Cited 1 time in webofscience Cited 2 time in scopus

Balancing Computation Loads and Optimizing Input Vector Loading in LSTM Accelerators

Title
Balancing Computation Loads and Optimizing Input Vector Loading in LSTM Accelerators
Authors
Park, JunkiYi, WooseokAhn, DaehyunKung, JaehaKim, Jae-Joon
DGIST Authors
Park, Junki; Yi, Wooseok; Ahn, Daehyun; Kung, Jaeha; Kim, Jae-Joon
Issue Date
2020-09
Citation
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 39(9), 1889-1901
Type
Article
Article Type
Article
Author Keywords
Sparse matricesLogic gatesHardwareComputer architectureClocksHistoryStandardsAcceleratorscomputer architecturehardwaremachine learningrecurrent neural networks (RNNs)
ISSN
0278-0070
Abstract
The long short-term memory (LSTM) is a widely used neural network model for dealing with time-varying data. To reduce the memory requirement, pruning is often applied to the weight matrix of the LSTM, which makes the matrix sparse. In this paper, we present a new sparse matrix format, named rearranged compressed sparse column (RCSC), to maximize the inference speed of the LSTM hardware accelerator. The RCSC format speeds up the inference by: 1) evenly distributing the computation loads to processing elements (PEs) and 2) reducing the input vector load miss within the local buffer. We also propose a hardware architecture adopting hierarchical input buffer to further reduce the pipeline stalls which cannot be handled by the RCSC format alone. The simulation results for various datasets show that combined use of the RSCS format and the proposed hardware requires 2x smaller inference runtime on average compared to the previous work.
URI
http://hdl.handle.net/20.500.11750/12562
DOI
10.1109/TCAD.2019.2926482
Publisher
Institute of Electrical and Electronics Engineers
Related Researcher
  • Author Kung, Jaeha Intelligent Digital Systems Lab
  • Research Interests 딥러닝, 가속하드웨어, 저전력 하드웨어, 고성능 시스템
Files:
There are no files associated with this item.
Collection:
Department of Information and Communication EngineeringIntelligent Digital Systems Lab1. Journal Articles


qrcode mendeley

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE