ACS Publications. Most Trusted. Most Cited. Most Read
Self-Powered Dye-Sensitized Solar-Cell-Based Synaptic Devices for Multi-Scale Time-Series Data Processing in Physical Reservoir Computing
More

OR SEARCH CITATIONS

My Activity
Publications

Figure 1Loading Img
Download Hi-Res ImageDownload to MS-PowerPointCite This:ACS Appl. Mater. Interfaces 2024, XXXX, XXX, XXX-XXX
  • Open Access
Functional Inorganic Materials and Devices

Self-Powered Dye-Sensitized Solar-Cell-Based Synaptic Devices for Multi-Scale Time-Series Data Processing in Physical Reservoir Computing
Click to copy article linkArticle link copied!

  • Hiroaki Komatsu
    Hiroaki Komatsu
    Department of Applied Electronics, Graduate School of Advanced Engineering, Tokyo University of Science, Katsushika, Tokyo 125-8585, Japan
    More by Hiroaki Komatsu
  • Norika Hosoda
    Norika Hosoda
    Department of Applied Electronics, Graduate School of Advanced Engineering, Tokyo University of Science, Katsushika, Tokyo 125-8585, Japan
    More by Norika Hosoda
  • Takashi Ikuno*
    Takashi Ikuno
    Department of Applied Electronics, Graduate School of Advanced Engineering, Tokyo University of Science, Katsushika, Tokyo 125-8585, Japan
    *Email: [email protected]
    More by Takashi Ikuno
    Orcidhttps://orcid.org/0000-0002-5322-9373
Open PDFSupporting Information (1)

ACS Applied Materials & Interfaces

Cite this: ACS Appl. Mater. Interfaces 2024, XXXX, XXX, XXX-XXX
Click to copy citationCitation copied!
https://doi.org/10.1021/acsami.4c11061
Published October 28, 2024

© 2024 The Authors. Published by American Chemical Society. This publication is licensed under

CC-BY 4.0 .

Abstract

Click to copy section linkSection link copied!
High Resolution Image
Download MS PowerPoint Slide

Physical reservoir computing (PRC) using synaptic devices has attracted attention as a promising edge artificial intelligence device. To handle time-series data on various time scales, it is necessary to fabricate devices with the desired time scale. In this study, we fabricated a dye-sensitized solar-cell-based synaptic device with controllable time constants by changing the light intensity. This device showed synaptic features, such as paired-pulse facilitation and paired-pulse depression, in response to light intensity. Moreover, we found that the high computational performance of the time-series data processing task was achieved by changing the light intensity, even when the input pulse width was varied. In addition, the fabricated device can be used for motion recognition tasks. This study paves the way for realizing multiple time-scale PRC.

This publication is licensed under

CC-BY 4.0 .
  • cc licence
  • by licence
© 2024 The Authors. Published by American Chemical Society

Subjects

what are subjects

Keywords

what are keywords

Introduction

Click to copy section linkSection link copied!
With the rapid growth of sensor networks, (1) new technologies can be developed to predict emergency events such as earthquakes, (2) volcanic eruptions, (3) heart attacks, (4) and buried pipeline failures (5) using artificial intelligence (AI). As the number of sensors increases, several challenges emerge, including higher network load, delayed data transfer, and increased power consumption on the server. (6) To address these challenges, in-sensor edge AI devices that incorporate AI functionality within the sensor are in demand. To predict emergency events in advance, time-series data need to be processed by AI. Feed-forward artificial neural networks (ANNs), a common AI technology, are not inherently designed for time-series data processing. (7) Therefore, reservoir computing (RC) has attracted considerable attention because it is specifically designed for time-series data processing. RC utilizes a reservoir layer with random coupling weights instead of a hidden layer in the feed-forward ANN, enabling time-series data to be handled with a lower power consumption. Among RC frameworks, physical reservoir computing (PRC), which uses physical devices as a reservoir layer, can integrate sensors and AI. Thus, PRC has garnered interest as a promising in-sensor edge AI technology. Thus, far, spintronic devices, (8,9) memristor devices, (10,11) ionic devices, (12,13) and photonic devices (14,15) have been reported to serve as reservoir layers in PRC.
The reservoir layer in PRC requires short-term memory (STM), nonlinearity, and high dimensionality. (16) Artificial synapses, which mimic synapses utilizing physical devices, are gaining interest as a reservoir layer for the PRC because synaptic plasticity, a fundamental synaptic function, achieves STM. Specifically, the state of the reservoir depends on past and current inputs. Moreover, approximately 90% of the information received by the human brain comes from vision. (17) Therefore, PRC with optoelectronic artificial synapses is expected to realize high recognition and real-time processing capabilities similar to those of the human visual system.
Thus, far, PRC with optoelectronic synaptic devices using various materials such as InGaZnO, (18) cellulose-ZnO nanocomposites, (19) and GaOx (20) have been reported. However, these devices operate based on a photocurrent, resulting in high power consumption due to the need to apply a bias voltage. Therefore, self-powered photovoltaic devices are preferred devices for reducing power consumption because these devices can be driven by incident light input signals. Thus, far, PRC with self-powered optoelectronic devices has only utilized band bending mechanisms, and no systematic research has been conducted on PRC with photovoltaics-based artificial optoelectronic synapse devices using typical solar cells (i.e., silicon, copper indium gallium selenide, dye-sensitized, and perovskite solar cells). Lao et al. proposed a self-powered optoelectronic PRC system using band bending at the interface between semiconductors and insulators. (21) This device showed 99.97% and 100% accuracy in face recognition and vehicle flow tasks, respectively. Additionally, the decay time constant of the device could be varied from 5.4 to 11.8 s by changing the number of light pulses. However, it was not possible to change the time constant significantly. Signals for monitoring infrastructure, the natural environment, and health conditions contain time-series data across various time scales. To handle these times-series data, the response time of the reservoir layer should be designed to match the time scale of the input data. Scholars have reported that dye-sensitized solar cells (DSCs) exhibit a time constant of the transient response at the open-circuit voltage (Voc) that depends on the light intensity (P). (22) Therefore, the response time can be adjusted according to the input data without altering the device structure. Furthermore, a DSC requires almost zero operating energy and functions as an in-sensor device because it operates using the energy of the input light. In addition, the carrier transport kinetics in a DSC are complex and involve multiple electrochemical reactions. (23) Consequently, DSCs can be expected to possess the nonlinearity required for the reservoir layer. (16)
In this study, we developed DSC-based synaptic devices to realize multiple-time scale PRC. Using a laser, we measured the transient voltage response as a function of P. This device exhibits synaptic features, such as paired-pulse facilitation (PPF) and paired-pulse depression (PPD). Moreover, it can be controlled by changing P. To evaluate the computational performance of our PRC system with DSC-based synaptic devices on various time scales, we conducted a STM task and a parity check (PC) task with different input pulse widths (Tp). We found that Tp with maximum memory capacity in the STM and PC tasks varied depending on P. This result suggests that the operating time scale of our PRC system can be altered by changing P. This novel PRC system can be applied to the processing of time-series data across various time scales. As a demonstration, a motion recognition task was also performed. Our PRC system can recognize human actions with high accuracy and can be used in intelligent camera applications in the future.

Results and Discussion

Click to copy section linkSection link copied!
We fabricated DSCs with squarylium derivative-based dyes (SQ2) that exhibit an absorption range of 550–700 nm. Figure 1a depicts the device and molecular structures of the SQ2. Figure 1b presents the incident photon-to-current conversion efficiency (IPCE) for the DSC. The spectrum shows two absorption ranges: 305–365 and 530–700 nm. The former matched the band edge absorption of TiO2 (3.2 eV), (24) whereas the latter was consistent with the absorption spectrum of SQ2. Then, a 658 nm laser was employed as the input light to distinguish between the injected electrons from the dye and the excited electrons in TiO2. Figure 1c displays the J–V curve as a function of the P. As expected, Voc and the short-circuit current (Jsc) increase with P. Figure 1d presents Voc and Jsc as functions of P.

Figure 1

Figure 1. Fundamental characteristics of the fabricated DSC. (a) Device structure of the DSC and molecule structure of SQ2. (b) Incident photon-to-current conversion efficiency of the DSC. (c) J–V curves of the DSCs (λ: 658 nm). (d) Voc and Jsc as a function of the incident light intensity, P.

High Resolution Image
Download MS PowerPoint Slide
We investigated the transient voltage response with respect to P, as illustrated in Figure 2a. At low P, such as 0.05 mW, it takes 5 s to reach Voc. By contrast, at high P, such as 15 mW, it takes 45 ms to reach Voc. To further explore the effect of P on the transient voltage response, we estimated the average rise time (τrise) and decay time (τdecay) constants, as shown in Figure 2b,c. The τrise and τdecay curves were fitted using single and double exponential functions, respectively. When P increases from 0.1 to 10 mW, τrise dramatically decreases from 0.75 s to 8.9 ms. Moreover, τdecay decreases with P in the range of 0.1–1 mW. Above 1 mW, τdecay remains constant within the 1–10 mW range, regardless of P. This feature of varying time constants with P is unique among solar cells. The typical transient voltage response of polycrystalline Si solar cells does not change with P (Figure S1). The rise time constant remains constant regardless of P, in contrast to DSC (Figure S2).

Figure 2

Figure 2. Transient voltage responses. (a) Normalized transient voltage responses as a function of P. (b,c) Rise time constant and decay time constant as a function of P. (d) Band diagram of TiO2 at high and low P levels when light irradiation starts. (e) Band diagram of TiO2 at high and low P levels when the light irradiation at a steady state is stopped.

High Resolution Image
Download MS PowerPoint Slide
For DSCs, scholars have reported that τrise decreases with increasing P. (22) This phenomenon can be attributed to the fact that the diffusion coefficient (Dn) of free electrons in TiO2 depends on P. In other words, Dn depends on the electron density (nc) in the conduction band of TiO2. Carrier transport in the TiO2 layer in DSCs can be interpreted by using the multiple trapping (MT) model. This model has been adopted for nanostructured semiconductors, drawing from extensive experience with disordered semiconductors. (25) By applying the MT model, solving the continuity equation for the electrons injected into the TiO2 layer, and considering trapping and detrapping processes, we obtain Dn as
Dn=NLNcαncα−1
(1)
where NL is the total density of localized sites (per unit volume), NC is the effective density of the conduction band states, and α is a parameter that is obtained by dividing the temperature (T) by the tailing parameter T0 (α = T/T0). We assumed that traps had an exponential distribution. Dn is also known to depend on nc. Specifically, at high P, high diffusion coefficients are obtained owing to an increase in nc, which can shorten τrise. Figure 2d depicts the energy diagram for a DSC when light irradiation starts at low and high P, respectively. Upon light irradiation, electrons are excited from the highest occupied molecular orbital level to the lowest unoccupied molecules orbital level in the dye. Then, the electrons are injected into TiO2. At low P, Dn is reduced because many electrons are trapped in comparison to the conduction band electrons. Thus, τrise is longer at low P. By contrast, Dn increases at high P because fewer electrons are trapped than conduction band electrons. Therefore, τrise is shorter at high P.
When light irradiation at a steady state is stopped, the voltage decays with time. The electron lifetime (τn) can be expressed as
τn=−kBTe(dVdt)−1
(2)
where V is the open-circuit voltage decay curve of DSC and kBT/e is the thermal energy, in which kB and e are the Boltzmann constant and elementary charge, respectively. (26) Therefore, τdecay depends on τn, which governs the recombination process. No electron recombination occurs in TiO2 owing to the absence of hole carriers. Possible recombination processes include either transferring electrons to the oxidized dye or transferring electrons to the acceptor in the electrolyte, as depicted in Figure 2e. Numerous studies have assumed that oxidized molecules are rapidly regenerated, resulting in recombination occurring only with the acceptor in the electrolyte. (23) Salafsky et al. reported that the recombination rate with the acceptor ranges from a few hundred milliseconds to a few seconds. (27) In addition, Huang et al. found that the reaction with I2– can be rate-limiting. (28) Consequently, in our work, the negligible change shown by τdecay in response to P can be attributed to the rate-limiting reaction with I3–. Considering this concept, the voltage response time of DSC can be altered by changing the P. This characteristic suggests that the response times of DSC-based synaptic devices and in PRC applications can be controlled by simply adjusting the P.
Next, we investigated the effect of P on the synaptic features of the DSC-based synaptic devices. Short-term plasticity is a fundamental synaptic characteristic that is believed to be closely related to the ability to decode temporal information. (29) Among them, PPF and PPD, which are facilitation or depression induced by two successive pulses, are typically utilized to evaluate the characteristics of synaptic devices. (30) Figure 3a shows a typical transient voltage response induced by low and high P (λ: 658 nm, Tp: 100 ms, and P: 0.2, 5.0 mW). At low P, the voltage increases with successive light irradiation. By contrast, at high P, the voltage was saturated with the first pulse and did not increase with the second light irradiation. This is because short time constants with high P values resulted in saturation during the first pulse. Once saturated with the first pulse, there is no room to increase with the second pulse. Therefore, PPF is expected to occur when Tp is less than time constant. To investigate the PPF in the DSC-based synaptic device, we measured the PPF index, which is defined as V2/V1 × 100, as displayed in Figure 3a. Figure 3b shows the PPF index as a function of ΔT with various P values. The PPF index increases with decreasing ΔT. When ΔT is fixed, the PPF index increases with decreasing P, reaching a maximum of 223% (P: 0.075 mW, ΔT: 10–4 s). Figure 3c shows the average PPF index as a function of Tp and ΔT with various P values. Besides ΔT, Tp also influences the PPF index. Interestingly, the PPF index distribution, which exhibits a high PPF index above 100%, shifts with the changes in P. For example, when P = 10 mW, no facilitation takes place, leading to a 100% PPF index at Tp = 0.1 s. However, upon reduction of P to 0.075 mW, facilitation occurs, resulting in a 175% PPF index at Tp = 0.1 s. When Tp is in the range of subseconds to tens of milliseconds, no facilitation occurs at high P; however, facilitation takes place when P is reduced. This result indicates that the response time, which exhibits synaptic features, can be modified by adjusting P.

Figure 3

Figure 3. Synaptic characteristics in the DSC. (a) Transient voltage responses induced by two optical pulses with high and low P (λ: 658 nm, Tp: 100 ms, ΔT: 100 ms, and P: 0.2, 5.0 mW). (b) PPF index as a function of ΔT (Tp: 100 ms, P: 0.075–2.5 mW). (c) Tp dependence of PPF index as functions of P and ΔT. (d) Transient voltage response induced by two optical pulses with different P values (λ: 658 nm, Tp: 100 ms, ΔT: 100 ms, P1: 5 mW, and P2: 1–15 mW). (e) PPF index map for P1 with various P2 (λ: 658 nm, Tp: 100 ms, P1: 1–15 mW, and P2: 1–15 mW). (f) Transient voltage responses induced by 10 optical pulses with various P values (Tp: 1 ms, P1: 1–15 mW). (g) Switching facilitation and depression of the DSC by changing P (P: 1, 10 mW).

High Resolution Image
Download MS PowerPoint Slide
To achieve depression, we changed P at the first (P1) and second pulses (P2). Figure 3d illustrates the transient voltage response at a constant P1 with various P2 values (P1: 5 mW, P2: 1–15 mW). When P2 is smaller than P1 (e.g., P2 = 1–2 mW), V2 is lower than V1. Therefore, PPD is achieved at P2 < P1. By contrast, V2 is greater than V1 (e.g., P2 = 5–15 mW), which implies that PPF is achieved at P2 > P1. Figure 3e shows the average PPF index map for P1 with various P2 values (λ: 658 nm, Tp: 100 ms, P1: 1–15 mW, and P2: 1–15 mW). In addition, the PPF index is approximately 100% when P1 and P2 have comparable values. When P2 > P1, the PPF index exceeds 100%, indicating facilitation. By contrast, when P2 < P1, the PPF index is less than 100%, suggesting that depression occurs.
Figure 3f displays the transient voltage response induced by 10 light pulses. The first five pulses were irradiated at P = 5 mW, and the remaining five pulses were irradiated at various P values. Facilitation or depression occurs in response to P. Figure 3g shows the transient voltage response when P is changed every 10 pulses. Depression and facilitation occur in response to P. Therefore, the accumulation and forgetting of time-series changes in P are reflected as variations in the voltage. This feature is known as fading memory, which is a desirable characteristic for the reservoir layer in the PRC. Consequently, DSC-based synaptic devices can be applied to the reservoir layer in the PRC.
To verify the feasibility of our DSC-based synaptic device for PRC, a standard benchmark task for time-series processing was performed. We used STM and PC tasks. (31,32) The STM task, which assesses the ability to reproduce past input, is used as a benchmark task to evaluate short-term memorability. The PC task, which assesses the ability to reproduce exclusive or past input, is utilized as a benchmark task to evaluate nonlinearity. The operating scheme of a PRC with a DSC-based synaptic device is illustrated in Figure 4a. Initially, random binary input waveforms are converted into light pulses. Then, light pulses are irradiated on the device using a laser. The transient voltage is measured and fed into the NN to calculate the output weight (Wout). The target waveform is then calculated using Wout. For each task, the input data (yinput) and target data (ytarget) are described by the following equations
ytarget(t)=yinput(t−d):STMtask
(3)
ytarget(t)=[∑d=0dmaxyinput(t−d)](mod⁡2):PCtask
(4)

Figure 4

Figure 4. Time-series data processing task. (a) Operating scheme of the time-series data processing task with a DSC-based synaptic device. (b,c) System output of the STM and PC tasks (d: 1, Tp: 100 ms, P: 1 mW, and number of virtual nodes: 125). (d,e) Forgetting curves of the STM and PC tasks with various P (d: 1, Tp: 100 ms, P: 0.075–10 mW, and number of virtual nodes: 125).

High Resolution Image
Download MS PowerPoint Slide
Figure 4b,c shows the system outputs for the STM and PC tasks for a DSC-based synaptic device (d: 1, Tp: 100 ms, P: 1 mW, and virtual node: 125). One can estimate the memory capacity (C) using
C=∑d=1dmax[Corr(ytarget(t),yinput(t))]2
(5)
where “corr” denotes the correlation coefficient. A larger C value indicates that more past input is retained or that more nonlinearities are present in the reservoir. Figure 4d,e displays the square of the correlation coefficient as a function of delay for the STM and PC tasks with various P (d: 1, Tp: 100 ms, P: 0.075–10 mW; and virtual node: 125). The PRC system with DSC-based synaptic device exhibits maximum values of C in the STM task (CSTM) and PC task (CPC) for 1.31 and 1.13, respectively. Compared to previous literature on various physical reservoirs, our CSTM values are smaller than those reported values (CSTM: 1–35). (19,33−38) Meanwhile, the CPC is comparable to those reported values (CSTM: 0.4–1.7). (19,35,36) In addition, to verify the capability of a DSC-based synaptic device for PRC, STM and PC tasks were performed with the input signal itself. The PRC system without DSC-based synaptic device exhibits nearly zero value of CSTM and CPC, as shown in Figure S3. We found that CSTM and CPC are higher when the light input has medium to high intensities compared to low intensity. This indicates that longer time constants of the device do not necessarily lead to higher STM or nonlinearity. Previous studies have shown that high CSTM is exhibited when the time constants of the device and Tp are comparable. (33,39) Therefore, we investigated the correlation between the pulse width of light at each irradiation intensity and both CSTM and CPC.
We fabricated eight samples and characterized their CSTM values and CPC. Figure 5a,b shows average CSTM and CPC as a function of Tp at each light intensity, respectively. At a light intensity of 0.075 mW, CSTM reached a maximum value of 1.31 at Tp = 500 ms. While the value remained almost unchanged for Tp longer than 500 ms, we observed the steep decline in CSTM for Tp shorter than 500 ms. At light intensities of 1 mW or higher, relatively high CSTM was maintained even in the region where Tp was less than 100 ms. In other words, within the time range of this graph, with increasing light intensity, the range of pulse widths that yield high CSTM values expands. CPC showed a similar trend. As the irradiation intensity increased, the range of pulse widths that exhibited high CPC values also expanded. Within the time range of this graph, it was observed that higher irradiation intensities resulted in higher CPC values (with a maximum of 1.13 at P = 10 mW).

Figure 5

Figure 5. Time-series data processing task as a function of Tp with various P. Error bar represents standard error of the mean. (a) Average CSTM as a function of Tp. (b) Average CPC as a function of Tp. (c) Transient voltage response induced by random binary pulses with various Tp (P: 10 mW, Tp: 1–1000 ms). (d) Histogram of the transient voltage response with various Tp (P: 10 mW, Tp: 1–1000 ms).

High Resolution Image
Download MS PowerPoint Slide
We conducted a more detailed investigation to understand why the memory capacity changes with respect to Tp. Figure 5c shows the output voltage in response to random input pulses (Tp = 1–1000 ms) used to characterize the STM task. The light intensity was 10 mW. It was observed that the range of output voltage varies depending on Tp. Figure 5d displays the histogram of the output voltage. The maximum value of all the data represents the Voc of the DSC. The results showed that the longer the Tp, the wider the range of the output voltage. Looking into the details, it appears that when Tp is smaller than the time constant of the device (τrise = 8.9 ms, and τdecay = 2.3 and 79 ms at P = 10 mW), the range of output voltage becomes narrower. A wider output voltage range corresponds to a higher CSTM. The reason might be that the wide range of output voltage would increase the variation in the values of the virtual nodes, resulting in an improvement in reproducing past input. These results suggest that a device design capable of achieving a wide range of output voltage across various light irradiation intensities and time scales is crucial for realizing high memory capacity. In applications where the irradiation intensity is fixed, a design that corresponds to the time scale of the incident light will be required. Additionally, from another perspective, when the pulse width and light intensity of the input light signal change simultaneously, the output voltage profile becomes non-exponential, which suggests that a physical reservoir with high nonlinearity can be constructed.
The fabricated DSC-based synaptic device exhibits the unique capability of output voltage values based on the accumulation and decay of time-series variations in P. This characteristic suggests that the device may be well-suited for processing time-series data related to P. To validate this, we conducted a motion recognition task using time-series changes in P as an input. The objective of this task was to classify the actions depicted in a video. In this experiment, a single DSC was utilized.
Figure 6a shows a schematic diagram of the motion recognition task using a PRC system using a DSC-based synaptic device. Human motions were captured by a camera and then binarized. Next, the video was split into eight parts, and the time-series data of the average luminance [L(t)] for each strip were obtained. L(t) was converted into P [P(t)] and then input into the DSC through a laser. The time scale of human motion is on the order of subseconds to seconds. Therefore, P was set to 0.1–1 mW with subsecond to second time constants. The next input was applied after waiting for complete relaxation of the voltage. The obtained voltage values were fed into a NN for motion recognition. Figure 6b displays the transient voltage responses with eight divided motion videos (jump). Figure 6c shows the confusion matrix of the motion recognition task. All actions (i.e., bend, jump, run, side, wave1, and wave2) are distinguished with over 80% accuracy. The total accuracy is 92%, indicating that motion recognition is successfully achieved by using the fabricated device. For comparison, the results of the motion recognition task in different machine-learning conditions are presented (Figure S4).

Figure 6

Figure 6. Demonstration of motion recognition task using a PRC system with the DSC-based synaptic device. (a) Schematic diagram of the motion recognition task using the DSC. (b) Transient voltage response of the DSC output with eight divided motion movies (jump). (c) Confusion matrix of the motion recognition tasks.

High Resolution Image
Download MS PowerPoint Slide
To verify the effectiveness of the fabricated device as part of a PRC system, we also conducted a motion recognition task without a DSC-based synaptic device. Figure S5 shows the results of this task, highlighting the performance comparison. The calculated accuracy of the PRC system without DSC-based synaptic device saturated at 17%. This result underscores the potential of DSC-based synaptic devices within the PRC system. We believe that our device, which has significantly fewer pixels than a modern camera, could achieve recognition tasks comparable to the human visual system by utilizing the synaptic responses of the DSC-based synaptic device. For this demonstration, we used laser light as an ideal light source due to the limited amount of training data. For practical applications, environmental light could serve as a light source if the volume of training data is increased. In addition, our device can be configured in a two-dimensional array to provide spatial resolution, enabling the realization of advanced neuromorphic vision systems.

Conclusions

Click to copy section linkSection link copied!
In this study, we fabricated DSC-based synaptic devices to achieve PRC. By adjusting P, these devices can significantly control τrise from 8.9 ms to 0.75 s. This modulation is attributed to the changing electron diffusion coefficient, which is influenced by the exponential trap distribution. Leveraging these characteristics, we demonstrated that synaptic features could be precisely controlled by varying the intensity of the incident light, achieving zero power consumption. To verify the feasibility of PRC using DSC-based synaptic devices, we conducted time-series data processing tasks, such as STM and PC tasks, using various pulse widths and light intensities. Both CSTM and CPC reached 1.31 and 1.13, respectively. This finding indicates that these devices are photovoltaic-based artificial synapse devices that can be driven by incident light input signals. Additionally, we observed that high C values can be achieved at P values at which τrise and Tp are comparable. Our findings also reveal that the operating time scale of PRC with DSC-based synaptic devices can be altered by adjusting the P. Furthermore, the device demonstrated a high accuracy of 92% in motion recognition tasks. This study represents the first time that the potential application of self-powered artificial synapses with solar cells to PRC has been investigated. Our findings pave the way for the realization of multiple-time scale PRC, highlighting the potential for advanced applications in edge AI and neuromorphic computing.

Experimental Section

Click to copy section linkSection link copied!

Device Fabrication

A fluorine-doped tin oxide (FTO) glass substrate (size: 2.5 × 2.5 cm, conductivity: 7 Ω/□, NPV-CFT2–7C, AS ONE CORPORATION, Japan) was cleaned with acetone, ethanol, and deionized water. Then, TiO2 paste (PST-18NR, JGC Catalysts and Chemicals Ltd., Japan) was coated onto the FTO glass substrate using the doctor blade method. Then, the TiO2 electrode was heated at 150 °C for 30 min, followed by heating at 450 °C for 30 min. Typically, the thickness of the TiO2 layer was 3 μm. The TiO2 electrode was immersed in a 0.1 mM squarylium derivative-based dye (SQ2, Solaronix S.A., Aubonne, Switzerland) solution in acetonitrile (014-00386, FUJIFILM Wako Pure Chemical Corporation, Japan) for 24 h. Next, the dye-coated TiO2 electrode was assembled with a thermoplastic spacer (HIMILAN, Dow-Mitsui Polychemicals Company, Ltd., Japan) and filled with 0.15 M triiodide solution (Z-150, Solaronix S.A., Aubonne, Switzerland) as the electrolyte. Finally, a Pt plate (size: 2 × 2 cm) was assembled as the counter electrode. The active electrode area was typically 1.5 cm2. All processes were conducted under ambient pressure and room temperature.

Device Characterization

IPCE spectra were obtained using a Peccell Technologies S10AC system with a 150 W xenon lamp. We set the step interval to 5 nm and the delay to 2 s. Electrical measurements were performed using a source meter (2400, Keithley, OH, USA), DAQ device (USB-6366, National Instruments, TX, USA), and laser diode (λ = 658 nm, L658P040, THORLABS Inc., USA). The laser spot diameter was less than 4 mm. The radiation power ranged from 0.01 to 15 mW. The time constants were obtained from the transient open-circuit voltage induced by laser irradiation (Tp: 5 s). All processes were conducted under ambient pressure and room temperature.

Time-Series Data Processing Task

The input waveform consisted of 1000 random binary signals, where 0 and 1 represented the ON and OFF states of the laser irradiation, respectively. Tp ranged from 1 to 1000 ms. The number of virtual nodes was set to 125. The washout was set to 200 pulses. The technical details have been previously presented elsewhere. (19,40) We fabricated eight devices and performed STM and PC tasks with two data sets, respectively.

Motion Recognition Task

Motion recognition data sets were derived from three persons with six actions (i.e., bend, wave1, wave2, jump, run, and side). Bend, wave1, and wave2 represent crunching, waving one hand, and waving both hands, respectively. Jump, run, and side represent actions to move from one edge of the screen to the other by jumping, running, or moving sideways, respectively. Motions were recorded using a commercial camera (EOS Kiss X9, Canon Inc., Japan) at 60 FPS. The video duration varies slightly depending on the type of video. The video was converted into a binary format at 30FPS. Subsequently, the video was divided into eight strips. To convert the video into a time variation of the laser P, we calculated the average luminance Pave(t) from P(x, y, t), which is the luminance of each pixel in each strip. Then, to highlight the motion features, we calculated Pave(t)−PminPmax. Pmin and Pmax are the minimum and maximum values of P(t), respectively. The obtained time-series data were fed into the laser, and the transient response of Voc was measured. Pmax was set to 1 mW. To preserve the motion speed, Tp was set to 0.033 s. The readout 0 s was defined as the time at which the video ended. We measured Voc at a readout of 10 ms and fed it into a simple NN with one layer. We used an open-source library (Keras) for NN processing. We used cross-entropy loss functions. The NN consists of a 1 × 8 input layer and 1 × 6 output layer and no hidden layers. Softmax function and RMSprop were used as activation functions and optimizer, respectively. The video database consisted of 24 videos per person in one action. We used 57 videos per action for training and 15 videos for testing.

Data Availability

Click to copy section linkSection link copied!

The data sets used and/or analyzed during this study are available from the corresponding author on reasonable request.

Supporting Information

Click to copy section linkSection link copied!

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsami.4c11061.

  • Transient response of polycrystalline Si solar cells, rise time constant of Si solar cells and DSC as a function of P, result of time-series processing task with and without DSC-based synaptic device, motion recognition accuracy with various machine learning algorithms, and motion recognition accuracy with and without DSC-based synaptic devices (PDF)

Terms & Conditions

Most electronic Supporting Information files are available without a subscription to ACS Web Editions. Such files may be downloaded by article for research use (if there is a public use license linked to the relevant article, that license may permit other uses). Permission may be obtained from ACS for other uses through requests via the RightsLink permission system: http://pubs.acs.org/page/copyright/permissions.html.

Author Information

Click to copy section linkSection link copied!
  • Corresponding Author
  • Authors
    • Hiroaki Komatsu - Department of Applied Electronics, Graduate School of Advanced Engineering, Tokyo University of Science, Katsushika, Tokyo 125-8585, Japan
    • Norika Hosoda - Department of Applied Electronics, Graduate School of Advanced Engineering, Tokyo University of Science, Katsushika, Tokyo 125-8585, Japan
  • Author Contributions

    H.K. and T.I. conceived the research and experiments. H.K, N.H., and T.I. contributed experiments, analyses, and visualization. H.K. contributed the development of measurement equipment and analysis software. H.K. wrote the manuscript with contributions from T.I. T.I. supervised the research and revised the manuscript. All authors have read and agreed to the published version of the manuscript.

  • Notes
    The authors declare no competing financial interest.

Acknowledgments

Click to copy section linkSection link copied!

The authors express their gratitude to Tatsuya Yamamoto, Naoki Kiyota, and Prof. Morio Nagata from Tokyo University of Science for their technical support with the IPCE measurements. This work was partially supported by JST, the establishment of university fellowships towards the creation of science technology innovation (grant number JPMJFS2144). Additional support was provided by JST SPRING (grant number JPMJSP2151). We would like to thank Editage (www.editage.com) for English language editing.

Abbreviations

Click to copy section linkSection link copied!
AI

artificial intelligence

ANN

artificial neural network

DSC

dye-sensitized solar cell

FTO

fluorine-doped tin oxide

IPCE

incident photon-to-current conversion efficiency

MT

multiple trapping

PC

parity check

PPD

paired pulse depression

PPF

paired pulse facilitation

PRC

physical reservoir computing

RC

reservoir computing

STM

short-term memory

STP

short-term plasticity

References

Click to copy section linkSection link copied!

This article references 40 other publications.

  1. 1
    Alam, M.; Tehranipoor, M. M.; Guin, U. TSensors Vision, Infrastructure and Security Challenges in Trillion Sensor Era. J. Hardw. Syst. Secur. 2017, 1, 311– 327,  DOI: 10.1007/s41635-017-0028-8
    Google Scholar
    There is no corresponding record for this reference.
  2. 2
    Banna, M. H. A.; Taher, K. A.; Kaiser, M. S.; Mahmud, M.; Rahman, M. S.; Hosen, A. S. M. S.; Cho, G. H. Application of Artificial Intelligence in Predicting Earthquakes: State-of-the-art and Future Challenges. IEEE Access 2020, 8, 192880– 192923,  DOI: 10.1109/ACCESS.2020.3029859
    Google Scholar
    There is no corresponding record for this reference.
  3. 3
    Anantrasirichai, N.; Biggs, J.; Albino, F.; Hill, P.; Bull, D. Application of Machine Learning to Classification of Volcanic Deformation in Routinely Generated InSAR Data. J. Geophys. Res. Solid Earth 2018, 123, 6592– 6606,  DOI: 10.1029/2018JB015911
    Google Scholar
    There is no corresponding record for this reference.
  4. 4
    Domyati, A.; Memon, Q. Machine Learning Based Improved Heart Disease Detection With Confidence. Int. J. Online Biomed. Eng. 2023, 19, 130– 143,  DOI: 10.3991/ijoe.v19i08.37417
    Google Scholar
    There is no corresponding record for this reference.
  5. 5
    Wang, W.; Wang, Y.; Zhang, B.; Shi, W.; Li, C.-Q. Failure Prediction of Buried Pipe Network With Multiple Failure Modes and Spatial Randomness of Corrosion. Int. J. Pressure Vessels Piping 2021, 191, 104367,  DOI: 10.1016/j.ijpvp.2021.104367
    Google Scholar
    There is no corresponding record for this reference.
  6. 6
    Sun, X.; Ansari, N. EdgeIoT: Mobile Edge Computing for the Internet of Things. IEEE Commu. Mag. 2016, 54, 22– 29,  DOI: 10.1109/mcom.2016.1600492cm
    Google Scholar
    There is no corresponding record for this reference.
  7. 7
    Brezak, D.; Bacek, T.; Majetic, D.; Kasac, J.; Novakovic, B. A comparison of feed-forward and recurrent neural networks in time series forecasting. In Proceedings of the IEEE/IAFE Computational Intelligence for Financial Engineering , 2012. DOI: 10.1109/CIFER.1997 .
    Google Scholar
    There is no corresponding record for this reference.
  8. 8
    Torrejon, J.; Riou, M.; Araujo, F. A.; Tsunegi, S.; Khalsa, G.; Querlioz, D.; Bortolotti, P.; Cros, V.; Yakushiji, K.; Fukushima, A.; Kubota, H.; Yuasa, S.; Stiles, M. D.; Grollier, J. Neuromorphic Computing With Nanoscale Spintronic Oscillators. Nature 2017, 547, 428– 431,  DOI: 10.1038/nature23011
    Google Scholar
    8
    Neuromorphic computing with nanoscale spintronic oscillators
    Torrejon, Jacob; Riou, Mathieu; Araujo, Flavio Abreu; Tsunegi, Sumito; Khalsa, Guru; Querlioz, Damien; Bortolotti, Paolo; Cros, Vincent; Yakushiji, Kay; Fukushima, Akio; Kubota, Hitoshi; Yuasa, Shinji; Stiles, Mark D.; Grollier, Julie
    Nature (London, United Kingdom) (2017), 547 (7664), 428-431CODEN: NATUAS; ISSN:0028-0836. (Nature Publishing Group)
    Neurons in the brain behave as nonlinear oscillators, which develop rhythmic activity and interact to process information. Taking inspiration from this behavior to realize high-d., low-power neuromorphic computing will require very large nos. of nanoscale nonlinear oscillators. A simple estn. indicates that to fit 108 oscillators organized in a two-dimensional array inside a chip the size of a thumb, the lateral dimension of each oscillator must be smaller than one micrometre. However, nanoscale devices tend to be noisy and to lack the stability that is required to process data in a reliable way. For this reason, despite multiple theor. proposals and several candidates, including memristive and superconducting oscillators, a proof of concept of neuromorphic computing using nanoscale oscillators has yet to be demonstrated. Here we show exptl. that a nanoscale spintronic oscillator (a magnetic tunnel junction) can be used to achieve spoken-digit recognition with an accuracy similar to that of state-of-the-art neural networks. We also det. the regime of magnetization dynamics that leads to the greatest performance. These results, combined with the ability of the spintronic oscillators to interact with each other, and their long lifetime and low energy consumption, open up a path to fast, parallel, on-chip computation based on networks of oscillators.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2sXht1ent7bL&md5=444c9c101e6c4acc1739e32f5ba02aa4
  9. 9
    Nakane, R.; Tanaka, G.; Hirose, A. Reservoir Computing With Spin Waves Excited in a Garnet Film. IEEE Access 2018, 6, 4462– 4469,  DOI: 10.1109/ACCESS.2018.2794584
    Google Scholar
    There is no corresponding record for this reference.
  10. 10
    Midya, R.; Wang, Z.; Asapu, S.; Zhang, X.; Rao, M.; Song, W.; Zhuo, Y.; Upadhyay, N.; Xia, Q.; Yang, J. J. Reservoir Computing Using Diffusive Memristors. Adv. Intell. Syst. 2019, 1, 1900084,  DOI: 10.1002/aisy.201900084
    Google Scholar
    There is no corresponding record for this reference.
  11. 11
    Zhang, G.; Qin, J.; Zhang, Y.; Gong, G.; Xiong, Z.-Y.; Ma, X.; Lv, Z.; Zhou, Y.; Han, S.-T. Functional Materials for Memristor-based Reservoir Computing: Dynamics and Applications. Adv. Funct. Mater. 2023, 33, 2302929,  DOI: 10.1002/adfm.202302929
    Google Scholar
    There is no corresponding record for this reference.
  12. 12
    Nishioka, D.; Tsuchiya, T.; Namiki, W.; Takayanagi, M.; Imura, M.; Koide, Y.; Higuchi, T.; Terabe, K. Edge-of-chaos Learning Achieved by Ion-electron–coupled Dynamics in an Ion-gating Reservoir. Sci. Adv. 2022, 8, eade1156  DOI: 10.1126/sciadv.ade1156
    Google Scholar
    There is no corresponding record for this reference.
  13. 13
    Koh, S.-G.; Shima, H.; Naitoh, Y.; Akinaga, H.; Kinoshita, K. Reservoir Computing With Dielectric Relaxation at an Electrode–ionic Liquid Interface. Sci. Rep. 2022, 12, 6958,  DOI: 10.1038/s41598-022-10152-9
    Google Scholar
    13
    Reservoir computing with dielectric relaxation at an electrode-ionic liquid interface
    Koh, Sang-Gyu; Shima, Hisashi; Naitoh, Yasuhisa; Akinaga, Hiroyuki; Kinoshita, Kentaro
    Scientific Reports (2022), 12 (1), 6958CODEN: SRCEC3; ISSN:2045-2322. (Nature Portfolio)
    A phys. reservoir device with tunable transient dynamics is strongly required to process time-series data with various timescales generated in the edge region. In this study, we proposed using the dielec. relaxation at an electrode-ionic liq. (IL) interface as the phys. reservoir by making the most of designable physicochem. properties of ILs. The transient dynamics of a Au/IL/Au reservoir device were characterized as a function of the alkyl chain length of cations in the IL (1-alkyl-3-methylimidazolium bis(trifluoromethane sulfonyl)imide). By considering a weighted sum of exponentials expressing a superposition of Debye-type relaxations, the transient dynamics were well reconstructed. Although such complex dynamics governed by multiple relaxation processes were obsd., each extd. relaxation time scales with a power law as a function of IL's viscosity detd. by the alkyl chain length of cations. This indicates that the relaxation processes are characterized by bulk properties of the ILs that obey the widely received Vogel-Fulcher-Tammann law. We demonstrated that the 4-bit time-series signals were transformed into the 16 classifiable data, and the data transformation, which enables to achieve higher accuracy in an image classification task, can be easily optimized according to the features of the input signals by controlling the IL's viscosity.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BB38XhtFGnsrnJ&md5=bb69fcba63e6f6860bd36b7588bc7e79
  14. 14
    Van der Sande, G.; Brunner, D.; Soriano, M. C. Advances in Photonic Reservoir Computing. Nanophotonics 2017, 6, 561– 576,  DOI: 10.1515/nanoph-2016-0132
    Google Scholar
    There is no corresponding record for this reference.
  15. 15
    Larger, L.; Soriano, M. C.; Brunner, D.; Appeltant, L.; Gutierrez, J. M.; Pesquera, L.; Mirasso, C. R.; Fischer, I. Photonic Information Processing Beyond Turing: an Optoelectronic Implementation of Reservoir Computing. Opt. Express 2012, 20, 3241– 3249,  DOI: 10.1364/OE.20.003241
    Google Scholar
    There is no corresponding record for this reference.
  16. 16
    Cucchi, M.; Abreu, S.; Ciccone, G.; Brunner, D.; Kleemann, H. Hands-on Reservoir Computing: a Tutorial for Practical Implementation. Neuromorph. Comput. Eng. 2022, 2, 032002,  DOI: 10.1088/2634-4386/ac7db7
    Google Scholar
    There is no corresponding record for this reference.
  17. 17
    Pocock, D. C. D. Sight and Knowledge. Trans. Inst. Br. Geogr. 1981, 6, 385– 393,  DOI: 10.2307/621875
    Google Scholar
    There is no corresponding record for this reference.
  18. 18
    Yang, Y.; Cui, H.; Ke, S.; Pei, M.; Shi, K.; Wan, C.; Wan, Q. Reservoir Computing Based on Electric-double-layer Coupled InGaZnO Artificial Synapse. Appl. Phys. Lett. 2023, 122, 043508,  DOI: 10.1063/5.0137647
    Google Scholar
    There is no corresponding record for this reference.
  19. 19
    Komatsu, H.; Hosoda, N.; Kounoue, T.; Tokiwa, K.; Ikuno, T. Disposable and Flexible Paper-based Optoelectronic Synaptic Devices for Physical Reservoir Computing. Adv. Electron. Mater. 2024, 10, 2300749,  DOI: 10.1002/aelm.202300749
    Google Scholar
    There is no corresponding record for this reference.
  20. 20
    Zhang, Z.; Zhao, X.; Zhang, X.; Hou, X.; Ma, X.; Tang, S.; Zhang, Y.; Xu, G.; Liu, Q.; Long, S. In-sensor Reservoir Computing System for Latent Fingerprint Recognition With Deep Ultraviolet Photo-synapses and Memristor Array. Nat. Commun. 2022, 13, 6590,  DOI: 10.1038/s41467-022-34230-8
    Google Scholar
    20
    In-sensor reservoir computing system for latent fingerprint recognition with deep ultraviolet photo-synapses and memristor array
    Zhang, Zhongfang; Zhao, Xiaolong; Zhang, Xumeng; Hou, Xiaohu; Ma, Xiaolan; Tang, Shuangzhu; Zhang, Ying; Xu, Guangwei; Liu, Qi; Long, Shibing
    Nature Communications (2022), 13 (1), 6590CODEN: NCAOBW; ISSN:2041-1723. (Nature Portfolio)
    Abstr.: Detection and recognition of latent fingerprints play crucial roles in identification and security. However, the sepn. of sensor, memory, and processor in conventional ex-situ fingerprint recognition system seriously deteriorates the latency of decision-making and inevitably increases the overall computing power. In this work, a photoelectronic reservoir computing (RC) system, consisting of DUV photo-synapses and nonvolatile memristor array, is developed to detect and recognize the latent fingerprint with in-sensor and parallel in-memory computing. Through the Ga-rich design, we achieve amorphous GaOx (a-GaOx) photo-synapses with an enhanced persistent photocond. (PPC) effect. The PPC effect, which induces nonlinearly tunable cond., renders the a-GaOx photo-synapses an ideal deep UV (DUV) photoelectronic reservoir, thus mapping the complex input vector into a dimensionality-reduced output vector. Connecting the reservoirs and a memristor array, we further construct an in-sensor RC system for latent fingerprint identification. The system maintains over 90% recognition accuracy for latent fingerprint within 15% stochastic noise level via the proposed dual-feature strategy. This work provides a subversive prototype system of DUV in-sensor RC for highly efficient recognition of latent fingerprints.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BB38XivVWqs7jL&md5=89e7e52cbfd94bf6d8f3bacf500ed0aa
  21. 21
    Lao, J.; Yan, M.; Tian, B.; Jiang, C.; Luo, C.; Xie, Z.; Zhu, Q.; Bao, Z.; Zhong, N.; Tang, X.; Sun, L.; Wu, G.; Wang, J.; Peng, H.; Chu, J. Ultralow-power Machine Vision With Self-powered Sensor Reservoir. Adv. Sci. 2022, 9, 2106092,  DOI: 10.1002/advs.202106092
    Google Scholar
    There is no corresponding record for this reference.
  22. 22
    Nakade, S.; Saito, Y.; Kubo, W.; Kanzaki, T.; Kitamura, T.; Wada, Y.; Yanagida, S. Laser-induced Photovoltage Transient Studies on Nanoporous TiO2 Electrodes. J. Phys. Chem. B 2004, 108, 1628– 1633,  DOI: 10.1021/jp036786f
    Google Scholar
    22
    Laser-Induced Photovoltage Transient Studies on Nanoporous TiO2 Electrodes
    Nakade, S.; Saito, Y.; Kubo, W.; Kanzaki, T.; Kitamura, T.; Wada, Y.; Yanagida, S.
    Journal of Physical Chemistry B (2004), 108 (5), 1628-1633CODEN: JPCBFK; ISSN:1520-6106. (American Chemical Society)
    At open circuit, pulsed-laser-induced photovoltage transients of nanoporous TiO2 electrodes filled with an electrolyte were studied using various pulse intensities and electrode thicknesses. The transients are interpreted using the electron diffusion in the electrodes. This is compared to the studies of electron transport, which was performed using photocurrent transients at short circuit. A similar light intensity dependence of electron diffusion is found from the photovoltage transients. Electron diffusion coeffs. are estd. by fitting of the photovoltage transients with the soln. obtained by solving a diffusion equation with a simple boundary condition. The fitted values of the electron diffusion coeffs. at open circuit are lower than those derived from the photocurrent transients at short circuit. The difference suggests that Fermi level gradient, caused by nonuniform electron distribution in the electrode under short-circuit conditions, has significant influence on the electron transport. Obsd. photovoltages are plotted with corresponding electron densities in electrodes, showing logarithmic relation. Difference and advantages of the photovoltage transient measurements in compassion to short-circuit transient measurements are discussed.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2cXotFyg&md5=9da42997f18a40c1ff259c23d22c26ed
  23. 23
    Hagfeldt, A.; Boschloo, G.; Sun, L.; Kloo, L.; Pettersson, H. Dye-sensitized Solar Cells. Chem. Rev. 2010, 110, 6595– 6663,  DOI: 10.1021/cr900356p
    Google Scholar
    23
    Dye-Sensitized Solar Cells
    Hagfeldt, Anders; Boschloo, Gerrit; Sun, Licheng; Kloo, Lars; Pettersson, Henrik
    Chemical Reviews (Washington, DC, United States) (2010), 110 (11), 6595-6663CODEN: CHREAY; ISSN:0009-2665. (American Chemical Society)
    A review on dye-sensitized solar cells (DSCs). Some brief notes on solar energy in general and DSC in particular are given, followed by a discussion of the operational principles of DSC (energetics and kinetics). Then, the development of material components and some specific exptl. techniques to characterize DSC are described. The current status of module development is also discussed, and finally a brief future outlook is given.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3cXhtFChs77M&md5=e6727377e1d3eec4c6c6d78276ff77a1
  24. 24
    Gärtner, M.; Dremov, V.; Müller, P.; Kisch, H. Bandgap Widening of Titania Through Semiconductor Support Interactions. ChemPhysChem 2005, 6, 714– 718,  DOI: 10.1002/cphc.200400185
    Google Scholar
    There is no corresponding record for this reference.
  25. 25
    Bisquert, J.; Vikhrenko, V. S. Interpretation of the Time Constants Measured by Kinetic Techniques in Nanostructured Semiconductor Electrodes and Dye-sensitized Solar Cells. J. Phys. Chem. B 2004, 108, 2313– 2322,  DOI: 10.1021/jp035395y
    Google Scholar
    25
    Interpretation of the Time Constants Measured by Kinetic Techniques in Nanostructured Semiconductor Electrodes and Dye-Sensitized Solar Cells
    Bisquert, Juan; Vikhrenko, Vyacheslav S.
    Journal of Physical Chemistry B (2004), 108 (7), 2313-2322CODEN: JPCBFK; ISSN:1520-6106. (American Chemical Society)
    The processes of charge sepn., transport, and recombination in dye-sensitized nanocryst. TiO2 solar cells are characterized by certain time consts. These are measured by small perturbation kinetic techniques, such as intensity modulated photocurrent spectroscopy (IMPS), intensity modulated photovoltage spectroscopy (IMVS), and electrochem. impedance spectroscopy (EIS). The electron diffusion coeff., Dn, and electron lifetime, τn, obtained by these techniques are usually found to depend on steady-state Fermi level or, alternatively, on the carrier concn. The authors study the phys. origin of such dependence, using a general approach that consists on reducing the general multiple trapping kinetic-transport formalism, to a simpler diffusion formalism, which is valid in quasi-static conditions. The authors describe in detail a simple kinetic model for diffusion, trapping, and interfacial charge transfer of electrons, and the authors demonstrate the compensation of trap-dependent factors when forming steady-state quantities such as the diffusion length, Ln, or the electron cond., σn.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2cXls1Wrsg%253D%253D&md5=6a3bd1aeae7c1ba7d8a6085708c4e855
  26. 26
    Zaban, A.; Greenshtein, M.; Bisquert, J. Determination of the Electron Lifetime in Nanocrystalline Dye Solar Cells by Open-circuit Voltage Decay Measurements. ChemPhysChem 2003, 4, 859– 864,  DOI: 10.1002/cphc.200200615
    Google Scholar
    26
    Determination of the electron lifetime in nanocrystalline dye solar cells by open-circuit voltage decay measurements
    Zaban, Arie; Greenshtein, Miri; Bisquert, Juan
    ChemPhysChem (2003), 4 (8), 859-864CODEN: CPCHFT; ISSN:1439-4235. (Wiley-VCH Verlag GmbH & Co. KGaA)
    The decay of photovoltage at open-circuit conditions in dye-sensitized solar cells constituted a very simple procedure to obtain the revealed information on the electron lifetime and the parameter β that governed the change of the lifetime. The electron lifetime depended on the open-circuit voltage exponentially in broad terms from 20 ms to 20 s, when the voltage decrease was about 0.6 V. The parameter β had an av. value of 1.4 and a residual variation with the photopotential.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD3sXmvVegt70%253D&md5=c38856bbe64cfa7057ec993c92d182f0
  27. 27
    Salafsky, J. S.; Lubberhuizen, W. H.; van Faassen, E.; Schropp, R. E. I. Charge Dynamics Following Dye Photoinjection Into a TiO2 Nanocrystalline Network. J. Phys. Chem. B 1998, 102, 766– 769,  DOI: 10.1021/jp973204j
    Google Scholar
    27
    Charge dynamics following dye photoinjection into a TiO2 nanocrystalline network
    Salafsky, J. S.; Lubberhuizen, W. H.; van Faassen, E.; Schropp, R. E. I.
    Journal of Physical Chemistry B (1998), 102 (5), 766-769CODEN: JPCBFK; ISSN:1089-5647. (American Chemical Society)
    Electrons can be injected into TiO2 nanocrystals by a surface-adsorbed Ru-based dye in a well-known photosensitization process. Here we present time-resolved microwave cond. measurements of the decay of electrons out of an elec. interconnected network of such TiO2 nanocrystals, following charge injection, and in the presence of an electron donor to the dye cation that prevents direct recombination. The time scale for this decay process is hundreds of milliseconds to seconds and explains the high current collection efficiency in devices based on these materials, since the long time scale allows for slow charge transport through the network.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK1cXivFSltg%253D%253D&md5=f1c51ad16523a9bbf083bb00de636d9e
  28. 28
    Huang, S. Y.; Schlichthörl, G.; Nozik, A. J.; Grätzel, M.; Frank, A. J. Charge Recombination in Dye-sensitized Nanocrystalline TiO2 Solar Cells. J. Phys. Chem. B 1997, 101, 2576– 2582,  DOI: 10.1021/jp962377q
    Google Scholar
    28
    Charge Recombination in Dye-Sensitized Nanocrystalline TiO2 Solar Cells
    Huang, S. Y.; Schlichthoerl, G.; Nozik, A. J.; Graetzel, M.; Frank, A. J.
    Journal of Physical Chemistry B (1997), 101 (14), 2576-2582CODEN: JPCBFK; ISSN:1089-5647. (American Chemical Society)
    Charge recombination between dye-sensitized nanocryst. TiO2 electrodes and the I3-/I- couple in nonaq. soln. is described. The sensitizer was [RuL2(NCS)2] (L = 2,2'-bipyridyl-4,4'-dicarboxylic acid). An apparent inequality between the dark current and the recombination current is ascribed to a voltage shift caused by a potential drop at the SnO2/TiO2 interface, ohmic losses in the SnO2 and TiO2, and an overpotential for the redox reaction at the Pt counter electrode. Treating the dye-coated TiO2 electrodes with pyridine derivs. (4-tert-butylpyridine, 2-vinylpyridine, or poly(2-vinylpyridine)) improves significantly both the open-circuit photovoltage Voc (from 0.57 to 0.73 V) and the cell conversion efficiency (from 5.8 to 7.5%) at a radiant power of 100 mW/cm2 (AM 1.5) with respect to the untreated electrode. An anal. expression relating Voc to the interfacial recombination kinetics is derived, and its limitations are discussed. Anal. of Voc vs radiant power data with this expression indicates that the pyridine compds. may lower the back-electron-transfer rate const. by 1-2 orders of magnitude. The pyridines are found to have no significant effect on the recombination mechanism and kinetics of electron injection from excited dye mols. to TiO2. Studies of the dye-covered electrodes show that the rate of recombination is second order in I3- concn., which is attributed to the dismutation reaction 2I2- → I3- + I- with I2 as the electron acceptor in the back-reaction. Mass-transport theory is applied to understand the dependence of the short-circuit photocurrent on the radiant power at low I3- concn. and to calc. the diffusion coeff. of I3- ions (7.6 × 10-6 cm2/s) in the porous TiO2 structure. The dependence of other cell parameters on the I3- concn. is also investigated.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK2sXisVCjsr8%253D&md5=18ea100186e4d09f3e9b423afe2b6e04
  29. 29
    Buonomano, D. V. Decoding Temporal Information: a Model Based on Short-term Synaptic Plasticity. J. Neurosci. 2000, 20, 1129– 1141,  DOI: 10.1523/JNEUROSCI.20-03-01129.2000
    Google Scholar
    29
    Decoding temporal information: a model based on short-term synaptic plasticity
    Buonomano, Dean V.
    Journal of Neuroscience (2000), 20 (3), 1129-1141CODEN: JNRSDS; ISSN:0270-6474. (Society for Neuroscience)
    In the current paper it is proposed that short-term plasticity and dynamic changes in the balance of excitatory-inhibitory interactions may underlie the decoding of temporal information, i.e., the generation of temporally selective neurons. Our initial approach was to simulate excitatory-inhibitory disynaptic circuits. Such circuits were composed of a single excitatory and inhibitory neuron and incorporated short-term plasticity of EPSPs and IPSPs and slow IPSPs. We first showed that it is possible to tune cells to respond selectively to different intervals by changing the synaptic wts. of different synapses in parallel. In other words, temporal tuning can rely on long-term changes in synaptic strength and does not require changes in the time consts. of the temporal properties. When the units studied in disynaptic circuits were incorporated into a larger single-layer network, the units exhibited a broad range of temporal selectivity ranging from no interval tuning to interval-selective tuning. The variability in temporal tuning relied on the variability of synaptic strengths. The network as a whole contained a robust population code for a wide range of intervals. Importantly, the same network was able to discriminate simple temporal sequences. These results argue that neural circuits are intrinsically able to process temporal information on the time scale of tens to hundreds of milliseconds and that specialized mechanisms, such as delay lines or oscillators, may not be necessary.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD3cXhtFGqtbk%253D&md5=29b2abf0e336eda12a1869303e476cde
  30. 30
    Chen, Z.-L.; Xiao, Y.; Huang, W.-Y.; Jiang, Y.-P.; Liu, Q.-X.; Tang, X.-G. In-sensor Reservoir Computing Based on Optoelectronic Synaptic Devices. Appl. Phys. Lett. 2023, 123, 100501,  DOI: 10.1063/5.0160599
    Google Scholar
    There is no corresponding record for this reference.
  31. 31
    Bertschinger, N.; Natschläger, T. Real-time Computation at the Edge of Chaos in Recurrent Neural Networks. Neural Comput. 2004, 16, 1413– 1436,  DOI: 10.1162/089976604323057443
    Google Scholar
    31
    Real-time computation at the edge of chaos in recurrent neural networks
    Bertschinger Nils; Natschlager Thomas
    Neural computation (2004), 16 (7), 1413-36 ISSN:0899-7667.
    Depending on the connectivity, recurrent networks of simple computational units can show very different types of dynamics, ranging from totally ordered to chaotic. We analyze how the type of dynamics (ordered or chaotic) exhibited by randomly connected networks of threshold gates driven by a time-varying input signal depends on the parameters describing the distribution of the connectivity matrix. In particular, we calculate the critical boundary in parameter space where the transition from ordered to chaotic dynamics takes place. Employing a recently developed framework for analyzing real-time computations, we show that only near the critical boundary can such networks perform complex computations on time series. Hence, this result strongly supports conjectures that dynamical systems that are capable of doing complex computational tasks should operate near the edge of chaos, that is, the transition from ordered to chaotic dynamics.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A280%3ADC%252BD2c3mvVKqtg%253D%253D&md5=ac880a747148f620516b88e557c12926
  32. 32
    Jaeger, H. Short term memory in echo state networks. Technical Report GMD Report 2001.
    Google Scholar
    There is no corresponding record for this reference.
  33. 33
    Lee, G.; Kang, C.; Kim, S.; Park, Y.; Shin, E. J.; Cho, B. J. Physical Reservoir Based on a Leaky-FeFET Using the Temporal Memory Effect. IEEE Electron Device Lett. 2024, 45, 108– 111,  DOI: 10.1109/LED.2023.3335142
    Google Scholar
    There is no corresponding record for this reference.
  34. 34
    Liao, Z.; Yamahara, H.; Terao, K.; Ma, K.; Seki, M.; Tabata, H. Short-term memory capacity analysis of Lu3Fe4Co0.5Si0.5O12-based spin cluster glass towards reservoir computing. Sci. Rep. 2023, 13, 5260,  DOI: 10.1038/s41598-023-32084-8
    Google Scholar
    There is no corresponding record for this reference.
  35. 35
    Sakai, K.; Yagi, M.; Ito, M.; Shirakashi, J.-i. Memory properties of electromigrated Au nanogaps to realize reservoir computing. Appl. Phys. Lett. 2021, 119, 084101,  DOI: 10.1063/5.0055352
    Google Scholar
    There is no corresponding record for this reference.
  36. 36
    Shimizu, Y.; Minegishi, K.; Tanaka, H.; Hasegawa, T. Ag2S island network reservoir that works with direct optical signal inputs. Jpn. J. Appl. Phys. 2023, 62, SG1001,  DOI: 10.35848/1347-4065/acab0a
    Google Scholar
    There is no corresponding record for this reference.
  37. 37
    Shirmohammadli, V.; Bahreyni, B. Physics-based approach to developing physical reservoir computers. Phys. Rev. Res. 2024, 6, 033055,  DOI: 10.1103/PhysRevResearch.6.033055
    Google Scholar
    There is no corresponding record for this reference.
  38. 38
    Namiki, W.; Nishioka, D.; Yamaguchi, Y.; Tsuchiya, T.; Higuchi, T.; Terabe, K. Experimental Demonstration of High-Performance Physical Reservoir Computing with Nonlinear Interfered Spin Wave Multidetection. Adv. Intell. Syst. 2023, 5, 2300228,  DOI: 10.1002/aisy.202300228
    Google Scholar
    There is no corresponding record for this reference.
  39. 39
    Komatsu, H.; Ogawa, T.; Hosoda, N.; Ikuno, T. Correlation Between PPF Index and Short-term Memory in Physical-reservoir Computing With Artificial Synapse Using Simple RC-series Circuit. AIP Adv. 2024, 14, 035026,  DOI: 10.1063/5.0199941
    Google Scholar
    There is no corresponding record for this reference.
  40. 40
    Tsunegi, S.; Taniguchi, T.; Nakajima, K.; Miwa, S.; Yakushiji, K.; Fukushima, A.; Yuasa, S.; Kubota, H. Physical Reservoir Computing Based on Spin Torque Oscillator With Forced Synchronization. Appl. Phys. Lett. 2019, 114, 164101,  DOI: 10.1063/1.5081797
    Google Scholar
    40
    Physical reservoir computing based on spin torque oscillator with forced synchronization
    Tsunegi, Sumito; Taniguchi, Tomohiro; Nakajima, Kohei; Miwa, Shinji; Yakushiji, Kay; Fukushima, Akio; Yuasa, Shinji; Kubota, Hitoshi
    Applied Physics Letters (2019), 114 (16), 164101/1-164101/5CODEN: APPLAB; ISSN:0003-6951. (American Institute of Physics)
    We investigated phys. reservoir computing (RC) using a vortex-type spin torque oscillator (STO) as a resource of nonlinear dynamics, which is essential for processing information in time-series data. Forced synchronization was used to suppress the thermal fluctuation of the oscillation trajectory of the STO. We examd. the memory property of the STO dynamics, called short-term memory (STM), by using a virtual node technique. The STM capacity increased about twofold compared with that obtained without forced synchronization. The performance index for the nonlinear transformation of the STO also increased; it was evaluated in a parity-check task. The results prove that the synchronized STO has great potential for phys. RC based on nanotechnol. (c) 2019 American Institute of Physics.
    >> More from SciFinder ®
    https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC1MXot1Wju74%253D&md5=73528308c4da45f6db61a5ae683338cc

Cited By

Click to copy section linkSection link copied!

This article has not yet been cited by other publications.

Download PDF
Go to ACS Applied Materials & Interfaces
Get e-Alerts
Get e-Alerts

ACS Applied Materials & Interfaces

Cite this: ACS Appl. Mater. Interfaces 2024, XXXX, XXX, XXX-XXX
Click to copy citationCitation copied!
https://doi.org/10.1021/acsami.4c11061
Published October 28, 2024

© 2024 The Authors. Published by American Chemical Society. This publication is licensed under

CC-BY 4.0 .

Article Views

827

Altmetric

-

Citations

-
Learn about these metrics

Article Views are the COUNTER-compliant sum of full text article downloads since November 2008 (both PDF and HTML) across all institutions and individuals. These metrics are regularly updated to reflect usage leading up to the last few days.

Citations are the number of other articles citing this article, calculated by Crossref and updated daily. Find more information about Crossref citation counts.

The Altmetric Attention Score is a quantitative measure of the attention that a research article has received online. Clicking on the donut icon will load a page at altmetric.com with additional details about the score and the social media presence for the given article. Find more information on the Altmetric Attention Score and how the score is calculated.

  • Abstract

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 1

    Figure 1. Fundamental characteristics of the fabricated DSC. (a) Device structure of the DSC and molecule structure of SQ2. (b) Incident photon-to-current conversion efficiency of the DSC. (c) J–V curves of the DSCs (λ: 658 nm). (d) Voc and Jsc as a function of the incident light intensity, P.

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 2

    Figure 2. Transient voltage responses. (a) Normalized transient voltage responses as a function of P. (b,c) Rise time constant and decay time constant as a function of P. (d) Band diagram of TiO2 at high and low P levels when light irradiation starts. (e) Band diagram of TiO2 at high and low P levels when the light irradiation at a steady state is stopped.

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 3

    Figure 3. Synaptic characteristics in the DSC. (a) Transient voltage responses induced by two optical pulses with high and low P (λ: 658 nm, Tp: 100 ms, ΔT: 100 ms, and P: 0.2, 5.0 mW). (b) PPF index as a function of ΔT (Tp: 100 ms, P: 0.075–2.5 mW). (c) Tp dependence of PPF index as functions of P and ΔT. (d) Transient voltage response induced by two optical pulses with different P values (λ: 658 nm, Tp: 100 ms, ΔT: 100 ms, P1: 5 mW, and P2: 1–15 mW). (e) PPF index map for P1 with various P2 (λ: 658 nm, Tp: 100 ms, P1: 1–15 mW, and P2: 1–15 mW). (f) Transient voltage responses induced by 10 optical pulses with various P values (Tp: 1 ms, P1: 1–15 mW). (g) Switching facilitation and depression of the DSC by changing P (P: 1, 10 mW).

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 4

    Figure 4. Time-series data processing task. (a) Operating scheme of the time-series data processing task with a DSC-based synaptic device. (b,c) System output of the STM and PC tasks (d: 1, Tp: 100 ms, P: 1 mW, and number of virtual nodes: 125). (d,e) Forgetting curves of the STM and PC tasks with various P (d: 1, Tp: 100 ms, P: 0.075–10 mW, and number of virtual nodes: 125).

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 5

    Figure 5. Time-series data processing task as a function of Tp with various P. Error bar represents standard error of the mean. (a) Average CSTM as a function of Tp. (b) Average CPC as a function of Tp. (c) Transient voltage response induced by random binary pulses with various Tp (P: 10 mW, Tp: 1–1000 ms). (d) Histogram of the transient voltage response with various Tp (P: 10 mW, Tp: 1–1000 ms).

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 6

    Figure 6. Demonstration of motion recognition task using a PRC system with the DSC-based synaptic device. (a) Schematic diagram of the motion recognition task using the DSC. (b) Transient voltage response of the DSC output with eight divided motion movies (jump). (c) Confusion matrix of the motion recognition tasks.

    High Resolution Image
    Download MS PowerPoint Slide

  • References

    This article references 40 other publications.

    1. 1
      Alam, M.; Tehranipoor, M. M.; Guin, U. TSensors Vision, Infrastructure and Security Challenges in Trillion Sensor Era. J. Hardw. Syst. Secur. 2017, 1, 311– 327,  DOI: 10.1007/s41635-017-0028-8
      There is no corresponding record for this reference.
    2. 2
      Banna, M. H. A.; Taher, K. A.; Kaiser, M. S.; Mahmud, M.; Rahman, M. S.; Hosen, A. S. M. S.; Cho, G. H. Application of Artificial Intelligence in Predicting Earthquakes: State-of-the-art and Future Challenges. IEEE Access 2020, 8, 192880– 192923,  DOI: 10.1109/ACCESS.2020.3029859
      There is no corresponding record for this reference.
    3. 3
      Anantrasirichai, N.; Biggs, J.; Albino, F.; Hill, P.; Bull, D. Application of Machine Learning to Classification of Volcanic Deformation in Routinely Generated InSAR Data. J. Geophys. Res. Solid Earth 2018, 123, 6592– 6606,  DOI: 10.1029/2018JB015911
      There is no corresponding record for this reference.
    4. 4
      Domyati, A.; Memon, Q. Machine Learning Based Improved Heart Disease Detection With Confidence. Int. J. Online Biomed. Eng. 2023, 19, 130– 143,  DOI: 10.3991/ijoe.v19i08.37417
      There is no corresponding record for this reference.
    5. 5
      Wang, W.; Wang, Y.; Zhang, B.; Shi, W.; Li, C.-Q. Failure Prediction of Buried Pipe Network With Multiple Failure Modes and Spatial Randomness of Corrosion. Int. J. Pressure Vessels Piping 2021, 191, 104367,  DOI: 10.1016/j.ijpvp.2021.104367
      There is no corresponding record for this reference.
    6. 6
      Sun, X.; Ansari, N. EdgeIoT: Mobile Edge Computing for the Internet of Things. IEEE Commu. Mag. 2016, 54, 22– 29,  DOI: 10.1109/mcom.2016.1600492cm
      There is no corresponding record for this reference.
    7. 7
      Brezak, D.; Bacek, T.; Majetic, D.; Kasac, J.; Novakovic, B. A comparison of feed-forward and recurrent neural networks in time series forecasting. In Proceedings of the IEEE/IAFE Computational Intelligence for Financial Engineering , 2012. DOI: 10.1109/CIFER.1997 .
      There is no corresponding record for this reference.
    8. 8
      Torrejon, J.; Riou, M.; Araujo, F. A.; Tsunegi, S.; Khalsa, G.; Querlioz, D.; Bortolotti, P.; Cros, V.; Yakushiji, K.; Fukushima, A.; Kubota, H.; Yuasa, S.; Stiles, M. D.; Grollier, J. Neuromorphic Computing With Nanoscale Spintronic Oscillators. Nature 2017, 547, 428– 431,  DOI: 10.1038/nature23011
      8
      Neuromorphic computing with nanoscale spintronic oscillators
      Torrejon, Jacob; Riou, Mathieu; Araujo, Flavio Abreu; Tsunegi, Sumito; Khalsa, Guru; Querlioz, Damien; Bortolotti, Paolo; Cros, Vincent; Yakushiji, Kay; Fukushima, Akio; Kubota, Hitoshi; Yuasa, Shinji; Stiles, Mark D.; Grollier, Julie
      Nature (London, United Kingdom) (2017), 547 (7664), 428-431CODEN: NATUAS; ISSN:0028-0836. (Nature Publishing Group)
      Neurons in the brain behave as nonlinear oscillators, which develop rhythmic activity and interact to process information. Taking inspiration from this behavior to realize high-d., low-power neuromorphic computing will require very large nos. of nanoscale nonlinear oscillators. A simple estn. indicates that to fit 108 oscillators organized in a two-dimensional array inside a chip the size of a thumb, the lateral dimension of each oscillator must be smaller than one micrometre. However, nanoscale devices tend to be noisy and to lack the stability that is required to process data in a reliable way. For this reason, despite multiple theor. proposals and several candidates, including memristive and superconducting oscillators, a proof of concept of neuromorphic computing using nanoscale oscillators has yet to be demonstrated. Here we show exptl. that a nanoscale spintronic oscillator (a magnetic tunnel junction) can be used to achieve spoken-digit recognition with an accuracy similar to that of state-of-the-art neural networks. We also det. the regime of magnetization dynamics that leads to the greatest performance. These results, combined with the ability of the spintronic oscillators to interact with each other, and their long lifetime and low energy consumption, open up a path to fast, parallel, on-chip computation based on networks of oscillators.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2sXht1ent7bL&md5=444c9c101e6c4acc1739e32f5ba02aa4
    9. 9
      Nakane, R.; Tanaka, G.; Hirose, A. Reservoir Computing With Spin Waves Excited in a Garnet Film. IEEE Access 2018, 6, 4462– 4469,  DOI: 10.1109/ACCESS.2018.2794584
      There is no corresponding record for this reference.
    10. 10
      Midya, R.; Wang, Z.; Asapu, S.; Zhang, X.; Rao, M.; Song, W.; Zhuo, Y.; Upadhyay, N.; Xia, Q.; Yang, J. J. Reservoir Computing Using Diffusive Memristors. Adv. Intell. Syst. 2019, 1, 1900084,  DOI: 10.1002/aisy.201900084
      There is no corresponding record for this reference.
    11. 11
      Zhang, G.; Qin, J.; Zhang, Y.; Gong, G.; Xiong, Z.-Y.; Ma, X.; Lv, Z.; Zhou, Y.; Han, S.-T. Functional Materials for Memristor-based Reservoir Computing: Dynamics and Applications. Adv. Funct. Mater. 2023, 33, 2302929,  DOI: 10.1002/adfm.202302929
      There is no corresponding record for this reference.
    12. 12
      Nishioka, D.; Tsuchiya, T.; Namiki, W.; Takayanagi, M.; Imura, M.; Koide, Y.; Higuchi, T.; Terabe, K. Edge-of-chaos Learning Achieved by Ion-electron–coupled Dynamics in an Ion-gating Reservoir. Sci. Adv. 2022, 8, eade1156  DOI: 10.1126/sciadv.ade1156
      There is no corresponding record for this reference.
    13. 13
      Koh, S.-G.; Shima, H.; Naitoh, Y.; Akinaga, H.; Kinoshita, K. Reservoir Computing With Dielectric Relaxation at an Electrode–ionic Liquid Interface. Sci. Rep. 2022, 12, 6958,  DOI: 10.1038/s41598-022-10152-9
      13
      Reservoir computing with dielectric relaxation at an electrode-ionic liquid interface
      Koh, Sang-Gyu; Shima, Hisashi; Naitoh, Yasuhisa; Akinaga, Hiroyuki; Kinoshita, Kentaro
      Scientific Reports (2022), 12 (1), 6958CODEN: SRCEC3; ISSN:2045-2322. (Nature Portfolio)
      A phys. reservoir device with tunable transient dynamics is strongly required to process time-series data with various timescales generated in the edge region. In this study, we proposed using the dielec. relaxation at an electrode-ionic liq. (IL) interface as the phys. reservoir by making the most of designable physicochem. properties of ILs. The transient dynamics of a Au/IL/Au reservoir device were characterized as a function of the alkyl chain length of cations in the IL (1-alkyl-3-methylimidazolium bis(trifluoromethane sulfonyl)imide). By considering a weighted sum of exponentials expressing a superposition of Debye-type relaxations, the transient dynamics were well reconstructed. Although such complex dynamics governed by multiple relaxation processes were obsd., each extd. relaxation time scales with a power law as a function of IL's viscosity detd. by the alkyl chain length of cations. This indicates that the relaxation processes are characterized by bulk properties of the ILs that obey the widely received Vogel-Fulcher-Tammann law. We demonstrated that the 4-bit time-series signals were transformed into the 16 classifiable data, and the data transformation, which enables to achieve higher accuracy in an image classification task, can be easily optimized according to the features of the input signals by controlling the IL's viscosity.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BB38XhtFGnsrnJ&md5=bb69fcba63e6f6860bd36b7588bc7e79
    14. 14
      Van der Sande, G.; Brunner, D.; Soriano, M. C. Advances in Photonic Reservoir Computing. Nanophotonics 2017, 6, 561– 576,  DOI: 10.1515/nanoph-2016-0132
      There is no corresponding record for this reference.
    15. 15
      Larger, L.; Soriano, M. C.; Brunner, D.; Appeltant, L.; Gutierrez, J. M.; Pesquera, L.; Mirasso, C. R.; Fischer, I. Photonic Information Processing Beyond Turing: an Optoelectronic Implementation of Reservoir Computing. Opt. Express 2012, 20, 3241– 3249,  DOI: 10.1364/OE.20.003241
      There is no corresponding record for this reference.
    16. 16
      Cucchi, M.; Abreu, S.; Ciccone, G.; Brunner, D.; Kleemann, H. Hands-on Reservoir Computing: a Tutorial for Practical Implementation. Neuromorph. Comput. Eng. 2022, 2, 032002,  DOI: 10.1088/2634-4386/ac7db7
      There is no corresponding record for this reference.
    17. 17
      Pocock, D. C. D. Sight and Knowledge. Trans. Inst. Br. Geogr. 1981, 6, 385– 393,  DOI: 10.2307/621875
      There is no corresponding record for this reference.
    18. 18
      Yang, Y.; Cui, H.; Ke, S.; Pei, M.; Shi, K.; Wan, C.; Wan, Q. Reservoir Computing Based on Electric-double-layer Coupled InGaZnO Artificial Synapse. Appl. Phys. Lett. 2023, 122, 043508,  DOI: 10.1063/5.0137647
      There is no corresponding record for this reference.
    19. 19
      Komatsu, H.; Hosoda, N.; Kounoue, T.; Tokiwa, K.; Ikuno, T. Disposable and Flexible Paper-based Optoelectronic Synaptic Devices for Physical Reservoir Computing. Adv. Electron. Mater. 2024, 10, 2300749,  DOI: 10.1002/aelm.202300749
      There is no corresponding record for this reference.
    20. 20
      Zhang, Z.; Zhao, X.; Zhang, X.; Hou, X.; Ma, X.; Tang, S.; Zhang, Y.; Xu, G.; Liu, Q.; Long, S. In-sensor Reservoir Computing System for Latent Fingerprint Recognition With Deep Ultraviolet Photo-synapses and Memristor Array. Nat. Commun. 2022, 13, 6590,  DOI: 10.1038/s41467-022-34230-8
      20
      In-sensor reservoir computing system for latent fingerprint recognition with deep ultraviolet photo-synapses and memristor array
      Zhang, Zhongfang; Zhao, Xiaolong; Zhang, Xumeng; Hou, Xiaohu; Ma, Xiaolan; Tang, Shuangzhu; Zhang, Ying; Xu, Guangwei; Liu, Qi; Long, Shibing
      Nature Communications (2022), 13 (1), 6590CODEN: NCAOBW; ISSN:2041-1723. (Nature Portfolio)
      Abstr.: Detection and recognition of latent fingerprints play crucial roles in identification and security. However, the sepn. of sensor, memory, and processor in conventional ex-situ fingerprint recognition system seriously deteriorates the latency of decision-making and inevitably increases the overall computing power. In this work, a photoelectronic reservoir computing (RC) system, consisting of DUV photo-synapses and nonvolatile memristor array, is developed to detect and recognize the latent fingerprint with in-sensor and parallel in-memory computing. Through the Ga-rich design, we achieve amorphous GaOx (a-GaOx) photo-synapses with an enhanced persistent photocond. (PPC) effect. The PPC effect, which induces nonlinearly tunable cond., renders the a-GaOx photo-synapses an ideal deep UV (DUV) photoelectronic reservoir, thus mapping the complex input vector into a dimensionality-reduced output vector. Connecting the reservoirs and a memristor array, we further construct an in-sensor RC system for latent fingerprint identification. The system maintains over 90% recognition accuracy for latent fingerprint within 15% stochastic noise level via the proposed dual-feature strategy. This work provides a subversive prototype system of DUV in-sensor RC for highly efficient recognition of latent fingerprints.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BB38XivVWqs7jL&md5=89e7e52cbfd94bf6d8f3bacf500ed0aa
    21. 21
      Lao, J.; Yan, M.; Tian, B.; Jiang, C.; Luo, C.; Xie, Z.; Zhu, Q.; Bao, Z.; Zhong, N.; Tang, X.; Sun, L.; Wu, G.; Wang, J.; Peng, H.; Chu, J. Ultralow-power Machine Vision With Self-powered Sensor Reservoir. Adv. Sci. 2022, 9, 2106092,  DOI: 10.1002/advs.202106092
      There is no corresponding record for this reference.
    22. 22
      Nakade, S.; Saito, Y.; Kubo, W.; Kanzaki, T.; Kitamura, T.; Wada, Y.; Yanagida, S. Laser-induced Photovoltage Transient Studies on Nanoporous TiO2 Electrodes. J. Phys. Chem. B 2004, 108, 1628– 1633,  DOI: 10.1021/jp036786f
      22
      Laser-Induced Photovoltage Transient Studies on Nanoporous TiO2 Electrodes
      Nakade, S.; Saito, Y.; Kubo, W.; Kanzaki, T.; Kitamura, T.; Wada, Y.; Yanagida, S.
      Journal of Physical Chemistry B (2004), 108 (5), 1628-1633CODEN: JPCBFK; ISSN:1520-6106. (American Chemical Society)
      At open circuit, pulsed-laser-induced photovoltage transients of nanoporous TiO2 electrodes filled with an electrolyte were studied using various pulse intensities and electrode thicknesses. The transients are interpreted using the electron diffusion in the electrodes. This is compared to the studies of electron transport, which was performed using photocurrent transients at short circuit. A similar light intensity dependence of electron diffusion is found from the photovoltage transients. Electron diffusion coeffs. are estd. by fitting of the photovoltage transients with the soln. obtained by solving a diffusion equation with a simple boundary condition. The fitted values of the electron diffusion coeffs. at open circuit are lower than those derived from the photocurrent transients at short circuit. The difference suggests that Fermi level gradient, caused by nonuniform electron distribution in the electrode under short-circuit conditions, has significant influence on the electron transport. Obsd. photovoltages are plotted with corresponding electron densities in electrodes, showing logarithmic relation. Difference and advantages of the photovoltage transient measurements in compassion to short-circuit transient measurements are discussed.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2cXotFyg&md5=9da42997f18a40c1ff259c23d22c26ed
    23. 23
      Hagfeldt, A.; Boschloo, G.; Sun, L.; Kloo, L.; Pettersson, H. Dye-sensitized Solar Cells. Chem. Rev. 2010, 110, 6595– 6663,  DOI: 10.1021/cr900356p
      23
      Dye-Sensitized Solar Cells
      Hagfeldt, Anders; Boschloo, Gerrit; Sun, Licheng; Kloo, Lars; Pettersson, Henrik
      Chemical Reviews (Washington, DC, United States) (2010), 110 (11), 6595-6663CODEN: CHREAY; ISSN:0009-2665. (American Chemical Society)
      A review on dye-sensitized solar cells (DSCs). Some brief notes on solar energy in general and DSC in particular are given, followed by a discussion of the operational principles of DSC (energetics and kinetics). Then, the development of material components and some specific exptl. techniques to characterize DSC are described. The current status of module development is also discussed, and finally a brief future outlook is given.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3cXhtFChs77M&md5=e6727377e1d3eec4c6c6d78276ff77a1
    24. 24
      Gärtner, M.; Dremov, V.; Müller, P.; Kisch, H. Bandgap Widening of Titania Through Semiconductor Support Interactions. ChemPhysChem 2005, 6, 714– 718,  DOI: 10.1002/cphc.200400185
      There is no corresponding record for this reference.
    25. 25
      Bisquert, J.; Vikhrenko, V. S. Interpretation of the Time Constants Measured by Kinetic Techniques in Nanostructured Semiconductor Electrodes and Dye-sensitized Solar Cells. J. Phys. Chem. B 2004, 108, 2313– 2322,  DOI: 10.1021/jp035395y
      25
      Interpretation of the Time Constants Measured by Kinetic Techniques in Nanostructured Semiconductor Electrodes and Dye-Sensitized Solar Cells
      Bisquert, Juan; Vikhrenko, Vyacheslav S.
      Journal of Physical Chemistry B (2004), 108 (7), 2313-2322CODEN: JPCBFK; ISSN:1520-6106. (American Chemical Society)
      The processes of charge sepn., transport, and recombination in dye-sensitized nanocryst. TiO2 solar cells are characterized by certain time consts. These are measured by small perturbation kinetic techniques, such as intensity modulated photocurrent spectroscopy (IMPS), intensity modulated photovoltage spectroscopy (IMVS), and electrochem. impedance spectroscopy (EIS). The electron diffusion coeff., Dn, and electron lifetime, τn, obtained by these techniques are usually found to depend on steady-state Fermi level or, alternatively, on the carrier concn. The authors study the phys. origin of such dependence, using a general approach that consists on reducing the general multiple trapping kinetic-transport formalism, to a simpler diffusion formalism, which is valid in quasi-static conditions. The authors describe in detail a simple kinetic model for diffusion, trapping, and interfacial charge transfer of electrons, and the authors demonstrate the compensation of trap-dependent factors when forming steady-state quantities such as the diffusion length, Ln, or the electron cond., σn.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2cXls1Wrsg%253D%253D&md5=6a3bd1aeae7c1ba7d8a6085708c4e855
    26. 26
      Zaban, A.; Greenshtein, M.; Bisquert, J. Determination of the Electron Lifetime in Nanocrystalline Dye Solar Cells by Open-circuit Voltage Decay Measurements. ChemPhysChem 2003, 4, 859– 864,  DOI: 10.1002/cphc.200200615
      26
      Determination of the electron lifetime in nanocrystalline dye solar cells by open-circuit voltage decay measurements
      Zaban, Arie; Greenshtein, Miri; Bisquert, Juan
      ChemPhysChem (2003), 4 (8), 859-864CODEN: CPCHFT; ISSN:1439-4235. (Wiley-VCH Verlag GmbH & Co. KGaA)
      The decay of photovoltage at open-circuit conditions in dye-sensitized solar cells constituted a very simple procedure to obtain the revealed information on the electron lifetime and the parameter β that governed the change of the lifetime. The electron lifetime depended on the open-circuit voltage exponentially in broad terms from 20 ms to 20 s, when the voltage decrease was about 0.6 V. The parameter β had an av. value of 1.4 and a residual variation with the photopotential.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD3sXmvVegt70%253D&md5=c38856bbe64cfa7057ec993c92d182f0
    27. 27
      Salafsky, J. S.; Lubberhuizen, W. H.; van Faassen, E.; Schropp, R. E. I. Charge Dynamics Following Dye Photoinjection Into a TiO2 Nanocrystalline Network. J. Phys. Chem. B 1998, 102, 766– 769,  DOI: 10.1021/jp973204j
      27
      Charge dynamics following dye photoinjection into a TiO2 nanocrystalline network
      Salafsky, J. S.; Lubberhuizen, W. H.; van Faassen, E.; Schropp, R. E. I.
      Journal of Physical Chemistry B (1998), 102 (5), 766-769CODEN: JPCBFK; ISSN:1089-5647. (American Chemical Society)
      Electrons can be injected into TiO2 nanocrystals by a surface-adsorbed Ru-based dye in a well-known photosensitization process. Here we present time-resolved microwave cond. measurements of the decay of electrons out of an elec. interconnected network of such TiO2 nanocrystals, following charge injection, and in the presence of an electron donor to the dye cation that prevents direct recombination. The time scale for this decay process is hundreds of milliseconds to seconds and explains the high current collection efficiency in devices based on these materials, since the long time scale allows for slow charge transport through the network.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK1cXivFSltg%253D%253D&md5=f1c51ad16523a9bbf083bb00de636d9e
    28. 28
      Huang, S. Y.; Schlichthörl, G.; Nozik, A. J.; Grätzel, M.; Frank, A. J. Charge Recombination in Dye-sensitized Nanocrystalline TiO2 Solar Cells. J. Phys. Chem. B 1997, 101, 2576– 2582,  DOI: 10.1021/jp962377q
      28
      Charge Recombination in Dye-Sensitized Nanocrystalline TiO2 Solar Cells
      Huang, S. Y.; Schlichthoerl, G.; Nozik, A. J.; Graetzel, M.; Frank, A. J.
      Journal of Physical Chemistry B (1997), 101 (14), 2576-2582CODEN: JPCBFK; ISSN:1089-5647. (American Chemical Society)
      Charge recombination between dye-sensitized nanocryst. TiO2 electrodes and the I3-/I- couple in nonaq. soln. is described. The sensitizer was [RuL2(NCS)2] (L = 2,2'-bipyridyl-4,4'-dicarboxylic acid). An apparent inequality between the dark current and the recombination current is ascribed to a voltage shift caused by a potential drop at the SnO2/TiO2 interface, ohmic losses in the SnO2 and TiO2, and an overpotential for the redox reaction at the Pt counter electrode. Treating the dye-coated TiO2 electrodes with pyridine derivs. (4-tert-butylpyridine, 2-vinylpyridine, or poly(2-vinylpyridine)) improves significantly both the open-circuit photovoltage Voc (from 0.57 to 0.73 V) and the cell conversion efficiency (from 5.8 to 7.5%) at a radiant power of 100 mW/cm2 (AM 1.5) with respect to the untreated electrode. An anal. expression relating Voc to the interfacial recombination kinetics is derived, and its limitations are discussed. Anal. of Voc vs radiant power data with this expression indicates that the pyridine compds. may lower the back-electron-transfer rate const. by 1-2 orders of magnitude. The pyridines are found to have no significant effect on the recombination mechanism and kinetics of electron injection from excited dye mols. to TiO2. Studies of the dye-covered electrodes show that the rate of recombination is second order in I3- concn., which is attributed to the dismutation reaction 2I2- → I3- + I- with I2 as the electron acceptor in the back-reaction. Mass-transport theory is applied to understand the dependence of the short-circuit photocurrent on the radiant power at low I3- concn. and to calc. the diffusion coeff. of I3- ions (7.6 × 10-6 cm2/s) in the porous TiO2 structure. The dependence of other cell parameters on the I3- concn. is also investigated.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK2sXisVCjsr8%253D&md5=18ea100186e4d09f3e9b423afe2b6e04
    29. 29
      Buonomano, D. V. Decoding Temporal Information: a Model Based on Short-term Synaptic Plasticity. J. Neurosci. 2000, 20, 1129– 1141,  DOI: 10.1523/JNEUROSCI.20-03-01129.2000
      29
      Decoding temporal information: a model based on short-term synaptic plasticity
      Buonomano, Dean V.
      Journal of Neuroscience (2000), 20 (3), 1129-1141CODEN: JNRSDS; ISSN:0270-6474. (Society for Neuroscience)
      In the current paper it is proposed that short-term plasticity and dynamic changes in the balance of excitatory-inhibitory interactions may underlie the decoding of temporal information, i.e., the generation of temporally selective neurons. Our initial approach was to simulate excitatory-inhibitory disynaptic circuits. Such circuits were composed of a single excitatory and inhibitory neuron and incorporated short-term plasticity of EPSPs and IPSPs and slow IPSPs. We first showed that it is possible to tune cells to respond selectively to different intervals by changing the synaptic wts. of different synapses in parallel. In other words, temporal tuning can rely on long-term changes in synaptic strength and does not require changes in the time consts. of the temporal properties. When the units studied in disynaptic circuits were incorporated into a larger single-layer network, the units exhibited a broad range of temporal selectivity ranging from no interval tuning to interval-selective tuning. The variability in temporal tuning relied on the variability of synaptic strengths. The network as a whole contained a robust population code for a wide range of intervals. Importantly, the same network was able to discriminate simple temporal sequences. These results argue that neural circuits are intrinsically able to process temporal information on the time scale of tens to hundreds of milliseconds and that specialized mechanisms, such as delay lines or oscillators, may not be necessary.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD3cXhtFGqtbk%253D&md5=29b2abf0e336eda12a1869303e476cde
    30. 30
      Chen, Z.-L.; Xiao, Y.; Huang, W.-Y.; Jiang, Y.-P.; Liu, Q.-X.; Tang, X.-G. In-sensor Reservoir Computing Based on Optoelectronic Synaptic Devices. Appl. Phys. Lett. 2023, 123, 100501,  DOI: 10.1063/5.0160599
      There is no corresponding record for this reference.
    31. 31
      Bertschinger, N.; Natschläger, T. Real-time Computation at the Edge of Chaos in Recurrent Neural Networks. Neural Comput. 2004, 16, 1413– 1436,  DOI: 10.1162/089976604323057443
      31
      Real-time computation at the edge of chaos in recurrent neural networks
      Bertschinger Nils; Natschlager Thomas
      Neural computation (2004), 16 (7), 1413-36 ISSN:0899-7667.
      Depending on the connectivity, recurrent networks of simple computational units can show very different types of dynamics, ranging from totally ordered to chaotic. We analyze how the type of dynamics (ordered or chaotic) exhibited by randomly connected networks of threshold gates driven by a time-varying input signal depends on the parameters describing the distribution of the connectivity matrix. In particular, we calculate the critical boundary in parameter space where the transition from ordered to chaotic dynamics takes place. Employing a recently developed framework for analyzing real-time computations, we show that only near the critical boundary can such networks perform complex computations on time series. Hence, this result strongly supports conjectures that dynamical systems that are capable of doing complex computational tasks should operate near the edge of chaos, that is, the transition from ordered to chaotic dynamics.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A280%3ADC%252BD2c3mvVKqtg%253D%253D&md5=ac880a747148f620516b88e557c12926
    32. 32
      Jaeger, H. Short term memory in echo state networks. Technical Report GMD Report 2001.
      There is no corresponding record for this reference.
    33. 33
      Lee, G.; Kang, C.; Kim, S.; Park, Y.; Shin, E. J.; Cho, B. J. Physical Reservoir Based on a Leaky-FeFET Using the Temporal Memory Effect. IEEE Electron Device Lett. 2024, 45, 108– 111,  DOI: 10.1109/LED.2023.3335142
      There is no corresponding record for this reference.
    34. 34
      Liao, Z.; Yamahara, H.; Terao, K.; Ma, K.; Seki, M.; Tabata, H. Short-term memory capacity analysis of Lu3Fe4Co0.5Si0.5O12-based spin cluster glass towards reservoir computing. Sci. Rep. 2023, 13, 5260,  DOI: 10.1038/s41598-023-32084-8
      There is no corresponding record for this reference.
    35. 35
      Sakai, K.; Yagi, M.; Ito, M.; Shirakashi, J.-i. Memory properties of electromigrated Au nanogaps to realize reservoir computing. Appl. Phys. Lett. 2021, 119, 084101,  DOI: 10.1063/5.0055352
      There is no corresponding record for this reference.
    36. 36
      Shimizu, Y.; Minegishi, K.; Tanaka, H.; Hasegawa, T. Ag2S island network reservoir that works with direct optical signal inputs. Jpn. J. Appl. Phys. 2023, 62, SG1001,  DOI: 10.35848/1347-4065/acab0a
      There is no corresponding record for this reference.
    37. 37
      Shirmohammadli, V.; Bahreyni, B. Physics-based approach to developing physical reservoir computers. Phys. Rev. Res. 2024, 6, 033055,  DOI: 10.1103/PhysRevResearch.6.033055
      There is no corresponding record for this reference.
    38. 38
      Namiki, W.; Nishioka, D.; Yamaguchi, Y.; Tsuchiya, T.; Higuchi, T.; Terabe, K. Experimental Demonstration of High-Performance Physical Reservoir Computing with Nonlinear Interfered Spin Wave Multidetection. Adv. Intell. Syst. 2023, 5, 2300228,  DOI: 10.1002/aisy.202300228
      There is no corresponding record for this reference.
    39. 39
      Komatsu, H.; Ogawa, T.; Hosoda, N.; Ikuno, T. Correlation Between PPF Index and Short-term Memory in Physical-reservoir Computing With Artificial Synapse Using Simple RC-series Circuit. AIP Adv. 2024, 14, 035026,  DOI: 10.1063/5.0199941
      There is no corresponding record for this reference.
    40. 40
      Tsunegi, S.; Taniguchi, T.; Nakajima, K.; Miwa, S.; Yakushiji, K.; Fukushima, A.; Yuasa, S.; Kubota, H. Physical Reservoir Computing Based on Spin Torque Oscillator With Forced Synchronization. Appl. Phys. Lett. 2019, 114, 164101,  DOI: 10.1063/1.5081797
      40
      Physical reservoir computing based on spin torque oscillator with forced synchronization
      Tsunegi, Sumito; Taniguchi, Tomohiro; Nakajima, Kohei; Miwa, Shinji; Yakushiji, Kay; Fukushima, Akio; Yuasa, Shinji; Kubota, Hitoshi
      Applied Physics Letters (2019), 114 (16), 164101/1-164101/5CODEN: APPLAB; ISSN:0003-6951. (American Institute of Physics)
      We investigated phys. reservoir computing (RC) using a vortex-type spin torque oscillator (STO) as a resource of nonlinear dynamics, which is essential for processing information in time-series data. Forced synchronization was used to suppress the thermal fluctuation of the oscillation trajectory of the STO. We examd. the memory property of the STO dynamics, called short-term memory (STM), by using a virtual node technique. The STM capacity increased about twofold compared with that obtained without forced synchronization. The performance index for the nonlinear transformation of the STO also increased; it was evaluated in a parity-check task. The results prove that the synchronized STO has great potential for phys. RC based on nanotechnol. (c) 2019 American Institute of Physics.
      >> More from SciFinder ®
      https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC1MXot1Wju74%253D&md5=73528308c4da45f6db61a5ae683338cc

  • Supporting Information

    Supporting Information

    The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsami.4c11061.

    • Transient response of polycrystalline Si solar cells, rise time constant of Si solar cells and DSC as a function of P, result of time-series processing task with and without DSC-based synaptic device, motion recognition accuracy with various machine learning algorithms, and motion recognition accuracy with and without DSC-based synaptic devices (PDF)


    Terms & Conditions

    Most electronic Supporting Information files are available without a subscription to ACS Web Editions. Such files may be downloaded by article for research use (if there is a public use license linked to the relevant article, that license may permit other uses). Permission may be obtained from ACS for other uses through requests via the RightsLink permission system: http://pubs.acs.org/page/copyright/permissions.html.