English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The demand to process vast amounts of data generated from state-of-the-arthigh resolution cameras has motivated novel energy-efficient on-device AIsolutions. Visual data in such cameras are usually captured in the form ofanalog voltages by a sensor pixel array, and then converted to the digitaldomain for subsequent AI processing using analog-to-digital converters (ADC).Recent research has tried to take advantage of massively parallel low-poweranalog/digital computing in the form of near- and in-sensor processing, inwhich the AI computation is performed partly in the periphery of the pixelarray and partly in a separate on-board CPU/accelerator. Unfortunately,high-resolution input images still need to be streamed between the camera andthe AI processing unit, frame by frame, causing energy, bandwidth, and securitybottlenecks. To mitigate this problem, we propose a novelProcessing-in-Pixel-in-memory (P2M) paradigm, that customizes the pixel arrayby adding support for analog multi-channel, multi-bit convolution, batchnormalization, and ReLU (Rectified Linear Units). Our solution includes aholistic algorithm-circuit co-design approach and the resulting P2M paradigmcan be used as a drop-in replacement for embedding memory-intensive first fewlayers of convolutional neural network (CNN) models withinfoundry-manufacturable CMOS image sensor platforms. Our experimental resultsindicate that P2M reduces data transfer bandwidth from sensors and analog todigital conversions by ~21x, and the energy-delay product (EDP) incurred inprocessing a MobileNetV2 model on a TinyML use case for visual wake wordsdataset (VWW) by up to ~11x compared to standard near-processing or in-sensorimplementations, without any significant drop in test accuracy.

P2M: A Processing-in-Pixel-in-Memory Paradigm for Resource-Constrained TinyML Applications