| SHAN Rui (山 蕊),SONG Yupeng,LIAO Wang.[J].高技术通讯(英文),2025,31(3):226~237 |
|
| A reconfigurable array with adaptive approximation using dual threshold accuracy |
| |
| DOI：10. 3772 / j. issn. 1006-6748. 2025. 03. 002 |
| 中文关键词: |
| 英文关键词: reconfigurable computing, approximate computing, low power, energy-efficient |
| 基金项目: |
| Author Name | Affiliation | | SHAN Rui (山 蕊) | (School of Electronic Engineering, Xi’an University of Posts and Telecommunications,Xi’an 710121, P. R. China) | | SONG Yupeng | | | LIAO Wang | |
|
| Hits: 20 |
| Download times: 25 |
| 中文摘要: |
| |
| 英文摘要: |
| With the growing demand for compute-intensive applications such as artificial intelligence (AI) and video processing, traditional reconfigurable array processors fail to meet the requirements of high-performance computing and related domains, primarily due to their high power consumption and low energy efficiency. To address this limitation, this paper proposes an accuracy-adaptive approximate reconfigurable array architecture featuring preset dual thresholds and support for four computational accuracy levels, enabling flexible adaptation to diverse application needs. The architecture integrates a self-adaptive mechanism that dynamically adjusts computational precision based on real-time error threshold feedback. To evaluate the proposed architecture, the you only look once version 5(YOLOv5) deep neural network algorithm is parallelized and deployed on the approximate reconfigurable array. Experimental results demonstrate that the architecture achieves an 18. 93% reduction in power consumption compared with conventional reconfigurable structures operating in full-precision mode. Additionally, the design exhibits superior energy efficiency and reduced computational resource utilization, thereby significantly enhancing the overall performance and applicability of reconfigurable array processors in power-sensitive scenarios. |
|
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
