Cellulose nanocrystal based bio‐memristor as a green artificial synaptic device for neuromorphic computing applications

Cellulose Nanocrystal Based Bio‐Memristor as a Green Artificial Synaptic Device for Neuromorphic Computing Applications See discussions, stats, and author profiles for this publication at https www researchgate netpublication355430751 Cellulose Nanocrystal Based Bio‐Memristor as a Green Artificial Synaptic Device for Neuromorphic Computing Applications Article in Advanced Materials Technologies October 2021 DOI 10 1002admt 202100744 CITATIONS 4 READS 153 11 authors, including Some of the au.

Cellulose Nanocrystal Based Bio‐Memristor as a Green Artificial Synaptic Device for Neuromorphic Computing Applications

Article  in  Advanced Materials Technologies · October 2021

DOI: 10.1002/admt.202100744

CITATIONS

4

READS

153

11 authors, including:

Some of the authors of this publication are also working on these related projects:

Development of Resistive Random Access Memory (RRAM) Devices View project

corrosion, metallurgy, surface coating View project

Tassawar Hussain

KU Leuven & IMEC.be

14PUBLICATIONS    95CITATIONS    

SEE PROFILE

Haider Abbas

Nanyang Technological University

31PUBLICATIONS    378CITATIONS     SEE PROFILE

Turgun Boynazarov

Sejong University

3PUBLICATIONS    6CITATIONS    

SEE PROFILE

Boncheol Ku

Hanyang University

16PUBLICATIONS    389CITATIONS     SEE PROFILE

www.advmattechnol.de ReseaRch aRticle

Cellulose Nanocrystal Based Bio-Memristor as a Green

Artificial Synaptic Device for Neuromorphic Computing

Applications

Tassawar Hussain, Haider Abbas, Chulmin Youn, Hojin Lee, Turgun Boynazarov,

Boncheol Ku, Yu-Rim Jeon, Hoonhee Han, Jong Hyeon Lee, Changhwan Choi,*

and Taekjib Choi*

DOI: 10.1002/admt.202100744

sustainable materials and green elec-tronics.[1,2] In particular, eco-friendly, renewable, bio-compatible, and bio-degra-dable natural biomaterials are potential alternatives to emerging green electronics that can reduce harmful electronic waste The use of natural biomaterials greatly aids the sustainable development of the electronics industry.[2–5] In fact, natural biomaterials such as (e.g., silk fibroin, spider silk, cellulose, chitosan, etc.) have been widely employed in a variety

of green-electronic systems, such as energy storage devices, biosensors, and bio-memristor, benefiting from unique biological structure, biocompatibility, biodegradability, transparency, and flex-ibility.[6–8] However, biomaterials often exhibit degradable and unstable perfor-mance due to their weak electrical func-tion Therefore, recent investigations into biocomposites containing one or more naturally-derived content combined with other functional materials have shown the improved performance of bioelectronic elements.[2,9–11] In addition, increasing demand for green information storage and computation technology has acceler-ated the rapid development of nonvolatile memory devices based on biocomposite materials Among the emerging nonvolatile memory technologies, resistive switching random access memory (RRAM), in which the resistance states can be switched between the high resistance state (HRS) and the low resistance state (LRS) by applying an electric field, has

Nanocomposites based on biomaterials are promising candidates for emerging

green- electronics benefiting from environment-friendly, renewable,

biocompat-ible, and biodegradable resources for sustainable research and development

Especially, the application of biocomposites-based memristor for simulating

artificial synapses called bio-memristor has further facilitated the progress of

ecologically benign bioelectronics In this study, the authors present that the

environment-friendly nanocomposites films, consisting of Ag nanoparticles

and cellulose nanocrystal (CNC)-based bio-memristor with excellent bipolar

resistive switching behavior can perform the artificial bio-synaptic emulation

with continuous resistance modulation for memory storage and

neuromor-phic computing applications The bio-memristor exhibits a large resistive

and reliable switching characteristics through the electrochemical formation/

rupture of Ag metallic filaments within the nanocomposite layer The device

presents coexistence of digital and analog switching properties favorable for

both nonvolatile digital memory and neuromorphic computing applications By

applying appropriate pulse stimulations to the device, the authors demonstrate

biological synaptic functions, including long-term potentiation/depression,

spike-rate-dependent plasticity, excitatory post-synaptic current, paired-pulse

facilitation, and paired-pulse depression Thus, this CNC-based bio-memristor

as an effective artificial synaptic device is beneficial towards the realization of

green-electronics and bio-inspired neuromorphic systems.

T Hussain, C Youn,[+] H Lee, T Boynazarov, T Choi

Hybrid Materials Research Center and Department of Nanotechnology

and Advanced Materials Engineering

Sejong University

Seoul 143-747, South Korea

E-mail: tjchoi@sejong.ac.kr

The ORCID identification number(s) for the author(s) of this article

can be found under https://doi.org/10.1002/admt.202100744

1 Introduction

Global concerns over environmental issues from growing

electronic waste along with a tremendous production of

sil-icon-based electronics have stimulated extensive research into

[+]Present address: Advanced Textile R&D Department, Korea Institute

of Industrial Technology, Ansan-si, 15588, South Korea

H Abbas, B Ku, Y.-R Jeon, H Han, C Choi Division of Materials Science and Engineering Hanyang University

Seoul 04763, South Korea E-mail: cchoi@hanyang.ac.kr

J H Lee Department of Chemistry Catholic University of Korea Bucheon, Gyeonggi 420-743, South Korea

gained intensive research interest due to its simple structure

with an insulating layer sandwiched between two conducting

electrodes.[12–15] This simple capacitor-like structure can provide

large-scale integration and high-density storage through the

fabrication of a 3D stacked crossbar array.  It is worth noting

that as a suitable material for implementing green nonvolatile

memory devices, biocomposites have proven resistive switching

devices with excellent performance Celano et  al represented

biodegradable RRAM devices consisting of a

nanocellulose-based resistive switching layer and a nano-paper substrate,

which exhibit bipolar resistive switching as well as multilevel

storage.[16] By using egg albumen as a switching layer, the

water-soluble and flexible RRAMs were reported with long

retention time and fast switching speed, in which the switching

performance was improved by the hybridization of metal

nano-particles (NPs).[17] On the other hand, besides RRAM based on

biocomposites, biomaterials such as lignin, carrageenan, and

collagen have been designed as bio-memristor for simulating

biological synaptic functions via analog resistive switching

behaviors.[18,19] Moreover, biocomposites-based RRAM is one

of the promising candidates for next-generation bio-memristor

with advantages of low power consumption, compatibility,

reli-ability, high switching ratio, and high storage density.[20] In

particular, bio-memristor composed of bio-nanocomposites

with both digital and analog switching characteristics is highly

desirable to realize biorealistic synaptic devices for

neuromor-phic computing that can overcome the limitations of von

Neu-mann computing.[21] However, there are very limited reports on

the co-existence of digital and analog switching bio-memristors

comprising biodegradable, eco-friendly, and green dielectric

materials as a primary element

A number of artificial synaptic devices have been proposed,

where the main switching layer is based on inorganic

mate-rials.[13,22–25] For example, Ohno et  al.[26] reported synaptic

behavior of Ag/Ag2S/nanogap/Pt device controlled by adjusting

repetition time of input pulses Serb et al.[27] reported synapses

with gradual, intrinsic, and multilevel resistive switching in

TiO2-based memristor Similarly, Wang et  al.[28] investigated

synaptic behavior of FeOx based memristor by manipulating

different analog characteristics via controlling compliance

cur-rent during the electroforming process However, these

inor-ganic metal oxide-based devices are not environment-friendly,

and new low-cost biodegradable organic materials should be

replaced with conventional materials in electronic synaptic

devices for implantable and wearable biomedical

applica-tions.[18,29–31] In this regard, Park et  al.[18] reported artificial

synapses in lignin-based memristor devices, lignin is

bio-degradable and the abundant renewable material extracted

from plants Li et al.[32] reported a bio-memristor based synaptic

device with a hybrid structure of Ag/HfO2/BSA:Au/Pt They

used hybrid biomaterial of bovine serum doped with nano-gold

(BSA:Au) and HfO2 double layers as insulating switching layer

materials Similarly, Li et al.[33] reported synaptic plasticity in a

biodegradable organic conducting polymer of PEDOT:PSS with

Ag/PEDOT:PSS/Ta device structure

Here, we report cellulose nanocrystal (CNC)-based

nanocom-posite as a bio-memristor for both memory storage and synaptic

application Cellulose nanofiber (CNF), one of the abundant and

emerging green materials derived from plant cells, is the focus

of modern research on a variety of novel high-tech material applications.[34–36] CNF, having a nanometric diameter below

50 nm and micrometric length with alternating crystalline and amorphous structure, provide excellent mechanical properties, including high stiffness, good tensile strength, and high surface area The surfaces of CNFs with the abundant hydroxyl (-OH) functional groups also provide greater flexibility for modifica-tion and incorporamodifica-tion of various other funcmodifica-tional groups for the desired applications.[34,37,38] We have successfully utilized CNCs into fabricating a bio-memristor for both memory storage (bio-RRAM), and neuromorphic computing application as an artificial synaptic device.[16] CNFs were modified by TEMPO oxi-dation and periodate oxioxi-dation processes for the hybridization of silver (Ag+) ions along its 1D CNC structure Our bio-memristor showed the coexistence of both digital and analog switching characteristics Besides a stable and reliable bipolar switching behavior, the features of bio-synaptic functioning such as long-term potentiation (LTP), long-long-term depression (LTD), spike rate-dependent plasticity (SRDP), specifically paired-pulse facili-tation (PPF), paired-pulse depression (PPD), and post-tetanic potentiation (PTP) are demonstrated using analog-voltage bias

to consecutive conductance modulation Ionic excitatory post-synaptic current (EPSC) is performed with the decay function of silver conducting filament Through the ultralow operation volt-ages for both digital and analog switching, our bio-memristors are beneficial for highly efficient synaptic devices

2 Results and Discussion

CNF insulating material was used as a precursor in our mem-ristor device which was modified for growing in situ AgNPs along with its chemical structure CNF was subjected to suc-cessive TEMPO-mediated oxidation and Periodate oxidation after which AgNPs were in situ-grown on it by tollen’s

reac-tion Figure 1a schematically illustrates the chemical reactions,

modification steps and NPs of silver grown on CNCs and the detailed method is explained in the synthesis section CNFs are fiber bundles with alternative crystalline and amorphous regions, by TEMPO oxidation we only get individual TCNC with carbonyl and aldehyde functional groups, and by perio-date oxidation, we tend to increase the dialdehyde functional groups in the obtained CNCs.[39–42] The thin rod-like structure

of nanocrystals can be observed by the TEM image of our pre-pared Tempo-oxidized CNCs as shown in Figure 1b To confirm the uniform attachment of AgNPs on CNCs a small amount of the final solution is drop-casted on FTO and dried at room tem-perature, after which it was analyzed with EDX for its elemental mapping, which confirms the uniform distribution of AgNPs along the coating as shown in Figure 1c

To investigate the chemical properties of CNF, surface anal-ysis of pristine CNF, TCNC, and dialdehyde-TCNC (DAC) was

performed using FTIR-ATR analysis as shown in Figure 2a The

FTIR results of pristine CNF can be clearly distinguished from TCNC and DAC, where the infrared spectra of both TCNC and DAC have the negative functional groups stretching such as (CH, CC, COC, CO, CH, and OCH) in addition

to (OH) functional group of pristine CNF.[43–45] In all samples, the broad bands around 3100–3600 cm−1 region are assigned

www.advancedsciencenews.com www.advmattechnol.de

to OH stretching vibrations, and in the final DAC sample,

the band around 2970 cm−1 refers to CH stretching

vibra-tions, around 1635 cm−1 it can refer to the offset of absorbed

water OH band or a CC aromatic band, at 1450 cm−1 it can

attribute to HCH and OCH in-plane bending vibrations,

and around 1380 cm−1 it is the C-H deformation vibration.[43,46]

The stretching at 1085 cm−1 refers to the COC stretching of

glucose and at 1045 cm−1 it reflects to the CO stretching of

pyranose ring vibrations All these functional groups belong to

TCNC and DAC crystals confirming successive

TEMPO-oxida-tion and Periodate oxidaTEMPO-oxida-tion The increased peak intensity of

hemiacetal at 881 cm−1 and CH stretching in the FTIR results

of final DAC demonstrates that the TCNC is oxidized to DAC

having a higher amount of aldehyde functional groups.[47,48]

These functional groups were introduced by the discussed

oxi-dation reactions so that the AgNPs can be easily and successfully

attached along the length of CNCs at the aldehyde sites.[46,49,50]

Free silver (Ag+) cations of the tollen’s reagent [Ag(NH3)2]+OH−

reacts with an aldehyde (CHO) and carbonyl (CO)

functional groups and is precipitated at these sites as NPs which

later facilitates the formation of conducting filaments.[51,52] The

particles zeta-potential (ζ) of parent material and after each

modification steps were measured and a noticeable expected

trend in electro-negativity shift is observed as shown in

Figure 2b The mean value of zeta-potential is shifted towards

a more electronegative value from CNF ≈ (−33.2 ± 5.27 mV)

to TCNC ≈ (−71.3 ± 4.80 mV) when CNF is transformed

because of the evident reason of replacing some of the (OH)

functional groups with more electronegative functional groups

of carboxylate by tempo-oxidation along nanocrystals After the periodate oxidation the mean value of zeta-potential (ζ) is

shifted to DAC ≈ (−63.8 ± 5.28 mV) which is because of the increase in the introduction of aldehyde group so, the replace-ment of carboxylate group with a less electronegative aldehyde functional group along TCNC length And at the final stage as

a result of tollen’s reaction, the free silver-ions (Ag+) precipita-tion at the aldehyde sites along the nanocrystals, which further shifts the zeta-potential towards a less electronegative value of AgNPs-TCNC ≈ (−44.1 ± 5.56 mV) This shift in zeta-potential (ζ) value after tollen’s reaction also indicates the successful

attachment of free silver ions along the nanocrystals

The schematic presentation of the complete memristor devices and the XRD characteristics of the switching layer is

presented in Figure 3 Figure  3a shows the schematic

struc-tures of the biological synapse and the memristor device with

Ag and FTO top and bottom electrodes and AgNPs-TCNC (nan-opaper) as a sandwiched switching layer of MIM configuration The structure of our bio-memristor correlates with that of the biological synapse The crystallinity of the coated AgNPs-TCNC insulating layer on the FTO substrate prior to the deposition

of the top electrode was observed by XRD analysis The XRD graph of AgNPs-TCNC film is indexed with standard JCPDS patterns of Ag (JCPDS: 04-0783), and cellulose Iβ as shown

with the diffraction peaks of 2θ at 15.3⁰ (110), 19.3⁰ (110), and 22.8⁰ (200).[46,53,54] The crystallographic planes of cubic Ag in the Ag-TCNC nanocomposite were identified by the observed diffraction peaks of 2θ at 38.1⁰ (111), and 44.5⁰ (200), further

Figure 1 a) Schematic illustration of chemical modification reactions, and attachment of Ag-NPs along CNC structure, b) The TEM image of prepared

Tempo-oxidized CNCs from CNF, and c) EDS elemental mapping analysis of the deposited Ag-TCNC thin film on FTO showing uniform distribution

of Ag-NPs on nanocrystals

confirming the crystalline nature of attached metallic silver

on the CNC surfaces.[46,55] As discussed in the introduction

section, our bio-memristor device showed a repeatable dual

switching behavior of digital switching and analog switching

with the coexistence of both types of switching The switching

behavior could be controlled by controlling the switching

volt-ages The digital switching and analog switching characteristics

could be achieved by tuning the switching voltages above ±0.15

and below ±0.10 V, respectively The digital switching

character-istics are utilized for digital memory applications Whereas, the

analog switching behavior is exploited to emulate the important

biological synaptic functions of the brain for neuromorphic

computing

2.1 Digital Switching for Memory Applications

The fabricated memristor device with 4.37 µm thickness of the

insulating layer and 100 µm2 size silver top electrode of ≈150 nm

thickness was subjected to measurement at room

tempera-ture and atmospheric pressure The FTO bottom electrode was

grounded, whereas the silver top electrode was connected with

the tungsten probe tip to control the applied bias voltage during

the current–voltage (I–V) measurements We applied

alter-native positive and negative voltage sweeps from 0 to ±0.5 V

for the digital switching A high-voltage electroforming

pro-cess was needed to initiate the switching The electroforming

process is a phenomenon of resistive memories where the ini-tial conductive filaments begin to grow by the progression of oxidized metal ions migration within the insulating medium by the application of applied voltage.[56] As a result of the contin-uous growth of these filaments, the top and bottom conducting electrodes connect together by virtue of which the insulating high resistance (OFF) state changes to conductive low resist-ance (ON) state It is usually a preparatory step in most of the memristors before the actual switching cycles and voltages for

SET (ON) and RESET (OFF) are observed SET voltage (VSET)

is the value of the voltage at which the current level abruptly changes from a high resistance value to a low resistance value transforming the device from OFF state to ON state, and the sandwiched insulating medium between the top and bottom conducting electrode can now conduct current by the growth

of these conductive filaments within the medium While the

RESET voltage (VRESET) represents the value of voltage at which the opposite phenomenon occurs and the conducting fila-ments break down and no more high level of current can pass through the insulating medium and its resistance state abruptly changes from lower resistance values to high resistance value which means the device is again transformed to OFF state The digital switching characteristics of the device and its repeatability and reliability for nonvolatile memory

applica-tions are presented in Figure 4 The device needed an initial

electroforming process to initiate the resistive switching The

I–V curve for the forming process is shown in Figure S1,

Sup-porting Information The forming voltage was found to be

≈−1.6 V Figure  4a illustrates switching cycles after the initial forming process The change of HRS to LRS after the com-pletion of forming was reversed by applying the positive bias voltage sweep (RESET) Subsequently, after the reversal of forming, the device was SET to LRS by negative voltage sweep (0 to −0.5 V) and RESET back again to HRS by positive voltage sweep (0 to +0.5 V), while this time lower VSET than the first forming voltage was observed By applying the alternative posi-tive and negaposi-tive voltage sweeps from 0 V to ±0.5 V, a sharp digital SET/RESET can be seen To check the effect of compli-ance current, the complicompli-ance current was increased from 10−3

to 10−2 A after 100 switching cycles for the same device and a compliance-free digital switching behavior was observed for the subsequent 150 continuous cycles (Figure S2, Supporting Infor-mation) The switching stability and cycle-to-cycle variability

of our memristor device were evaluated by the data endurance characteristics, as shown in Figure  4b,c The DC endurance characteristics were evaluated up to 200 continuous switching cycles, as shown in Figure 4b The device showed good endur-ance properties, maintaining a decent ON/OFF resistendur-ance ratio of ≈104 at a readout voltage of 0.04 V Moreover, the pulse endurance characteristics were tested to evaluate the endur-ance characteristics of the device for a higher number of cycles The device presented repeatable bipolar switching without any failure for 104 cycles, as presented in Figure  4c The LRS was relatively stable, whereas variations in HRS were observed The variation in HRS is attributed to the stochastic behavior of the conducting filament rupture during the RESET process.[13] For pulse endurance measurement, the SET and RESET pulses with pulse amplitudes of −1 and +1 V were applied, followed by

a small read pulse of 0.05 V The pulse widths of both SET and

Figure 2 a) FTIR-ATR spectra of CNF, TCNC, and DAC materials,

and b) Zeta-Potential behavior of particles of CNF, TCNC, DAC, and

AgNPs-TCNC

www.advancedsciencenews.com www.advmattechnol.de

RESET pulses were 2 ms and that of the read pulse was 0.3 ms

Moreover, the study of the device-to-device variability is a

pre-requisite for mass production and is essential for the practical

applications of the devices The digital switching

characteris-tics of different devices were tested to evaluate the

device-to-device variability for the basic digital switching properties of the

devices The typical I–V characteristics of five different devices

showing the device-to-device switching variability for the digital

switching are presented in Figure S3, Supporting Information

It is noted that all of the devices exhibited digital switching with

negligible variations in the switching voltages and the HRS

and LRS levels This shows that the CNF-based devices exhibit

better device-to-device repeatability for digital switching

con-firming the suitability of the devices for nonvolatile memory

applications Furthermore, the nonvolatile switching behavior

was evaluated with the retention time measurements The

data retention characteristics of the device are presented in

Figure 4d It is noted that both ON and OFF states of the device

were maintained for more than 104 s confirming the nonvolatile

nature of our memristor device The device maintained an ON/

OFF ratio of ≈104 over the time The average VSET and VRESET

values calculated from continuous switching cycles were −0.187

and +0.211 V, respectively For VSET and VRESET the standard

deviation was measured to be 0.0187 and 0.0859, respectively,

which indicates a minimal and acceptable variation of switching voltages during switching cycles The statistical distributions of

operational VSET and VRESET and resistance levels during HRS and LRS are also plotted in Figure 4e,f as the form of a

cumu-lative probability to show the distribution of VSET and VRESET values and HRS and LRS levels recorded during these switching cycles If we see the distribution of the SET and RESET voltages over the repeated DC sweep cycles we can see that the average RESET voltage during the repeated DC sweeps is higher than the SET voltage For the repeated DC sweep cycles, the RESET voltages are distributed from about −0.17–−0.28 V, whereas the SET voltages are distributed as about 0.16–0.42 V The cycle-to-cycle variability in the RESET voltage is higher than that of the SET voltage This higher variation in the RESET voltage is attributed to the stochastic behavior of the conductive filament

rupture during the RESET process The small variation in VSET

can be understood from Figure  4a, displaying that after the initial forming voltage of −1.6 V the VSET of the first cycle was (−0.26 V), which gradually reduced with the increasing cycles to

a certain point The VSET observed for the 1st, 20th, 50th, and 100th cycle was −0.26, −0.23, −0.23, and −0.19 V, respectively, but a random trend was observed during the RESET process A similar trend is observed for cycles measured at a higher com-pliance current of 10−2 A (Figure S2, Supporting Information)

Figure 3 a) Schematic diagrams of the device structure and biological synapse presenting a correlation between the memristor and biological synapse

b) XRD pattern of final AgNPs-TCNC film on FTO

The VSET showed a slightly decreasing trend as the number of

cycles increases The reason for this behavior we believe is the

formation of multiple filaments as a result of continuous

reduc-tion of Ag+ ions as Ag+ + e− → Ag (reduction) and therefore,

the SET voltage for the later repeated cycles is slightly lower

than that of the initial cycles While upon RESET the filaments

rapture at bottom electrode side by the effect of redox reaction,

thus Ag oxidizes again into Ag+ ions as Ag → Ag+ + e−

(oxi-dation) The device was further evaluated to see if the device

will turn on when there are enough switching cycles For this,

we first tested the pulse endurance of the device by repeating

the switching for 104 cycles The device showed a stable

endur-ance without any larger degradation in the device switching

After that, the DC I–V characteristics were measured

imme-diately after 104 repeated cycles The I–V curve following the

104 repeated pulse switching cycles is added in Figure 4a This

shows that the device can still turn on, although the SET voltage

has reduced compared to the initial cycles A slight increase in

the HRS current is also observed Hence, all the

above-men-tioned results satisfy the reliability of our device for nonvolatile

memory application with good endurance and data retention

characteristics, high ON/OFF ratio, and acceptable variation of

operating voltages for digital SET and RESET

2.2 Analog Switching for Biorealistic Synaptic Emulation

The analog switching characteristics of the device were uti-lized to mimic the plasticity functions of the biological syn-apses, which is essential for the realization of neuromorphic computing systems The analog switching behaviors were tested under DC voltage sweeps and pulse measurements

for the emulation of synaptic functions as shown in Figure 5

The DC voltage was applied to the Ag top electrode and controlled carefully to prevent it from direct digital switching, and the sweeping voltage was kept below ±0.10 V As shown

in Figure  5a the I–V curve response can be seen when the

consecutive 6 negative voltage sweeps and consecutive 6 posi-tive voltage sweeps of ±0.07 V were applied In the response of consecutive negative voltage sweeps, the absolute value of the current level after each sweep increases stepwise from curve 1

to 6, in an analog fashion which is similar to synaptic potentia-tion of biological synapses Whereas, the current level similarly decreases upon the application of consecutive positive voltage sweeps, from curve 7 to 12, in an analog stepwise fashion, which is considered as a counterpart of synaptic depression

in biological synapses This gradual increase and decrease of conductance level at readout voltage of ±0.04 V during these 12

Figure 4 Digital switching characteristics of the bio-memristor for nonvolatile memory storage applications a) Digital switching I–V characteristics of

AgNPs-TCNC bio-memristor under current compliance of 10−3 A b) The endurance characteristics of the device presenting good repeatability in the ON and OFF states for 200 DC sweep cycles c) Pulse endurance characteristics of the device for 104 pulse cycles d) Data retention characteristics of the device maintaining a memory window of ≈104 for more than ≈104 s e) Statistical distribution data of operating voltages for SET and RESET processes and f) HRS and LRS resistance distributions for 200 continuous switching cycles

www.advancedsciencenews.com www.advmattechnol.de

consecutive voltage sweeps are shown in Figure 5b This shows

the potentiation and depression behavior of the device due to

the consecutive voltage sweeps acting as the action potential

The retention characteristics of the four highest conductance

states were evaluated for 10 min, as shown in Figure S4,

Sup-porting Information For each conductance state, the device

current was read at a reading voltage of 0.04 V for 600 s with

an interval of 50 s between each reading measurement event

The higher states could maintain the conductance state

how-ever, the lowest state could not maintain for a longer time This

is attributed to the formation of very small conductive filaments

which cannot sustain for a longer time Moreover, the analog

switching characteristics were investigated in different devices

to evaluate the device-to-device variability which is critical for

practical applications of the devices Figure S5, Supporting

Information, shows the incremental current modulation in

five different devices depicting the device-to-device variations

during the analog switching It can be seen that the CNF-based

devices exhibit good device-to-device repeatability for analog

switching Realizing the continuous and gradual change in the

current level upon the applied DC sweeps, we tried to

under-stand more clearly by further exploring the transient electrical

characteristics upon applying input pulses to the device The 10

consecutive pulses (±0.10 V, 30 ms) with opposite polarities are

applied to measure the current response with time as shown

in Figure  5c The increase of current values upon 5

succes-sive negative bias pulses as well as decrease of current values

upon 5 successive positive bias pulses with time demonstrate

the resultant potentiation and depression of conductance of the

device This potentiation/depression behavior is the imitation

of successive variable synaptic weight and connection strength

in biological synapses Usually, the human brain experiences

an enormous number of synapses (≈1015) between presynaptic neurons and postsynaptic neurons of connected neural net-works, which serve all the computations, and facilitates our entire memory blocks This fundamental phenomenon of our brain is known as synaptic plasticity Synaptic plasticity is an activity-dependent process where synaptic weight is modulated

by the input stimuli within the synaptic cleft between synaptic connections of the neural network.[57–59] When the presynaptic neurons receive an action potential (electrical stimuli) Ca2+ ions influx are transported to presynaptic vesicles through voltage-gated Ca2+ ion channels, and neurotransmitters are released

by the presynaptic-neurons Upon reaching the postsynaptic neurons, the receptors receive these neuro-transmitters, and the electrical signal is transferred forward The weak stimulus lasts only hundreds of milliseconds, and this state is termed as short term plasticity (STP), but by the repetition of receiving action potentials (electrical stimulus), the Ca2+ ion influx is pro-longed and the synaptic conduction through neurotransmitters

is enhanced, which causes structural changes in the synaptic connection known as long term plasticity (LTP).[60] Our bio-memristor as an artificial synaptic device mimics the above mechanism utilizing attached AgNPs AgNPs oxidize into Ag+ ions as a counterpart of Ca2+ ions influx, activated by voltage pulses, which is equivalent to action potential (presynaptic spiking) of biological synapses To check the long-term potenti-ation and depression, two pulse trains with successive negative

Figure 5 Analog switching characteristics of bio-memristor a) I–V characteristics of the device during 6 consecutive negative and 6 positive sweeps

b) Changes of conductance levels during consecutive sweeps c) Current-time responses with applied bias pulses of opposite polarity (±0.10 V, 30 ms) d) Potentiation and depression of conductance levels after 30 consecutive negative and 30 positive pulses of (−0.10 V, 1 ms) and (+0.10 V, 3 ms), respectively

and positive pulses (-0.10 V, 1 ms) and (0.10 V, 3 ms) were

applied and as a result of which the potentiation and

depres-sion in the current levels can be observed in Figure 5d By the

response of each negative voltage pulse, the silver ions attached

to nanocrystals are oxidized and accumulate to grow the

con-ductive filaments incrementally, reducing the disconnection

gap between the filaments and the bottom electrode due to

which the conductance level increases with each passing pulse

The opposite phenomenon of filament dissolution occurs with

each pulse of positive polarity

In the biological system, synaptic plasticity is considered as

the basis for learning and memory functions where synaptic

weight is adjusted according to the presynaptic and postsynaptic

stimulations The emulation of such synaptic plasticity

char-acteristics is desirable in artificial electronic synapses to build

neuromorphic systems Our bio-memristor synaptic device, as

shown in Figure 6, successfully emulated some of the

impor-tant synaptic functions One of the imporimpor-tant synaptic plasticity

functions is the SRDP For the emulation of SRDP functions

with our electronic synaptic device, 5 sets of 10 pulse trains with

different intervals but the same pulse height (−0.10 V) and width

(30 ms) were applied to check the corresponding responses The

pulse scheme diagram and the device corresponding response

in terms of conductance with each pulse number are shown

in Figures 6a,b, respectively The pulse stimulations with pulse intervals of 5 and 1 s showed no increment in the current level, while the pulse stimulation train with intervals of 500 ms showed the least increment in the current level The stimulation with pulse intervals of 100 and 50 ms showed the highest incre-ment in current levels In biological synapses, the presynaptic stimulation causes the influx of Ca+ ions, resulting in the release

of neurotransmitters temporarily, after which it is recovered But if the second or third identical stimulation is received before the Ca+ ions recovery, the response of the post-synapse will be larger than the response it showed to the first one In our case,

by the reduction of pulse interval during measurements, we can see a similar effect in terms of increased current conduc-tion, and this effect is known as PPF By continuation of these identical stimulations (pulses) in the form of a pulse train with suitable intervals, it will cause a gradual increase in conduction (synaptic transmission), and this effect is called PTP To make this behavior of PPF and PTP prominently visible, we plotted the increased current in the form of change in current (∆I) in

Figure  6c, which is calculated by subtracting the current (I N,

N  = 1, 2, 3, 4…10) from I1 Both PPF and PTP behaviors for each pulse train with different intervals are highlighted Moreover, the characteristics of the PPF index and PPD index were fur-ther studied by applying a pair of pulses with a varying interval

Figure 6 The synaptic function characteristics of Ag/AgNPs-TCNC/FTO bio-memristor device a) Pulse train scheme of 10 pulses with different

inter-vals b) Measured current levels for 10 consecutive pulses with different pulse interinter-vals c) The difference in current levels (∆I) after each consecutive pulse for each mentioned pulse train d) The demonstration of the PPF behavior of the device showing the dependence of the PPF index on the inter-spike interval e) The demonstration of the PPD behavior of the device f) EPSC characteristics with a single voltage pulse stimulation (−0.1 V, 10 ms)

www.advancedsciencenews.com www.advmattechnol.de

between both pulse spikes The PPF index and PPD index were

calculated as a ratio of the current response of the device due to

the second pulse to that of the first pulse (I2/I1) The PPF index

characteristics of the device are presented in Figure  6d The

paired-pulse stimulations with smaller pulse intervals showed

the largest PPF index whereas the dual pulse stimuli with

sig-nificantly larger intervals depicted no or little current facilitation

showing a smaller PPF index The device also showed similar

behavior for the PPD The PPD index showed a clear

depend-ence on the inter-spiking interval, as shown in Figure 6e This

confirms that our device can faithfully mimic the short-term

plasticity behaviors of the biological synapses

The change in the conductance of the device as a result

of successive pulses is because of the temporal interaction

between applied stimulations and its ionic EPSC, which also

gives us information about the decay function of the

con-ducting filament For measuring an EPSC, a single electrical

pulse of (−0.1 V, 10 ms) was applied as a stimulus source and

a readout pulse of 0.04 V was continuously applied after a

single stimulation to observe the resultant behavior as a

post-synaptic current The postpost-synaptic current decayed

gradu-ally over time after an abrupt increase with input spike as

shown in Figure 6f Moreover, the EPSC characteristics of the

device were further evaluated by varying the amplitudes of

the applied stimuli The EPSC characteristics for the pulses

with different pulse amplitudes are presented in Figure S6,

Supporting Information With the interpretation of the above

results and discussion, our bio-memristor can successfully

work as an artificial synaptic device

Since the device is prepared with the silver top electrode, it

is expected that the electroforming and the SET phenomenon could occur by applying positive voltage bias on the silver top electrode because of the ionization of silver near the positive bias electrode and the subsequent Ag+ ions migration under electric field but the opposite case was observed The device showed the electroforming and SET phenomenon with the application of negative voltage bias on the top electrode We propose that this phenomenon is due to the high Ag+ ions con-centration precipitated on nanocrystals The schematic presen-tation of the proposed digital and analog switching mechanisms

is presented in Figure 7 The switching mechanism is

real-ized by the effect of redox reactions at the electrode interfaces causing oxidation of Ag atoms and reduction of Ag+ ions in the switching process The attached silver of AgNPs-TCNC dielec-tric layer near the FTO bottom electrode is oxidized into Ag+ ions as the FTO acts as an opposite positive electrode in this case The ionized silver ions near the positive FTO bottom elec-trode start migrating under the effect of the electric field, and

by reaching the negative top electrode the Ag+ ions reduce to

Ag by receiving the electrons from the top electrode illustrated

as Ag+ + e− → Ag (reduction) With the successive metallization

of silver ions and the accumulation process of metallic silver, stable conductive filaments are formed connecting top and bottom electrodes The thickness of these formed conducting multi-filaments grows with increasing voltage during voltage sweep With the reversal of the applied voltage polarity, the silver filament metal starts oxidizing into Ag+ ions dissolving

in the dielectric layer near the negative bottom electrode (FTO)

Figure 7 The schematic diagram of the switching mechanism to explain the proposed mechanisms of filament formation and rupture during the

process of a) digital switching and b) analog switching