Exploiting Neural-Network Statistics for Low-Power DNN Inference
Specialized compute blocks have been developed for efficient nn execution. However, due to the vast amount of data and parameter movements, the interconnects and on-chip memories form another bottleneck, impairing power and performance. This work addresses this bottleneck by contributing a low-power...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2024-01-01
|
| Series: | IEEE Open Journal of Circuits and Systems |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10498075/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832592864385695744 |
|---|---|
| author | Lennart Bamberg Ardalan Najafi Alberto Garcia-Ortiz |
| author_facet | Lennart Bamberg Ardalan Najafi Alberto Garcia-Ortiz |
| author_sort | Lennart Bamberg |
| collection | DOAJ |
| description | Specialized compute blocks have been developed for efficient nn execution. However, due to the vast amount of data and parameter movements, the interconnects and on-chip memories form another bottleneck, impairing power and performance. This work addresses this bottleneck by contributing a low-power technique for edge-AI inference engines that combines overhead-free coding with a statistical analysis of the data and parameters of neural networks. Our approach reduces the power consumption of the logic, interconnect, and memory blocks used for data storage and movements by up to 80% for state-of-the-art benchmarks while providing additional power savings for the compute blocks by up to 39 %. These power improvements are achieved with no loss of accuracy and negligible hardware cost. |
| format | Article |
| id | doaj-art-69f456498f434569bfd63a9cc0ff980d |
| institution | Kabale University |
| issn | 2644-1225 |
| language | English |
| publishDate | 2024-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Open Journal of Circuits and Systems |
| spelling | doaj-art-69f456498f434569bfd63a9cc0ff980d2025-01-21T00:02:51ZengIEEEIEEE Open Journal of Circuits and Systems2644-12252024-01-01517818810.1109/OJCAS.2024.338821010498075Exploiting Neural-Network Statistics for Low-Power DNN InferenceLennart Bamberg0https://orcid.org/0000-0003-4673-8310Ardalan Najafi1https://orcid.org/0000-0002-6529-4084Alberto Garcia-Ortiz2https://orcid.org/0000-0002-6461-3864NXP Semiconductors, Hamburg, GermanyNXP Semiconductors, Hamburg, GermanyIntegrated Digital Systems, ITEM Institute, University of Bremen, Bremen, GermanySpecialized compute blocks have been developed for efficient nn execution. However, due to the vast amount of data and parameter movements, the interconnects and on-chip memories form another bottleneck, impairing power and performance. This work addresses this bottleneck by contributing a low-power technique for edge-AI inference engines that combines overhead-free coding with a statistical analysis of the data and parameters of neural networks. Our approach reduces the power consumption of the logic, interconnect, and memory blocks used for data storage and movements by up to 80% for state-of-the-art benchmarks while providing additional power savings for the compute blocks by up to 39 %. These power improvements are achieved with no loss of accuracy and negligible hardware cost.https://ieeexplore.ieee.org/document/10498075/Artificial intelligenceedge-AI inferencelow-power codinglow-power digital designneural networks |
| spellingShingle | Lennart Bamberg Ardalan Najafi Alberto Garcia-Ortiz Exploiting Neural-Network Statistics for Low-Power DNN Inference IEEE Open Journal of Circuits and Systems Artificial intelligence edge-AI inference low-power coding low-power digital design neural networks |
| title | Exploiting Neural-Network Statistics for Low-Power DNN Inference |
| title_full | Exploiting Neural-Network Statistics for Low-Power DNN Inference |
| title_fullStr | Exploiting Neural-Network Statistics for Low-Power DNN Inference |
| title_full_unstemmed | Exploiting Neural-Network Statistics for Low-Power DNN Inference |
| title_short | Exploiting Neural-Network Statistics for Low-Power DNN Inference |
| title_sort | exploiting neural network statistics for low power dnn inference |
| topic | Artificial intelligence edge-AI inference low-power coding low-power digital design neural networks |
| url | https://ieeexplore.ieee.org/document/10498075/ |
| work_keys_str_mv | AT lennartbamberg exploitingneuralnetworkstatisticsforlowpowerdnninference AT ardalannajafi exploitingneuralnetworkstatisticsforlowpowerdnninference AT albertogarciaortiz exploitingneuralnetworkstatisticsforlowpowerdnninference |