Browse by author
Lookup NU author(s): Adrian WheeldonORCiD, Professor Rishad Shafik, Tousif Rahman, Jie lei, Professor Alex Yakovlev
Energy efficiency continues to be the core design challenge for artificial intelligence hardware (AI) designers. In this paper, we propose a new AI hardware architecture targeting IoT applications. The architecture is founded on the principle of learning automata, defined using propositional logic. The logic-based underpinning enables low energy footprints as well as high learning accuracy during training and inference, which are crucial requirements for efficient AI with long operating life. We present the first insights into this new architecture in the form of a custom-designed integrated circuit (IC) for pervasive applications. Fundamental to this circuit is systematic encoding of binarized input data fed into maximally parallel logic blocks. The allocation of these blocks is optimized through a design exploration and automation flow using FPGA-based fast prototypes and software simulations. The design flow allows for expedited hyperparameter search for meeting the conflicting requirements of energy frugality and high accuracy. Extensive validations on the hardware implementation of the new architecture using single and multi-class ML datasets show potential for significantly lower energy compared with the existing AI hardware architectures. In addition, we demonstrate test accuracy and robustness matching the software implementation, outperforming other state-of-the-art ML algorithms.
Author(s): Wheeldon A, Shafik R, Rahman T, Lei J, Yakovlev A, Granmo OC
Publication type: Article
Publication status: Published
Journal: Philosophical Transactions of the Royal Society A
Year: 2020
Volume: 378
Issue: 2182
Print publication date: 16/10/2020
Online publication date: 14/09/2020
Acceptance date: 03/06/2020
Date deposited: 13/07/2020
ISSN (print): 1471-2962
ISSN (electronic): 1364-503X
Publisher: The Royal Society
URL: https://doi.org/10.1098/rsta.2019.0593
DOI: 10.1098/rsta.2019.0593
Altmetrics provided by Altmetric
