Lu et al.

Microsystems & Nanoengineering (2020)6:84

https://doi.org/10.1038/s41378-020-00189-z
Microsystems & Nanoengineering
www.nature.com/micronano

ARTICLE Open Access

Biological receptor-inspired flexible artificial

synapse based on ionic dynamics
Qifeng Lu1, Fuqin Sun1, Lin Liu2, Lianhui Li1, Yingyi Wang2, Mingming Hao1, Zihao Wang1, Shuqi Wang1 and
Ting Zhang1,3

Abstract
The memristor has been regarded as a promising candidate for constructing a neuromorphic computing platform that
is capable of confronting the bottleneck of the traditional von Neumann architecture. Here, inspired by the working
mechanism of the G-protein-linked receptor of biological cells, a novel double-layer memristive device with reduced
graphene oxide (rGO) nanosheets covered by chitosan (an ionic conductive polymer) as the channel material is
constructed. The protons in chitosan and the functional groups in rGO nanosheets imitate the functions of the ligands
and receptors of biological cells, respectively. Smooth changes in the response current depending on the historical
applied voltages are observed, offering a promising pathway toward biorealistic synaptic emulation. The memristive
behavior is mainly a result of the interaction between protons provided by chitosan and the defects and functional
groups in the rGO nanosheets. The channel current is due to the hopping of protons through functional groups and is
limited by the traps in the rGO nanosheets. The transition from short-term to long-term potentiation is achieved, and
learning-forgetting behaviors of the memristor mimicking those of the human brain are demonstrated. Overall, the
bioinspired memristor-type artificial synaptic device shows great potential in neuromorphic networks.
1234567890():,;
1234567890():,;
1234567890():,;
1234567890():,;

Introduction processing unit with a conventional von Neumann

Inspired by the massive parallelism, robust computa- architecture limits the speed of current computer sys-
tion, fault tolerance, and energy efficiency of the human tems7. However, the human brain, containing ~1011
brain, neuromorphic computing has attracted a tre- neurons and 1015 synapses, only consumes less than
mendous upsurge of research interest since synaptic 20 W, making it more efficient than any other computing
behaviors were mimicked by Mead in 1996 using a system8. Therefore, a computing system that can poten-
floating-gate silicon metal-oxide-semiconductor transis- tially emulate the neuromorphic network and overcome
tor1–5. The implementation of neural networks at the the von Neumann bottleneck is critically required.
software level with algorithms, such as Google’s Alpha Go Memristors acting as a synapse unit, resembling their
and IBM Watson, based on complementary metal-oxide biological counterpart in terms of structure, can emulate
semiconductor technology requires a large amount of several functions of the synaptic behaviors of the human
resources6. In addition, the tremendous amount of data brain. In addition, memristor-based artificial synapses can
processing between the memory and the central be used as basic building blocks to achieve large-scale
neural network parallelism.
Thus, a great amount of attention has been attracted
Correspondence: Ting Zhang ([email protected])
1 and much effort has been made to improve the perfor-
i-Lab, Key Laboratory of Multifunctional Nanomaterials and Smart Systems,
Suzhou Institute of Nano-Tech and Nano-Bionics (SINANO), Chinese Academy mance of memristors. Generally, memristors are a type of
of Sciences (CAS), 398 Ruoshui Road, 215123 Suzhou, PR China
2
two-terminal device with a layer of memristive material
Department of Health and Environmental Sciences, Xi’an Jiaotong Liverpool
sandwiched between metal electrodes in either a vertical
University, 111 Ren’ai Road, 215123 Suzhou, PR China
Full list of author information is available at the end of the article or planar structure. The working mechanism of the
These authors contributed equally: Qifeng Lu, Fuqin Sun

© The Author(s) 2020

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction
in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if
changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If
material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain
permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
Lu et al. Microsystems & Nanoengineering (2020)6:84 Page 2 of 10

devices is highly related to the selection of both mem- insulation and proton conduction properties20. The pro-
ristive and electrode materials. Recently, binary metal tons, hopping between the channel layers, behave like the
oxides, oxide perovskites, polymers, and 2D materials ligands of biological cells. Reduced graphene oxide (rGO)
have been widely used as memristive layers in the con- nanosheets were used as the bottom-layer material, and
struction of memristors9–12. Each category of material has the functional groups and defects in the nanosheets
advantages in terms of the working mechanism and the behave like the receptors of biological cells. The number
material properties, which is beneficial to applications in of functional groups and defects in the rGO nanosheets
specific areas with the suitable selection of the materials. can be well controlled by tuning the degree of the che-
According to different working mechanisms, memristors mical redox reaction21,22. As a result, the device exhibits
can be categorized into filament-type and barrier-type history-dependent memristive behaviors with continuous
devices8. With regard to filament-type memristors, the changes in the conductance states, meeting the funda-
conductance of the devices is controlled by the formation mental requirements for mimicking the functions of a
and rupture of the filament. Undesired abrupt con- biological synapse. Consequently, a series of synaptic
ductance changes and cycle-to-cycle variation are characteristics, including short-term potentiation (STP),
observed due to the random distribution of the filaments long-term potentiation (LTP), spike-rate-dependent
in the insulator caused by the stochastic nature of ion plasticity (SRDP), and learning-forgetting behaviors,
migration13. Different from filament-type memristors, toward biorealistic synaptic emulation were successfully
barrier-type memristors generally modulate the con- realized.
ductance states by the defect effect, which can overcome
the electroforming randomness, ensuring reproducibility, Results and discussion
and device-to-device uniformity14,15. In addition, a series Device structure and properties
of continuous conductance states capable of emulating Figure 1a shows the schematic diagrams for two neu-
biosynaptic functions can be modulated by external sti- rons connected by a synapse in the nervous system and
muli. Therefore, use of the barrier-type memristor is the corresponding working mechanism. Transmission of
considered a more suitable approach to emulate synaptic synaptic signals between neurons is a complex process
behaviors. that starts with the opening of voltage-controlled ion
Based on the above analysis, the design of memristive channels by external stimuli23. In brief, the diffusion of
devices, whose conductance states can be modulated by Ca2+ ions leads to the release of neurotransmitters from
the defect effect, is a strategy to emulate biorealistic synaptic vesicles, and the ions bind to the receptor sites of
synaptic behaviors. In addition, the high efficiency of Na+-gated ion channels at postsynaptic neurons. As a
information processing and transmission in biological result, the Na+-gated ion channels will open, and Na+ can
nervous systems provides an approach to designing diffuse inside the cell, which leads to a positively charged
memristive devices with high performance based on the membrane potential. The neuron will fire an action
metrics of biological systems. For example, a double-layer potential once the membrane potential reaches the
memristive device can be conceived based on inspiration threshold. This process contributes to the information
from the working mechanism and structure of the G- flow and signal processing of the nervous system24. Based
protein-linked receptors of biological cells16,17, in which on the structure and types of receptors (acceptors), the
neural signals are transmitted by ligands from the extra- synapses in the biological system can be classified into
cellular environment in conjunction with receptors from various categories. Taking the G-protein-linked receptor
transmembrane proteins located in the cytomembrane. In as an example, the ligands from the extracellular envir-
the designed device, mobile carriers provided by the top onment react with the receptors from transmembrane
layer are able to mimic the functions of the ligands from proteins located in the cytomembrane (circled by the red
the extracellular environment, while functional groups dashed lines in Fig. 1a-II), and an ionic channel is formed
and defects in the bottom layer can behave like the to transmit the signals. Inspired by the working principle
receptors from transmembrane proteins. However, to the of the G-protein-linked receptor, a double-layer mem-
best of our knowledge, emulation of synaptic behaviors ristive device consisting of rGO and CS as bottom-layer
using this bioinspired working mechanism has not been and top-layer materials, respectively, was fabricated, as
reported even though the use of various materials, such as shown in Fig. 1b. In the designed memristive device, the
binary oxides, polymers, and perovskites, has been protons in CS and the functional groups in rGO are able
attempted in the fabrication of memristive devices9,11,18,19. to imitate the functions of the ligands and receptors of
Herein, we designed and fabricated a double-layer biological cells, respectively. Thus, the synaptic behavior
memristive device in which chitosan (CS), an abundant, can be mimicked using this device by monitoring the
nontoxic, biodegradable, and biocompatible polymer, was conductance change of the memristor. The detailed fab-
selected as the top-layer material due to its electronic rication process and related parameters are included in
Lu et al. Microsystems & Nanoengineering (2020)6:84 Page 3 of 10

a I
b
Synapse
Vesicle
Na+
Preneuron rGO/CS

Ti/Au

se Post-synapse
Pre-synap
K+
Neurotransmitter
Artificial synapse
c

II
r
la Ligand
e llu NH3+
r ac
xt ane Receptor
E br
Postneuron em
m
to
Cy r
la COO–
l lu
ce Enzyme or
tra G-protein
In ion channel

G-protein-linked receptor

d e 0.3
f 750
7.5 rGO CS rGO/CS
2
5.0 2 0.2 2 500

Current (nA)
Current (nA)
Current (nA)

2.5 0.1 250 1

1
1
0.0 0.0 0
3 3 3
–2.5 –0.1 –250
4
–5.0 4 –0.2 –500 4

–7.5 –0.3 –750

–3 –2 –1 0 1 2 3 –3 –2 –1 0 1 2 3 –3 –2 –1 0 1 2 3
Voltage (V) Voltage (V) Voltage (V)

Fig. 1 Structures and working mechanism of biological synapses and electrical properties of the receptor-inspired counterpart.
a-I Schematic diagram for the connection of two neurons by a synapse in the biological system and a-II the working principle of the G-protein-linked
receptor. b Schematic diagram and c optical image of the flexible memristors. CS and rGO were used as the top and bottom layers of the device
channels, respectively. IV characteristics for the devices with d rGO, e CS, and f rGO covered by CS as the channel materials

Supplementary Figs. S1–S3. In addition, to emulate the dependent on the applied voltages. In addition, the bend
neuromorphic functions of the brain, a matrix of the test shown in Supplementary Fig. S4 indicates the flex-
designed double-layer memristor was fabricated, and ibility and mechanical reliability of the devices.
the optical image is presented in Fig. 1c. After fabrication
of the devices, the current–voltage (IV) characteristics of Physical and electrical characteristics
the devices with rGO, CS, and rGO covered with CS as To explore the underlying mechanism of the mem-
the channel materials were investigated, as presented in ristive behaviors of the device, the physical and electrical
Fig. 1d, e, f, respectively. No hysteresis in the IV curves is properties of the rGO nanosheets, the CS membrane and
observed for the devices with rGO or CS as the channel the interactions between them were investigated in
material, and the conductance is independent of the detail. From the atomic force microscopy (AFM) images
applied voltage. Regarding the device with rGO/CS as the in Fig. 2a–d, there are many nanoholes in the rGO
channel material, a hysteretic IV characteristic is nanosheets, with an average size of ~1 μm, which lead to
obtained, which is a main feature of memristive devices. a number of defects (more information about the size
The results imply the occurrence of an interaction and thickness of the nanosheets is presented in Supple-
between the CS membrane and rGO nanosheets, which mentary Figs. S5 and S6). In this research, since the
leads to the change in device conductance being channel length of the fabricated device is 40 μm and the
Lu et al. Microsystems & Nanoengineering (2020)6:84 Page 4 of 10

1.5 1.5
a b c d
1.0 1.0

Height (nm)

Height (nm)
0.5 0.5

0.0 0.0

–0.5 –0.5
0 1 2 3 4 5 0.0 0.2 0.4 0.6 0.8 1.0
1.0 µm 200.0 nm
µm)
Length (µ Length (µm)

e f
CS N 1s Amine
CS + rGO N-H*
Transmittance

Intensity (a.u.)
N-H* Protonated amine
CS

rGO+CS N 1s Amine

rGO
Protonated amine

4000 3500 3000 2500 2000 1500 1000 406 404 402 400 398 396
Wavelength (cm–1) Binding energy (eV)

g 10–6
h 10–6
Current (A)
Current (A)

10–7 10–7
l ll
l ll lll

10–8 10–8
0.1 1 10 0.1 1 10
Voltage (V) Voltage (V)

Fig. 2 Characterization of the materials for the fabrication of the artificial synapses and the underlying mechanism for the memristive
behaviors. a–d AFM images and the corresponding heights of rGO nanosheets. rGO nanosheets with an average size of ~1 μm were obtained.
e Comparison of FTIR spectra for rGO, CS, and rGO/CS samples. Changes arising from antisymmetric –NH3+ deformation and –N–H bending
vibrations were observed. The (f) N 1s peak in the XPS spectra shows that extent of protonation of amine groups increases from 2.96 to 14.25% for CS
and rGO/CS samples. g, h demonstrate the forward and reverse IV curves of the memristor plotted in log scale, which are governed by SCLC

size of the rGO nanosheets is ~1 μm, it can be inferred 1628–1633 cm−1 and 1524–1529 cm−1 are observed;
that there are a number of homojunctions caused by these bands are assigned to amide I and antisymmetric
overlaps of the nanosheets and intersecting grain –NH3+ deformation and to amide II and –N–H bending
boundaries in the channel. The homojunctions and grain vibrations, respectively28. Thus, the changes are con-
boundaries as well as the defects inside the rGO sidered to be caused by the protonation of amide, which
nanosheets work as traps and impede the flow of current is also supported by the N 1s peaks of the XPS spectra
through the channel. By contrast, from the XPS spectra shown in Fig. 2f, indicating that the extent of protona-
shown in Supplementary Fig. S7, a considerable number tion of amine groups increases from 2.96 to 14.25%. No
of functional groups are observed in the rGO nanosh- obvious change in the C 1s peaks is observed for the XPS
eets, which will contribute to the current induced by the spectra collected from the CS and CS/rGO samples, as
hopping of the carriers in the channel25–27. In addition, shown in Supplementary Fig. S8. Furthermore, the
from the comparison of the FTIR spectra for CS and increased conductance of the devices with humidity
rGO/CS in Fig. 2e, obvious changes in the bands at (Supplementary Fig. S9) is additional evidence for the
Lu et al. Microsystems & Nanoengineering (2020)6:84 Page 5 of 10

above analysis, as the extent of protonated amino groups Working mechanism of the artificial synapse
of CS is influenced by the ambient humidity29,30. Based on the above discussion, a schematic diagram for
Based on the above results, it can be inferred that the the working mechanism of the memristor is demonstrated
current in the channel is caused by the hopping of protons in Fig. 3a, and a corresponding model is proposed. The
through the functional groups27,31. The hopping process hopping process of protons through functional groups is
resembles the interaction between the ligands and similar to the interaction of ligands and receptors in
receptors in biological systems. In other words, the for- biological cells, as indicated in the right corner of Fig. 3a.
mation of current based on ionic dynamics in the mem- Protons from the CS membrane act as ligands from the
ristor is similar to the working mechanism of G-protein- extracellular environment, while the functional groups in
linked receptors. In contrast, traps formed by the grain the rGO nanosheets behave like the receptors from
boundaries, homojunctions, and defects will capture the transmembrane proteins. The current induced by the
carriers and impede the current, as mentioned above32. hopping of protons in the device channel mimics the
From the slopes of the IV curve in regions I and II of transmission of neuron signals. The band gap of the rGO
Fig. 2g, as the voltage increases, the current changes from nanosheets is estimated to be 3.5 eV from the UV-vis
the low-exponent space-charge-limited current (SCLC) absorption spectra (Supplementary Fig. S10). Initially, it is
affected by shallow traps to the high-exponent SCLC tentatively hypothesized that the traps impede the carriers
affected by deep traps. More injected carriers will fill the in the channel due to the barrier heights32,36,37. Thus, the
trapping sites in order of the deeper energy levels, as memristor is modeled as a number of face-to-face diodes
indicated in region III, with the further increase in voltage, with a resistor between them, as shown in Fig. 3b. When a
which leads to the dramatic increase in conductance33,34. positive voltage is applied to the left terminal of the
Because holes are the main carriers in this device, their device, as presented in Fig. 3c, hopping of the protons
injection, transport, and trapping processes are analyzed through the functional groups occurs under the external
to provide a possible explanation for the underlying electric field. A portion of the injected protons is captured
mechanism. The CS membrane serves as a proton reser- by the trapping sites. With increasing applied voltage, the
voir, and the current is formed due to the hopping of traps are gradually filled in the order of the energy levels.
protons through the functional groups in the channel. Then, the barrier heights (or the number of diodes in the
However, the traps in the rGO nanosheets acting as proposed model) gradually decrease, as indicated in
barriers will impede the current, and therefore, the Fig. 3c, which leads to a decrease in the impedance.
resistance is relatively large in region I of the forward IV Therefore, the slope of the channel current is enhanced
characteristic, as shown in Fig. 2g. As the voltage due to the decreased impedance. As a result, the forward
increases, protons captured at the trapping sites diminish trace labeled 1 in Fig. 3f is obtained. In the reverse sweep,
the existing trap concentration gradually in the order of the carriers trapped in the deep energy levels cannot be
the energy levels, and the resistance decreases, which is thoroughly released with decreasing applied voltage,
supported by the increasing slopes of the fitting curves in which indicates that a portion of the traps are always filled
regions II and III of Fig. 2g34,35. Thus, the modulation of and that the diminished barrier height cannot be recov-
conductance can be achieved as a function of the trapped ered to the initial states. Thus, the reverse trace labeled 2
carrier concentration. As a result, the conductance in Fig. 3f with a current larger than the forward trace is
increases continuously depending on the history of achieved. The same working principle is also applicable to
operation, indicating the typical memristive property. In the case of a negative sweep of the IV curves, as shown in
the reverse trace, protons in the shallow traps are released Fig. 3d.
as the voltage decreases. However, protons trapped in the
deep trapping energy sites cannot escape easily. Thus, a Synaptic behaviors of the artificial synapse
more conductive channel is obtained than that forward After understanding the underlying mechanism of the
sweeping due to the unreleased carriers in the trapping memristor, a possible electronic artificial synapse that
sites. As a sequence, an obvious hysteretic loop is gener- can imitate the behaviors of biological synapses is
ated, as illustrated in Fig. 2f. In the reverse IV curves, demonstrated. Compared with a biological synapse, the
another linear relation is fitted on the log scale, as shown two electrodes of the fabricated device are analogous to
in region I of Fig. 2h, which correlates with the lower- the presynaptic and postsynaptic terminals. The synaptic
exponent SCLC. Therefore, it can be concluded that the strength, or synaptic plasticity, in biological systems,
physical mechanisms for memristive behaviors are mainly which is believed to be a result of the number of neu-
attributed to the SCLC transition caused by the charge rotransmitters released at the synapse, is determined by
trapping/detrapping process with different filling ratios of presynaptic Ca2+ flux induced by an action potential38,39.
protons in the trapping sites of the rGO nanosheets as a Thus, a temporal change in synaptic strength can be
function of the operation history. achieved as a function of the stimuli. Similarly, the
Lu et al. Microsystems & Nanoengineering (2020)6:84 Page 6 of 10

a
OH OH OH
O O
HO O

n
O O
HO HO OH

sa
HO
NH2 NH3
ito
NH2
Ch
n

O OH

HO
COOH
HO HO
OH O
OH O
OH

OH
OH
OH
Ligand
O
rG

NH3+
Receptor
OH

OH
O
COO–
Water Proton

b c d
Metal CS/rGO Metal Metal CS/rGO Metal Metal CS/rGO Metal

– –
+ +

Df1 Dr1 Df1 Dr1 Df1 Dr1

Df2 Dr2 Df2 Dr2 Df2 Dr2
R0 R0 R0
Df3 Dr3 Df3 Dr3 Df3 Dr3

Dfn Drn Dfn Drn Dfn Drn

traps traps filled by carriers

Fig. 3 Proposed model for working mechanism of the artificial synapse. a Schematic diagram for the working mechanism of the memristor. The
current is formed due to the hopping of protons provided CS through functional groups and limited by the traps in rGO nanosheets. b, c, d show a
proposed model consisting of a number of face-to-face diodes for the memristor working in zero bias, forward bias, and reverse bias conditions,
respectively

synaptic weight of an artificial synapse, which is defined pulse trains with a pulse width larger than 50 ms40.
by the conductance of the memristor, can be modulated Therefore, a pulse width of 30 ms is selected to investi-
by the successive stimuli of external applied pulses. gate other synaptic behaviors. The SRDP was examined
Therefore, first, the plasticity characteristics of the arti- by varying the frequency and amplitude of the pulses.
ficial synapse by 100 consecutive pulses with an ampli- For consecutive pulses with an amplitude of 2 V and a
tude of 2 V and pulse widths of 5, 10, 20, 30, and 50 ms width of 30 ms, there is almost no increase in the
were investigated, as presented in Fig. 4a. In the case of response current at a given frequency of 1.1 Hz. How-
100 successive stimuli with a pulse width smaller than ever, when the frequency increases to 9 Hz, the current
5 ms, no obvious increase in the response current, that is, increase rate is larger than that of previous conditions
synaptic potentiation, is observed. With increasing pulse with lower frequencies, as illustrated in Fig. 4b. This
width, an enhancement in synaptic potentiation is behavior is related to the competition between the
obtained. However, when the pulse width is larger than relaxation time of the artificial synapse and the interval
50 ms, some fluctuations in the response current occur. between two pulses. When the amplitude of pulses
This phenomenon is probably attributed to different increases from 1 to 2 V, the current increase rate
facilitation ratios caused by the variations in relaxation becomes even larger, as illustrated in Fig. 4c, which
time for each stimulus, which is dependent on the con- corresponds to the filling of traps with higher energy
ductance mechanism, when the device is stimulated by levels, as discussed above.
Lu et al. Microsystems & Nanoengineering (2020)6:84 Page 7 of 10

a 4.0
1600 5 ms 10 ms 20 ms 30 ms 50 ms
3.5
1400
3.0
1200
Current (nA)

2.5

Voltage (V)
1000 0 50 100 150 200 0 50 100 150 200 250 0 75 150 225 300 0 100 200 300 400 0 100 200 300 400 500
Time (ms) Time (ms) Time (ms) Time (ms) Time (ms)
2.0
800

600 1.5

400 1.0

200 0.5

0 0.0
0 50 100 200 250 300 400 450 500 600 650 700 800 850 900
Pulse number

b 600 c 1000 2.5

Voltage Voltage
550 2.5 900
Current Current
500
800 2.0
450 1.1 Hz 2.3 Hz 9 Hz
2.0
700
400
Current (nA)

Current (nA)
Voltage (V)

Voltage (V)
350 600 1.5
1.5
300 500
250 400 1.0
1.0
200
300
150
0.5 200 0.5
100
50 100
0 0.0 0 0.0
0 10 20 30 40 60 65 70 75 80 102 105 0 50 100 150 200 250 300
Time (s) Pulse number

Fig. 4 Synaptic behaviors of the artificial synapse. a Plasticity characteristics of the artificial synapse by 100 consecutive pluses with pulse widths
of 5, 10, 20, 30, and 50 ms. b, c demonstrate the SRDP plasticity behavior of the artificial synapse

Emulation of learning and forgetting process shown in Fig. 5a. The STP is replicated in Supplementary
If the external stimuli are removed after the potentia- Fig. S11 of the Supplementary materials, indicating the
tion of artificial synapses, the conductance will decay to its stability of the artificial synapse. In addition, STP can be
initial low value, dependent on the number of applied trained into a permanent change in the conductance state
pulses. If the conductance decays to its initial value called LTP. In electronic devices, the transition from STP
rapidly (within several seconds), it is termed STP8. The to LTP is monitored by the decrease in conductance
process is related to the release of trapped carriers in the during the decay time. As shown in Fig. 5b, the con-
trapping sites, and the relaxation follows the exponential ductance is maintained at a higher value than the initial
law with time described by the equation below31,41: state for a long time (more than 10 min) after the removal
  β  of the applied pulses, when the pulse width of stimuli
t increases to 500 ms. This phenomenon represents the
I ¼ ðI0  I1 Þ exp  þ I1 ð1Þ
τ transition from STP to LTP.
In biological systems, as the Ebbinghaus forgetting
where t is the relaxation time, τ is the time constant, I0 is curve describes in the early period, information is lost
the initial current level, I∞ is the current level at infinite quickly when it is initially learned, and the rate at which
time, and β is the stretch index, with a value between these memories are lost is dependent on the frequency the
0 and 1. information is reviewed42. In addition, the multistore
In this research, an STP behavior with a relaxation time model of human memory proposed by Richard C.
constant of 10.5 s is achieved after 100 consecutive pulses Atkinson and Richard Shiffrin in 1958 states that new
with an amplitude of 2 V and a pulse width of 30 ms, as information coming from outside the environment will be
Lu et al. Microsystems & Nanoengineering (2020)6:84 Page 8 of 10

a 2.5 b
30 ms
0.7
Applied voltage (V)

2.0

1.5 0.6
1.0

Conductance (µS)
0.5
0.5
0.4
0.0
0.33
STP 0.3
Conductance (µS)

0.30
relaxation time LTP
0.27
0.2
0.24
0.21
0.1
0.18
0.15
0.0
–9000 –6000 –3000 0 3000 6000 9000 12000 0 30 60 90 225 300 570 600
Time (ms) Time (s)

c d
Learning Learning
w/o 30 times w/o 30 times

60 times 100 times 60 times 100 times

Forgetting Forgetting
1s 5s 10 s 30 s

10 s 15 s 5 mins 10 mins

Fig. 5 Emulation of learning and forgetting process with the designed artificial synapse. a, b show the STP and LTP behavior of the artificial
synapse. A demonstration of STM and LTM of the artificial synapses by putting letters “LH” and “X” into the matrix is presented in (c) and (d)

selected as short-term memory (STM) and long-term mimicking both STP and LTP and its transition process,
memory (LTM). Both studies indicate that the retention has the potential for the realization of STM and LTM and,
of the memory level is dependent on the training times in turn, the construction of a neuromorphic network. To
and similar to the transition from STP to LTP in mem- prove the ability of STM and LTM emulation, a simula-
ristive devices demonstrated in this research43. Therefore, tion was implemented by inputting the letters “LH” and
the artificial synapse developed in this study, capable of “X” into the memristor matrix, as shown in Fig. 5c. “LH”
Lu et al. Microsystems & Nanoengineering (2020)6:84 Page 9 of 10

was stimulated by 100 consecutive pulses with an ampli- ammonia solution (28 wt% in water, Sinopharm, China)
tude of 2 V and a pulse width of 30 ms, and “X” was and 4 μL hydrazine solution (35 wt% in water, Wuxi
trained with the same number of pulses but for a longer Zhanwang, China) for chemical reduction of GO. The
pulse width of 500 ms. Both “LX” and “X” were gradually weight ratio of hydrazine to GO was ~6:10. After fully
memorized by the devices with increasing spike number, mixing, the flask was immersed into an oil bath at 90 °C
which represented the dynamic learning process. After for 1 h with magnetic stirring and cyclic cooling water.
learning 100 times, the applied stimuli were removed, and After complete reaction in the flask, an rGO dispersion
the forgetting process was investigated. For the devices was obtained, which was used as the precursor for
stimulated by the pulses with a shorter pulse width, that channel materials of memristors without any further
is, 30 ms, the forgetting process was faster than that with a treatment.
pulse width of 500 ms, and therefore, the “LX” was almost
completely forgotten in 10 s, which was consistent with Fabrication of the memristors on flexible substrates
the relaxation time shown in Fig. 5a. Regarding the A diluted polyimide (PI) solution was spin-coated on
devices trained by pulses with a longer pulse width of cleaned glass and cured at 300 °C for 1 h. Then, electrodes
500 ms, the letter “X” can be memorized for more than (Ti/Au) were deposited by sputtering and patterned using
10 min, which can be considered LTM. The above a photolithograph and lift-off method on the PI sub-
demonstration indicates that basic learning and forgetting strates. Next, an aqueous rGO dispersion was drop-cast
behaviors, which are performed by a complex neural net- on the sample, and the active regions were defined using
work in the human brain, can be successfully emulated photolithography and O2 plasma etching. To reduce the
using the memristor matrix44. In addition, the continuous contact resistance between the electrodes and rGO
conductance states of the device modulated by the pulse nanosheets, the sample was annealed at 150 °C for 90 min
stimuli provide a solid foundation for implementing the in ambient argon. Both IV characteristics of the device
neural network at the hardware level. All the results imply before and after annealing and the isolation between
that memristor-based artificial synapses can be regarded as devices are illustrated in Supplementary Fig. S2. After-
building blocks of artificial neural networks, representing ward, the 2% CS solution (5 μL) was dropped on top of the
an essential step toward neuromorphic applications. active regions. Finally, the PI film was peeled off from the
glass. The detailed fabrication process is demonstrated in
Conclusions Supplementary Fig. S1 of the Supplemental materials.
In conclusion, a bioinspired double-layered memristor
matrix was fabricated with rGO nanosheets prepared by Device characterizations
chemical redox reactions and natural CS biopolymers. The size and surface morphology of rGO nanosheets
Distinct from previously reported memristors, the mem- were investigated by AFM (Bruker Dimension, USA) and
ristive device developed in this research is inspired by the scanning electron microscopy (SEM, Hitachi Regulus,
working principle of the G-protein-linked receptors of Japan) at an accelerating voltage of 5 kV. The thickness of
biological cells, and the underlying mechanism is based on the rGO nanosheets could also be determined from the
the interaction between rGO and CS. Studies on the phy- AFM results. FTIR spectra were collected via a Thermo
sical and electrical properties of the devices and materials Nicolet iN 10 spectrometer (Thermo-Fisher, USA) with the
prove that the memristive behavior is due to the interaction laser operating at 77 K in a liquid nitrogen environment.
between protons provided by CS and the defects and XPS analysis was carried out using a Thermo Scientific
functional groups in the rGO nanosheets. Learning- EXCALAB 250 XI system (Thermo-Fisher, USA). Raman
forgetting processes based on the STP and LTP proper- spectra of the materials were collected using a Horiba-JY
ties of the devices were demonstrated, which showed great Labram HR system (HORIBA JY, France). A Perkin Elmer
potential in emulating the functions of the human brain. Lambda 25 spectrometer (Perkin Elmer, USA) was used to
Overall, artificial synapses offer a new avenue for the con- investigate the absorbance of rGO nanosheets. All the
struction of neuromorphic computation systems based on electrical measurements were performed at 25 °C and 40%
devices with bionic structures and working mechanisms. humidity using an Agilent B1500A semiconductor para-
meter analyzer (Keysight, USA).
Materials and methods
Synthesis of rGO aqueous solutions
Acknowledgements
Graphene oxide (GO) was exfoliated from natural gra- The authors acknowledge the funding support from the National Key R&D
phite (350 mesh, Aladdin, China) by a modified Hummers Program of China (No. 2018YFB1304700), the National Natural Science
method45. Then, a 10 μg/mL GO dispersion in aqueous Foundation of China (No. 61574163, No. 61801473), the Science Foundation for
Distinguished Young Scholars of Jiangsu Province, China (Nos. BK20170008
solution was prepared by ultrasonication for 1 h. The and BK20160011), the National Natural Science Foundation of China (No.
obtained GO aqueous solution was mixed with 74 μL 61801473), and the NANO-X Workstation scientifically supported this research.
Lu et al. Microsystems & Nanoengineering (2020)6:84 Page 10 of 10

Author details 18. Wu, C., Kim, T. W., Choi, H. Y., Strukov, D. B. & Yang, J. J. Flexible three-
1
i-Lab, Key Laboratory of Multifunctional Nanomaterials and Smart Systems, dimensional artificial synapse networks with correlated learning and trainable
Suzhou Institute of Nano-Tech and Nano-Bionics (SINANO), Chinese Academy memory capability. Nat. Commun. 8, 752 (2017).
of Sciences (CAS), 398 Ruoshui Road, 215123 Suzhou, PR China. 2Department 19. Xiao, Z. & Huang, J. Energy-efficient hybrid perovskite memristors and synaptic
of Health and Environmental Sciences, Xi’an Jiaotong Liverpool University, 111 devices. Adv. Electron. Mater. 2, 1600100 (2016).
Ren’ai Road, 215123 Suzhou, PR China. 3Center for Excellence in Brain Science 20. Zhu, L. Q., Sun, J., Wu, G. D., Zhang, H. L. & Wan, Q. Self-assembled dual in-
and Intelligence Technology, Chinese Academy of Sciences, 200031 Shanghai, plane gate thin-film transistors gated by nanogranular SiO2 proton conductors
PR China for logic applications. Nanoscale 5, 1980–1985 (2013).
21. Yeh, T. F., Chan, F. F., Hsieh, C. T. & Teng, H. Graphite oxide with different
Author contributions oxygenated levels for hydrogen and oxygen production from water under
Q.L., F.S., and T.Z. designed the project and experiments. Q.L., L.L., T.Z., and L.L. illumination: the band positions of graphite oxide. J. Phys. Chem. C 115,
prepared the paper. Y.W., M.H., Z.W., T.Z., and S.W. analyzed and discussed the 22587–22597 (2011).
data. All authors reviewed the paper. 22. Chang, H. et al. Regulating infrared photoresponses in reduced graphene
oxide phototransistors by defect and atomic structure control. ACS Nano 7,
Conflict of interest 6310–6320 (2013).
The authors declare that they have no conflict of interest. 23. Kuzum, D., Yu, S. & Wong, H. S. Synaptic electronics: materials, devices and
applications. Nanotechnology 24, 382001 (2013).
24. Drachman, D. A. Do we have brain to spare? Neurology 64, 2004–2005 (2005).
Supplementary information accompanies this paper at https://doi.org/
25. Tu, N. D. K., Choi, J., Park, C. R. & Kim, H. Remarkable conversion between n-and
10.1038/s41378-020-00189-z.
p-type reduced graphene oxide on varying the thermal annealing tempera-
ture. Chem. Mater. 27, 7362–7369 (2015).
Received: 11 May 2020 Accepted: 11 June 2020 26. Krishnamoorthy, K., Veerapandian, M., Yun, K. & Kim, S. J. The chemical and
structural analysis of graphene oxide with different degrees of oxidation.
Carbon 53, 38–49 (2013).
27. Qiu, H. et al. Hopping transport through defect-induced localized states in
molybdenum disulphide. Nat. Commun. 4, 2642 (2013).
References 28. Osman, Z. & Arof, A. K. FTIR studies of chitosan acetate based polymer elec-
1. Diorio, C. et al. Neuromorphic electronic systems. IEEE Trans. Electron. Devices trolytes. Electrochim. Acta 48, 993–999 (2003).
43, 1972–1980 (1996). 29. Lagaron, J. M., Fernandez-Saiz, P. & Ocio, M. J. Using ATR-FTIR spectro-
2. Rountree, C., Raghunathan, A., Troy, J. & Saggere, L. Prototype chemical scopy to design active antimicrobial food packaging structures based
synapse chip for spatially patterned neurotransmitter stimulation of the retina on high molecular weight chitosan polysaccharide. J. Agr. Food Chem.
ex vivo. Microsyst. Nanoeng. 3, 17052 (2017). 55, 2554–2562 (2007).
3. Sarkar, D. et al. Mimicking biological synaptic functionality with an indium 30. Aider, M. Chitosan application for active bio-based films production and
phosphide synaptic device on silicon for scalable neuromorphic computing. potential in the food industry. LWT-Food Sci. Technol. 43, 837–842 (2010).
ACS Nano 12, 1656–1663 (2018). 31. Sturman, B., Podivilov, E. & Gorkunov, M. Origin of stretched exponential
4. Xie, D. et al. Coplanar multigate MoS2 electric-double-layer transistors for relaxation for hopping-transport models. Phys. Rev. Lett. 91, 176602 (2003).
neuromorphic visual recognition. ACS Appl. Mater. Interfaces 10, 25943–25948 32. Van Der Zande, A. M. et al. Grains and grain boundaries in highly crystalline
(2018). monolayer molybdenum disulphide. Nat. Mater. 12, 554–561 (2013).
5. Wang, Z. et al. Memristors with diffusive dynamics as synaptic emulators for 33. Odagawa, A. et al. Colossal electroresistance of a Pr0.7Ca0.3MnO3 thin film at
neuromorphic computing. Nat. Mater. 16, 101–108 (2017). room temperature. Phys. Rev. B 70, 224403 (2004).
6. Wong, H. S. & Salahuddin, S. Memory leads the way to better computing. Nat. 34. Shang, D. S. et al. Effect of carrier trapping on the hysteretic current-voltage
Nanotechnol. 10, 191–194 (2015). characteristics in Ag/La0.7Ca0.3MnO3/Pt heterostructures. Phys. Rev. B 73,
7. Afifi, A., Ayatollahi, A. & Raissi, F. STDP implementation using memristive 245427 (2006).
nanodevice in CMOS-nano neuromorphic networks. IEICE Electron. Expr. 6, 35. Yoon, J. H. et al. Pt/Ta2O5/HfO2-x/Ti resistive switching memory competing
148–153 (2009). with multilevel NAND flash. Adv. Mater. 27, 3811–3816 (2015).
8. Kim, S. G., Han, J. S., Kim, H., Kim, S. Y. & Jang, H. W. Recent advances in 36. Wang, L. et al. Long-term homeostatic properties complementary to
memristive materials for artificial synapses. Adv. Mater. Technol. 3, 1800457 hebbian rules in cupc-based multifunctional memristor. Sci. Rep. 6, 35273
(2018). (2016).
9. Padovani, A., Woo, J., Hwang, H. & Larcher, L. Understanding and opti- 37. Schmidt, H. et al. Transport properties of monolayer MoS2 grown by chemical
mization of pulsed SET operation in HfOx-based RRAM devices for neu- vapor deposition. Nano Lett. 14, 1909–1913 (2014).
romorphic computing applications. IEEE Electron. Device Lett. 39, 672–675 38. Atluri, P. P. & Regehr, W. G. Determinants of the time course of facilitation at
(2018). the granule cell to Purkinje cell synapse. J. Neurosci. 16, 5661–5671 (1996).
10. Shi, J., Ha, S. D., Zhou, Y., Schoofs, F. & Ramanathan, S. A correlated nickelate 39. Lamprecht, R. & LeDoux, J. Structural plasticity and memory. Nat. Rev. Neurosci.
synaptic transistor. Nat. Commun. 4, 1–9 (2013). 5, 45–54 (2004).
11. De Burgt, Y. V. et al. A non-volatile organic electrochemical device as a low- 40. Gao, B. et al. Ultra-low-energy three-dimensional oxide-based electronic
voltage artificial synapse for neuromorphic computing. Nat. Mater. 16, synapses for implementation of robust high-accuracy neuromorphic com-
414–418 (2017). putation systems. ACS Nano 8, 6998–7004 (2014).
12. Jiang, J. et al. 2D MoS2 neuromorphic devices for brain‐like computational 41. Zhu, L. Q., Wan, C. J., Guo, L. Q., Shi, Y. & Wan, Q. Artificial synapse network on
systems. Small 13, 1700933 (2017). inorganic proton conductor for neuromorphic systems. Nat. Commun. 5, 3158
13. Thomas, A. Memristor-based neural networks. J. Phys. D Appl. Phys. 46, 093001 (2014).
(2013). 42. Ebbinghaus, H. Memory: A Contribution to Experimental Psychology (trans:
14. Waser, R., Dittmann, R., Staikov, G. & Szot, K. redox-based resistive switching Ruger, Henry A. & Bussenius, Clara E.) (Teachers College, Columbia University,
memories-nanoionic mechanisms, prospects, and challenges. Adv. Mater. 21, New York, 1913).
2632–2663 (2009). 43. Bower, G. in The Psychology of Learning and Motivation: Advances in Research
15. Bagdzevicius, S., Maas, K., Boudard, M. & Burriel, M. Interface-type resistive and Theory. Vol. 1999 (eds Peche, D. & Zwaan R.) 125–173 (Academic Press,
switching in perovskite materials. J. Electroceram. 39, 157–184 (2017). New York, 2002).
16. Campbell, A. P. & Smrcka, A. V. Targeting G protein-coupled receptor signalling 44. Jo, S. H. et al. Nanoscale memristor device as synapse in neuromorphic sys-
by blocking G proteins. Nat. Rev. Drug Discov. 17, 789 (2018). tems. Nano Lett. 10, 1297–1301 (2010).
17. Lefkowitz, R. J. A brief history of G-protein coupled receptors (Nobel Lecture). 45. Hummers, W. S. Jr. & Offeman, R. E. Preparation of graphitic oxide. J. Am. Chem.
Angew. Chem. Int. Ed. 52, 6366–6378 (2013). Soc. 80, 1339–1339 (1958).