and multimodal mode). (d) Confusion
matrix of simulation classification output versus the expected outputs. The result shows the good classification ability of the
MFSN-based system. Reproduced with permission from [212], copyright 2022, Wiley. (e) Three different types of 3 × 3 kernel
operation (edge detection, sharpen and soften) are performed. (f) Corrupted letter images are generated by adding Gaussian noise
(δ = 0.5) to images from the eye layer. (g) Letter images are denoised by inserting a denoising layer that includes memristor
crossbar arrays, as described in the block diagram. Reproduced with permission from [221], copyright 2022, Springer Nature.

the data. TRNG must be unpredictable, statistically As shown in Figure 11(a), Zhu et al. reported an
independent, and uniformly distributed. Kim et al. gen­ artificial multimodal sensory system consisting of an
erated real random numbers using the random oscilla­ array of multimodal fusion spiking neurons (MFSN)
tion behavior of NbOx-mott memristors, which operating in the spiking domain and an SNN classifier
exhibited self-clocking, fast, and variable tolerance for processing multimodal sensory input without los­
[211]. The random number generation rate of the ing unimodal information (Figure 11(b)). This MFSN
device could be at least 40 kb/s which was the fastest perceived multimodal sensory information and con­
record compared to previous volatile memristor-based verted it to spikes without using a conversion module,
TRNG devices [211]. reducing the complexity of circuit and power con­
sumption. By feeding the spikes into the SNN classi­
fier, the results shown that patterns with multiple
5.2 Analog calculations
modes endowed a higher recognition rate (93%) to
Analog calculations can be divided into the following two identifying cup features than only temperature
aspects: 1. Artificial sensory system. To mimicking bio­ (72.5%) or pressure (67%) mode, as demonstrated in
logical sensing systems, sensor devices are further inte­ Figure 11(c). The confusion matrix (Figure 11(d))
grated with the memristor storage and computing unit, shown the simulated classification output compared
integrating sensing, computing, storage to build to the expected output (the color bars show the counts
a sensing-memory-computing integrated processing of the output neurons), demonstrating the excellent
unit [212]. 2. Neuromorphic brain-like computing. The classification capability of the MFSN-based sys­
conductance changes in most memristors originate from tem [212].
ionic motion, which are very similar to the processes Besides, memristors are expected to simulate the
involving neurons and synapses in the brain. Therefore, reconstruction of biological neural networks and rea­
memristor is an important component for building arti­ lize neuromorphic brain-like computing. Choi et al.
ficial neural networks for future applications such as reported stackable heterogenous integrated chips,
computer vision [126,192,194,213–215], speech recogni­ which used optoelectronic device arrays for chip-to-
tion [216,217], driverless vehicles [218], robotics [219], chip communication and neuromorphic cores based
finance [220], etc. on memristor crossbar array for highly parallel data
Sci. Technol. Adv. Mater. 24 (2023) 16 Y. XIAO et al.

Figure 12. Design methodology and main features of the NeuRRAM chip. (a) Cross-layer co-optimizations across the full stack of
the design enable NeuRRAM to simultaneously deliver high versatility, computational efficiency and software-comparable
inference accuracy. (b) Micrograph of the NeuRRAM chip. (c) Reconfigurability in various aspects of the design enables
NeuRRAM to implement diverse AI models for a wide variety of applications. (d) Comparison of EDP, a commonly used energy-
efficiency and performance metric among recent RRAM-based CIM hardware. (e) Fully hardware-measured inference accuracy on
NeuRRAM is comparable to software models quantized to 4-bit weights across various AI benchmarks. Reproduced with
permission from [226], copyright 2022, Springer Nature.

processing. They created a system with stackable and system which can efficiently process spatiotem­
replaceable chips and successfully executed three 3 × 3 poral signals in real time [223]. For a RRAM
core operations (vertical edge detection, sharpening compute-in-memory (CIM) chip to be widely
and softening) for image processing (Figure 11(e)). adopted in real-world AI applications, it needs to
To test the denoising ability of the system, the authors simultaneously provide high energy efficiency,
generated corrupted letter images by adding Gaussian flexibility to support different AI model architec­
noise (δ = 0.5) to the image from the eye layer tures and software-comparable inference accuracy.
(Figure 11(f)). As shown in Figure 11(g), the letter Hitherto, there has not been a study aimed at
images were then denoised by inserting a denoising simultaneously improving all these three aspects
layer including a memristor crossbar array. The results of a design [224,225]. Through collaborative opti­
showed that the stackable and replaceable chips exhib­ mization from algorithms and architectures to cir­
ited excellent immunity to high noise level [221]. cuits and devices (Figure 12(a)), as show in
Zhong et al. reported a dynamic memristor-based Figure 12(b), Wan et al. reported a NeuRRAM –
parallel reservoir computing system. The system a RRAM-based CIM chip [226]. The chips with
achieved a low word error rate of 0.4% in speech reconfiguring CIM cores were compatible with
digit recognition and a low normalized root-mean- different model architectures and their energy effi­
square error of 0.046 in time series prediction of ciency was two-times better than previous state-of
Hénon graphs. This result outperformed most existing -the-art RRAM-CIM chips (Figure 12(c,d)). Fully
hardware-based reservoir computing systems and hardware-measured inference accuracy on
software-based reservoir computing systems for the NeuRRAM was comparable to software models
Hénon graph prediction task [217]. quantized to 4-bit weights across a variety of AI
Liang et al. developed a hardware prototype for tasks (Figure 12(e)), including accuracy of 99% on
near-sensor computing, chaotic time-series predic­ MNIST and 85.7% on CIFAR-10 image classifica­
tion and handwriting classification [222]. Zhong tion, 84.7% on Google speech command recogni­
et al. reported a fully analogue reservoir computing tion [226].
Sci. Technol. Adv. Mater. 24 (2023) 17 Y. XIAO et al.

6. Challenges and prospects National Natural Science Foundation of China (62274058),

Strategic Priority Research Program of the Chinese Academy
This review discussed various RS materials for memris­ of Sciences under Grant No. XDB44000000, Hubei Province
tors, including electrodes, binary oxides, perovskite, Key Research and Development Program under Grant
organics, and 2D materials. Besides, the role in the No. 2022BAA020, Open Project of China-Poland Belt and
memristor were systematically discussed. The construc­ Road Joint Laboratory of Measurement and Control
Technology (MCT202104).
tion of the shaped electrode, the design of functional
layer and other factors influencing the device perfor­
mance are analyzed. The excellent and advanced perfor­
mances of the memristor from the frontier researches are Disclosure statement
illustrated. The applications of memristors in logic and No potential conflict of interest was reported by the authors.
analog are presented. The prospects and challenges of
memristors are highlighted, providing further develop­
ments and new insights into in-memory computing
Funding
based on memristors. Despite decades of intensive
research, memristors have been greatly developed. This work was supported by the National Natural Science
However, on the road to commercialization of memris­ Foundation of China [62274058], Strategic Priority
tors, there are still many issues to be solved. Research Program of the Chinese Academy of Sciences
under Grant No. XDB44000000, Hubei Province Key
Even for the same material with sandwich struc­ Research and Development Program under Grant
ture, different research groups may propose different No. 2022BAA020, Open Project of China-Poland Belt and
resistive mechanisms, and this phenomenon may be Road Joint Laboratory of Measurement and Control
caused by differences in the manufacturing process. Technology [MCT202104].
More systematic work is needed to elucidate the resis­
tive mechanism and failure mechanism of memristors
in the future. Currently, except for HfOx and TaOx, Notes on contributors
the vast majority of RS materials are not yet compa­
Yongyue Xiao is currently a Ph.D. can­
tible with CMOS processes, which limits the practical
didate in Faculty of Physics and
applications. Therefore, the integration technology Electronic Science at Hubei
should support conventional CMOS compatible mate­ University. His research interest
rials and CMOS-incompatible materials for a wider focuses on binary oxide based memris­
range of applications in the future. tor and neuromorphic computing
The conductance linearity, power consumption, applications.
device-to-device (spatial) and cycle-to-cycle (temporal)
characteristics of most memristors are not ideal, so
device performances are needed to be further boost up. Bei Jiang is currently a lecture at Hubei
In addition, the human sensory system is a fusion of University. She received her Ph.D. degree
multiple senses. Therefore, developing device systems in Physics from Wuhan University. Her
research interests are focused on electro­
with multi-sensory fusion and diverse processing cap­ nic and optoelectronic devices based on
abilities is a promising direction for future applications. 2D materials.
As for system-level optimization, dedicated algorithms
and peripheral circuits for memristor-based neuro­
morphic computing should be developed in order to Zihao Zhang is currently a master stu­
fully exploit the performance of particular devices. dent in faculty of physics and electronic
science at Hubei University. His
In the future, brain-like applications based on mem­ research interest focuses on oxide-
ristors need to be further expanded, such as the use of based memristor for synaptic plasticity
bionic brain-like computing devices to build brain- and artificial nociceptor.
computer interface, which can provide new research
ideas for medical electronics, biomedicine, and other
fields. Based on the functional characteristics of the
biosensory nervous system, an efficient and intelligent
information perception system with biological reality Cong Ye received her Ph.D. in material
physics from the Institute of Solid State
can be established to create smarter bio-inspired robots. Physics, Chinese Academy of Sciences.
She is currently a professor at Hubei
University, China. Her major research
Acknowledgments interests include semiconductor device,
resistive random access memory, neu­
Yongyue Xiao, Bei Jiang and Zihao Zhang contributed equally romorphic device, two-dimensional
to this work. The authors acknowledge the grants from material and device, first-principle computing.
Sci. Technol. Adv. Mater. 24 (2023) 18 Y. XIAO et al.

References neuromorphic computing. ACS Appl Mater

Interfaces. 2021;13(44):52822–52832. DOI:10.1021/
[1] Neumann Jv. The principles of large-scale computing acsami.1c18329
machines. Ann History Comput. 1988;10(4):243–256. [20] Ahn Y, Shin HW, Lee TH, et al. Effects of a Nb
[2] Wulf WA, McKee SA. Hitting the memory wall: nanopin electrode on the resistive random-access
implications of the obvious. SIGARCH Comput memory switching characteristics of NiO thin films.
Archit News. 1995;23(1):20–24. Nanoscale. 2018;10(28):13443–13448. DOI:10.1039/
[3] Backus J. Can programming be liberated from the C8NR02986E
von Neumann style? A functional style and its [21] Ismail M, Mahata C, Kim S. Forming-free Pt/Al2O3
algebra of programs. Commun ACM. 1978;21 /HfO2/HfAlOx/TiN memristor with controllable mul­
(8):613–641. tilevel resistive switching and neuromorphic charac­
[4] Horowitz M. 1.1 Computing’s energy problem (and teristics for artificial synapse. J Alloys Compd.
what we can do about it). 2014 IEEE International 2022;892:162141.
Solid-State Circuits Conference Digest of Technical [22] Lin P, Li C, Wang Z, et al. Three-dimensional mem­
Papers (ISSCC); San Francisco: IEEE; 2014. p. 10–14. ristor circuits as complex neural networks. Nat
[5] Schaller RR. Moore’s law: past, present and future. Electron. 2020;3(4):225–232. DOI:10.1038/s41928-
IEEE Spectr. 1997;34(6):52–59. 020-0397-9
[6] Mutlu O, Ghose S, Gómez-Luna J, et al. Processing [23] Zhou Y, Liu D, Wang J, et al. Black phosphorus based
data where it makes sense: enabling in-memory multicolor light-modulated transparent memristor
computation. Microprocess Microsyst. 2019;67:28–41. with enhanced resistive switching performance. ACS
[7] Merolla PA, Arthur JV, Alvarez-Icaza R, et al. Appl Mater Interfaces. 2020;12(22):25108–25114.
A million spiking-neuron integrated circuit with DOI:10.1021/acsami.0c04493
a scalable communication network and interface. [24] Chen Z, Huang W, Zhao W, et al. Nonvolatile mem­
Science. 2014;345(6197):668–673. DOI:10.1126/ ory: ultrafast multilevel switching in Au/YIG/n-Si
science.1254642 RRAM (Adv. Electron. Mater. 2/2019). Adv Electron
[8] Indiveri G, Liu SC. Memory and information proces­ Mater. 2019;5(2):1970008. DOI:10.1002/aelm.
sing in neuromorphic systems. Proc IEEE. 2015;103 201970008
(8):1379–1397. [25] Shin HS, Song H, Kim DH, et al. Multimode synaptic
[9] Stone HS. A Logic-in-memory computer. IEEE Trans operation of a HfAlOx-based memristor as
Comput. 1970;C-19(1):73–78. a metaplastic device for neuromorphic applications.
[10] Kautz WH. Cellular logic-in-memory arrays. IEEE ACS Appl Electron Mater. 2022;4(8):3786–3793.
Trans Comput. 1969;C-18(8):719–727. DOI:10.1021/acsaelm.2c00320
[11] Chua L. Memristor-The missing circuit element. [26] Fu Y, Zhou Y, Huang X, et al. Reconfigurable synaptic
IEEE Trans Circuit Theory. 1971;18(5):507–519. and neuronal functions in a V/VOx/HfWOx/Pt mem­
[12] Strukov DB, Snider GS, Stewart DR, et al. The missing ristor for nonpolar spiking convolutional neural
memristor found. Nature. 2008;453(7191):80–83. network. Adv Funct Mater. 2022;32(23):2270130.
DOI:10.1038/nature06932 DOI:10.1002/adfm.202270130
[13] Pei J, Deng L, Song S, et al. Towards artificial general [27] Yan X, Zhao J, Liu S, et al. Memristor with Ag-cluster-
intelligence with hybrid Tianjic chip architecture. doped TiO2 films as artificial synapse for neuroin­
Nature. 2019;572(7767):106–111. DOI:10.1038/ spired computing. Adv Funct Mater. 2018;28
s41586-019-1424-8 (1):1705320. DOI:10.1002/adfm.201705320
[14] Li S, Pam M-E, Li Y, et al. Wafer-scale 2D hafnium [28] Zhao M, Sun Y, Yan L, et al. Memristor with BiVO4
diselenide based memristor crossbar array for nanoparticle as artificial synapse for neuroinspired
energy-efficient neural network hardware. Adv computing. Appl Phys Lett. 2022;120(9):093501.
Mater. 2022;34(25):2103376. DOI:10.1002/adma. DOI:10.1063/5.0079418
202103376 [29] Liu Y, Wu L, Liu Q, et al. Topochemical synthesis of
[15] Wu X, Dang B, Wang H, et al. Spike-enabled audio copper phosphide nanoribbons for flexible optoelec­
learning in multilevel synaptic memristor array-based tronic memristors. Adv Funct Mater. 2021;32
spiking neural network. Adv Intell Syst. 2022;4 (14):2110900. DOI:10.1002/adfm.202110900
(3):2100151. DOI:10.1002/aisy.202100151 [30] Xu R, Jang H, Lee MH, et al. Vertical MoS2
[16] Peng ZH, Wu FC, Jiang L, et al. HfO2-based memris­ double-layer memristor with electrochemical metalli­
tor as an artificial synapse for neuromorphic comput­ zation as an atomic-scale synapse with switching
ing with tri-layer HfO2/BiFeO3/HfO2 design. Adv thresholds approaching 100 mV. Nano Lett. 2019;19
Funct Mater. 2021;31(48):2107131. DOI:10.1002/ (4):2411–2417. DOI:10.1021/acs.nanolett.8b05140
adfm.202107131 [31] Tsai SC, Lo HY, Huang CY, et al. Structural analysis
[17] Poddar S, Zhang Y, Gu L, et al. Down-scalable and and performance in a dual-mechanism conductive
ultra-fast memristors with ultra-high density filament memristor. Adv Electron Mater. 2021;7
three-dimensional arrays of perovskite quantum (10):2100605. DOI:10.1002/aelm.202100605
wires. Nano Lett. 2021;21(12):5036–5044. DOI:10. [32] Wang J, Pan X, Luo W, et al. Voltage-programmable
1021/acs.nanolett.1c00834 negative differential resistance in memristor of
[18] Kim M-H, Park H-L, Kim M-H, et al. Fluoropolymer- single-crystalline lithium niobate thin film. Appl
based organic memristor with multifunctionality for Phys Lett. 2022;120(3):032901. DOI:10.1063/5.
flexible neural network system. npj Flexible Electron. 0070132
2021;5(1):34. DOI:10.1038/s41528-021-00132-w [33] Pi S, Li C, Jiang H, et al. Memristor crossbar arrays
[19] Huang CH, Chang H, Yang TY, et al. Artificial with 6-nm half-pitch and 2-nm critical dimension.
synapse based on a 2D-SnO2 memtransistor with Nat Nanotechnol. 2019;14(1):35–39. DOI:10.1038/
dynamically tunable analog switching for s41565-018-0302-0
Sci. Technol. Adv. Mater. 24 (2023) 19 Y. XIAO et al.

[34] Zrinski I, Minenkov A, Cancellieri C, et al. Mixed oxide-based resistive switching memory. Adv
anodic oxides for forming-free memristors revealed Electron Mater. 2019;5(5):1800833. DOI:10.1002/
by combinatorial screening of hafnium-tantalum aelm.201800833
system. Appl Mater Today. 2022;26:101270. [50] Ye C, Deng T, Zhang J, et al. Enhanced resistive
[35] Wang J, Cao G, Sun K, et al. Alloy electrode engineer­ switching performance for bilayer HfO2/TiO2 resis­
ing in memristors for emulating the biological tive random access memory. Semicond Sci Technol.
synapse. Nanoscale. 2022;14(4):1318–1326. DOI:10. 2016;31(10):105005. DOI:10.1088/0268-1242/31/10/
1039/D1NR06144E 105005
[36] Woo KS, Kim J, Han J, et al. A high-speed true [51] Tsai T-M, Huang H-C, Tai Y-H, et al. Origin of
random number generator based on a CuxTe1−x dif­ hopping conduction in Sn-doped silicon oxide
fusive memristor. Adv Intell Syst. 2021;3(7):2100062. RRAM with supercritical CO2 fluid treatment. IEEE
DOI:10.1002/aisy.202100062 Electron Device Lett. 2012;33(12):1693–1695.
[37] Wang Y, Wu L, Liu G, et al. Non-destructive photo­ DOI:10.1109/LED.2012.2217932
voltaic reading of interface type memristors using [52] Xia Q, Wu J, Pan C-H, et al. Impact of forming
graphene as transparent electrode. J Alloys Compd. compliance current on storage window induced by
2018;740:273–277. a gadolinium electrode in oxide-based resistive ran­
[38] Min SY, Cho WJ. Resistive switching characteristic
dom access memory. IEEE Trans Electron Devices.
improvement in a single-walled carbon nanotube
2018;65(1):96–100. DOI:10.1109/TED.2017.2775104
random network embedded hydrogen silsesquioxane
[53] Choi BJ, Torrezan AC, Norris KJ, et al. Electrical
thin films for flexible memristors. Int J Mol Sci.
performance and scalability of Pt dispersed SiO2
2021;22(7):3390.
[39] Mahata C, Kang M, Kim S. Multi-level analog resis­ nanometallic resistance switch. Nano Lett. 2013;13
tive switching characteristics in tri-layer HfO2/Al2O3 (7):3213–3217. DOI:10.1021/nl401283q
/HfO2 based memristor on ITO electrode. [54] Prezioso M, Merrikh-Bayat F, Hoskins BD, et al.
Nanomaterials. 2020;10(10):2069. Training and operation of an integrated neuro­
[40] Liu L, Cheng ZQ, Jiang B, et al. Optoelectronic arti­ morphic network based on metal-oxide memristors.
ficial synapses based on two-dimensional Nature. 2015;521(7550):61–64. DOI:10.1038/
transitional-metal trichalcogenide. ACS Appl Mater nature14441
Interfaces. 2021;13(26):30797–30805. DOI:10.1021/ [55] Li Y, Chu J, Duan W, et al. Analog and digital bipolar
acsami.1c03202 resistive switching in solution-combustion-processed
[41] Ye C, Wu J-J, Pan C-H, et al. Boosting the perfor­ NiO memristor. ACS Appl Mater Interfaces. 2018;10
mance of resistive switching memory with (29):24598–24606. DOI:10.1021/acsami.8b05749
a transparent ITO electrode using supercritical fluid [56] Le V-Q, Do T-H, Retamal JRD, et al. Van der Waals
nitridation. RSC Adv. 2017;7(19):11585–11590. heteroepitaxial AZO/NiO/AZO/muscovite (ANA/
DOI:10.1039/C7RA01104K muscovite) transparent flexible memristor. Nano
[42] Tao D-W, Chen J-B, Jiang Z-J, et al. Making reversi­ Energy. 2019;56:322–329.
ble transformation from electronic to ionic resistive [57] Wang Y, Wang W, Zhang C, et al. A digital–analog
switching possible by applied electric field in an integrated memristor based on a ZnO NPs/CuO NWs
asymmetrical Al/TiO2/FTO nanostructure. Appl heterostructure for neuromorphic computing. ACS
Surf Sci. 2020;502:144124. Appl Electron Mater. 2022;4(7):3525–3534. DOI:10.
[43] Wang S, Dong X, Xiong Y, et al. CsFamapbibr photo­ 1021/acsaelm.2c00495
electric memristor based on ion-migration induced [58] Huang C-H, Tang Y, Matsuzaki K, et al. Resistive
memristive behavior. Adv Electron Mater. 2021;7 switching memory effects in p-type
(5):2100014. DOI:10.1002/aelm.202100014 hydrogen-treated CuO nanowire. Appl Phys Lett.
[44] Hussain T, Abbas H, Youn C, et al. Cellulose nano­ 2020;117(4):043502. DOI:10.1063/5.0010839
crystal based bio-memristor as a green artificial [59] Zhang L, Huang H, Ye C, et al. Exploration of highly
synaptic device for neuromorphic computing appli­ enhanced performance and resistive switching
cations. Adv Mater Technol. 2022;7(2):2100744. mechanism in hafnium doping ZnO memristive
DOI:10.1002/admt.202100744 device. Semicond Sci Technol. 2018;33(8):085013.
[45] Zhou GD, Sun B, Hu XF, et al. Negative photocon­ DOI:10.1088/1361-6641/aacff1
ductance effect: an extension function of the TiOx [60] Lin Y, Liu J, Shi J, et al. Nitrogen-induced ultralow
-based memristor. Adv Sci. 2021;8(13):2003765. power switching in flexible ZnO-based memristor for
DOI:10.1002/advs.202003765 artificial synaptic learning. Appl Phys Lett. 2021;118
[46] Shan X, Zhao C, Wang X, et al. Plasmonic optoelec­ (10):103502. DOI:10.1063/5.0036667
tronic memristor enabling fully light-modulated [61] Liu L, Xiong W, Liu YX, et al. Designing
synaptic plasticity for neuromorphic vision. Adv Sci. high-performance storage in HfO2/BiFeO3 memristor
2022;9(6):2104632. DOI:10.1002/advs.202104632 for artificial synapse applications. Adv Electron
[47] Deng T, Ye C, Wu J, et al. Improved performance of Mater. 2020;6(2):1901012. DOI:10.1002/aelm.
ITO/TiO2/HfO2/Pt random resistive accessory mem­ 201901012
ory by nitrogen annealing treatment. Microelectron [62] Hu M, Graves CE, Li C, et al. Memristor-based analog
Reliab. 2016;57:34–38. computation and neural network classification with
[48] Wu F, Cao P, Peng Z, et al. Memristor based on TiOx a dot product engine. Adv Mater. 2018;30
/Al2O3 bilayer as flexible artificial synapse for neuro­ (9):1705914. DOI:10.1002/adma.201705914
morphic electronics. IEEE Trans Electron Devices. [63] Lee TS, Choi C. Improved analog switching charac­
2022;69(1):375–379. DOI:10.1109/TED.2021.3128841 teristics of Ta2O5-based memristor using indium tin
[49] Zhang R, Huang H, Xia Q, et al. Role of oxygen oxide buffer layer for neuromorphic computing.
vacancies at the TiO2/HfO2 interface in flexible Nanotechnology. 2022;33(24):245202.
Sci. Technol. Adv. Mater. 24 (2023) 20 Y. XIAO et al.

[64] Kim YS, An J, Jeon JB, et al. Ternary logic with [80] Xiao Z, Huang J. Energy-efficient hybrid perovskite
stateful neural networks using a bilayered TaOx - memristors and synaptic devices. Adv Electron
based memristor exhibiting ternary states. Adv Sci. Mater. 2016;2(7):1600100.
2022;9(5):2104107. DOI:10.1002/advs.202104107 [81] Zhao X, Wang Z, Li W, et al. Photoassisted electro­
[65] Du C, Cai F, Zidan MA, et al. Reservoir computing forming method for reliable low-power organic–inor­
using dynamic memristors for temporal information ganic perovskite memristors. Adv Funct Mater.
processing. Nat Commun. 2017;8(1):2204. DOI:10. 2020;30(17):1910151. DOI:10.1002/adfm.201910151
1038/s41467-017-02337-y [82] Kumar M, Patel M, Park DY, et al. Switchable
[66] Shin J, Kang M, Kim S. Gradual conductance mod­ two-terminal transparent optoelectronic devices
ulation of Ti/WOx/Pt memristor with based on 2D perovskite. Adv Electron Mater. 2019;5
self-rectification for a neuromorphic system. Appl (2):1800662. DOI:10.1002/aelm.201800662
Phys Lett. 2021;119(1):012102. [83] Schmitt R, Kubicek M, Sediva E, et al. Accelerated
[67] Wei X, Huang H, Ye C, et al. Exploring the role of ionic motion in amorphous memristor oxides for
nitrogen incorporation in ZrO2 resistive switching nonvolatile memories and neuromorphic
film for enhancing the device performance. J Alloys computing. Adv Funct Mater. 2019;29(5):1804782.
Compd. 2019;775:1301–1306. DOI:10.1002/adfm.201804782
[68] He P, Ye C, Wu J, et al. Effect of sputtering atmo­ [84] John RA, Yantara N, Ng SE, et al. Diffusive and drift
sphere on the characteristics of ZrOx resistive switch­ halide perovskite memristive barristors as nociceptive
ing memory. Semicond Sci Technol. 2017;32 and synaptic emulators for neuromorphic computing.
(5):055016. DOI:10.1088/1361-6641/aa637a Adv Mater. 2021;33(15):2007851. DOI:10.1002/adma.
[69] Liu Y, Ye C, Chang KC, et al. A robust and low-power 202007851
bismuth doped tin oxide memristor derived from [85] Yan K, Chen B, Hu H, et al. First fiber-shaped non-
coaxial conductive filaments. Small. 2020;16 volatile memory device based on hybrid organic–
(46):2004619. DOI:10.1002/smll.202004619 inorganic perovskite. Adv Electron Mater. 2016;2
[70] Lee M-J, Lee CB, Lee D, et al. A fast, high- (8):1600160. DOI:10.1002/aelm.201600160
endurance and scalable non-volatile memory [86] Zhang X, Zhao X, Shan X, et al. Humidity effect on
device made from asymmetric Ta2O5−x/TaO2−x resistive switching characteristics of the CH3NH3PbI3
bilayer structures. Nat Mater. 2011;10(8):625–630. memristor. ACS Appl Mater Interfaces. 2021;13
DOI:10.1038/nmat3070 (24):28555–28563. DOI:10.1021/acsami.1c05590
[71] Chen YY, Goux L, Clima S, et al. Endurance/ [87] Wang SX, Zhang ZJ, Xiong YX, et al. Low power
Retention trade-off on HfO2/metal cap 1T1R bipolar consumption photoelectric coupling perovskite
RRAM. IEEE Trans Electron Devices. 2013;60 memristor with adjustable threshold voltage.
(3):1114–1121. DOI:10.1109/TED.2013.2241064 Nanotechnology. 2021;32(37):375201. DOI:10.1088/
[72] Chevallier CJ, Siau CH, Lim SF, et al. A 0.13 µm 64 1361-6528/ac0667
Mb multi-layered conductive metal-oxide memory. [88] Li Y, Xu W, Mussakhanuly N, et al. Homologous
2010 IEEE International Solid-State Circuits bromides treatment for improving the open-circuit
Conference - (ISSCC); San Francisco: IEEE; 2010. p. voltage of perovskite solar cells. Adv Mater. 2022;34
260–261. (6):2106280. DOI:10.1002/adma.202106280
[73] Hong X, Loy DJ, Dananjaya PA, et al. Oxide-based [89] Yang K, Liao Q, Huang J, et al. Intramolecular non­
RRAM materials for neuromorphic computing. covalent interaction-enabled dopant-free
J Mater Sci. 2018;53(12):8720–8746. DOI:10.1007/ hole-transporting materials for high-performance
s10853-018-2134-6 inverted perovskite solar cells. Angew Chem Int Ed.
[74] Kim YB, Lee SR, Lee D, et al. Bi-layered RRAM with 2022;61(2):e202113749. DOI:10.1002/anie.202113749
unlimited endurance and extremely uniform [90] Pan A, Ma X, Huang S, et al. CsPbBr3 perovskite
switching. Symposium on VLSI Technology - Digest nanocrystal grown on MXene nanosheets for
of Technical Papers; Honolulu, HI: IEEE; 2011. p. enhanced photoelectric detection and photocatalytic
52–53. CO2 reduction. J Phys Chem Lett. 2019;10
[75] Xue X, Yang J, Zhang Y, et al. A 28 nm 512 Kb (21):6590–6597. DOI:10.1021/acs.jpclett.9b02605
adjacent 2T2R RRAM PUF with interleaved cell mir­ [91] Shrestha S, Fischer R, Matt GJ, et al. High-
roring and self-adaptive splitting for extremely low bit performance direct conversion X-ray detectors
error rate of cryptographic key. 2019 IEEE Asian Solid- based on sintered hybrid lead triiodide perovskite
State Circuits Conference (A-SSCC); Macao: IEEE; wafers. Nat Photonics. 2017;11(7):436–440. DOI:10.
2019. p. 29–32. 1038/nphoton.2017.94
[76] Akkerman QA, Manna L. What defines a halide [92] Li Z, Chen Z, Yang Y, et al. Modulation of recombi­
perovskite? ACS Energy Lett. 2020;5(2):604–610. nation zone position for quasi-two-dimensional blue
[77] Zhu T, Yang Y, Gong X. Recent advancements and perovskite light-emitting diodes with efficiency
challenges for low-toxicity perovskite materials. ACS exceeding 5%. Nat Commun. 2019;10(1):1027.
Appl Mater Interfaces. 2020;12(24):26776–26811. DOI:10.1038/s41467-019-09011-5
[78] Yang J-Q, Wang R, Wang Z-P, et al. Leaky [93] Xu W, Hu Q, Bai S, et al. Rational molecular passiva­
integrate-and-fire neurons based on perovskite mem­ tion for high-performance perovskite light-emitting
ristor for spiking neural networks. Nano Energy. diodes. Nat Photonics. 2019;13(6):418–424. DOI:10.
2020;74:104828. 1038/s41566-019-0390-x
[79] Chen J, Feng Z, Luo M, et al. High-performance [94] Cheng XF, Qian WH, Wang J, et al. Environmentally
perovskite memristor by integrating a tip-shape robust memristor enabled by lead-free double perovs­
contact. J Mater Chem C. 2021;9(43):15435–15444. kite for high-performance information storage. Small.
DOI:10.1039/D1TC04164A 2019;15(49):e1905731. DOI:10.1002/smll.201905731
Sci. Technol. Adv. Mater. 24 (2023) 21 Y. XIAO et al.

[95] Subramanian A, Tiwale N, Kisslinger K, et al. [112] Xiong W, Zhu LQ, Ye C, et al. Flexible poly(vinyl
Reduced stochastic resistive switching in alcohol)–graphene oxide hybrid nanocomposite
organic-inorganic hybrid memristors by based cognitive memristor with Pavlovian-condi­
vapor-phase infiltration. Adv Electron Mater. 2022;8 tioned reflex activities. Adv Electron Mater. 2020;6
(7):2200172. DOI:10.1002/aelm.202200172 (5):1901402.
[96] Zhou J, Li W, Chen Y, et al. A monochloro copper [113] Tang Y, Lei P, Liao K, et al. Observation of nonvola­
phthalocyanine memristor with high-temperature tile resistive switching behaviors in 2D layered InSe
resilience for electronic synapse applications. Adv nanosheets through controllable oxidation. Appl
Mater. 2021;33(5):2006201. DOI:10.1002/adma. Phys Lett. 2021;119(13):133103.
202006201 [114] Wang K, Chen J, Yan X. Mxene Ti3C2 memristor for
[97] Goswami S, Thompson D, Williams RS, et al. Colossal neuromorphic behavior and decimal arithmetic opera­
current and voltage tunability in an organic memris­ tion applications. Nano Energy. 2021;79:105453.
tor via electrode engineering. Appl Mater Today. [115] Yalagala B, Sahatiya P, Mattela V, et al. Ultra-low cost,
2020;19:100626. large area graphene/MoS2-based piezotronic memris­
[98] Feng Y, Gao X, Zhong Y-N, et al. Solution-processed tor on paper: a systematic study for both direct cur­
polymer thin-film memristors with an electrochromic rent and alternating current inputs. ACS Appl
feature and frequency-dependent synaptic plasticity. Electron Mater. 2019;1(6):883–891.
Adv Intell Syst. 2019;1(3):1900022. [116] Lee H-S, Sangwan VK, Rojas WAG, et al. Dual-gated
[99] van de Burgt Y, Lubberman E, Fuller EJ, et al. A MoS2 memtransistor crossbar array. Adv Funct
non-volatile organic electrochemical device as a Mater. 2020;30(45):2003683.
low-voltage artificial synapse for neuromorphic [117] Nam JH, Oh S, Jang HY, et al. Low power MoS2/Nb2
computing. Nat Mater. 2017;16(4):414–418. O5 memtransistor device with highly reliable hetero­
[100] Song Y-W, Song M-K, Choi D, et al. Encapsulation- synaptic plasticity. Adv Funct Mater. 2021;31
enhanced switching stability of MoS2 memristors. (40):2104174.
J Alloys Compd. 2021;885:161016. [118] Choi S, Choi J-W, Kim JC, et al. Energy-efficient
[101] Liu J, Yang F, Cao L, et al. A robust nonvolatile three-terminal SiOx memristor crossbar array enabled
resistive memory device based on a freestanding by vertical Si/graphene heterojunction barristor.
ultrathin 2D imine polymer film. Adv Mater. Nano Energy. 2021;84:105947.
2019;31(28):1902264. [119] Chen Q, Lin M, Wang Z, et al. Low power parylene‐
[102] Xu X, Cho EJ, Bekker L, et al. A bioinspired artificial based memristors with a graphene barrier layer for
injury response system based on a robust polymer flexible electronics applications. Adv Electron Mater.
memristor to mimic a sense of pain, sign of injury, 2019;5(9):1800852.
and healing. Adv Sci. 2022;9(15):2200629. [120] Xiang D, Liu T, Xu J, et al. Two-dimensional multibit
[103] Zhang B, Chen W, Zeng J, et al. 90% yield production optoelectronic memory with broadband spectrum
of polymer nano-memristor for in-memory distinction. Nat Commun. 2018;9(1):2966.
computing. Nat Commun. 2021;12(1):1984. [121] Yang H, Xiao M, Cui Y, et al. Nonvolatile memristor
[104] Park H-L, Kim M-H, Kim H, et al. Self-selective organic based on heterostructure of 2D room-temperature
memristor by engineered conductive nanofilament dif­ ferroelectric α-In2Se3 and WSe2. Sci China Inf Sci.
fusion for realization of practical neuromorphic 2019;62(12):220404.
system. Adv Electron Mater. 2021;7(8):2100299. [122] Wu E, Xie Y, Wang S, et al. Tunable and nonvolatile
[105] Cha J-H, Jang BC, Oh J, et al. Highly reliable synaptic multibit data storage memory based on MoTe2/boron
cell array based on organic–inorganic hybrid bilayer nitride/graphene heterostructures through contact
stack toward precise offline learning. Adv Intell Syst. engineering. Nanotechnology. 2020;31(48):485205.
2022;4(6):2200018. [123] Sun L, Ding M, Li J, et al. Phase-controlled
[106] Xu X, Zhou X, Wang T, et al. Robust DNA-bridged large-area growth of MoTe 2 and MoTe2-xOx
memristor for textile chips. Angew Chem Int Ed. /MoTe2 heterostructures for tunable memristive
2020;59(31):12762–12768. behavior. Appl Surf Sci. 2019;496:143687.
[107] Qin S, Dong R, Yan X, et al. A reproducible [124] Li H, Ye L, Xu J. Novel continuously-tunable mem­
write-(read)n-erase and multilevel bio-memristor ristor based on few-layers black phosphorus/MoS2
based on DNA molecule. Org Electron. heterojunction. 2016 IEEE 16th International
2015;22:147–153. Conference on Nanotechnology (IEEE-NANO);
[108] Yan X, Li X, Zhou Z, et al. Flexible transparent Sendai: IEEE; 2016. p. 861–864.
organic artificial synapse based on the tungsten/egg [125] Hsieh Y-L, W-H S, Huang C-C, et al. Solution-
albumen/indium tin oxide/polyethylene terephthalate processed black phosphorus nanoflakes for integrat­
memristor. ACS Appl Mater Interfaces. 2019;11 ing nonvolatile resistive random access memory and
(20):18654–18661. the mechanism unveiled. Nanotechnology. 2019;30
[109] Yan X, Zhou Z, Zhao J, et al. Flexible memristors as (44):445702.
electronic synapses for neuro-inspired computation [126] Lee S, Peng R, Wu C, et al. Programmable black
based on scotch tape-exfoliated mica substrates. Nano phosphorus image sensor for broadband optoelectro­
Res. 2018;11(3):1183–1192. nic edge computing. Nat Commun. 2022;13(1):1485.
[110] He X, Zhang J, Wang W, et al. Transient resistive [127] Li Q, Liu M, Zhang Y, et al. Hexagonal boron nitride–
switching devices made from egg albumen dielectrics graphene heterostructures: synthesis and interfacial
and dissolvable electrodes. ACS Appl Mater properties. Small. 2016;12(1):32–50.
Interfaces. 2016;8(17):10954–10960. [128] Park H, Mastro MA, Tadjer MJ, et al. Programmable
[111] Zhang WG, Gao H, Deng CS, et al. An ultrathin multilevel memtransistors based on van der Waals
memristor based on a two-dimensional WS2/MoS2 heterostructures. Adv Electron Mater. 2019;5
heterojunction. Nanoscale. 2021;13(26):11497–11504. (10):1900333.
Sci. Technol. Adv. Mater. 24 (2023) 22 Y. XIAO et al.

[129] Liu C, Chen H, Wang S, et al. Two-dimensional [147] Wang S, Dang B, Sun J, et al. Physically transient
materials for next-generation computing diode with ultrathin tunneling layer as selector for
technologies. Nat Nanotechnol. 2020;15(7):545–557. bipolar one diode-one resistor memory. IEEE
[130] Park WI, Yoon JM, Park M, et al. Self-assembly- Electron Device Lett. 2021;42(5):700–703.
induced formation of high-density silicon oxide [148] Chen J, Pan WQ, Li Y, et al. High-precision sym­
memristor nanostructures on graphene and metal metric weight update of memristor by gate voltage
electrodes. Nano Lett. 2012;12(3):1235–1240. ramping method for convolutional neural network
[131] Sun J, Li YB, Ye Y, et al. 3D geometric engineering of accelerator. IEEE Electron Device Lett. 2020;41
the double wedge-like electrodes for filament-type (3):353–356.
RRAM device performance improvement. IEEE [149] Wang R, Shi T, Zhang X, et al. Implementing in-situ
Access. 2020;8:4924–4934. self-organizing maps with memristor crossbar arrays
[132] Wang J, Li L, Huyan H, et al. Highly uniform resistive for data mining and optimization. Nat Commun.
switching in HfO2 films embedded with ordered 2022;13(1):2289.
metal nanoisland arrays. Adv Funct Mater. 2019;29 [150] Zheng HX, Chen MC, Chang TC, et al. Overcoming
(25):1808430. limited resistance in 1T1R RRAM caused by pinch-off
[133] Chang K, Yu X, Liu B, et al. Quantum-dots optimized voltage during reset process. IEEE Trans Electron
electrode for high-stability transient memristor. IEEE Devices. 2019;66(11):4706–4709.
Electron Device Lett. 2021;42(6):824–827. [151] Wang Z, Li C, Song W, et al. Reinforcement learning
[134] Hota MK, Bera MK, Bera MK. Switching mechanism with analogue memristor arrays. Nat Electron. 2019;2
in Au nanodot-embedded Nb2O5 memristors. (3):115–124.
J Nanosci Nanotechnol. 2014;14(5):3538–3544. [152] Minguet Lopez J, Rafhay Q, Dampfhoffer M, et al.
[135] Huang XD, Li Y, Li HY, et al. Forming-free, fast, 1S1R optimization for high-frequency inference on
uniform, and high endurance resistive switching binarized spiking neural networks. Adv Electron
from cryogenic to high temperatures in W/AlOx/Al2 Mater. 2022;8(8):2200323.
O3/Pt bilayer memristor. IEEE Electron Device Lett. [153] Upadhyay NK, Sun W, Lin P, et al. A memristor with
2020;41(4):549–552. low switching current and voltage for 1S1R integra­
[136] Cho B, Kim TW, Song S, et al. Rewritable switching of tion and array operation. Adv Electron Mater. 2020;6
one diode-one resistor nonvolatile organic memory (5):1901411.
devices. Adv Mater. 2010;22(11):1228–1232. [154] Zhou J, Cai F, Wang Q, et al. Very low-programming-
[137] Sun Y, Zhao X, Song C, et al. Performance-enhancing current RRAM with self-rectifying characteristics.
selector via symmetrical multilayer design. Adv Funct IEEE Electron Device Lett. 2016;37(4):404–407.
Mater. 2019;29(13):1808376. [155] Adam GC, Hoskins BD, Prezioso M, et al.
[138] Kim H, Mahmoodi MR, Nili H, et al. 4K-memristor 3-D memristor crossbars for analog and neuro­
analog-grade passive crossbar circuit. Nat Commun. morphic computing applications. IEEE Trans
2021;12(1):5198. Electron Devices. 2017;64(1):312–318.
[139] Yin J, Zeng F, Wan Q, et al. Adaptive crystallite [156] Rahaman SZ, Maikap S, Chen WS, et al. Repeatable
kinetics in homogenous bilayer oxide memristor for unipolar/bipolar resistive memory characteristics and
emulating diverse synaptic plasticity. Adv Funct switching mechanism using a Cu nanofilament in
Mater. 2018;28(19):1706927. a GeOx film. Appl Phys Lett. 2012;101(7):073106.
[140] Li H-Y, Huang X-D, Yuan J-H, et al. Controlled [157] Zhang X, Wu C, Lv Y, et al. High-performance flex­
memory and threshold switching behaviors in ible polymer memristor based on stable filamentary
a heterogeneous memristor for neuromorphic switching. Nano Lett. 2022;22(17):7246–7253.
computing. Adv Electron Mater. 2020;6(8):2000309. [158] Lee HY, Chen YS, Chen PS, et al. Evidence and solu­
[141] Li C, Belkin D, Li Y, et al. Efficient and self-adaptive tion of over-RESET problem for HfOX based resistive
in-situ learning in multilayer memristor neural memory with sub-ns switching speed and high
networks. Nat Commun. 2018;9(1):2385. endurance. 2010 International Electron Devices
[142] Xu H, Zhai X, Wang Z, et al. An epitaxial synaptic Meeting; San Francisco: IEEE; 2010. p. 19.17.11–
device made by a band-offset BaTiO3/Sr2IrO4 bilayer 19.17.14.
with high endurance and long retention. Appl Phys [159] Wang M, Cai S, Pan C, et al. Robust memristors based
Lett. 2019;114(10):102904. on layered two-dimensional materials. Nat Electron.
[143] Saleem A, Kumar D, Singh A, et al. Oxygen vacancy 2018;1(2):130–136.
transition in HfOx-based flexible, robust, and synap­ [160] Poddar S, Zhang Y, Chen Z, et al. Image processing
tic bi-layer memristor for neuromorphic and wear­ with a multi-level ultra-fast three dimensionally inte­
able applications. Adv Mater Technol. 2022;7 grated perovskite nanowire array. Nanoscale Horiz.
(7):2101208. 2022;7(7):759–769.
[144] Linn E, Rosezin R, Kügeler C, et al. Complementary [161] Torrezan AC, Strachan JP, Medeiros-Ribeiro G, et al.
resistive switches for passive nanocrossbar memories. Sub-nanosecond switching of a tantalum oxide
Nat Mater. 2010;9(5):403–406. memristor. Nanotechnology. 2011;22(48):485203.
[145] Kim TW, Zeigler DF, Acton O, et al. All-organic [162] Lu Y-F, Li H-Y, Li Y, et al. A high-performance
photopatterned one diode-one resistor cell array for Ag/TiN/HfOx/HfOy/HfOx/Pt diffusive memristor
advanced organic nonvolatile memory applications. for calibration-free true random number generator.
Adv Mater. 2012;24(6):828–833. Adv Electron Mater. 2022;8(9):2200202.
[146] Li YT, Long SB, Lv HB, et al. Novel self-compliance [163] Yao P, Wu HQ, Gao B, et al. Fully
bipolar 1D1R memory device for high-density RRAM hardware-implemented memristor convolutional
application. 2013 5th IEEE International Memory neural network. Nature. 2020;577(7792):641–646.
Workshop; Monterey, California: IEEE; 2013. p. [164] Ye C, Zhan C, Tsai T-M, et al. Low-power bipolar
184–187. resistive switching TiN/HfO2/ITO memory with
Sci. Technol. Adv. Mater. 24 (2023) 23 Y. XIAO et al.

self-compliance current phenomenon. Appl Phys switching transformation in all Ti-based memristor
Express. 2014;7(3):034101. devices. Appl Phys Lett. 2020;117(7):073504.
[165] Ye C, Xu Z, Chang K-C, et al. Hafnium nanocrystals [182] Wang C, Mao G-Q, Huang M, et al. HfOx/AlOy
observed in a HfTio compound film bring about superlattice-like memristive synapse. Adv Sci. 2022;9
excellent performance of flexible selectors in memory (21):2201446.
integration. Nanoscale. 2019;11(43):20792–20796. [183] Saleem A, Simanjuntak FM, Chandrasekaran S, et al.
[166] Guan H, Sha J, Zhang Z, et al. Optical and oxide Transformation of digital to analog switching in TaOx
modification of CsFamapbibr memristor achieving -based memristor device for neuromorphic
low power consumption. J Alloys Compd. applications. Appl Phys Lett. 2021;118(11):112103.
2022;891:162096. [184] Yang K, Duan Q, Wang Y, et al. Transiently chaotic
[167] Wang J, Wang J, Zhang J, et al. Ultralow-power simulated annealing based on intrinsic nonlinearity of
synaptic transistors based on Ta2O5/Al2O3 bilayer memristors for efficient solution of optimization
dielectric for algebraic arithmetic. Adv Electron problems. Sci Adv. 2020;6(33):eaba9901.
Mater. 2022;8(5):2100922. [185] Yang JM, Jung YK, Lee JH, et al. Asymmetric carrier
[168] Ni R, Yang L, Huang XD, et al. Controlled transport in flexible interface-type memristor enables
majority-inverter graph logic with highly nonlinear, artificial synapses with sub-femtojoule energy
self-rectifying memristor. IEEE Trans Electron
consumption. Nanoscale Horiz. 2021;6(12):987–997.
Devices. 2021;68(10):4897–4902.
[186] Zhang C, Ye WB, Zhou K, et al. Bioinspired artificial
[169] Cheong WH, Jeon JB, In JH, et al. Demonstration of
sensory nerve based on nafion memristor. Adv Funct
neuromodulation-inspired stashing system for
Mater. 2019;29(20):1808783.
energy-efficient learning of spiking neural network
[187] Ge J, Zhang S, Liu Z, et al. Flexible artificial nocicep­
using a self-rectifying memristor Array. Adv Funct
tor using a biopolymer-based forming-free
Mater. 2022;32(29):2200337.
memristor. Nanoscale. 2019;11(14):6591–6601.
[170] Choi S, Kim Y, Nguyen TV, et al. Low-power
[188] Qiao N, Wei P, Xing Y, et al. Rapid charge storage and
self-rectifying memristive artificial neural network
release at etching-assist electret in organic transistors
for near internet-of-things sensor computing. Adv
for memories, photodetectors, and artificial synapses.
Electron Mater. 2021;7(6):2100050.
Adv Mater Interfaces. 2022;9(19):2200713.
[171] Lv J, Wang S, Li F, et al. A physically transient
[189] Kapur O, Guo D, Reynolds J, et al. Back-end-of-line
self-rectifying and analogue switching memristor
SiC-based memristor for resistive memory and arti­
synapse. IEEE Electron Device Lett. 2021;42
ficial synapse. Adv Electron Mater. 2022;8(9):220
(11):1599–1602.
[172] Ren SG, Ni R, Huang XD, et al. Pt/Al2O3/TaOX/Ta 0312.
self-rectifying memristor with record-low operation [190] Ali A, Abbas H, Hussain M, et al. Thickness-
current (<2pa), low power (fJ), and high scalability. dependent monochalcogenide GeSe-based CBRAM
IEEE Trans Electron Devices. 2022;69(2):838–842. for memory and artificial electronic synapses. Nano
[173] Zhao X, Zhang K, Hu K, et al. Self-rectifying Al2O3 Res. 2022;15(3):2263–2277.
/TaOx memristor with gradual operation at low cur­ [191] Hu L, Yang J, Wang J, et al. All-optically controlled
rent by interfacial layer. IEEE Trans Electron Devices. memristor for optoelectronic neuromorphic
2021;68(12):6100–6105. computing. Adv Funct Mater. 2021;31(4):2005582.
[174] Choi S, Jang S, Moon J-H, et al. A self-rectifying TaOy [192] Seo S, Jo SH, Kim S, et al. Artificial optic-neural
/nanoporous TaOx memristor synaptic array for synapse for colored and color-mixed pattern
learning and energy-efficient neuromorphic systems. recognition. Nat Commun. 2018;9(1):5106.
NPG Asia Mater. 2018;10(12):1097–1106. [193] Yu J, Yang X, Gao G, et al. Bioinspired
[175] Li C, Han L, Jiang H, et al. Three-dimensional cross­ mechano-photonic artificial synapse based on
bar arrays of self-rectifying Si/SiO2/Si memristors. graphene/MoS2 heterostructure. Sci Adv. 2021;7(12):
Nat Commun. 2017;8(1):15666. eabd9117.
[176] Lanza M, Sebastian A, Lu WD, et al. Memristive [194] Sun L, Wang Z, Jiang J, et al. In-sensor reservoir
technologies for data storage, computation, encryp­ computing for language learning via
tion, and radio-frequency communication. Science. two-dimensional memristors. Sci Adv. 2021;7(20):
2022;376(6597):eabj9979. eabg1455.
[177] Liu H, Hua Q, Yu R, et al. A bamboo-like GaN [195] Zhou Z, Pei Y, Zhao J, et al. Visible light responsive
microwire-based piezotronic memristor. Adv Funct optoelectronic memristor device based on CeOx/ZnO
Mater. 2016;26(29):5307–5314. structure for artificial vision system. Appl Phys Lett.
[178] Lappalainen J, Mizsei J, Huotari M. Neuromorphic 2021;118(19):191103.
thermal-electric circuits based on phase-change VO2 [196] Woo J, Moon K, Song J, et al. Optimized program­
thin-film memristor elements. J Appl Phys. 2019;125 ming scheme enabling linear potentiation in fila­
(4):044501. mentary HfO2 RRAM synapse for neuromorphic
[179] Pan WQ, Chen J, Kuang R, et al. Strategies to improve systems. IEEE Trans Electron Devices. 2016;63
the accuracy of memristor-based convolutional (12):5064–5067.
neural networks. IEEE Trans Electron Devices. [197] Kang J, Kim T, Hu S, et al. Cluster-type analogue
2020;67(3):895–901. memristor by engineering redox dynamics for
[180] Woo J, Moon K, Song J, et al. Improved synaptic high-performance neuromorphic computing. Nat
behavior under identical pulses using AlOx/HfO2 Commun. 2022;13(1):4040.
bilayer RRAM array for neuromorphic systems. [198] Jung PY, Panda D, Chandrasekaran S, et al. Enhanced
IEEE Electron Device Lett. 2016;37(8):994–997. switching properties in TaOx memristors using diffu­
[181] Chang L-Y, Simanjuntak FM, Hsu C-L, et al. sion limiting layer for synaptic learning. IEEE
Suboxide interface induced digital-to-analog J Electron Devices Soc. 2020;8:110–115.
Sci. Technol. Adv. Mater. 24 (2023) 24 Y. XIAO et al.

[199] Simanjuntak FM, Chandrasekaran S, Lin C-C, et al. [213] Hsu C-L, Saleem A, Singh A, et al. Enhanced linearity
ZnO2/ZnO bilayer switching film for making fully in CBRAM synapse by post oxide deposition anneal­
transparent analog memristor devices. APL Mater. ing for neuromorphic computing applications. IEEE
2019;7(5):051108. Trans Electron Devices. 2021;68(11):5578–5584.
[200] Yoon JH, Zhang J, Lin P, et al. A low-current and [214] John RA, Yantara N, Ng YF, et al. Ionotronic halide
analog memristor with Ru as mobile species. Adv perovskite drift-diffusive synapses for low-power
Mater. 2020;32(9):1904599. neuromorphic computation. Adv Mater. 2018;30
[201] Borghetti J, Snider GS, Kuekes PJ, et al. ‘Memristive’ (51):1805454.
switches enable ‘stateful’ logic operations via material [215] Wang Y, Gong Y, Yang L, et al. Mxene-ZnO mem­
implication. Nature. 2010;464(7290):873–876. ristor for multimodal in-sensor computing. Adv
[202] Mao JY, Wu S, Ding G, et al. A van der Waals inte­ Funct Mater. 2021;31(21):2100144.
grated damage-free memristor based on layered 2D [216] Seo S, Kang BS, Lee JJ, et al. Artificial van der Waals
hexagonal boron nitride. Small. 2022;18(12):2106253. hybrid synapse and its application to acoustic pattern
[203] Gao B, Lin B, Pang Y, et al. Concealable physically recognition. Nat Commun. 2020;11(1):3936.
unclonable function chip with a memristor array. Sci [217] Zhong YA, Tang JS, Li XY, et al. Dynamic
Adv. 2022;8(24):eabn7753. memristor-based reservoir computing for high-
[204] Krishnaprasad A, Dev D, Han SS, et al. MoS2 efficiency temporal signal processing. Nat
synapses with ultra-low variability and their imple­ Commun. 2021;12(1):408.
mentation in Boolean logic. ACS Nano. 2022;16 [218] Wang Y, Gong Y, Huang S, et al. Memristor-based
(2):2866–2876. biomimetic compound eye for real-time collision
[205] Yuan R, Ma M, Xu L, et al. Efficient 16 Boolean logic detection. Nat Commun. 2021;12(1):5979.
and arithmetic based on bipolar oxide memristors. Sci [219] Zhu Y, Wu C, Xu Z, et al. Light-emitting memristors
China Inf Sci. 2020;63(10):202401. for optoelectronic artificial efferent nerve. Nano Lett.
[206] Sun J, Wang Z, Wang S, et al. Physical unclonable 2021;21(14):6087–6094.
functions based on transient form of memristors for [220] Xia Q, Yang JJ. Memristive crossbar arrays for
emergency defenses. IEEE Electron Device Lett. brain-inspired computing. Nat Mater. 2019;18
2022;43(3):378–381. (4):309–323.
[207] Yang L, Cheng L, Li Y, et al. Cryptographic key [221] Choi C, Kim H, Kang J-H, et al. Reconfigurable het­
generation and in situ encryption in one-transistor- erogeneous integration using stackable chips with
one-resistor memristors for hardware security. Adv embedded artificial intelligence. Nat Electron. 2022;5
Electron Mater. 2021;7(5):2001182. (6):386–393.
[208] Song Y, Wu Q, Wang X, et al. Two memristors-based [222] Liang X, Zhong Y, Tang J, et al. Rotating neurons for
XOR logic demonstrated with encryption/decryption. all-analog implementation of cyclic reservoir
IEEE Electron Device Lett. 2021;42(9):1398–1401. computing. Nat Commun. 2022;13(1):1549.
[209] Gao B, Lin B, Li X, et al. A unified PUF and TRNG [223] Zhong Y, Tang J, Li X, et al. A memristor-based
design based on 40-nm RRAM with high entropy and analogue reservoir computing system for real-time
robustness for IoT security. IEEE Trans Electron and power-efficient signal processing. Nat Electron.
Devices. 2022;69(2):536–542. 2022;5(10):672–681.
[210] Yang F, Wang Y, Wang C, et al. High-entropy true [224] Xue C-X, Chiu Y-C, Liu T-W, et al. A
random number generator based on memristor reset CMOS-integrated compute-in-memory macro based
switching. IEEE Electron Device Lett. 2022;43 on resistive random-access memory for AI edge
(9):1459–1462. devices. Nat Electron. 2021;4(1):81–90.
[211] Kim GM, In JH, Kim YS, et al. Self-clocking fast and [225] Narayanan P, Ambrogio S, Okazaki A, et al. Fully
variation tolerant true random number generator on-chip MAC at 14 nm enabled by accurate
based on a stochastic mott memristor. Nat row-wise programming of PCM-based weights and
Commun. 2021;12(1):2906. parallel vector-transport in duration-format. IEEE
[212] Zhu J, Zhang X, Wang R, et al. A heterogeneously Trans Electron Devices. 2021;68(12):6629–6636.
integrated spiking neuron array for multimode-fused [226] Wan W, Kubendran R, Schaefer C, et al. A compute-in-
perception and object classification. Adv Mater. memory chip based on resistive random-access
2022;34(24):2200481. memory. Nature. 2022;608(7923):504–512.