Toward All-Digital Time-Domain Neural Network Accelerators for In-Sensor Processing Applications

Ahmed M. Mohey*, Marko Kosunen, Jussi Ryynänen, Martin Andraud

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

55 Downloads (Pure)

Abstract

Deep Neural Network (DNN) accelerators are increasingly integrated into sensing applications, such as wearables and sensor networks, to provide advanced in-sensor processing capabilities. Given wearables’ strict size and power requirements, minimizing the area and energy consumption of DNN accelerators is a critical concern. In that regard, computing DNN models in the time domain is a promising architecture, taking advantage of both technology scaling friendliness and efficiency. Yet, time-domain accelerators are typically not fully digital, limiting the full benefits of time-domain computation. In this work, we propose a time-domain multiply and accumulate (MAC) circuitry enabling an all-digital with a small size and low energy consumption to target in-sensor processing. The proposed MAC circuitry features a simple and efficient architecture without dependencies on analog non-idealities such as leakage and charge errors. It is implemented in 22nm FD-SOI technology, occupying 35 μm×35 μm while supporting multi-bit inputs (8-bit) and weights (4-bit). The power dissipation is 46.61 μW at 500MHz, and 20.58 μW at 200MHz. Combining 32 MAC units achieves an average power efficiency, area efficiency and normalized efficiency of 0.45 TOPS/W and 75 GOPS/mm2, and 14.4 1b-TOPS/W.
Original languageEnglish
Title of host publication2023 IEEE Nordic Circuits and Systems Conference, NorCAS 2023 - Proceedings
EditorsJari Nurmi, Peeter Ellervee, Peter Koch, Farshad Moradi, Ming Shen
PublisherIEEE
Pages1-6
Number of pages6
ISBN (Electronic)979-8-3503-3757-0
ISBN (Print)979-8-3503-3758-7
DOIs
Publication statusPublished - 1 Nov 2023
MoE publication typeA4 Conference publication
EventIEEE Nordic Circuits and Systems Conference - Aalborg, Denmark, Aalborg, Denmark
Duration: 31 Oct 20231 Nov 2023

Conference

ConferenceIEEE Nordic Circuits and Systems Conference
Abbreviated titleNorCAS
Country/TerritoryDenmark
CityAalborg
Period31/10/202301/11/2023

Keywords

  • Edge computing
  • Human activity recognition
  • Inertial measurement unit
  • In-sensor processing
  • Multiply-and-accumulate
  • Neural network accelerator
  • Smart sensor interface
  • Time-domain signal processing

Fingerprint

Dive into the research topics of 'Toward All-Digital Time-Domain Neural Network Accelerators for In-Sensor Processing Applications'. Together they form a unique fingerprint.

Cite this