The Shanghai Institute of Optics and Photonics, Chinese Academy of Sciences has made progress in the field of multi-dimensional multiplexed optical computing.
2026-09-07 17:54Chips & Hardware🔥 42.2 heat score
1sources
1days unfolding
42.2heat score
3mentions
SummaryAI generated
On September 7, 2026, the team led by Researcher Xie Peng from the Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, made progress in the field of multi-dimensional multiplexed optical computing. Based on research using the “Meteor-1” chip, they developed a multi-dimensional multiplexed optical computing architecture that combines wavelength and spatial multiplexing. This architecture enables parallel processing of multiple tasks and multimodal information, and its feasibility was verified in remote sensing image registration and recognition. The system consists of three independent optical computing cores, with each core capable of parallel computation for 108 wavelengths, achieving a computational parallelism of 324 and matrix network sharing for 2 tasks. Experiments demonstrated the performance of various optical convolution operators over a wide spectral range, with root mean square errors for both electronic benchmarks falling below 0.08; the accuracy of remote sensing image recognition reached 84.5%. This work verified the feasibility of expanding the computing power of optical computing through multi-dimensional processing of wavelength, space, and tasks. Related results were published in *Laser&PhotonicsReviews*.
The team led by Researcher Xie Peng from the Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, has made progress in the field of multi-dimensional optical computing. The relevant findings were published in *Laser&PhotonicsReviews*. Based on research using the “Meteor-1” chip, the team developed a multi-dimensional optical computing architecture that integrates wavelength division and spatial division复用, enabling parallel processing of multiple tasks and modalities. This architecture was verified in remote sensing image registration and recognition. The system consists of three independent optical computing cores, with each core capable of parallel computation for 108 wavelengths, achieving a computational parallelism of 324 and matrix network sharing for 2 tasks. Experiments demonstrated the performance of various optical convolution operators over a wide spectral range, with root-mean-square errors for both electronic benchmarks falling below 0.08; the accuracy of remote sensing image recognition reached 84.5%. This work验证了 the feasibility of expanding optical computing capabilities through multi-dimensional information processing in terms of wavelength, space, and tasks.