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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Convolutional neural networks (CNNs) have recently demonstrated superiorquality for computational imaging applications. Therefore, they have greatpotential to revolutionize the image pipelines on cameras and displays.However, it is difficult for conventional CNN accelerators to supportultra-high-resolution videos at the edge due to their considerable DRAMbandwidth and power consumption. Therefore, finding a further memory- andcomputation-efficient microarchitecture is crucial to speed up this comingrevolution.  In this paper, we approach this goal by considering the inference flow,network model, instruction set, and processor design jointly to optimizehardware performance and image quality. We apply a block-based inference flowwhich can eliminate all the DRAM bandwidth for feature maps and accordinglypropose a hardware-oriented network model, ERNet, to optimize image qualitybased on hardware constraints. Then we devise a coarse-grained instruction setarchitecture, FBISA, to support power-hungry convolution by massiveparallelism. Finally,we implement an embedded processor---eCNN---whichaccommodates to ERNet and FBISA with a flexible processing architecture. Layoutresults show that it can support high-quality ERNets for super-resolution anddenoising at up to 4K Ultra-HD 30 fps while using only DDR-400 and consuming6.94W on average. By comparison, the state-of-the-art Diffy uses dual-channelDDR3-2133 and consumes 54.3W to support lower-quality VDSR at Full HD 30 fps.Lastly, we will also present application examples of high-performance styletransfer and object recognition to demonstrate the flexibility of eCNN.

eCNN: A Block-Based and Highly-Parallel CNN Accelerator for Edge Inference