File Download

There are no files associated with this item.

  • Find it @ UNIST can give you direct access to the published full text of this article. (UNISTARs only)

Views & Downloads

Detailed Information

Cited time in webofscience Cited time in scopus
Metadata Downloads

Full metadata record

DC Field Value Language
dc.contributor.advisor Lee, Kyuho -
dc.contributor.author Jang, Wu-Young -
dc.date.accessioned 2025-04-04T13:51:38Z -
dc.date.available 2025-04-04T13:51:38Z -
dc.date.issued 2025-02 -
dc.description.abstract A super resolution (SR) accelerator with variable bit compression is proposed for 4K restoration with > 60 frames- per-second (fps) in mobile video applications. Although demands of 4K video contents grows dramatically, huge feature maps of 4K SR with low bit precision prohibit its real time operation due to the large external memory access (EMA) bandwidth that recent mobile processors cannot support. Despite the effort of previous SR processors to reduce EMA by quantizing both feature maps and weights, they suffered from the trade-off between EMA and network performance drop. To facilitate > 60 fps 4K SR on mobile processor, this work proposes key contributions at both software and hardware level: the tile-based distribution-aware statistical encoding that results in 61.2% compression rate without information loss; and an energy-efficient SR processor that supports variable bit compression that reduces power consumption and and EMA by 34.2% and 89.7%, respectively. Designed with 28 nm CMOS technology, the proposed system can accelerate ×2 scale 4K image restoration at 68.7 fps. It shows 3.69 TOPS of peak performance and 10.1 TOPS/W of energy efficiency, achieving 2.8× higher energy efficiency than the previous SR processor. -
dc.description.degree Master -
dc.description Graduate School of Artificial Intelligence -
dc.identifier.uri https://scholarworks.unist.ac.kr/handle/201301/86592 -
dc.identifier.uri http://unist.dcollection.net/common/orgView/200000868970 -
dc.language ENG -
dc.publisher Ulsan National Institute of Science and Technology -
dc.subject Accelerator -
dc.subject Super Resolution -
dc.subject Variable Length Compression -
dc.title An Energy-Efficient 4K Super Resolution Accelerator with Variable Length Compression -
dc.type Thesis -

qrcode

Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.