Reservoir computing in a lithium-based magneto-ionic device
Sreeveni Das, Rhodri Mansell, Aarne Piha, Lukáš Flajšman, Maria-Andromachi Syskaki, Jürgen Langer, Sebastiaan van Dijken · 2025
Read Reservoir computing in a lithium-based magneto-ionic device by Sreeveni Das, Rhodri Mansell, Aarne Piha, Lukáš Flajšman, Maria-Andromachi Syskaki, Jürgen Langer, Sebastiaan van Dijken (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-08346.txt.
Introduction
Sreeveni Das and colleagues demonstrate reservoir computing in a lithium based magneto ionic device, where computation happens inside the material itself. A voltage controlled crossbar stack with a cobalt iron boron layer and a lithium conducting film produces magnetic domain patterns that change with input voltage. A chaotic Mackey Glass time series is encoded as voltage, and Fourier transforms of the domain patterns form the reservoir output.
Input rate, output smoothing, state size, and training length all shape forecasting skill. Short term prediction favors smoothed, low dimensional states with little training, while prediction near the series delay time prefers raw, high dimensional states and longer training. Faster inputs reduce memory capacity and nonlinear processing strength.
Catalog record
| CREATOR | Sreeveni Das, Rhodri Mansell, Aarne Piha, Lukáš Flajšman, Maria-Andromachi Syskaki, Jürgen Langer, Sebastiaan van Dijken |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 30 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-08346
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Reservoir computing in a lithium-based magneto-ionic device Sreeveni Das, Rhodri Mansell, Aarne Piha, Lukáš Flajšman, Maria-Andromachi Syskaki, Jürgen Langer, Sebastiaan van Dijken Abstract In-materio computing exploits the intrinsic physical dynamics of materials to perform complex computations, enabling low-power, real-time data processing by embedding computation directly within physical layers. Here, we demonstrate a voltage-controlled magneto-ionic device that functions as a reservoir computer capable of forecasting chaotic time series. The device consists of a crossbar structure with a Ta/CoFeB/Ta/MgO/Ta bottom electrode and a LiPON/Pt top electrode. A chaotic Mackey-Glass time series is encoded into a voltage signal applied to the device, while 2D Fourier transforms of voltage-dependent magnetic domain patterns form the output. Performance is influenced by the input rate, smoothing of the output, the number of elements in the reservoir state vector, and the training duration. We identify two distinct computational regimes: short-term prediction is optimized using smoothed, low-dimensional states with minimal training, whereas prediction around the Mackey-Glass delay time benefits from unsmoothed, high-dimensional states and extended training. Reservoir computing metrics reveal that slower input rates are more tolerant to output smoothing, while faster input rates degrade both memory capacity and nonlinear processing. These findings demonstrate the potential of magneto-ionic systems for neuromorphic computing and offer design principles for tuning performance in response to input signal characteristics. I Introduction Physical reservoir computing is a promising neuromorphic computing paradigm that uses the intrinsic dynamics of physical systems to perform temporal information processing and pattern recognition tasks Lukoševičius and Jaeger 2009 ; Tanaka et al. 2019 ; Yan et al. 2024 . Unlike recurrent neural networks, which require extensive training of internal weights, reservoir computing simplifies the process by training only the output weights. Various physical implementations have been demonstrated using optical Duport et al. 2012 ; Van der Sande et al. 2017 , optoelectronic Paquot et al. 2012 ; Liu et al. 2022 , mechanical Fernando and Sojakka 2003 ; Dion et al. 2018 , electronic Du et al. 2017 ; Moon et al. 2019 ; Zhong et al. 2021 , and magnetic systems Torrejon et al. 2017 ; Bourianoff et al. 2018 ; Watt and Kostylev 2020 ; Allwood et al. 2023 ; Lee et al. 2024 ; Everschor-Sitte et al. 2024 , each exploiting distinct temporal dynamics and offering prospects for high-speed, energy-efficient computation Yan et al. 2024 . Magneto-ionic devices, systems that modulate magnetic properties through voltage-driven ion migration, hold significant promise for neuromorphic computing. These devices naturally exhibit short-term plasticity, mimicking the behavior of biological synapses Mishra et al. 2019 ; Monalisha et al. 2023 ; Monalisha et al. 2024 ; Bernard et al. 2025 ; Das et al. 2025 . Furthermore, the interplay between ionic transport and magnetization dynamics gives rise to nonlinear responses and memory effects, making magneto-ionic systems well-suited for reservoir computing Monalisha et al. 2024 ; Das et al. 2025 . Magneto-ionic effects have been demonstrated across a wide range of material systems involving various mobile ions, including oxygen Bi et al. 2014 ; Bauer et al. 2015 , hydrogen Tan et al. 2018 ; Huang et al. 2021 , nitrogen de Rojas et al. 2020 ; de Rojas et al. 2022 , and lithium Dasgupta et al. 2014 ; Ameziane et al. 2022 ; Ameziane et al. 2023a . Reversible voltage-induced ion migration enables dynamic control over key magnetic properties, such as magnetic moment Dasgupta et al. 2014 ; de Rojas et al. 2020 ; de Rojas et al. 2022 , magnetic anisotropy Bi et al. 2014 ; Bauer et al. 2015 ; Tan et al. 2018 ; Ameziane et al. 2022 , the Dzyaloshinskii-Moriya interaction Srivastava et al. 2018 , and the Ruderman–Kittel–Kasuya–Yosida (RKKY) interaction Kossak et al. 2023 ; Ameziane et al. 2023a . Low-voltage magneto-ionic actuation has also been employed to reversibly control skyrmion nucleation and annihilation Fillion et al. 2022 ; Ameziane et al. 2023b . Figure 1: (a) Schematic of the magneto-ionic device, comprising a crossbar junction with a layered Ta/CoFeB/Ta/MgO/Ta bottom electrode and LiPON/Pt top electrode. Application of a positive voltage to the top electrode drives Li + ions from the LiPON layer toward the bottom electrode, enabling voltage-controlled modulation of the magnetic domain state in the CoFeB layer. (b) Out-of-plane magnetic hysteresis loops measured by MOKE microscopy under different applied voltages, illustrating voltage-induced changes in magnetic behavior. (c)-(e) MOKE microscopy images showing the magnetic domain structure in the CoFeB layer at a fixed out-of-plane magnetic field of 0.3 mT under applied voltages of (c) -2 V, (d) 0 V, and (e) +2 V. Figure 2: (a) Delay-embedded trajectory of the Mackey-Glass signal used in this study. The inset shows the same 100-step segment of the signal as a time series. (b) Representative voltage waveforms used to drive the magneto-ionic reservoir, obtained by linearly mapping the Mackey-Glass signal onto a voltage range of -2 V to +2 V. The input rates shown are 0.04 (top), 0.20 (middle), and 0.41 (bottom) Mackey-Glass steps per MOKE microscopy frame. (c)-(e) Zoom-ins of MOKE microscopy images captured at the (c) 1000th, (d) 1050th, and (e) 1100th frame in a video recorded at an input rate of 0.20 Mackey-Glass steps per frame. Device architectures inspired by solid-state batteries and supercapacitors, such as LiCoO 2 /LiPON/Co and LiPON/CoFeB heterostructures Ameziane et al. 2022 ; Ameziane et al. 2023b ; Monalisha et al. 2024 , have enabled reversible modulation of magnetic properties via lithium ion migration. Lithium insertion into the magnetic layer induces significant changes in magnetic anisotropy, leading to pronounced modifications of the magnetic domain structure. These devices also demonstrate excellent voltage cycling endurance, exceeding 10 5 cycles Ameziane et al. 2023b , underscoring their potential for use in reservoir computing applications. The Mackey-Glass (MG) equation is widely used as a benchmark to evaluate the temporal processing capabilities of reservoir computing systems Soriano et al. 2015 ; Lee et al. 2024 ; Sun et al. 2023 . Originally developed to model delayed physiological feedback, this delay differential equation generates chaotic behavior for specific parameter regimes Mackey and Glass 1977 . The form used here is: \frac{\textrm{d}x}{\textrm{d}t}=\frac{ax(t-\tau)}{1+x(t-\tau)^{n}}-bx(t), (1) where x is the Mackey-Glass signal, t is time, and a,b,\tau and n are control parameters. In this work, we implement a reservoir computing system based on a magneto-ionic device, in which voltage-driven ion migration induces the nonlinear dynamics required for temporal information processing. The system’s performance is evaluated using a Mackey-Glass time-series prediction task, serving as a representative benchmark for chaotic signal processing. We show how input rate, state-vector dimensionality, and causal smoothing modify prediction ability, yielding task-adaptive operating points without hardware changes. II Results and Discussion Figure 3: (a) Radially averaged 2D Fourier transforms of the full-size versions of the MOKE microscopy images from Fig. 2, recorded at an input rate of 0.20 Mackey-Glass steps per frame. (b) Temporal evolution of the reservoir signal for selected wavenumber components as a function of video frame number, recorded at an input rate of 0.20 Mackey-Glass steps per frame. The corresponding input signal is shown for comparison. (c) Effect of causal Savitzky-Golay filtering on the reservoir states for an input rate of 0.20 Mackey-Glass steps per frame. Reservoir responses are shown for second-order polynomial smoothing with window