BitBlade: Energy-Efficient Variable Bit-Precision Hardware Accelerator for Quantized Neural Networks
- Authors
- Ryu, Sungju; Kim, Hyungjun; Yi, Wooseok; Kim, Eunhwan; Kim, Yulhwa; Kim, Taesu; Kim, Jae-Joon
- Issue Date
- Jun-2022
- Publisher
- IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
- Keywords
- Computer architecture; Neural networks; Hardware acceleration; Adders; Arrays; Random access memory; Throughput; Bit-precision scaling; bitwise summation; channel-first and pixel-last tiling (CFPL); channel-wise aligning; deep neural network; hardware accelerator; multiply-accumulate unit
- Citation
- IEEE JOURNAL OF SOLID-STATE CIRCUITS, v.57, no.6, pp.1924 - 1935
- Journal Title
- IEEE JOURNAL OF SOLID-STATE CIRCUITS
- Volume
- 57
- Number
- 6
- Start Page
- 1924
- End Page
- 1935
- URI
- http://scholarworks.bwise.kr/ssu/handle/2018.sw.ssu/41722
- DOI
- 10.1109/JSSC.2022.3141050
- ISSN
- 0018-9200
- Abstract
- We introduce an area/energy-efficient precisionscalable neural network accelerator architecture. Previous precision-scalable hardware accelerators have limitations such as the under-utilization of multipliers for low bit-width operations and the large area overhead to support various bit precisions. To mitigate the problems, we first propose a bitwise summation, which reduces the area overhead for the bit-width scaling. In addition, we present a channel-wise aligning scheme (CAS) to efficiently fetch inputs and weights from on-chip SRAM buffers and a channel-first and pixel-last tiling (CFPL) scheme to maximize the utilization of multipliers on various kernel sizes. A test chip was implemented in 28-nm CMOS technology, and the experimental results show that the throughput and energy efficiency of our chip are up to 7.7x and 1.64x higher than those of the state-of-the-art designs, respectively. Moreover, additional 1.5-3.4x throughput gains can be achieved using the CFPL method compared to the CAS.
- Files in This Item
- There are no files associated with this item.
- Appears in
Collections - College of Information Technology > ETC > 1. Journal Articles
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.