Novel view synthesis is an essential functionality for enabling immersive experiences in various Augmented- and Virtual-Reality (AR/VR) applications, for which Neural Radiance Field (NeRF) has emerged as the state-of-the-art (SOTA) technique. In particular, generalizable NeRFs have gained increasing popularity thanks to their cross-scene generalization capability, which enables NeRFs to be instantly serviceable for new scenes without per-scene training. Despite their promise, generalizable NeRFs aggravate the prohibitive complexity of NeRFs due to their required extra memory accesses needed to acquire scene features, causing NeRFs' ray marching process to be memory-bounded. To tackle this dilemma, existing sparsity-exploitation techniques for NeRFs fall short, because they require knowledge of the sparsity distribution of the target 3D scene which is unknown when generalizing NeRFs to a new scene. To this end, we propose Gen-NeRF, an algorithm-hardware co-design framework dedicated to generalizable NeRF acceleration, which aims to win both rendering efficiency and generalization capability in NeRFs. To the best of our knowledge, Gen-NeRF is the first to enable real-time generalizable NeRFs, demonstrating a promising NeRF solution for next-generation AR/VR devices. On the algorithm side, Gen-NeRF integrates a coarse-then-focus sampling strategy, leveraging the fact that different regions of a 3D scene contribute differently to the rendered pixels depending on where the objects are located in the scene, to enable sparse yet effective s ampling. In addition, Gen-NeRF replaces the ray transformer, which is generally included in SOTA generalizable NeRFs to enhance density estimation, with a novel Ray-Mixer module to reduce workload heterogeneity. On the hardware side, Gen-NeRF highlights an accelerator micro-architecture dedicated to accelerating the resulting model workloads from our Gen-NeRF algorithm to maximize the data reuse opportunities among different rays by making use of their epipolar geometric relationship. Furthermore, our Gen-NeRF accelerator features a customized dataflow to enhance data locality during point-to-hardware mapping and an optimized scene feature storage strategy to minimize memory bank conflicts across camera rays of NeRFs. Extensive experiments validate the effectiveness of our proposed Gen-NeRF framework in enabling real-time and generalizable novel view synthesis.

Biomedical imaging plays a pivotal role in the diagnosis and management of various diseases, offering invaluable insights into the human body’s anatomy and intricate physiological processes.^{1 – 4} Traditional imaging modalities, such as X-ray [Fig. 1(a)(i)] and ultrasound (US) [Fig. 1(a)(ii)], have long been the cornerstones of medical diagnostics, each endowed with unique strengths and limitations.^5,8,9 Photoacoustic tomography (PAT)^{10 – 14} is a medical imaging technique that employs both optical and acoustic energy, as shown in Fig. 1(a)(iii–iv).⁶ PAT, based on photoacoustic (PA) effect [Fig. 1(b)], transforms absorbed light energy into sound waves.¹⁵ As shown in Fig. 1(a), PAT, providing high-resolution imaging of breast cancer,^7,16,17 has recently been approved by the Food and Drug Administration as a complementary tool to X-ray mammography and US for breast cancer diagnosis and screening.¹⁸

Eye tracking has become an essential human-machine interaction modality in virtual and augmented reality (VR/AR) applications requiring high throughput (e.g., >240 FPS), small form-factor, and enhanced visual privacy. Existing eye tracking systems have adopted bulky lens-based cameras, and thus suffer from both a large form-factor and high communication cost between the camera and backend processor. This work presents a camera, algorithm, and accelerator co-designed lensless eye tracking system dubbed EyeCoD, which to the best of our knowledge is the first to provide a general front-end eye tracking solution for AR/VR while satisfying the requirements for both high throughput and smaller form-factor. Specifically, EyeCoD integrates system-, algorithm-, and accelerator-level techniques to boost system efficiency without sacrificing the eye tracking accuracy. We believe that our EyeCoD system will pave the way for next-generation eye tracking solutions in VR/AR and can shed light on future innovations for intelligent imaging systems.

Wearable photoacoustic imaging devices hold great promise for continuous health monitoring and point-of-care diagnostics. However, the large data volume generated by high-density transducer arrays presents a major challenge for realizing compact and power-efficient wearable systems. This article presents a photoacoustic imaging receiver (RX) that embeds compressive sensing directly into the hardware to address this bottleneck. The RX integrates 16 AFEs and four matrix-vector-multiplication (MVM) SAR ADCs that perform energy- and area-efficient analog-domain compression. The architecture achieves a $4\times $ – $8\times $ reduction in output data rate while preserving low-loss full-array information. The MVM SAR ADC executes passive and accurate MVM using user-defined programmable ternary weights. Two signal reconstruction methods are implemented: 1) an optimization approach using the fast iterative shrinkage-thresholding algorithm and 2) a learning-based approach employing implicit neural representation. Fabricated in 65-nm CMOS, the chip achieves an ADC’s signal-to-noise-distortion ratio (SNDR) of 57.5 dB at 20.41 MS/s, with an analog front-end (AFE) input-referred noise of 3.5 nV/ $\mathbf {(\text {Hz})^{1/2}}$ . MVM linearity measurements show $R^{2} \gt 0.999$ across a wide range of weights and input amplitudes. The system is validated through phantom imaging experiments, demonstrating high-fidelity image reconstruction under up to $8\times $ compression. The RX consumes 5.83 mW/channel and supports a general ternary-weighted measurement matrix, offering a compelling solution for next-generation miniaturized, wearable PA imaging systems.

Quantum many-body scarred systems host nonthermal excited eigenstates immersed in a sea of thermal ones. In cases where exact expressions for these special eigenstates are not known, it is computationally demanding to distinguish them from their exponentially many thermal neighbors. We propose a matrix-product-state (MPS) algorithm, dubbed DMRG-S, to extract such states at system sizes far beyond the scope of exact diagonalization. Using this technique, we obtain scarred eigenstates in Rydberg-blockaded chains of up to 80 sites and perform a finite-size scaling study to address the lingering question of the stability for the N\'eel state revivals in the thermodynamic limit. Our method also provides a systematic way to obtain exact MPS representations for scarred eigenstates near the target energy without a priori knowledge. In particular, we find several new scarred eigenstates with exact MPS representations in kinetically constrained spin and clock models. The combination of numerical and analytical investigations in our work provides a new methodology for future studies of quantum many-body scars.

Current intraoperative imaging technologies in neurosurgery are limited by a trade-off between molecular specificity and imaging depth. Fluorescence imaging offers high molecular sensitivity but is restricted to superficial tissue (~1–2 mm), whereas ultrasound (US) provides deeper imaging without molecular contrast. Photoacoustic imaging (PAI) overcomes this limitation by converting optical absorption into acoustic signals, enabling depth-resolved molecular imaging. Here, we report the first-in-human intraoperative application of PAI during brain tumor surgery. A clinical photoacoustic/ultrasound (PA/US) system with a handheld linear-array probe and LED illumination was used to image endogenous hemoglobin contrast at 850 nm and exogenous indocyanine green (ICG) contrast at 820 and 940 nm. Imaging was performed at the cortical surface, on ex vivo tumor specimens, and within the surgical cavity. PAI visualized vascular architecture and enabled depth-resolved molecular imaging to ~1.5 cm, extending beyond the penetration depth of fluorescence imaging and establishing a new paradigm for intraoperative neurosurgical guidance.