Molecular dynamics (MD) simulations with ab initio accuracy have become indispensable for studying microscopic phenomena. Although graph neural network (GNN)-based MD models provide higher accuracy than deep neural network (DNN)-based models, their complex message-passing architectures introduce substantial computational and communication bottlenecks for large-scale inference. In this paper, we propose AtomFlow, a holistic acceleration for a representative GNN-based MD model on the Fugaku supercomputer. Our key optimizations include: (1) we design a fused domain-specific EC kernel that avoids materializing large edge-related intermediate tensors; (2) we design an inter-process communication operator that leverages Fugaku’s uTofu interconnect to accelerate atomic-feature exchange; and (3) we redesign the inference workflow with a framework-free implementation to avoid the overhead of the PyTorch framework on Fugaku. Experimental results show that AtomFlow achieves up to a 9.77 × speedup over the baseline implementation, reaching up to 26.14 nanoseconds per day for a 576,000-atom system with 12,000 nodes on Fugaku.
Yiming Du, Mingzhen Li, Lijun Liu et al.· 0 citations
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 resistors 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 by-product of electronic devices, can be used as a computational resource for edge-computing applications.
Seita Umemoto, Yuki Matsunaga, Yasuaki Ikeda et al.· Journal of Applied Physics· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
