Is Your AI Fast Enough to Run a Fusion Reactor?

2026-09-30

AI Summary

This study addresses the limitations in AI inference speed and temporal determinism for feedback control in nuclear fusion. Using the DIII-D tokamak as a testbed, we construct a cross-backend benchmarking framework to systematically evaluate the real-time inference performance of ten neural network architectures. The core contribution is a deployment backend selection methodology that co-optimizes model scale against control cycle time budgets. Through comparative experiments across deep learning frameworks and GPU/CPU inference backends, our results demonstrate that large-parameter models incur prohibitive inference latency on CPUs, establishing GPUs as essential hardware infrastructure for high-speed real-time control systems. This work provides systematic guidance for deploying AI-based controllers in fusion devices.

Abstract

Machine learning models are increasingly used in feedback control loops for nuclear fusion, where inference speed and predictable timing are critical. We summarize lessons from models deployed for control on the DIII-D tokamak and develop a benchmark to compare inference backends across ten neural networks and model components from fusion control and diagnostic pipelines. For models greater than five million parameters, the CPU backends take tens to thousands of milliseconds, while GPU inference is substantially faster, suggesting an upper limit on CPU-oriented development for control. These results show why the deployment backend must be selected together with the model and its control-cycle budget.

Problem

• nuclear fusion control
• machine learning inference speed
• real-time feedback control
• inference backend benchmarking

Innovation

• nuclear fusion control
• inference benchmark
• deployment backend
• real-time inference
• tokamak