A 22 nm 10 TOPS Mixed-Precision Neural Network SoC for Image Processing with Energy-Efficient Dilated Convolution Support

Research output: Contribution to book/Conference proceedings/Anthology/ReportConference contributionContributedpeer-review

Contributors

Abstract

Dilated convolutions have become popular in recent Deep Neural Networks (DNNs). However, they introduce additional zeros within the weights. To efficiently compute these layers within image processing DNNs, we implemented a 10 TOPS neural network SoC in 22 nm that supports dilated convolutions with reduced energy consumption. By exploiting this advantage, the energy can be reduced by a factor of 4.46 even for DNNs such as DeepLabV3+ with a majority of standard convolutional layers.

Details

Original languageEnglish
Title of host publication2024 IEEE Symposium in Low-Power and High-Speed Chips (COOL CHIPS)
PublisherInstitute of Electrical and Electronics Engineers (IEEE)
Number of pages3
ISBN (electronic)979-8-3503-8414-7
ISBN (print)979-8-3503-8415-4
Publication statusPublished - 19 Apr 2024
Peer-reviewedYes

Conference

Title27th IEEE Symposium in Low-Power and High-Speed Chips
Abbreviated titleCOOL CHIPS 27
Conference number27
Duration17 - 19 April 2024
Website
LocationThe University of Tokyo
CityTokyo
CountryJapan

External IDs

Scopus 85194142861