Title | ||
---|---|---|
A Novel Signed Bit-serial Fixed-point Accumulator with Configurable Overflow-Protection Precision |
Abstract | ||
---|---|---|
In the design of digital signal arithmetic circuits, the serial structure is widely used because of its high clock frequency, low power and area consumption. The bit-serial accumulator is the basic arithmetic unit used in the serial systems, and also the key unit for the performance of the whole system. For the requirements of a bit-serial accumulator for general digital signal processing (DSP) systems and machine learning (ML) algorithms, a novel scheme of the signed bit-serial fixed-point accumulator is proposed in this paper. For the overflow issue, an overflow-protection circuit with configurable precision is provided as the main contribution of the proposed scheme, which can achieve the tradeoff between accuracy and latency for the signed system design. In the case study, the proposed scheme is applied to the convolution operations in convolutional neural networks (CNNs). The experimental results show that the proposed scheme can achieve a good implementation performance with reconfigurability. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1109/ASICON47005.2019.8983432 | 2019 IEEE 13th International Conference on ASIC (ASICON) |
Keywords | Field | DocType |
serial structure,high clock frequency,bit-serial accumulator,basic arithmetic unit,serial systems,overflow issue,overflow-protection circuit,configurable precision,signed system design,configurable overflow-protection precision,digital signal arithmetic circuits,signed bit-serial fixed-point accumulator | Digital signal processing,Reconfigurability,Convolutional neural network,Computer science,Digital signal,Convolution,Systems design,Electronic engineering,Computer hardware,Accumulator (structured product),Clock rate | Conference |
ISSN | ISBN | Citations |
2162-7541 | 978-1-7281-0736-3 | 0 |
PageRank | References | Authors |
0.34 | 0 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Lin Li | 1 | 323 | 79.92 |
Qiu Huang | 2 | 0 | 0.34 |
Hu Jianhao | 3 | 96 | 20.56 |
Jienan Chen | 4 | 84 | 13.64 |