Parallel optical computing capable of 100‑wavelength multiplexing

Optical computing, as a representative of non-von Neumann architectures, offers natural advantages such as scalability, low power consumption, ultra-high speed, wide bandwidth, and high parallelism. It has emerged as a key technology in the post-Moore era for solving large-scale, high-dimensional data problems, including tensor operations and complex image processing. In recent years, academia and industry have focused on increasing the matrix size and optical clock speed of optical computing chips. However, further breakthroughs along these dimensions are proving extremely challenging. Therefore, enhancing parallelism has become a frontier direction in improving the performance of optical computing and a key step toward practical implementation.

 

In a new paper published in eLight, a team, led by Prof. Peng Xie from Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, and Prof Guangwei Hu from Nanyang Technological University (Singapore), introduced a groundbreaking optical computing architecture that dramatically boosts parallelism. Addressing key challenges such as high-density channel crosstalk, low-latency signal synchronization, and large-scale integration, the team developed a novel on-chip parallel optical computing system named “Liuxing-I”. It integrates multiple components, including a self-developed integrated microcavity optical frequency comb for multi-wavelength light sources, a reconfigurable high-performance optical computing chip, and a scalable, high-precision driving system. Under a 50 GHz modulation frequency, the system can reach a theoretical peak performance exceeding 2560 TOPS with energy efficiency surpassing 3.2 TOPS/W.

 

To overcome spectral dispersion and crosstalk issues, the researchers established a physical model for parallel optical computing and proposed a universal error correction method that improves wavelength-channel consistency to over 90%. A microcomb-based multi-wavelength source was designed to suppress inter-channel interference. Furthermore, the inverse design was used to enhance bandwidth and robustness, achieving over 40 nm device bandwidth.

 

Prof. Peng Xie remarked: “We demonstrated parallel optical computing system capable of 100-wavelength multiplexing, paving a new path to scaling up optical computility. This is like turning a single-lane highway into a superhighway capable of accommodating over a hundred lanes in parallel—dramatically increasing throughput without modifying the chip hardware. Each critical technical challenge was tackled by dedicated team members, following a point-to-line-to-surface research strategy. This systematic approach enabled us to move rapidly from basic research to full system integration.”

 

This study validates the feasibility of ultra-high parallelism in optical computing. By integrating multi-wavelength light sources, optical interconnection chips, computing chips, matrix drivers system, and hybrid photonic-electronic computing algorithms, the researchers validated scalable parallel optical computing architecture capable of 100-wavelength multiplexing. The proposed physical models and correction methods significantly improve consistency in high-density, parallel data processing. The successful implementation of this technology marks a significant leap in the computing capacity of photonic systems and represents a critical step toward the practical application of optical computing.

---

---