High-Accuracy Approximate Addition based on Carry Speculation
碩士 === 國立交通大學 === 資訊科學與工程研究所 === 106 === Approximate computing has been proposed as an approach to break through the limitations of traditional designs. Since nowadays a lot of applications can tolerate a certain degree of errors, it provides a new trade-off between computation quality and performan...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | en_US |
| Published: |
2017
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/npzmuc |
| Summary: | 碩士 === 國立交通大學 === 資訊科學與工程研究所 === 106 === Approximate computing has been proposed as an approach to break through the limitations of traditional designs. Since nowadays a lot of applications can tolerate a certain degree of errors, it provides a new trade-off between computation quality and performance. In this thesis, we propose a carry truncate adder (CTA), which can achieve high performance with limited area overhead and well-controlled low error magnitude. The experimental results show that with only 23% area overhead and small loss in accuracy, CTA can achieve up to 2.17X throughput improvement compared to traditional adder. The strength of CTA in error magnitude is also demonstrated using three error-tolerant applications. Experimental results show that the approximate addition based on CTA can achieve almost the same quality of results as accurate addition.
|
|---|