Reservoir Computing with Magnetic Thin Films

TL;DR

This work demonstrates a spintronic reservoir computing approach using micromagnetically simulated magnetic thin films (Co, Ni, Fe) to harness nonlinear spin dynamics and memory for temporal tasks. By coupling evolution in materio with reservoir computing, the authors optimize input mappings and material parameters to compete with small recurrent neural networks on benchmarks such as Laser, NARMA-10, and NARMA-30, even at compact film sizes. Task-independent measures (kernel rank and memory capacity) reveal trade-offs between nonlinearity and memory and show material- and size-dependent trends, while scaling insights emphasize that larger films alone do not guarantee better performance. The study provides a framework for material discovery and design of physical RC devices, with implications for fast, energy-efficient edge computing and future experimental realizations of magnetic reservoir hardware.

Abstract

Advances in artificial intelligence are driven by technologies inspired by the brain, but these technologies are orders of magnitude less powerful and energy efficient than biological systems. Inspired by the nonlinear dynamics of neural networks, new unconventional computing hardware has emerged with the potential to exploit natural phenomena and gain efficiency, in a similar manner to biological systems. Physical reservoir computing demonstrates this with a variety of unconventional systems, from optical-based to memristive systems. Reservoir computers provide a nonlinear projection of the task input into a high-dimensional feature space by exploiting the system's internal dynamics. A trained readout layer then combines features to perform tasks, such as pattern recognition and time-series analysis. Despite progress, achieving state-of-the-art performance without external signal processing to the reservoir remains challenging. Here we perform an initial exploration of three magnetic materials in thin-film geometries via microscale simulation. Our results reveal that basic spin properties of magnetic films generate the required nonlinear dynamics and memory to solve machine learning tasks (although there would be practical challenges in exploiting these particular materials in physical implementations). The method of exploration can be applied to other materials, so this work opens up the possibility of testing different materials, from relatively simple (alloys) to significantly complex (antiferromagnetic reservoirs).

Figures (10)

• Figure 1: (a) Reservoir computing model split into input, reservoir and output layers connected by adjustable weights. The reservoir is self-contained, typically featuring a sparse, recurrent network of processing nodes. (b) Schematic of simulated thin-film magnetic reservoir system, consisting of micromagnetic macro-cells with properties derived from atomistic values. Global input sources $u$ connect via weights $W_{in}$ to drive local magnetisation fields inducing spin oscillations. Each macro-cell's average magnetic moment produces a 3-d orientation vector $X_{xyz}$ forming a reservoir state. States are then combined via a linear readout function $W_{out}$ to produce the final system output $y$. (c) Impulse response of micromagnetic spin system. Signal injected in the centre of the film via the $z$-axis at 25 time-step intervals with 10 ps scanning frequency.
• Figure 2: Dynamics of Co magnetic film at different sizes. An input pulse is supplied at two locations on the film at $t=10$. Red indicates a positive magnetisation, and blue, negative.
• Figure 3: Performance of materials and reservoir networks on benchmark tasks. NMSE is used to compare equivalent-sized reservoirs (smaller is better). Multiple reservoir sizes are shown in columns, task are shown across rows. Each type of system is represented by colour (graph reservoir$=$purple; echo state reservoir$=$green; material$=$orange). The method used to create the reservoirs is given on the $x$-axis (random or evolved). The boxplots show the best reservoir from each of 20 runs of the search.
• Figure 4: Kernel rank (normalised by reservoir size) versus linear memory capacity for all materials, sizes and tasks. Materials are separated by colour: Ni (light green), Co (orange), and Fe (grey). Each column refers to a film size; each row refers to a task. Material reservoirs shown are the 20 best reservoirs per material per film size per task from the previous experiment.
• Figure 5: Parameters (input scaling $b$, damping $\alpha$, leak rate $a$) found by evolution for each material and film size, on each task.