Computing power is an important support for the development of artificial intelligence (AI). With the saturation of the Moore’s law, the development of emerging intelligent computing carriers and basic theories is imminent. Empowered by the high-speed and high parallelism of light propagation, optoelectronic intelligent computing has evolved as the potential for next-generation high-performance computing paradigm.

Our team has carried out original explorations of large-scale reconfigurable optoelectronic intelligent computing in terms of theory, architecture, algorithms, and systems. We have proposed the Fourier domain diffraction neural network, constructed the reconfigurable diffraction computing processor (DPU), developed the all-analog optoelectronic fusion computing chip ACCEL, and the large-scale general-purpose intelligent optoelectronic computing chip "TaiChi," achieving a thousand-fold increase in computing power and a million-fold increase in energy efficiency. Our works opened up new paths for exploring high-performance intelligent computing in the post-Moore era, which would be propelling a range of areas such as autonomous driving, edge computing, etc.

The related achievements have been published in top journals such as Nature (2023), Science(2024) and their series journals. More than 50 domestic and foreign patents have been granted. Related works have been selected as highly cited articles in ESI, cover articles, awarded the Top 10 Advances in Chinese Optics in 2022&2023, and awarded the Top 10 Advances in Semiconductor Research in China in 2023.

Representative publications

• Fully forward mode training for optical neural networks, Nature 2024
  

• Large-scale photonic chiplet Taichi empowers 160-TOPS/W artificial general intelligence, Science 2024

• Photonic neuromorphic architecture for tens-of-task lifelong learning, Light: Science & Applications 2024

• All-analog photo-electronic chip for high-speed vision tasks, Nature 2023

• Photonic unsupervised learning variational autoencoder for high-throughput and low-latency image transmission, Science Advances 2023

• Training large-scale optoelectronic neural networks with dual-neuron optical-artificial learning, Nature Communications 2023

• Ultrafast dynamic machine vision with spatiotemporal photonic computing, Science Advances 2023

• A multichannel optical computing architecture for advanced machine vision, Light: Science & Applications 2022

• Large-scale neuromorphic optoelectronic computing with a reconfigurable diffractive processing unit, Nature Photonics 2021

• In situ optical backpropagation training of diffractive optical neural networks, Photonics Research 2020

• Fourier-space diffractive deep neural network, Physical Review Letters 2019