A 28nm 0.22μJ/token memory-compute-intensity-aware CNN-Transformer accelerator with hybrid-attention-based layer-fusion and cascaded pruning for semantic-segmentation
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866909979611496448 |
|---|---|
| author | Dong, Pingcheng Tan, Yonghao Liu, Xuejiao Luo, Peng Liu, Yu Liang, Luhong Zhou, Yitong Pang, Di Yung, Man-To Zhang, Dong Huang, Xijie Liu, Shih-Yang Wu, Yongkun Tian, Fengshi Tsui, Chi-Ying Tu, Fengbin Cheng, Kwang-Ting |
| author_facet | Dong, Pingcheng Tan, Yonghao Liu, Xuejiao Luo, Peng Liu, Yu Liang, Luhong Zhou, Yitong Pang, Di Yung, Man-To Zhang, Dong Huang, Xijie Liu, Shih-Yang Wu, Yongkun Tian, Fengshi Tsui, Chi-Ying Tu, Fengbin Cheng, Kwang-Ting |
| contents | This work presents a 28nm 13.93mm2 CNN-Transformer accelerator for semantic segmentation, achieving 3.86-to-10.91x energy reduction over previous designs. It features a hybrid attention unit, layer-fusion scheduler, and cascaded feature-map pruner, with peak energy efficiency of 52.90TOPS/W (INT8). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_17555 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | A 28nm 0.22μJ/token memory-compute-intensity-aware CNN-Transformer accelerator with hybrid-attention-based layer-fusion and cascaded pruning for semantic-segmentation Dong, Pingcheng Tan, Yonghao Liu, Xuejiao Luo, Peng Liu, Yu Liang, Luhong Zhou, Yitong Pang, Di Yung, Man-To Zhang, Dong Huang, Xijie Liu, Shih-Yang Wu, Yongkun Tian, Fengshi Tsui, Chi-Ying Tu, Fengbin Cheng, Kwang-Ting Image and Video Processing This work presents a 28nm 13.93mm2 CNN-Transformer accelerator for semantic segmentation, achieving 3.86-to-10.91x energy reduction over previous designs. It features a hybrid attention unit, layer-fusion scheduler, and cascaded feature-map pruner, with peak energy efficiency of 52.90TOPS/W (INT8). |
| title | A 28nm 0.22μJ/token memory-compute-intensity-aware CNN-Transformer accelerator with hybrid-attention-based layer-fusion and cascaded pruning for semantic-segmentation |
| topic | Image and Video Processing |
| url | https://arxiv.org/abs/2512.17555 |