In the realm of energy and computing, a team of researchers from Aalto University in Finland has made a significant stride in the field of in-materio computing, which leverages the inherent physical properties of materials to perform complex computations. This approach promises to enable low-power, real-time data processing by integrating computation directly within physical systems. The team, led by Sreeveni Das and including Rhodri Mansell, Aarne Piha, Lukáš Flajšman, Maria-Andromachi Syskaki, Jürgen Langer, and Sebastiaan van Dijken, has demonstrated a voltage-controlled magneto-ionic device that functions as a reservoir computer capable of forecasting chaotic time series. Their findings were published in the journal Nature Communications.
The device in question consists of a crossbar structure with a Ta/CoFeB/Ta/MgO/Ta bottom electrode and a LiPON/Pt top electrode. The researchers encoded a chaotic Mackey-Glass time series into a voltage signal applied to the device, while 2D Fourier transforms of voltage-dependent magnetic domain patterns formed the output. The performance of the device was influenced by several factors, including the input rate, smoothing of the output, the number of elements in the reservoir state vector, and the training duration.
The study identified two distinct computational regimes. Short-term prediction was optimized using smoothed, low-dimensional states with minimal training. On the other hand, prediction around the Mackey-Glass delay time benefited from unsmoothed, high-dimensional states and extended training. The researchers also found that slower input rates were more tolerant to output smoothing, while faster input rates degraded both memory capacity and nonlinear processing.
These findings highlight the potential of magneto-ionic systems for neuromorphic computing, which mimics the brain’s architecture and functionality. For the energy sector, this research could pave the way for more efficient and powerful data processing systems that require less energy. This could be particularly beneficial for applications such as real-time monitoring and control of energy grids, predictive maintenance of energy infrastructure, and advanced energy management systems. The design principles identified in this study could also guide the development of future computing systems tailored to specific input signal characteristics, further enhancing their performance and efficiency.
In summary, the research conducted by the team at Aalto University represents a significant advancement in the field of in-materio computing. Their work not only demonstrates the potential of magneto-ionic devices for neuromorphic computing but also offers valuable insights for the energy sector, where efficient and powerful data processing systems are increasingly in demand.
This article is based on research available at arXiv.