Physical reservoir computing (PRC) has emerged as a low-cost machine-learning framework for edge computing applications, where real-time processing and energy efficiency are critical due to the rapid growth of data generated by Internet-of-Things devices. This study proposes a PRC scheme based on nonlinear heat-conduction phenomena arising from the nonlinear temperature-dependent thermal conductivity of materials and examines its computational properties using a one-dimensional model. Information is encoded as transient thermal inputs applied at one boundary, while the resulting temporal temperature responses at multiple spatial positions constitute the reservoir states. Nondimensionalization of the governing heat equation reveals key dimensionless parameters that govern dynamical memory and nonlinearity, enabling systematic exploration of the design space independent of specific physical units. Performance is evaluated using benchmark tasks, including the short-term memory (STM) task and the parity check (PC) task. The results demonstrate that memory capacity and nonlinear transformation capability can be tuned by adjusting the heat-input magnitude and the thermal diffusion time scale. The proposed heat-conduction-based PRC successfully performs both benchmark tasks, demonstrating the feasibility of implementing PRC through nonlinear thermal transport in materials with temperature-dependent thermal conductivity. These findings indicate that engineered thermal media can serve as effective physical reservoirs with material-level tunability for energy-efficient computing systems.
Physical reservoir computing (PRC) has attracted considerable attention as a low-training-cost framework for edge computing. In this study, we propose a PRC system that uses nonlinear heat conduction in an alumina ceramic plate. In the proposed system, the input information is applied as heat, and the transient temperature responses generated by heat conduction are used as reservoir states. The temperature dependences of the thermal conductivity and specific heat of alumina provide nonlinearity without relying on nonlinear sensor elements. The PRC performance was evaluated using temperature data obtained by finite-element analysis of an alumina disk. The results showed that the short-term memory task was improved by shortening the input step duration and increasing the input heat flux, whereas the parity-check task was enhanced by increasing the input heat flux. These results indicate that the input step duration primarily controls the memory, whereas the input heat flux controls the nonlinear transformation through temperature-dependent thermal diffusion. A demonstration experiment was performed using a fabricated alumina ceramic device with embedded resisters for heating and temperature sensing. The experimental results showed trends that were qualitatively consistent with the numerical simulations, demonstrating the feasibility of PRC using heat conduction in an actual solid-state thermal system. This study suggests that heat, which is generally treated as an unavoidable byproduct of electronic devices, can be used as a computational resource for edge-computing applications.
Yasuaki Ikeda, Seita Umemoto, Jun Hirotani et al.· AIP Publishing· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
