Pixel

Algorithm–Architecture Co-Design for Vertically Integrated AI and HPC Systems: Challenges, Advantages, and Future Outlook

Algorithm–Architecture Co-Design for Vertically Integrated AI and HPC Systems: Challenges, Advantages, and Future Outlook 150 150

Abstract:

Next-generation scientific instruments in high-energy physics, X-ray imaging, fusion diagnostics, and other data-intensive applications generate data rates that increasingly exceed practical communication bandwidth and power budgets, making the traditional approach of transferring raw data to remote computing systems unsustainable. This paper presents the Vertically Integrated AI for Sensing and High-Performance …

View on IEEE Xplore

An Intelligent Vision SoC With Fully Analog Sensing-to-Inference Pipeline and Linearity-Recovery RRAM Compute-in-Memory

An Intelligent Vision SoC With Fully Analog Sensing-to-Inference Pipeline and Linearity-Recovery RRAM Compute-in-Memory 150 150

Abstract:

Edge vision systems that integrate image sensing and neural-network inference on a single chip are constrained by inter-layer analog-to-digital converters (ADCs) and on-chip weight storage density, which together limit system efficiency and network scale. This article presents a fully analog vision system-on-chip (SoC) with 2-Mbit multi-level-cell (MLC) resistive random access …

View on IEEE Xplore

A Modeling/Rendering Unified 3DGS Processor Based on Locality-Optimized Computation and Reconfigurable Architecture

A Modeling/Rendering Unified 3DGS Processor Based on Locality-Optimized Computation and Reconfigurable Architecture 150 150

Abstract:

Traditional 3-D Gaussian-splatting (3DGS) modeling relies on iterative optimization, leading to substantial latency and limited generalizability. Feedforward 3DGS alleviates these limitations by generating Gaussians through a single network inference. However, most existing processors are optimized for optimization-based 3DGS and do not fully exploit the distinct computational characteristics of feedforward 3DGS. …

View on IEEE Xplore

A 28-nm CMOS 1-Chip In-Pixel Memory Backplane Circuit for 1800 × 1350 6652-PPI MicroLED AR Glasses With Per-Pixel Current Sensing and Compensation

A 28-nm CMOS 1-Chip In-Pixel Memory Backplane Circuit for 1800 × 1350 6652-PPI MicroLED AR Glasses With Per-Pixel Current Sensing and Compensation 150 150

Abstract:

This article presents a 28-nm CMOS 1-chip LED-on-Silicon (LEDoS) backplane (BP) for 6652-pixel-per-inch (PPI) microLED ( $mu $ LED) augmented reality (AR) displays. To address power and display quality challenges in AR glasses, the proposed BP employs two low-power techniques: self-refresh and partial update, and introduces four display quality enhancement methods: MSB …

View on IEEE Xplore

Raisu: A Radiation-Tolerant Neuromorphic Imager With In-Pixel Self-Healing and Unified Spiking Processing for Spaceborne Vision

Raisu: A Radiation-Tolerant Neuromorphic Imager With In-Pixel Self-Healing and Unified Spiking Processing for Spaceborne Vision 150 150

Abstract:

Spaceborne imaging systems operate under severe radiation and energy constraints, where single-event effects (SEEs), total ionizing dose (TID), and displacement damage dose (DDD) critically degrade reliability and performance. This work presents a radiation-tolerant neuromorphic imager prototype featuring a fully spike-based neuromorphic vision architecture with in-pixel self-healing capability and an integrated …

View on IEEE Xplore

TD-dAJC: A 100 MS/s 2 pJ/Pixel Time-Domain Weight and Integrating-MAC-Based Direct Analog to MJPEG Compression for Video Sensor Nodes

TD-dAJC: A 100 MS/s 2 pJ/Pixel Time-Domain Weight and Integrating-MAC-Based Direct Analog to MJPEG Compression for Video Sensor Nodes 150 150

Abstract:

Always-on, high-quality video sensors are increasingly required in wearables, surveillance, and Internet-of-Things (IoT) devices, where stringent energy constraints make efficient in-sensor compression necessary. These systems must support high-definition (HD) and 4 K video, yet conventional digital video compression pipelines consume significant power because high-speed analog-to-digital converters (ADCs) must digitize every pixel, …

View on IEEE Xplore

A 96 × 64 SPAD LiDAR Sensor With 6× Imaging FoM Promotion by AI-Assisted Fusion of 2-D and Edge-Guided Sparse 3-D Detection

A 96 × 64 SPAD LiDAR Sensor With 6× Imaging FoM Promotion by AI-Assisted Fusion of 2-D and Edge-Guided Sparse 3-D Detection 150 150

Abstract:

Flash-mode direct time-of-flight (dToF) light detection and ranging (LiDAR) is widely adopted in autonomous systems and robotics due to its capability for accurate depth measurement and robustness under challenging lighting conditions. However, scaling single-photon avalanche diode (SPAD)-based dToF pixel arrays toward higher resolution 3-D imaging introduces significant challenges in …

View on IEEE Xplore

QVGA CMOS LiDAR Sensor With nMOS-Only SPAD Analog Front-End and Area-Efficient Priority Histogramming TDC

QVGA CMOS LiDAR Sensor With nMOS-Only SPAD Analog Front-End and Area-Efficient Priority Histogramming TDC 150 150

Abstract:

This article presents a high-resolution (HR) CMOS light detection and ranging (LiDAR) sensor capable of generating quarter video graphics array (QVGA) ( $320~{\times }~240$ ) depth images. To support high pixel density, the sensor employs a compact 6-transistor (6-T) nMOS-only single-photon avalanche diode (SPAD) analog front-end (AFE) with a column-shared active recharge circuit. …

View on IEEE Xplore