Jiamin Wu

The rapid advancement of artificial intelligence (AI) imposes unprecedented speed and energy requirements on large-scale computation. Owing to its intrinsic high bandwidth, massive parallelism capacity and low energy consumption, optical computing is widely regarded as one of the most promising technologies to address AI computation demands. Here we offer a synthesis of photonic platforms that have already shown programmability, high integration density and stability, and that can support large-scale AI models, edge computing, and logic and scientific computing. We summarize the requirements that optical computing imposes on nanomaterials properties, architecture design, nanofabrication and large-scale integration. Finally, we analyse commercial systems and provide a perspective on how nanotechnology-informed solutions can improve performance in terms of computing speed, scalability and complexity in optical computing.

Volumetric time-lapse fluorescence microscopy enables panoramic visualization of biological structures and dynamics in living organisms. However, noise fundamentally limits the fidelity and quantitative analysis of fluorescence recordings. Current self-supervised denoising methods do not fully model the four-dimensional (4D) correlations of volumetric time-lapse data, thereby constraining restoration quality. Here we present SDT-4D, the first 4D self-supervised denoising framework to jointly incorporate lateral, axial, and temporal dependencies without requiring any clean images for training. We designed a unified 4D transformer architecture to capture long-range dependencies across 4D hypervoxels. This design enables SDT-4D to restore biological signals while preserving volumetric continuity and temporal dynamics. To improve reliability, we further developed a physics-informed uncertainty quantification (PhUQ) method and embedded it into SDT-4D, deriving pixel-wise confidence maps of restored images. In volumetric calcium imaging, SDT-4D achieves state-of-the-art denoising performance and reveals neuronal anatomy and calcium transients. In cellular morphodynamic imaging, SDT-4D restores cellular structures obscured by noise and facilitates downstream analysis of immune cell migration and glial morphology evolution. Together, these results demonstrate the effectiveness and reliability of SDT-4D for restoring and interpreting photon-limited biological dynamics.

Recent developments in imaging facilitate large-scale three-dimensional (3D) neuronal recording. While the resulting large datasets shed light on population-level neural coding, extracting neuronal calcium dynamics from 3D volumes remains more challenging than from two-dimensional images due to noise and scattering. Here we present DeepWonder3D, a general end-to-end pipeline for rapid and robust 3D neuronal extraction with high fidelity. Instead of processing voxel by voxel, DeepWonder3D works on the multiview projections of 3D imaging data obtained either digitally or optically through specific point spread functions and is therefore applicable to diverse techniques, including point-scanning microscopy, light-field microscopy and two-photon synthetic aperture microscopy. Integrating denoising, resolution registration, background removal, neuronal extraction and multiview fusion into a unified pipeline tailored for large-scale high-resolution datasets contaminated by noise and scattering, DeepWonder3D outperforms state-of-the-art methods in 3D localization accuracy with a tenfold reduction in computational costs, validated by numerical simulations and a hybrid two-photon/light-field imaging system. With the RUSH3D mesoscope, DeepWonder3D achieves high-fidelity 3D calcium extraction of tens of thousands of neurons across the mouse cortex within hours.

Simultaneously capturing subcellular dynamics across thousands of cells in mammals remains challenging because of trade-offs among spatial resolution, field of view, imaging speed and sample viability, particularly under high-NA (numerical aperture) conditions. Here we design a high-NA mesoscale objective lens and present a confocal volumetric mesoscope with an NA of 1.05, RUSH3D-HR, achieving ~390-nm resolution across a volume of 2.7 × 2.0 × 0.04 mm ³ at five volumes per second with low phototoxicity. By integrating our previously developed confocal scanning light-field framework with a parallelized deep-learning pipeline, RUSH3D-HR increases experimental efficiency by orders of magnitude. Long-term imaging at 70 billion voxels per second visualizes immune responses such as long-distance collective cell migrations during wound healing and large-scale migrasome generation and neutrophil–macrophage interactions during acute liver failure. Hour-long monitoring over 6,000 neutrophils reveals transient emergent swarming behaviors in mouse spleen during lipopolysaccharide-induced inflammation, which is suppressed with electroacupuncture at ST36, providing direct evidence for the immunomodulatory mechanism of peripheral electrical stimulation.

Fluorescence lifetime imaging microscopy (FLIM) enables quantitative measurement of molecular environments, interactions and protein conformations. However, a large number of photons are required for accurate lifetime determination, restricting its practical applications in fast deep-tissue imaging. Here, we present event-based first-photon FLIM (EFLIM), a self-supervised denoising method that infers fluorescence lifetime at extremely low light. By representing each excitation event as a binary process instead of histogram accumulation, EFLIM reduces photon requirement by over two orders of magnitude compared to state-of-the-art algorithms, leading to an apparent mean lifetime measurement below one photon per pixel with strong robustness to intensity artifacts. To demonstrate EFLIM's applicability, we observed transient intracellular dynamics of ligand-dependent molecular states, captured putative vesicle-mediated contacts between different lymphocytes through multiplexed imaging in a single spectral channel and achieved rapid label-free visualization of tumor heterogeneity in human glioma tissue. These results illustrate EFLIM's strong potential in neuroscience, cell biology, immunology and pathology by probing dynamic molecular processes in vivo.