Ocelli: Efficient Processing-in-Pixel Array Enabling Edge Inference of Ternary Neural Networks

Article Engineering, Electrical & Electronic

Senputing: An Ultra-Low-Power Always-On Vision Perception Chip Featuring the Deep Fusion of Sensing and Computing

Han Xu et al.

Summary: In this paper, a new sensor chip called Senputing is proposed, which operates in two modes to achieve hierarchical processing of BNN. This chip can achieve always-on coarse classification and fine-grained recognition functions, and has high energy efficiency compared to state-of-the-art work.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS (2022)

Add to Collection

Article Computer Science, Hardware & Architecture

MR-PIPA: An Integrated Multilevel RRAM (HfOx)-Based Processing-In-Pixel Accelerator

Minhaz Abedin et al.

Summary: This study proposes a pixel-embedded accelerator based on multi-level HfOx RRAM for real-time smart image processing at edge devices. By optimizing the compute pixel and nonvolatile weight storage, it significantly reduces the energy consumption of data conversion and transmission while maintaining accuracy.

IEEE JOURNAL ON EXPLORATORY SOLID-STATE COMPUTATIONAL DEVICES AND CIRCUITS (2022)

Add to Collection

Article Engineering, Electrical & Electronic

A Reconfigurable Convolution-in-Pixel CMOS Image Sensor Architecture

Ruibing Song et al.

Summary: The separation of data capture and analysis in modern vision systems has caused issues such as high power consumption and slow response. This paper proposes a CMOS sensor architecture based on PIP, which performs convolution operations before column readout, significantly reducing power consumption and improving resource utilization for AI on resource-limited sensing devices.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY (2022)

Add to Collection

Proceedings Paper Computer Science, Hardware & Architecture

TizBin: A Low-Power Image Sensor with Event and Object Detection Using Efficient Processing-in-Pixel Schemes

Sepehr Tabrizchi et al.

Summary: This paper proposes TizBin, a low-power processing in-sensor scheme with event and object detection capabilities, to enable data-intensive neural network tasks. TizBin offers unique features such as analog convolutions and non-volatile magnetic RAMs, reducing power consumption and achieving high efficiency.

2022 IEEE 40TH INTERNATIONAL CONFERENCE ON COMPUTER DESIGN (ICCD 2022) (2022)

Add to Collection

Proceedings Paper Engineering, Biomedical

Integrated Sensing and Computing using Energy-Efficient Magnetic Synapses

Shaahin Angizi et al.

Summary: This work presents a processing-in-sensor platform leveraging magnetic devices for real-time and smart image processing in AI devices, reducing power consumption while maintaining acceptable accuracy on image datasets.

PROCEEDINGS OF THE TWENTY THIRD INTERNATIONAL SYMPOSIUM ON QUALITY ELECTRONIC DESIGN (ISQED 2022) (2022)

Add to Collection

Article Engineering, Electrical & Electronic

A 0.5-V Real-Time Computational CMOS Image Sensor With Programmable Kernel for Feature Extraction

Tzu-Hsiang Hsu et al.

Summary: This article introduces a low-power, low-voltage dual-mode 0.5-V computational CMOS image sensor capable of real-time feature extraction without sacrificing frame rate. The sensor, validated to display raw and feature images at 480 frames per second, is applied in real-time edge feature detection and hand gesture recognition scenarios.

IEEE JOURNAL OF SOLID-STATE CIRCUITS (2021)

Add to Collection

Article Automation & Control Systems

An On-Chip Binary-Weight Convolution CMOS Image Sensor for Neural Networks

Woo-Tae Kim et al.

Summary: This paper introduces a CMOS image sensor capable of performing binary convolution, which greatly reduces memory usage and computational complexity. The sensor operates in two modes to generate high-quality images efficiently. Experimental evaluations show low read noise and a large dynamic range in the normal operation mode.

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS (2021)

Add to Collection

Proceedings Paper Engineering, Electrical & Electronic

A 4.57μW@120fps Vision System of Sensing with Computing for BNN-Based Perception Applications

Han Xu et al.

IEEE ASIAN SOLID-STATE CIRCUITS CONFERENCE (A-SSCC 2021) (2021)

Add to Collection

Article Engineering, Electrical & Electronic

NS-FDN: Near-Sensor Processing Architecture of Feature-Configurable Distributed Network for Beyond-Real-Time Always-on Keyword Spotting

Qin Li et al.

Summary: The proposed feature-configurable distributed network architecture (NS-FDN) optimizes energy consumption, latency, and recognition accuracy for always-on keyword spotting (KWS) applications. By splitting the conventional single network into distributed sub-networks, NS-FDN adapts to changing keyword demands in real scenes, improving recognition accuracy. Using a channel-independent training framework reduces speech feature redundancy, while configuring features reduces computing complexity and enhances processing speed for deeper optimization.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS (2021)

Add to Collection

Article Chemistry, Analytical