Energy-efficient XNOR-free In-Memory BNN Accelerator with Input Distribution Regularization

SRAM-based in-memory Binary Neural Network (BNN) accelerators are garnering interests as a platform for energy-efficient edge neural network computing thanks to their compactness in terms of hardware and neural network parameter size. However, previous works had to modify SRAM cells to support XNOR...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Digest of technical papers - IEEE/ACM International Conference on Computer-Aided Design S. 1 - 9
Hauptverfasser: Kim, Hyungjun, Oh, Hyunmyung, Kim, Jae-Joon
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: Association on Computer Machinery 02.11.2020
Schlagworte:
binary input data
computation error
Computational modeling
conventional 6T SRAM arrays
conversion method
energy conservation
Energy consumption
Energy efficiency
energy-efficient edge neural network computing
energy-efficient XNOR-free in-memory BNN accelerator
Handheld computers
inference energy efficiency
input distribution regularization
memory array
network training
neural chips
neural network parameter size
SRAM cells
SRAM chips
SRAM-based in-memory binary neural network accelerators
Training
Transistors
XNOR operations
ISSN:1558-2434
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:SRAM-based in-memory Binary Neural Network (BNN) accelerators are garnering interests as a platform for energy-efficient edge neural network computing thanks to their compactness in terms of hardware and neural network parameter size. However, previous works had to modify SRAM cells to support XNOR operations on memory array resulting in limited area and energy efficiencies. In this work, we present a conversion method which replaces the signed inputs (+1/-1) of BNN with the unsigned inputs (1/0) without computation error, and vice versa. The method enables BNN computing on conventional 6T SRAM arrays and improves area and energy efficiencies. We also demonstrate that further energy saving is possible by skewing the distribution of binary input data based on regularization during network training. Evaluation results show that the proposed techniques improve the inference energy efficiency by up to 9.4x for various benchmarks over previous works.
ISSN:1558-2434
DOI:10.1145/3400302.3415641