Browse by author
Lookup NU author(s): Komal Krishnamurthy, Omar AwfORCiD, Tian Lan, Dr Fei Xia, Professor Alex Yakovlev, Professor Rishad ShafikORCiD
Full text for this publication is not currently held within this repository. Alternative links are provided below where available.
Randomization constitutes a major part in stochastic machine learning (ML) training processes. Typically, linear feedback shift register (LFSR) is used for periodic pseudorandom bits generation. However, they are known to have high switching activity as well as complexity as the precision requirements increase. This work proposes an alternative method for randomization using variable duty cycle-based pulse generation. Key principles include generation of oscillating signal and obtaining a delayed version of the same leading to different duty cycles. Two designs for creating a variable duty cycle are proposed. The first technique utilizes a buffer-based delay element as a duty cycle regulator, whereas the second relies on a basic RC circuit. The shifted version of the same signal is generated and passed through logic gates providing varied pulse widths. The obtained signal’s pulse width is designed to allow the duty cycle to range between 20% and 90%. When compared to RC-based variation ±(8.5)%, buffer-based performs better against PVT variation with a significant small deviation ±(0.8)%. A multiplexer is used to pick the required duty cycle depending on the application’s necessity. The proposed circuit consumes low power for RC and buffer chain (34.4,58.72)μW respectively.
Author(s): Krishnamurthy K, Ghazal O, Lan T, Xia F, Yakovlev A, Shafik R
Publication type: Conference Proceedings (inc. Abstract)
Publication status: Published
Conference Name: 21st IEEE Interregional NEWCAS Conference (NEWCAS 2023)
Year of Conference: 2023
Pages: 1-5
Print publication date: 07/08/2023
Online publication date: 07/08/2023
Acceptance date: 12/04/2023
ISSN: 2474-9672
Publisher: IEEE
URL: https://doi.org/10.1109/NEWCAS57931.2023.10198108
DOI: 10.1109/NEWCAS57931.2023.10198108
Library holdings: Search Newcastle University Library for this item
ISBN: 9798350300246