Magnetoionics for Synaptic Devices and Neuromorphic Computing: Recent Advances, Challenges, and Future Perspectives

Abstract

With the advent of Big Data, traditional digital computing is struggling to cope with intricate tasks related to data classification or pattern recognition. To mitigate this limitation, software-based neural networks are implemented, but they are run in conventional computers whose operation principle (with separate memory and data-processing units) is highly inefficient compared to the human brain. Brain-inspired in-memory computing is achieved through a wide variety of methods, for example, artificial synapses, spiking neural networks, or reservoir computing. However, most of these methods use materials (e.g., memristor arrays, spintronics, phase change memories) operated with electric currents, resulting in significant Joule heating effect. Tuning magnetic properties by voltage-driven ion motion (i.e., magnetoionics) has recently emerged as an alternative energy-efficient approach to emulate functionalities of biological synapses: potentiation/depression, multilevel storage, or transitions from short-term to long-term plasticity. In this perspective, the use of magnetoionics in neuromorphic applications is critically reviewed, with emphasis on modulating synaptic weight through: 1) control of magnetization by voltage-induced ion retrieval/insertion; and 2) control of magnetic stripe domains and skyrmions in gated magnetic thin films adjacent to solid-state ionic supercapacitors. The potential prospects in this emerging research area together with a forward-looking discussion on future opportunities are provided.

Keywords: artificial synapses; brain‐inspired memories; magnetoionics; skyrmions.

Figure 1

a) Schematic illustration and electrical connection of the Co₃O₄‐based magnetic synapse in capacitor configuration. b) Schematics of the ionic distribution in the electrolyte at no‐gating condition. c) Schematics to illustrate the physical mechanism governing magnetoionics with extraction of oxygen ions upon negative gating.

Figure 2

a) M–μ₀ H hysteresis loops of 15 nm Co₃O₄ film in as‐prepared, treated (with V _G = –5 V for 1 h), and recovered (treated with V _G = 5 V for 1 h) states. The overlapping of M–μ₀ H loops of as‐prepared and recovered sample shows complete reversibility. X‐ray photoelectron spectroscopy analysis of 15 nm Co₃O₄ film is shown in panels b–d): b) Co 2p _3/2 spectra of the as‐prepared sample consisting of Co²⁺ and Co³⁺ peaks. c) Co 2p _3/2 spectra of the treated (V _G = –5 V, for 1 h) sample consisting of Co⁰ (metallic cobalt), Co²⁺, and Co³⁺ peaks. d) Co 2p _3/2 spectra of the recovered (V _G = 5 V, for 1 h), for which the Co⁰ metallic contribution vanishes, and only Co²⁺ and Co³⁺ peaks are again identified.

Figure 3

a) M versus t curves of 7 and 15 nm Co₃O₄ thin films treated with –5 V and measured by applying in‐plane magnetic field of μ₀ H = 1 T. b) M–μ₀ H hysteresis loops of both 7 and 15 nm films previously treated with –5 V for 40 min, showing a paramagnetic‐to‐ferromagnetic (OFF/ON) switching. Note that voltage was continuously applied during the measurement of the hysteresis loops. c) M–t curves of 7 and 15 nm Co₃O₄ thin films treated with V _G = –7 V for 20 min (green shadowed area), showing that the 7 nm film shows a volatile change of M after the voltage is turned off, whereas the 15 nm film exhibits a nonvolatile magnetoionic change. d) Dependence of M as a function of time for the 15 nm‐thick Co₃O₄ thin film measured at 0 V under different μ₀ H values (top panel), after having subjected the samples to prior voltage treatments using V _G = 0, −5, and −7 V for 20 min and subsequent demagnetization processes with an alternating magnetic field of decreasing amplitude (AC degauss).

Figure 4

a) Schematics of signal transmission across a biological synapse by ion migration. b) The M versus. V _G curve of the 15 nm‐thick magnetic synapse shows nonvolatile and reversible change in sample M with V _G sweeping.

Figure 5

STP realized in the 7 nm Co₃O₄ thin film. Excitatory postsynaptic behavior has been emulated by applying negative gate voltage pulses (V _G = –7 V) allowing the sample to magnetically relax in between them (at V _G = 0 V). It is experimentally demonstrated that potentiation becomes easier upon the repeated application of a negative voltage in the thin Co₃O₄‐based artificial synapse.

Figure 6

Long‐term plasticity realized in the 15 nm Co₃O₄ thin film with high retention. a) LTP: M progressively increases by applying a series of negative voltage pulses (V _G = –7 V, t _P = 15 s, t _D = 3 s); b) LTD: M decreases drastically by applying a series of positive voltage pulses (V _G = 7 V, t _P = 15 s, t _D = 3 s); c) simultaneous realization of one LTP + LTD cycle. d) Series of 50 potentiation/ depression cycles (see text for details) demonstrating good endurance of the effects.

Figure 7

a) SADP: The M–μ₀ H loops of 15 nm Co₃O₄ thin films treated with voltage pulses of different amplitudes (V _G = –5, –10, –15 V) for 20 min (“AP” corresponds to the as‐prepared state). b) Corresponding M _S values as a function of spike amplitude; c) Multilevel nonvolatile reversible magnetic states were realized for 15 nm Co₃O₄ thin film. A series of negative gate voltage pulses were applied to obtain a step‐wise increment of M. The magnetization was then brought back to the initial stage in a step‐wise manner by applying a series of positive gate voltage pulses.

Figure 8

a) Cross‐sectional schematic of the sample. The bottom Ta/Pt electrode is grounded, and a voltage is applied to the top Pt contact. b) Cyclic voltammogram of a junction taken at a voltage sweep rate of 50 mV s⁻¹. c–f) Polar MOKE hysteresis loops as a function of applied out‐of‐plane magnetic field for applied voltages of c) –2 V, d) 0 V, e) 1 V, and f) 1.5 V. g–j) Images with magnetic contrast obtained at 0 mT during the respective MOKE loop sweeps for the applied voltage indicated in the MOKE loop above each image. k–n) Images with magnetic contrast obtained at 0.73 mT during the MOKE loop sweeps at different applied voltages. All images are on the same scale, as indicated in (g).

Figure 9

a) The number of skyrmions or stripe domains as a function of time through a set sequence of the applied voltage which switches from –2 to 2 V and back to –2 V. b) The fractional area filled by skyrmions and stripes for the same sequence of voltages as in (a). c–f) MOKE image snapshots of the domain states at the times indicated in the image which correspond to the data in (a,b). The scale is indicated in (c). The applied magnetic field in all measurements is 0.68 mT.

Figure 10

a) The number of skyrmions obtained using a stepped voltage under 0.7 mT applied magnetic field. The number of skyrmions is extracted from images taken 10 s after each voltage step. b) The number of skyrmions at 0 V after a voltage pulse from 0 to 2.5 V or 3 V as a function of the pulse length under 0.65 mT applied magnetic field. c) The number of skyrmions as a function of time following a voltage pulse applied from –1.1 V to +3 V and back to –1.1 V for different pulse length. The length of the voltage pulse is indicated in the legend and the applied magnetic field is 0.7 mT. d) The number of skyrmions as a function of time following a voltage pulse applied from different starting voltages, as shown in the legend, to +7 V for 400 μs, and back to the starting voltage. The applied magnetic field is 0.7 mT.