BIST Design and Implementation for a Fixed-Point Arithmetic MAC Unit within a Systolic Array

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

Abstract

In embedded systems involving hardware accelerators to support embedded machine learning (ML) applications, both sequential and parallel operations are used to ensure the system can perform its required operations in the required time. For ML applications involving trained ML model inference, parallel (concurrent) hardware structures can be utilized to perform parallel multiplications usually involving matrix multiplication operations in hardware rather than software. In this paper, fault simulation, embedded instrument test access and built-in self-test (BIST) are considered in relation to the multiply-accumulate (MAC) unit that would be found within a systolic array. A fault simulation was performed on the pre-synthesis MAC unit Verilog HDL design module and used to develop a BIST unit accessible via an IEEE Std 1687 network. The design was prototyped using a Xilinx Artix-7 field programmable gate array (FPGA).

Original languageEnglish
Title of host publicationProceeding - 2023 International Electrical Engineering Congress, iEECON 2023
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages113-116
Number of pages4
ISBN (Electronic)9798350332360
DOIs
Publication statusPublished - 2023
Event2023 International Electrical Engineering Congress, iEECON 2023 - Krabi, Thailand
Duration: 8 Mar 202310 Mar 2023

Publication series

NameProceeding - 2023 International Electrical Engineering Congress, iEECON 2023

Conference

Conference2023 International Electrical Engineering Congress, iEECON 2023
Country/TerritoryThailand
CityKrabi
Period8/03/2310/03/23

Keywords

  • embedded instruments
  • fault simulation
  • FPGA
  • MAC
  • machine learning
  • test

Fingerprint

Dive into the research topics of 'BIST Design and Implementation for a Fixed-Point Arithmetic MAC Unit within a Systolic Array'. Together they form a unique fingerprint.

Cite this