# Fileset

[apexadb19bsupp1.pdf](https://mdr.nims.go.jp/filesets/6f2c9c3b-c27a-4846-839a-d1ef82121d51/download)

## Creator

[Hina Kitano](https://orcid.org/0009-0008-9132-0275), [Daiki Nishioka](https://orcid.org/0000-0002-3369-7700), [Kazuya Terabe](https://orcid.org/0000-0003-3988-3456), [Takashi Tsuchiya](https://orcid.org/0000-0002-6950-6160)

## Rights

[Creative Commons BY Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/)

## Other metadata

[Physical reservoir computing with graphene-based solid electric double layer transistor and the information processing capacity analysis](https://mdr.nims.go.jp/datasets/1c555c29-4898-4edd-b872-e7ebacdfadf2)

## Fulltext

1  Supplementary Information  Physical Reservoir Computing with Graphene-based Solid Electric Double Layer Transistor and the Information Processing Capacity Analysis  Hina Kitano1,2, Daiki Nishioka1,3, Kazuya Terabe1, and Takashi Tsuchiya1,2*  1Research Center for Materials Nanoarchitectonics (MANA), National Institute for Materials Science (NIMS), 1-1 Namiki, Tsukuba, Ibaraki, 305-0044, Japan.  2Department of Applied Physics, Faculty of Advanced Engineering, Tokyo University of Science, Katsushika, Tokyo 125-8585, Japan 3International Center for Young Scientists (ICYS), NIMS, 1-1 Namiki, Tsukuba, Ibaraki, 305-0044, Japan.   *E-mail: TSUCHIYA.Takashi@nims.go.jp      2  Device fabrication A monolayer graphene was grown using a chemical vapor deposition (CVD) on Cu foil and transferred on SiO2/Si substrate. The channel was fabricated with photolithography and dry etching. During etching, H2O gas was introduced at 10 Pa, and graphene was scraped at 100 W for 1 minute to for the channel shape. The IGR-EDLT were fabricated with six channels, lengths were 5, 20 and 100 μm, and widths were 30 and 80 μm. Au/Cr thin films (50/10 nm) were deposited as source and drain electrodes using photolithography and electron-beam evaporation. 400 nm of Li+ conducting amorphous Li-Nb-O (a-LN) and 100 nm Li+-hole mixed conducting LiCoO2 thin films were then deposited by pulsed laser deposition. Pt (50 nm) was also deposited on top of the LiCoO2 layer as a current collector.   Measurement method IGR-EDLT measurements were carried out at room temperature in a vacuum chamber evacuated. Probers were used to connect the IGR-EDLT in the chamber, and electrical measurements were carried out using the source measure unit and pulse measure unit of semiconductor parameter analyzer (4200A-SCS, Keithley).  IPC calculation method The memory capacity and nonlinearity of RC are quantitatively characterized by IPC. The target data ym(k) is an orthogonal polynomial that covered all linear and nonlinear combinations of inputs.  𝑦!(𝑘) =&𝑃"!,#[𝑢(𝑘 − 𝑑)]#$%&  (S1) Pn is an orthogonal polynomial of degree n’ (n’=1, 2, …), m, d, and D represent the polynomial index, delay and maximum delay, respectively. The input u(k) is a uniformly distributed random sequence. The capacity Cm of each component is calculated from the NMSE when reconstructing the target ym expressed by Eq. S1 from the reservoir state X(k), obtained by injecting u(k) into the reservoir as follows: 𝐶! = 1 − NMSE‘  (S2) The NMSE’ is the sum of squared errors divided by the target sum of squares. The total capacity Ctot is the sum of these component capacities.  𝐶'(' = 6 𝐶!)!%*  (S3) 3  M is the total number of indexes determined by the combination of order and delay length. The capacity by order, Cn, was calculated as the sum of the capacity by components where the total number of orders of interest is n: 𝐶" = 6 𝐶!!(")  (S4) m(n) represents all the indices of all orders n. To prevent overestimation of IPC, we used a surrogate method [S1, S2]. The threshold set at 1.5 times the surrogate capacity.  Electrical responses in each condition and the results of the NARMA2 task  Fig. S1. (a)ID response in VG pulse measurement. (b)Enlarged view of ID response in VG pulse measurement. (c)Target waveform and predict waveform in NARMA2 task.  The results of four representative points (Vmax = -0.8 V, Vmax = +0.2 V, Vmax = +1.2 V, and Vmax = +2.0 V) are shown in Fig. S1. As shown in Fig.S1(a), the response current ID to the applied VG pulses. The ID response becomes more complex as Vmax increases. Figure S1(b) is an enlarged view of Fig. S1(a), where Vmax = +1.2 V and Vmax = +2.0 V include the Dirac point in the VG range, and a quasi-second harmonic was generated. Figure S1(c) shows the calculation results of the NARMA2 task for each condition, among which Vmax 4  = +1.2 V was the best score.  Observation of quasi-second harmonic by the ID - VG cycle measurement   Fig. S2. (a)The ID-VG response to the triangular wave input (ch3). (b)The ID-VG response to the triangular wave input (ch1). (c)The ID response vs time (ch3). (d)The ID response vs time (ch1).  Figure S2(a) and (b) show the ID - VG response for ch1 (length was100 μm and width was 30 μm) and ch3 (length was5 μm and width was 30 μm) when the gate voltage (VG) of a transistor is swept continuously for 40 cycles within the range of -1.5 V to -0.5 V. With each cycle, the drain current increases in the negative voltage side of the Dirac point. Figure S3(c) and (d) show the same measurement results as in Fig. S3(a) and (b) but shown with time on the horizontal axis and ID on the vertical axis. In both Fig. S3(c) and Fig. S3(d), the quasi-second harmonic wave is observed. While it increases in ch3, it remains almost constant in ch1. As seen in the contrast behavior, the electrical response differs depending on the channel.  5   Fig. S3. (a)The ID - VG response for ch3. (b)The fast Fourier transform (FFT) spectrum of ch3. (c)The ID - VG response for ch1. (d)The FFT spectrum of ch1 6   The ID response to the triangular wave input shown in Fig. S2(c,d) was divided into four regions, and the fast Fourier transform (FFT) spectra were calculated as shown in Fig. S3. Figure S3(a) shows the divided ID response of ch3. In region 1, the quasi-second harmonic due to including the Dirac point in the VG range is observed. Corresponding to this response, overtones were clearly observed in the FFT spectrum of region 1, as shown in Fig. S3(b). As the quasi-second harmonic due to the Dirac point weakened from region 1 to 4, the FFT spectrum also showed a decrease in overtones. The overtones were also observed in the FFT spectrum corresponding to the ID response of ch1, as shown in Fig. S3(d). A similar decrease in overtones was observed in Fig. S3(d), although the decrement was not as pronounced as in ch3. The contrast difference in the two FFT spectra agrees well with the corresponding two ID responses in Fig. S3(a,c).  References [S1] J. Dambre, D. Verstraeten, B. Schrauwen, and S. Massar, Information processing capacity of dynamical systems. Scientific reports, 2(1), 514. (2012). [S2] S. Tsunegi, T. Kubota, A. Kamimaki, J. Grollier, V. Cros, K. Yakushiji, A. Fukushima, S. Yuasa, H. Kubota, K. Nakajima and T. Taniguchi, Information processing capacity of spintronic oscillator. Advanced Intelligent Systems, 5(9), 2300175 (2023).