Researchers build a digital twin for optical computing

A research team in China has proposed a Digital Twin Optical Computing System that lets optical computing tasks be trained, tested and optimized offline before deployment to physical hardware. The approach could reduce device bottlenecks, speed development and make optical computing research more shareable and reproducible. Why it matters: - Optical computing systems promise high bandwidth, low latency and parallel processing for AI workloads. - The new digital twin approach reduces dependence on scarce physical hardware during training and optimization. - The framework could shorten development cycles, support parallel task work and make optical computing easier to share across research teams. What happened: - Researchers proposed a Digital Twin Optical Computing System, or DT-OCS, that reproduces the input-output responses of a physical optical computing system on a digital platform. - The study was published online in Opto-Electronic Advances on April 21, 2026. - The work was carried out in China. The details: - DT-OCS is a system-level, measurement-driven digital surrogate for a physical optical computing system. - The model lets researchers run offline simulation, training and optimization without repeatedly occupying hardware. - The framework is designed to transfer optimized parameters from the digital environment to the physical optical computing system. - The researchers used a high-speed optical computing system with a silicon photonic feature-computing chip as the experimental platform. - The team demonstrated DT-OCS on image classification and sequential decision-making tasks. - The digital twin system supported offline training and final deployment for the Fashion-MNIST classification task. - Experimental results showed that physical-system performance closely matched the digital model’s predictions after training and optimization. - The framework also supported parallel development of multiple tasks. - The study made the DT-OCS framework and related task datasets openly available to the research community. - The paper lists DOI 10.29026/oea.2026.250254 . Between the lines: - The central shift is from hardware-tied experimentation to a reusable software model that can be tested before any physical run. - That matters because conventional optical computing development can be slowed by device queues, repeated tuning and online calibration. - The open-source release is meant to make optical computing more reproducible and easier to compare across tasks and platforms. - The study also argues for a future model in which optical computing platforms combine hardware with continuously updated digital twins. What’s next: - Researchers are expected to use DT-OCS to train, optimize and validate new tasks offline before moving to physical systems. - If the approach scales, optical computing platforms could become more collaborative, more scalable and less dependent on exclusive hardware access. - The broader goal is to move optical computing from a specialized lab setup to a general-purpose research resource.