DC Field | Value | Language |
---|---|---|
dc.contributor.author | Han, Donghyeon | ko |
dc.contributor.author | Im, Dongseok | ko |
dc.contributor.author | Park, Gwangtae | ko |
dc.contributor.author | Kim, Youngwoo | ko |
dc.contributor.author | Song, Seokchan | ko |
dc.contributor.author | Lee, Juhyoung | ko |
dc.contributor.author | Yoo, Hoi-Jun | ko |
dc.date.accessioned | 2022-05-24T05:01:01Z | - |
dc.date.available | 2022-05-24T05:01:01Z | - |
dc.date.created | 2022-05-24 | - |
dc.date.created | 2022-05-24 | - |
dc.date.issued | 2022-03 | - |
dc.identifier.citation | IEEE MICRO, v.42, no.2, pp.16 - 24 | - |
dc.identifier.issn | 0272-1732 | - |
dc.identifier.uri | http://hdl.handle.net/10203/296647 | - |
dc.description.abstract | In this article, an energy-efficient deep learning processor is proposed for deep neural network (DNN) training in mobile platforms. Conventional mobile DNN training processors suffer from high-bit precision requirement and high ReLU-dependencies. The proposed processor breaks through these fundamental issues by adopting three new features. It first combines the runtime automatic bit precision searching method addition to both conventional dynamic fixed-point representation and stochastic rounding to realize low-precision training. It adopts bit-slice scalable core architecture with the input skipping functionality to exploit bit-slice-level fine-grained sparsity. The iterative channel reordering unit helps the processor to maintain high core utilization by solving the workload unbalancing problem during zero-slice skipping. It finally achieves at least 4.4x higher energy efficiency compared with the conventional DNN training processors. | - |
dc.language | English | - |
dc.publisher | IEEE COMPUTER SOC | - |
dc.title | A Mobile DNN Training Processor With Automatic Bit Precision Search and Fine-Grained Sparsity Exploitation | - |
dc.type | Article | - |
dc.identifier.wosid | 000792913500011 | - |
dc.identifier.scopusid | 2-s2.0-85121785469 | - |
dc.type.rims | ART | - |
dc.citation.volume | 42 | - |
dc.citation.issue | 2 | - |
dc.citation.beginningpage | 16 | - |
dc.citation.endingpage | 24 | - |
dc.citation.publicationname | IEEE MICRO | - |
dc.identifier.doi | 10.1109/MM.2021.3135457 | - |
dc.contributor.localauthor | Yoo, Hoi-Jun | - |
dc.contributor.nonIdAuthor | Song, Seokchan | - |
dc.description.isOpenAccess | N | - |
dc.type.journalArticle | Article | - |
dc.subject.keywordAuthor | Throughput | - |
dc.subject.keywordAuthor | Neural networks | - |
dc.subject.keywordAuthor | Resource management | - |
dc.subject.keywordAuthor | Deep learning | - |
dc.subject.keywordAuthor | Computer architecture | - |
dc.subject.keywordAuthor | System-on-chip | - |
dc.subject.keywordAuthor | Energy efficiency | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.