Towards Rapid Exploration of Heterogeneous TinyML Systems using Virtual Platforms and TVM's UMA

Samira Ahmadifarsani, Rafael Stahl, Philipp Van Kempen, Daniel Mueller-Gritschneder, Ulf Schlichtmann

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

Abstract

The rapid setup of deep learning compilation toolchains for heterogeneous TinyML systems with a processor and dedicated ML accelerator is still at an early stage. Here, achieving the most optimal combination of targets for a TinyML application on ultra-low-power edge devices demands additional benchmarking solutions to estimate the final performance.Apache TVM's Universal Modular Accelerator (UMA) interface as an easy-to-use API is a promising speed-up approach to this scope. In this paper, we integrate a simple custom dedicated accelerator into TVM using UMA to offload the quantized convolution operators in order to demonstrate such an approach. Furthermore, we leverage MLonMCU tool and its capability of virtual prototyping to estimate and explore the performance improvement achieved by the accelerator.

Original languageEnglish
Title of host publicationProceedings - 2023 IEEE/ACM International Workshop on Compilers, Deployment, and Tooling for Edge AI, CODAI 2023
PublisherAssociation for Computing Machinery, Inc
Pages6-10
Number of pages5
ISBN (Electronic)9798400703379
DOIs
StatePublished - 21 Sep 2023
Event2023 IEEE/ACM International Workshop on Compilers, Deployment, and Tooling for Edge AI, CODAI 2023 - Hamburg, Germany
Duration: 21 Sep 2023 → …

Publication series

NameProceedings - 2023 IEEE/ACM International Workshop on Compilers, Deployment, and Tooling for Edge AI, CODAI 2023

Conference

Conference2023 IEEE/ACM International Workshop on Compilers, Deployment, and Tooling for Edge AI, CODAI 2023
Country/TerritoryGermany
CityHamburg
Period21/09/23 → …

Keywords

  • TinyML
  • TVM
  • UMA
  • virtual prototyping

Fingerprint

Dive into the research topics of 'Towards Rapid Exploration of Heterogeneous TinyML Systems using Virtual Platforms and TVM's UMA'. Together they form a unique fingerprint.

Cite this