betaFeedback

WorksPhysics and Astronomy

Stress-induced Artificial neuron spiking in Diffusive memristors

Abstract

Diffusive memristors owing to their ability to produce current spiking when a constant or slowly changing voltage is applied are competitive candidates for the development of artificial electronic neurons. These artificial neurons can be integrated into various prospective autonomous and robotic systems as sensors, e.g. ones implementing object grasping and classification. We report here Ag nanoparticle-based diffusive memristor prepared on a flexible polyethylene terephthalate (PET) substrate in which the electric spiking behaviour was induced by the electric voltage under an additional stimulus of external mechanical impact. By changing the magnitude and frequency of the mechanical impact, we are able to manipulate the spiking response of our artificial neuron. This functionality to control the spiking characterstics paves a pathway for the development of touch-perception sensors that can convert local pressure into electrical spikes for further processing in neural networks. We have proposed a mathematical model which captures the operation principle of the fabricated memristive sensors and qualitatively describes the measured spiking behaviour.

Introduction

Memristors as electrical circuit elements have only emerged recently [1]. Due to their potential to realise energy-efficient hardware for neuromorphic AI systems, memristive devices have received tremendous attention [2, 3, 4]. It has also been shown that memristors are capable to emulate certain biological neuron behaviour, which makes them promising for neuromorphic devices [5, 6]. Various types of memristive device structures such as Au/Ag/SiOx/Au\mathrm{Au/Ag/SiO_x/Au},[7, 8] Pt/Ag:SiOx/Pt\mathrm{Pt/Ag:SiO_x/Pt},[9, 10, 11, 12] TiN/HfOx/AlOx/Pt\mathrm{TiN/HfO_x/AlO_x/Pt},[13] Cu/ZnS/Pt\mathrm{Cu/ZnS/Pt},[14] Ag/SrTiO3/(La,Sr)MnO3\mathrm{Ag/SrTiO_3/(La,Sr)MnO_3},[15] Pt/NbOx/Pt\mathrm{Pt/NbO_x/Pt} [16] have been developed which utilizes the migration of metallic ions or oxygen vacancies resulting as the change in resistance between high resistance state (HRS) and low resistance state (LRS).

In this work, we focus on a subclass of so-called diffusive memristors with metallic Ag nanoparticles (NPs) embedded in insulating matrix of SiO2\mathrm{SiO_2} [7, 8, 9, 17]. For these memristive devices, the transition from HRS to LRS and vice versa occurs due to a field-induced diffusion of metallic NPs to coalesce together to form a conduction filament (CF) between the electrodes. Upon increasing the external voltage, a CF is completely formed between the electrodes above a threshold voltage (VtV_t) and is ruptured at voltage below the hold one (VhV_h). This type of volatile resistive switching has been utilized [9] to build artificial neurons, converting DC voltage to current spikes. The external voltage and temperature cause change of spiking regimes, interpsike intervals, and other spiking characteristics as discussed in Ref. [9, 10]. Therefore, we expect that other external parameters such as mechanical stress or pressure can affect NP dynamics [18, 19, 20], and thus, artificial neuron response, making memristor-based electronic components to be good candidates for neuromorphic sensors.

Recently, there has been significant interest in the development of flexible memristors [21, 22, 23, 24, 25, 26]. However, the main focus has been on studies of the effect of substrate flexibility on neuromorphic devices endurance and reliability. Some papers have also discussed fabrication of artificial neural networks on flexible memristors for neuromorphic computation,[23, 27] but most of these technologies rely on applying an external electric voltage. As far as we know, the impact of external mechanical forces on the spiking behavior of memristors has not yet been explored. Consequently, this research introduces a promising opportunity to use flexible volatile memristors as touch-perception sensors for various applications, including in neural networks.

The application of decoding tactile information from sensors into electrical signals is intriguing and has a significant potential for development of robotic systems and sensors. For example, various Field Programmable Gate Arrays (FPGA) based Spiking Neural Networks (SNN) have been employed to convert sensor data into spike patterns using artificial mechanoreceptors.[28, 29, 30, 31, 32] However, a major roadblock for the implementation of dense and efficient FPGA-SNN is the large power overhead. So there is a trade off between accuracy and energy efficiency [33]. Here, we report on spiking characteristics of Ag NPs-based flexible diffusive memristors and show spiking dependence on the time interval between successive mechanical impacts and the corresponding pressure. The observed mechanical stimulation into electrical response is highly desirable for development of energy effective wearable and implantable electronics that can mimic e-skin function such as a mechanoreceptor sensor.

Results and Discussion

Fabrication and characterization of flexible memristor

Diffusive memristors with structure Pt/Ag:SiOx_x/Ag (see Figure 1(a)) with 20%:80%Ag:Si atomic ratio were deposited on a polyethylene terephthalate (PET) substrate at room temperature using magnetron sputtering technique. A bottom Pt electrode layer, 30 nm thick, was deposited on a PET substrate. Following that, Ag and SiO2_2 were co-sputtered in Ar to create a switching layer with a nominal thickness of 100 nm. Lastly, using a shadow mask with circles 120 μ\mum in diameter, 5 nm of Ag was sputtered to form the top electrodes.

Fabricated diffusive memristor and IV characterization

Figure 1. Fabricated diffusive memristor and IV characterization. (a) A side view schematic of a diffusive memristor device, PET/Pt(30 nm)/Ag:SiOx_x(100 nm)/Ag (5 nm),with an attached voltage source. (b) Typical I-V characteristics of a fabricated memristor measured in both positive (blue line) and negative (red line) voltage cycles. The average of the measured I-V is shown in black. The arrows indicate the direction of voltage change.

The I-V loops of a device were recorded at room temperature employing a Keithley 4200 SCS analyzer equipped with an attached Everbeing probe stage. An external voltage was applied to the top electrode (see Figure 1(a)) and varied within the range of -2 V to 2 V at a sweep rate of 50 mV/s. The resulting I-V loops, shown in Figure 1(b), demonstrate clear volatile resistive switching behavior for both positive and negative voltage sweeps. The observed I-V characteristics are nearly mirror-symmetric, with a transition from a high resistance state (HRS) to a low resistance state (LRS) occurring at a threshold voltage of ∣Vt∣=0.65±0.05|V_t| = 0.65 \pm0.05 V.

To analyze the composition of the fabricated memristors, X-ray photoelectron spectroscopy (XPS) was performed. The XPS survey and the corresponding high-resolution spectra for oxygen OO 1s, silicon SiSi 2p and silver AgAg 3d are shown in Figure 2 (a-Figure 2d) respectively.

Surface characterization of the diffusive memristor

Figure 2. Surface characterization of the diffusive memristor. (a) XPS survey of the fabricated diffusive memristor, (b-d) High resolution XPS spectra showing O 1s, Si 2p, and Ag 3d with the highlighted areas showing the XPS fits.

The XPS survey scan spectra is shown in Figure 2 (a). Distinct peaks for Ag and Pt can be seen. The XPS spectrum for O 1s (Figure 2(b)) shows two peaks [12, 34] corresponding to silicon oxide (532.9eVeV) and silver-oxide (530.8eVeV) indicating partial oxidation of the silver NPs, while the remaining fraction of oxygen is contained in the dielectric silica matrix. This can also be attributed to oxidation of the top Ag layer which is even more likely given the surface sensitivity of the XPS. The Si 2p XPS spectrum (Figure 2(c)) shows that silicon is in Si4+Si^{4+} state and is part of silica. The Ag XPS 3d peak at 367.69eVeV (Figure 2(d)) corresponds to silver oxide (AgO) [12]. Therefore, we conclude that the Ag NPs are partially in metallic state and partially oxidized whilst being embedded in the dielectric matrix of SiO2SiO_{2} [9, 11]. Based on our XPS data, we can state that the prepared memristors are similar and there is no degradation or contamination of our diffusive memristor devices when deposited on PET substrates instead of silicon substrates.[9]

Influence of Mechanical Impact on Artificial Neuron Spiking

An artificial neuron device was constructed by connecting our diffusive memristor to an external resistor RL=70 kΩR_L=70\ \mathrm{k}\Omega and a capacitor Cp=1 nFC_p=1\ \mathrm{nF}, as depicted in Figure 3(a). The values of RLR_L and CpC_p were selected based on previous studies of similar devices, e.g. in Ref. [9, 12], where electrical spiking behavior was observed. When a constant DC voltage above a certain threshold was applied, for instance, Vext=0.6 VV_{\mathrm{ext}}=0.6\ \mathrm{V}, the artificial neuron began generating voltage spikes across the memristor, as illustrated in Figure 3(b). This demonstrates a spiking behavior similar to that previously reported for non-flexible diffusive memristors (see Ref. [9, 10]). Note that relatively large capacitance of 1 nF1\ \mathrm{nF} in an integrated circuit could be potentially replaced with a lower capacitance and a correspondingly higher load resistance RLR_L to maintain the same circuit time constant RLCR_LC. As above previous studies showed, this parameter is crucial as it determines the generation and characteristics of the voltage spikes. However, this proposed approach requires further experimental verifications.

Artificial memristive neuron

Figure 3. Artificial memristive neuron. (a) Electrical circuit of an artificial neuron with a diffusive memristor; (b) Measured memristor voltage spikes (red line) vs time for an external input voltage of 0.6 V (black line)

To investigate effects of mechanical stress on artificial neuron spiking behaviour for our flexible diffusive memristors, we have applied a direct mechanical impact on the memristor using an in-house built mechanical impact station as shown in Figure 4 (a) and (Figure 4b). A vertical mechanical impact up to 0.4 MPa0.4\ \mathrm{MPa} with a minimal delay between two impacts of 0.2 s0.2\ \mathrm{s} was applied to the top electrode using a pneumatically controlled stressor, whilst the memristor was covered with a 0.5 cm0.5\ \mathrm{cm} thick rubber layer and hosted on rubber covered platform [35]. For the experiments involving mechanical impacts, we slightly modified the configuration of the artificial neuron, cf. Figure 4(b) and Figure 3(a).

Stress-induced spiking measurement with experimental setup, circuit diagram, voltage spikes, I-V characteristics, and memristor voltage spikes

Figure 4. Stress-induced spiking.(a) Experimental setup showing the pneumatically controlled impact station. The memristor is placed on the platform with a white rubber cover. The stressor is periodically pressed onto our memristor from the top. (b) Circuit diagram for the stress-induced spiking measurement, (c) Measured voltage spikes for impact pressure 0.15 MPa (black) and 0.35 MPa (red) with a 0.2 s time delay between successive impacts. The memrsitor spikes (left black) from HRS to LRS as a response to the impact which eventually stops (right black) as it is stuck in a permanent LRS. The spiking action is retrieved by gradually increasing the impact (left and right red) from 0.15 MPa to 0.35 MPa in steps of 0.05 MPa.(d) I-V characteristics of the device measured post several impacts in both positive (blue line) and negative (red line) voltages showing asymmetry in VtV_t. The curved arrows indicate the direction of the change in device resistance. (e) Measured memristor voltage spikes (red line) vs time for an external input voltage of 1.9 V (black line) for the device after approximately 4200 impacts

Specifically, as illustrated in Figure 4(b), the memristor was connected in series with an external capacitor CC and a parallel leakage resistor RR, which improved the stability of the spiking measurements under repetitive impacts. The capacitance of CpC_p, previously used in the measurements without impact, illustrated in Figure 3(a), was replaced by the internal parasitic capacitance CiC_i of the memristor originating from the electrodes and the active layer. It is reasonably to assume that CiC_i changes as the device subjected to pressure. A constant voltage of 1.5 V1.5\,\mathrm{V} was applied to the neuron whilst the voltage drop across the resistor RR was monitored for voltage spikes. We believe that the role of the external load and parallel capacitance as in Figure 3a was fulfilled by the device contacts resistance combined with the external leakage resistance, and by the device internal capacitance as shown in Figure 4b, respectively. The external capacitor in series was used to stabilise the spiking dynamics as well.

The time dependence of voltage spikes in response to impacts of 0.15 MPa0.15\,\mathrm{MPa} (black line) and 0.35 MPa0.35\,\mathrm{MPa} (red line) applied and then interrupted for 0.2 s0.2\,\mathrm{s}, are shown in Figure 4 (c). It must be noted that the device did not show any spiking behaviour for impact lower than 0.15 MPa0.15\,\mathrm{MPa}. (See supplementary information Fig. S1). When the memristor is spiking, two different regimes have been observed. For an impact of 0.15 MPa0.15\,\mathrm{MPa} (Figure 4 (c)) the external capacitor voltage spikes occur from low to high values, which means the voltage across the memristor switches from high to low values, that is the memristor staying most of the time at HRS occasionally transits to LRS for a short time (Figure 4 (c) (left black)). Such spiking behaviour eventually stops after a series of impacts[Figure 4 (c)(right black)] that is, the device remains permanently in HRS. However, for impact pressure of 0.35 MPa0.35\,\mathrm{MPa}, the spiking behaviour gradually revives[Figure 4 (c)(left red)] after several impacts (approximately 4200) and the spikes occur from top to bottom (LRS to HRS for the memristor)[Figure 4 (c)(right red)].

Further to this, the I-V loop characteristic was re-measured, and showed asymmetry (Figure 4 (d)) in comparison to a fresh sample with no impact history (Figure 2 (b)). The spiking frequency has reduced as well (Figure 4 (e)). An explanation for this is discussed in the next section. However, the ability to recover to HRS with an I-V characteristic demonstrates that the device has a memory.

Artificial neuron spiking rates under repeated mechanical impact

We measured spiking rate at different time intervals between successive application of impact tpt_p (Figure 5). It is observed that the average spiking rate decreases for all impact pressure values, 0.2 MPa0.2\,\mathrm{MPa} - 0.4 MPa0.4\,\mathrm{MPa} as the time interval between successive impacts is increased. In this instance, we have implemented a fresh diffusive memristor layer without any prior impact history. This ensures that any potential impact damage (further discussed in next section and shown in Figure 6(b)(Figure 6d)) does not interfere with the spike rate analysis.

Spiking rate versus time interval between impacts at different impact pressures

Figure 5. Spiking rate: Average spiking rate (number of spike per second) vs time interval between the impact, at different impact pressures.

Effect of impact on sample surface

Figure 6. Effect of impact on sample surface (a)Schematic showing the migration of Ag NPs in the initial state and (b) due to impact induced deformation. (c,d) SEM photograph for our sample showing Ag NPs on the sample surface before and after several mechanical impacts.

The change of the spiking rates under the influence of the mechanical impact can have two origins. The first one is due to internal capacitance change as a result of impact and deformation. Since the internal device capacitance is directly involved in the time constant of the artificial neuron, we believe this effect is responsible for the spiking rate change as discussed in the next section. Secondly, the additional

pinning centres produced after repeated mechanical impacts are likely responsible for the transformation between two different spiking regimes as observed in Figure 4 (c) and explained in [9], that is, whether the memristor remains most time in HRS and occasionally switches to LRS or vice versa.

In order to understand how mechanical impact influences the Ag NP distributions, we obtained SEM images of conductive clusters in the memristor 6 (c) and (d). The observed significant change of the size and distribution of Ag NPs allows us to propose physical mechanisms and mathematical model described in the next section.

Model of artificial neuron spiking under repeated mechanical impact

When an external voltage (VextV_{ext}) is applied, the device internal capacitance (CiC_i) is charged until the voltage across the memristor reaches VtV_t, causing HRS to LRS switch. When in LRS, CiC_i starts discharging until voltage drops to VhV_h, causing it to reset to HRS. A series of such charging and discharging processes under the influence of VextV_{ext} generates a sequence of voltage spikes. [9]

To explain the effect of mechanical impact on the spiking behaviour of the fabricated artificial neurons, we argue below that two different physical mechanisms can affect spiking and its intensity: (i) changing the time constant of the artificial neuron capacitor and (ii) varying the electrochemical profile where Ag clusters diffuse.

Effects of pressure on the internal capacitance of a memristor

When an external mechanical pressure is applied, the memristor undergoes deformation. The internal capacitance of our memristor (CiC_i) is:

Ci=Aϵd(1)C_i = \frac{A\epsilon}{d} \tag*{(1)}

with an area AA, the thickness dd between the electrodes and dielectric permittivity ϵ\epsilon. The relative change in the internal capacitance due to small deformations can be estimated as:

ΔCiCi=Δϵϵ+ΔAA−Δdd(2)\frac{\Delta C_i}{C_i} = \frac{\Delta\epsilon}{\epsilon} + \frac{\Delta A}{A} - \frac{\Delta d}{d} \tag*{(2)}

The first term is the variation of the dielectric constant due to the impact, while the last two terms are the geometric changes to the studied memristor film.

The effect of deformation on the dielectric constant is called electrostriction enhancement of capacitance and is explained by the deformation-induced anisotropy of dielectric properties. A detailed explanation about this effect has been given in Ref. [35] and Ref. [36]. While the contributions of ΔA\Delta A and Δϵ\Delta\epsilon to ΔCi\Delta C_i are relatively small under the impact pressures used in the experiment, (see Supplementary information for estimation of dielectric constant estimations) the change in the effective distance between the electrodes due to pressure can be substantial. As shown in the comparison of Figure 6 (b),(Figure 6c) and (Figure 6d), the impact causes to form larger Ag clusters with characteristic sizes comparable to the value of d=100d = 100 nm. The presence of large and thicker metallic inserts between the capacitor plates significantly reduces the effective value of dd, leading to a notable change in capacitance. Consequently, this change in capacitance affects the charging time of the capacitor and, therefore, the current interspike intervals, which is in agreement with our measurements.

Another contribution to change of capacitance originates from the modified dynamics of Ag particles. It can be argued that the deformation process leads to a reduction in the effective distance between the nanoparticles (NPs), facilitating the formation of a conductive filament (CF). Furthermore, the rearrangement of silver (Ag) NPs alters the local electrical field distribution, contributing to changes in the dielectric constant (ϵ\epsilon).

Influence of pressure on the potential landscape for Ag clusters diffusion

The SEM images presented in Figure 6 (b) and (Figure 6c) illustrate the migration of Ag NPs toward the top electrode surface after repeated mechanical impacts (approximately 4200 impacts on individual devices. Consequently, an increased asymmetry in the current-voltage (IV) curves is observed following these impacts (Figure 4 (d)), indicating the generation of additional pinning centers for Ag NPs. Another contribution to the spiking rates changes originates due to mechanical impact influencing the diffusion of Ag NPs in the deformed layer, see the schematic in Figure 6 (b). This effect is similar to a change of the mobility of charge carriers in silicon under strain [37]. The influence of matrix lattice on a diffusion of conductive particles in memristors and modification of potential due to immobilised pinned charges, both of these effects can be influenced by pressure, have been considered in [38]. All these physical mechanisms can be reduced to a modification of electrochemical potential of Ag NPs and can be evidenced by re-distribution of Ag NPs.Indeed, this can be another important contribution to the spiking rate changes under repeated mechanical impact. This phenomenon is further reflected in the measured voltage spikes, where less frequent spikes with lower spiking amplitude are observed for higher input voltage (Figure 4 (e)).

Modelling artificial neuron spiking under mechanical impact

To model the spiking rate changes in our artificial neuron in response to external mechanical impact, we consider interplay between three degrees of freedom: diffusion of Ag clusters forming conducting filaments (CFs) Eqn. (3), Joule heat sinking to substrate Eqn. (3), and electric current dynamics in the circuits described by Kirchhoff equations Eqn. (3). To simplify the problem we consider the case when significant changes of the resistance occur due to only one Ag cluster sitting in the bottleneck of an almost formed CF. [9, 39] Under this assumption, the set of equations (see Supplementary information) describing an artificial neuron can be written as:

ηdxdt=−(1+ΔUf(t))∂U∂x+qVLV+2kBηTξ(t)dTdt=TChR(x)−κ(T−T0)τ(1+C2−C1C1f(t))dVdt=Vext−(1+RextR(x))V.(3)\begin{aligned} \eta\frac{dx}{dt} &= -\left(1+\Delta_U f(t)\right)\frac{\partial U}{\partial x}+q\frac{V}{L_V}+\sqrt{2k_B\eta T}\xi(t) \\ \frac{dT}{dt} &= \frac{T}{C_hR(x)}-\kappa(T-T_0) \\ \tau\left(1+\frac{C_2-C_1}{C_1}f(t)\right)\frac{dV}{dt} &= V_{\mathrm{ext}}-\left(1+\frac{R_{\mathrm{ext}}}{R(x)}\right)V. \tag*{(3)} \end{aligned}

In this case the resistance depends only on the location xx of the bottleneck Ag cluster in the gap between almost formed parts of the filament, and can be approximated as [9] R0=cosh⁡(x/λ)R_0=\cosh(x/\lambda) with the electron tunneling length λ\lambda and all distances normalised by the gap size and the resistances by the minimal memristor resistance as described in the Supplementary information.

Accordingly, the impact of pressure can be reduced to two main factors: the modulation of capacitance due to changes in the dielectric constant and the effective distance dd between terminals, as well as the modulation of the height of the effective electrochemical potential U(x)U(x). All these effects can be incorporated into the model by introducing explicit time dependence associated with the periodic (square wave) application of pressure. Specifically, we model the change in internal capacitance between C1C_1 and C2C_2, resulting in oscillations of the artificial neuron RC-time between two values τ=RextC1\tau=R_{\mathrm{ext}}C_1 and RextC2=τC2/C1R_{\mathrm{ext}}C_2=\tau C_2/C_1. Assuming that such a change in the capacitance occurs abruptly with switching period tpt_p, we can approximate f(t)=sign⁡(sin⁡(2πt/tp))f(t)=\operatorname{sign}(\sin(2\pi t/t_p)) with sign⁡(y)=0\operatorname{sign}(y)=0 if y>0y>0 and sign⁡(y)=−1\operatorname{sign}(y)=-1 if y<0y<0. As a simple model of electrochemical profile oscillations, we study the square-wave modulations of potential height as follows: U(x)→(1+ΔU(f(t)))U(x)U(x)\to(1+\Delta_U(f(t)))U(x). In this approximation potential abruptly changes from U(x)U(x) to (1+ΔU)U(x)(1+\Delta_U)U(x) with frequency 1/tp1/t_p.

Our simulations as shown in Figure 7 demonstrate that in both cases (modulation of capacitance only or modulation of electrochemical potential only), the rates can decrease with decreasing impact frequency. For modulation of potential height (Figure 7(a)), the interspike intervals are more noticeably affected, resulting in a rate decrease. For capacitance modulations (Figure 7(b)), a decay of spiking rates with increasing tpt_p occurs primarily because of the heights of the spikes decrease, causing some of them to fall below the detection threshold. This decrease is also accompanied by several ”damped” oscillations, similar to those observed in our experiments.

Simulated spiking rate and pressure-modulation simulations

Figure 7. (a) Simulated spiking rate (a) Average simulated spiking rate (number of spikes detected during time interval τ\tau using criteria threshold G=0.05G = 0.05) as a function of tpt_p (idle time between impact application) when the pressure affects (a) only RC-time (i.e., ΔU=0\Delta U = 0) with (C2−C1)/C1=0.8(C_2 - C_1)/C_1 = 0.8 and (d) only electrochemical potential (i.e., C1=C2C_1 = C_2) and ΔU=0.4\Delta U = 0.4. We simulated the dimensionless equations (see Supplementary materials where all simulation parameters are provided). Panels (b) and (c) [panel (e) and (f)] correspond to simulations (a) [(d)] for timpact/τ=0.05t_{\mathrm{impact}}/\tau= 0.05 and 2, shown in (a) [(d)] by blue and red circle symbols

Conclusion

In conclusion, we have fabricated diffusive memristors with silicon oxide and silver NPs on a flexible PET substrate. XPS analysis along with SEM microscopy confirms the incorporation of metallic Ag NPs into the SiOx\mathrm{SiO}_x dielectric matrix. Current spikes under constant electrical voltage and repeated mechanical impact showed that there is a strong dependence of current spikes on the impact pressure and frequency. We have observed that the spiking can abruptly stop after several impacts due to a permanent conductive state. The high-resistive state can be retrieved by increasing the pressure of the impact. We have also demonstrated the dependence of average spiking rate of our memristor on the time interval between successive impacts. These experimental finding were explained and reproduced by a numerical model that involves variation of the internal memristor capacitance or electrochemical potential upon repeated mechanical impact. The obtained dependence of the spiking rate on impact idle time tpt_p, at different pressures, in addition to the memory of its I-V characteristics opens a possibility to utilize this device as a pressure sensor which is able to estimate the touching strength and also perform in-material computation (e.g., reservoir computing) utilizing capability of memorising mechanical history.

The use of such memristors that can directly decode tactile information into spikes, similar to force sensors or pressure sensors, could have significant benefits in various robotics and in-material computing applications [40, 41, 42, 43, 44], such as artificial prosthetic hands and grasping in robotics. This could enable the robots and autonomous systems to obtain biological like perception without requiring external bridges or complex processing, potentially enhancing their ability to interact with the environment in a more human-like manner.

Methods

Sample preparation: The studied diffusive memristors were deposited on a Polyethylene terephthalate (PET) substrate at room temperature using magnetron sputtering technique. Bottom Pt electrode of 30 nm was deposited on the substrate at 50 W dc power. The diffusive memristor layer of nominal thickness 100 nm was made by co-sputtering from Ag and SiO2\mathrm{SiO}_2 targets at 20 W and 300 W respectively in 0.85 Pa Ar pressure. A thin 5 nm Ag reservoir layer was deposited which also acts as the top electrode. A rectangular shadow mask was used during the deposition to gain access to the bottom Pt electrode.

Electrical characterization: I-V characterization for the fabricated memristor was made using a Keithley 4200 SCS and an Everbeing probe station with 2 μ\mum tungsten tips. For the artificial neuron spiking measurement, a voltage pulse (0.6 V, 100 s) was applied to the device using a Rigol waveform generator and the device voltage was recorded using a PicoScope digital oscilloscope, whilst the memristor was connected in series to a load resistance RL=70kΩR_L = 70\mathrm{k}\Omega and in parallel to a capacitor CP=1C_P = 1 nF.

The measurements for the effect of impact on the memristor were performed using the in-house developed impact station discussed in the main text. For this measurement, the same memristor was put on the impact station test bench, with a rubber on top. The memristor was connected to an external RC block with R=1kΩR = 1\mathrm{k}\Omega and C=0.02 μFC = 0.02\ \mu\mathrm{F}.

Surface characterization: X-ray photoelectron spectroscopy (XPS) was performed using a Thermo K-Alpha system with an Al Kα\alpha monochromated (1486.6 eV) source with an overall energy resolution of 350 meV. The analysis area captured was approximately 100 μ\mum ×\times 200 μ\mum. A survey scan was first performed to preview the chemical composition, subsequent high-resolution scans were then performed on the elements of interest before fitting their peaks to identify elemental state. All scans were charge corrected to adventitious C1s (C-C, C-H) peak at 284.8 eV. Scanning electron microscopy (SEM) was carried out on a JEOL 7100F. Imaging was done at both 5 kV and 2 kV to achieve the best resolution and surface detail of small particles.

Acknowledgments

D.P. thanks Alan Turing Institute (London) for the postdoctoral endowment fund (2022 Cohort), which was essential to develop the experimental methodology. The authors would like to thank Dr. Sam Davis for the XPS characterization and SEM imaging and acknowledge the use of the facilities within the Loughborough Materials Characterisation Centre (LMCC). This work was supported by The Engineering and Physical Sciences Research Council (EPSRC), grant no. EP/S032843/1.

Competing interests

The authors declare no competing interests.

Author contributions

DP conceived the concept, prepared the samples, performed electrical and surface characterisation and wrote the first draft of the manuscript. DP,PB,AB,SS wrote the manuscript.YS,AA,PF designed the experiment and analyzed the measured data led by DP and PB. SS,AB performed the numerical simulations. All authors contributed to the discussion and polishing the manuscript.

References

  1. [1]Strukov, D.B., Snider, G.S., Stewart, D.R., Williams, R.S.: The missing memristor found. nature 453(7191), 80–83 (2008)DOI
  2. [2]Zidan, M.A., Strachan, J.P., Lu, W.D.: The future of electronics based on memristive systems. Nature electronics 1(1), 22–29 (2018)DOI
  3. [3]Wang, Z., Wu, H., Burr, G.W., Hwang, C.S., Wang, K.L., Xia, Q., Yang, J.J.: Resistive switching materials for information processing. Nature Reviews Materials 5(3), 173–195 (2020)DOI
  4. [4]Sangwan, V.K., Hersam, M.C.: Neuromorphic nanoelectronic materials. Nature nanotechnology 15(7), 517–528 (2020)DOI
  5. [5]Roy, K., Jaiswal, A., Panda, P.: Towards spike-based machine intelligence with neuromorphic computing. Nature 575(7784), 607–617 (2019)DOI
  6. [6]Zhang, W., Gao, B., Tang, J., Yao, P., Yu, S., Chang, M.-F., Yoo, H.-J., Qian, H., Wu, H.: Neuro-inspired computing chips. Nature electronics 3(7), 371–382 (2020)DOI
  7. [7]Wang, Z., Joshi, S., Savel’ev, S.E., Jiang, H., Midya, R., Lin, P., Hu, M., Ge, N., Strachan, J.P., Li, Z., et al.: Memristors with diffusive dynamics as synaptic emulators for neuromorphic computing. Nature materials 16(1), 101–108 (2017)DOI
  8. [8]Jiang, H., Belkin, D., Savel’ev, S.E., Lin, S., Wang, Z., Li, Y., Joshi, S., Midya, R., Li, C., Rao, M., et al.: A novel true random number generator based on a stochastic diffusive memristor. Nature communications 8(1), 882 (2017)DOI
  9. [9]Pattnaik, D.P., Ushakov, Y., Zhou, Z., Borisov, P., Cropper, M.D., Wijayantha, U.W., Balanov, A.G., Savel’ev, S.E.: Temperature control of diffusive memristor hysteresis and artificial neuron spiking. Phys. Rev. Appl. 19, 024065 (2023) https://doi.org/10.1103/PhysRevApplied.19.024065
  10. [10]Ushakov, Y., Akther, A., Borisov, P., Pattnaik, D., Savel’ev, S., Balanov, A.G.: Deterministic mechanisms of spiking in diffusive memristors. Chaos, Solitons & Fractals 149, 110997 (2021)DOI
  11. [11]Gabbitas, A., Pattnaik, D., Zhou, Z., Borisov, P.: Resistive switching study on diffusive memristors using electrochemical impedance spectroscopy. Journal of Physics D: Applied Physics 56(30), 305102 (2023)DOI
  12. [12]Cropper, M., Balanov, A., Savel’ev, S.: Gamma radiation-induced nanodefects in diffusive memristors and artificial neurons. Nanoscale (2023)DOI
  13. [13]Yu, S., Wu, Y., Jeyasingh, R., Kuzum, D., Wong, H.-S.P.: An electronic synapse device based on metal oxide resistive switching memory for neuromorphic computation. IEEE Transactions on Electron Devices 58(8), 2729–2737 (2011)DOI
  14. [14]Yu, J., Liang, L., Hu, L., Duan, H., Wu, W., Zhang, H., Gao, J., Zhuge, F., Chang, T., Cao, H.: Optoelectronic neuromorphic thin-film transistors capable of selective attention and with ultra-low power dissipation. Nano Energy 62, 772–780 (2019)DOI
  15. [15]Hu, H., Li, Y., Yang, Y., Lv, W., Yu, H., Lu, W., Dong, Y., Wen, Z.: Enhanced resistance switching in ultrathin ag/srtio3/(la, sr) mno3 memristors and their long-term plasticity for neuromorphic computing. Applied Physics Letters 119(2), 023502 (2021)DOI
  16. [16]Johnson, B., Brahim, K., Balanov, A., Savel’ev, S., Borisov, P.: Transition from noise-induced to self-sustained current spiking generated by a nbox thin film threshold switch. Applied Physics Letters 118(2), 023502 (2021)
  17. [17]Xia, Q., Yang, J.J.: Memristive crossbar arrays for brain-inspired computing. Nature materials 18(4), 309–323 (2019)DOI
  18. [18]Jeong, Y., Kim, S., Lu, W.D.: Utilizing multiple state variables to improve the dynamic range of analog switching in a memristor. Applied Physics Letters 107(17), 173105 (2015)DOI
  19. [19]Kim, S., Choi, S., Lu, W.: Comprehensive physical model of dynamic resistive switching in an oxide memristor. ACS nano 8(3), 2369–2376 (2014)DOI
  20. [20]Kim, S., Choi, S., Lee, J., Lu, W.D.: Tuning resistive switching characteristics of tantalum oxide memristors through si doping. ACS nano 8(10), 10262–10269 (2014)DOI
  21. [21]Wang, T.-Y., Meng, J.-L., Rao, M.-Y., He, Z.-Y., Chen, L., Zhu, H., Sun, Q.-Q., Ding, S.-J., Bao, W.-Z., Zhou, P., et al.: Three-dimensional nanoscale flexible memristor networks with ultralow power for information transmission and processing application. Nano letters 20(6), 4111–4120 (2020)DOI
  22. [22]Gergel-Hackett, N., Tedesco, J.L., Richter, C.A.: Memristors with flexible electronic applications. Proceedings of the IEEE 100(6), 1971–1978 (2011)
  23. [23]Yan, X., Zhou, Z., Zhao, J., Liu, Q., Wang, H., Yuan, G., Chen, J.: Flexible memristors as electronic synapses for neuro-inspired computation based on scotch tape-exfoliated mica substrates. Nano Research 11, 1183–1192 (2018)DOI
  24. [24]Qi, S.-S., Yang, F., Li, J.-C.: Bending effect on the synaptic behavior of wo3-based flexible memristor under variable temperatures. Applied Physics Letters 120(22), 223501 (2022)DOI
  25. [25]Yan, X., Wang, J., Zhao, M., Li, X., Wang, H., Zhang, L., Lu, C., Ren, D.: Artificial electronic synapse characteristics of a ta/ta2o5-x/al2o3/ingazno4 memristor device on flexible stainless steel substrate. Applied Physics Letters 113(1), 013503 (2018)DOI
  26. [26]Park, H.-L., Kim, M.-H., Lee, S.-H.: Introduction of interfacial load polymeric layer to organic flexible memristor for regulating conductive filament growth. Advanced Electronic Materials 6(10), 2000582 (2020)DOI
  27. [27]Ding, Q., Jiao, W., Wang, H., Zhang, X., Gao, S.: Study of a stretchable polymer for adjustable flexible organic memristor. IEEE Transactions on Electron Devices (2023)DOI
  28. [28]Misra, J., Saha, I.: Artificial neural networks in hardware: A survey of two decades of progress. Neurocomputing 74(1-3), 239–255 (2010)DOI
  29. [29]Li, P., Anwar Ali, H.P., Cheng, W., Yang, J., Tee, B.C.: Bioinspired prosthetic interfaces. Advanced Materials Technologies 5(3), 1900856 (2020)DOI
  30. [30]Abderrahmane, N., Lemaire, E., Miramond, B.: Design space exploration of hardware spiking neurons for embedded artificial intelligence. Neural Networks 121, 366–386 (2020)DOI
  31. [31]Shawahna, A., Sait, S.M., El-Maleh, A.: Fpga-based accelerators of deep learning networks for learning and classification: A review. ieee Access 7, 7823–7859 (2018)
  32. [32]Auge, D., Hille, J., Mueller, E., Knoll, A.: A survey of encoding techniques for signal processing in spiking neural networks. Neural Processing Letters 53(6), 4693–4710 (2021)DOI
  33. [33]Pham, Q.T., Nguyen, T.Q., Hoang, P.C., Dang, Q.H., Nguyen, D.M., Nguyen, H.H.: A review of snn implementation on fpga. In: 2021 International Conference on Multimedia Analysis and Pattern Recognition (MAPR), pp. 1–6 (2021). IEEE
  34. [34]Boronin, A., Koscheev, S., Zhidomirov, G.: Xps and ups study of oxygen states on silver. Journal of electron spectroscopy and related phenomena 96(1-3), 43–51 (1998)DOI
  35. [35]Shkel, Y.M., Ferrier, N.J.: Electrostriction enhancement of solid-state capacitance sensing. IEEE/ASME transactions on mechatronics 8(3), 318–325 (2003)DOI
  36. [36]Huang, J.-Q., Huang, Q.-A., Qin, M., Dong, W., Chen, X.: Strain effect of the dielectric constant in silicon dioxide. Journal of microelectromechanical systems 19(6), 1521–1523 (2010)
  37. [37]Sun, Y., Thompson, S., Nishida, T.: Physics of strain effects in semiconductors and metal-oxide-semiconductor field-effect transistors. Journal of Applied Physics 101(10), 104503 (2007)
  38. [38]Yi, W., Savel’Ev, S.E., Medeiros-Ribeiro, G., Miao, F., Zhang, M.-X., Yang, J.J., Bratkovsky, A.M., Williams, R.S.: Quantized conductance coincides with state instability and excess noise in tantalum oxide memristors. Nature communications 7(1), 11142 (2016)DOI
  39. [39]Savel’ev, S.E., Marchesoni, F., Bratkovsky, A.M.: Mesoscopic resistive switch: non-volatility, hysteresis and negative differential resistance. The European Physical Journal B 86, 1–6 (2013)
  40. [40]Xu, Z., Wu, C., Li, F., Chen, W., Guo, T., Kim, T.W.: Triboelectric electronic-skin based on graphene quantum dots for application in self-powered, smart, artificial fingers. Nano Energy 49, 274–282 (2018)DOI
  41. [41]Stoppa, M., Chioerilo, A.: Wearable electronics and smart textiles: A critical review. sensors 14(7), 11957–11992 (2014)DOI
  42. [42]Shi, J., Liu, S., Zhang, L., Yang, B., Shu, L., Yang, Y., Ren, M., Wang, Y., Chen, J., Chen, W., et al.: Smart textile-integrated microelectronic systems for wearable applications. Advanced materials 32(5), 1901958 (2020)DOI
  43. [43]Chen, B., Yang, H., Song, B., Meng, D., Yan, X., Li, Y., Wang, Y., Hu, P., Ou, T.-H., Barnell, M., et al.: A memristor-based hybrid analog-digital computing platform for mobile robotics. Science Robotics 5(47), 6938 (2020)DOI
  44. [44]Wang, S., Song, L., Chen, W., Wang, G., Hao, E., Li, C., Hu, Y., Pan, Y., Nathan, A., Hu, G., et al.: Memristor-based intelligent human-like neural computing. Advanced Electronic Materials 9(1), 2200877 (2023)DOI

Paper details

vlnc:W_abfyqdgm
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 5

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents