Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 40 of the 40 references it cites were found in a registry or an index; 28 of the 56 sources its claims cite were checked for support.

TinyML: Enabling of Inference Deep Learning Models on Ultra-Low-Power IoT Edge Devices for AI Applications

Open access
Alajlan, Norah N., Ibrahim, Dina M. · 2022 · Micromachines

Abstract

Recently, the Internet of Things (IoT) has gained a lot of attention, since IoT devices are placed in various fields. Many of these devices are based on machine learning (ML) models, which render them intelligent and able to make decisions. IoT devices typically have limited resources, which restricts the execution of complex ML models such as deep learning (DL) on them. In addition, connecting IoT devices to the cloud to transfer raw data and perform processing causes delayed system responses, exposes private data and increases communication costs. Therefore, to tackle these issues, there is a new technology called Tiny Machine Learning (TinyML), that has paved the way to meet the challenges of IoT devices. This technology allows processing of the data locally on the device without the need to send it to the cloud. In addition, TinyML permits the inference of ML models, concerning DL models on the device as a Microcontroller that has limited resources. The aim of this paper is to provide an overview of the revolution of TinyML and a review of tinyML studies, wherein the main contribution is to provide an analysis of the type of ML models used in tinyML studies; it also presents the details of datasets and the types and characteristics of the devices with an aim to clarify the state of the art and envision development requirements.

Overview

CoChat Summary

TL;DR

This review surveys TinyML, the deployment of compressed deep-learning inference on ultra-low-power IoT edge devices such as microcontrollers, and analyzes models, datasets, devices, frameworks, and limitations across 38 selected papers. It finds that TinyML offers energy, cost, latency, and privacy benefits, but remains constrained by device heterogeneity, limited processing power and memory, and a shortage of suitable DL models.

Key findings

  • Across the reviewed TinyML studies, image datasets were the most common data type at 46%, followed by physiological/behavioral metrics at 29% and audio at 11%; the most frequently used image datasets were ImageNET and Visual Wake Words, while Google Speech Commands was the most used audio dataset.
  • The most common models achieving best results were ANN at 21%, CNN at 12%, and MobileNet-V1 at 6%; non-neural ML models included Decision Tree at 9%, SVM at 6%, and KNN at 6%.
  • Device usage was dominated by STM32 microcontrollers: STM32-F746ZG appeared in 22% of studies and STM32-H743ZI2 in 20%; reviewed device costs ranged approximately from USD 16.00 to USD 28.00, and MCUs were described as typically small (~1 cm³), low power (~1 mW), and cheap (~$1).
  • Reported use-case results included ASL detection at 98.84% accuracy and 20 FPS after deployment, handwriting recognition at 100% prediction accuracy with 135.68 KB RAM, 668.97 KB flash, and 330 ms computation, medical face mask detection at 99.83% accuracy and 30 FPS with a 138 KB model, and MCUNet reaching 70.0% accuracy on STM32H743 with 0.49 MB SRAM and 0.9 MB flash.

Methods

  • This is a systematic literature review following PRISMA guidelines, organized around research questions on datasets, models, devices, application domains, frameworks/libraries, and limitations in TinyML.
  • Searches used Google Scholar, Clarivate’s Web of Science, Science Direct, IEEE, MDPI, and Springer; the first TinyML publication was dated 13 December 2019, and a total of 38 relevant published papers/articles were retrieved and screened using inclusion/exclusion criteria.
  • Data extraction covered each study’s main contribution, implementation details including datasets, models, devices, and evaluation criteria; findings were synthesized descriptively into dataset, model, and device categories.

Limitations

  • The paper states that TinyML faces device heterogeneity, with no standardized framework across manufacturers; lower processing power and speed than cloud-based systems; limited memory of only a few KB; and a lack of DL models that achieve high accuracy with small model size and low latency.
  • Evident from the text: as a review, the findings are descriptive counts and summaries from 38 selected studies rather than new experimental validation, and the exclusion criteria omitted studies on compression techniques such as quantization and pruning, memory reduction, TinyML services, technical reports, and website-published experiments.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/mi13060851

Cites

40 of 40 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 A framework of IoT applications with Cloud computing, Edge computing and TinyML.
Figure 1 A framework of IoT applications with Cloud computing, Edge computing and TinyML.CC BY · PubMed Central
Figure 2 Comparison between the main characteristics of devices powered either by microcontroller or microprocessor.
Figure 2 Comparison between the main characteristics of devices powered either by microcontroller or microprocessor.CC BY · PubMed Central
Figure 3 PRISMA flowchart for the study.
Figure 3 PRISMA flowchart for the study.CC BY · PubMed Central
Figure 4 Dataset type distribution in previous studies.
Figure 4 Dataset type distribution in previous studies.CC BY · PubMed Central
Figure 5 Machine and deep-learning models distributed in previous studies.
Figure 5 Machine and deep-learning models distributed in previous studies.CC BY · PubMed Central
Figure 6 TinyML devices distribution in previous studies.
Figure 6 TinyML devices distribution in previous studies.CC BY · PubMed Central

Verification

Claims the paper makes

23 Valid · 9 Not valid · 19 Not checked · 35 Own results and interpretations

18 more claims · 1 with their cited sources checked

Create a free account to see all 86 claims and to verify the 8 references still open.

See all claims — free account
  • Probably not a substantial claim. 45
  • It may state two separate claims as one. 39

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

3 of 3

Citations

Cites

Show all 37 references

Research works printed but not linked: 6

  1. Banbury, C.R., Reddi, V.J., Lam, M., Fu, W., Fazel, A., Holleman, J., Huang, X., Hurtado, R., Kanter, D., and Lokhmotov, A. (2020). Benchmarking TinyML Systems: Challenges and Direction. arXiv.
  2. Venzke, M., Klisch, D., Kubik, P., Ali, A., Missier, J.D., and Turau, V. (2020). Artificial Neural Networks for Sensor Data Classification on Small Embedded Systems. arXiv.
  3. Wong, A., Famouri, M., Pavlova, M., and Surana, S. (2020). TinySpeech: Attention Condensers for Deep Speech Recognition Neural Networks on Edge Devices. arXiv.
  4. Cai, H., Gan, C., Zhu, L., and Han, S. (2020). TinyTL: Reduce Memory, Not Parameters for Efficient On-Device Learning. arXiv.
  5. Vu, T.H., Olsson, C., Laptev, I., Oliva, A., and Sivic, J. (2014). Food-101–Mining Discriminative Components with Random Forests, Springer International Publishing. LNCS 8693.
  6. Chowdhery, A., Warden, P., Shlens, J., Howard, A., and Rhodes, R. (2019). Visual Wake Words Dataset. arXiv.

Cited by

See all 224

Metadata

Authors
Alajlan, N. N., & Ibrahim, D. M.
Affiliations
Qassim University (SA)
Corresponding author
Ibrahim, Dina M.
Venue
Micromachines
Publisher
MDPI AG
Published
2022-05-29
Volume
13
Issue
6
Pages
851
ISSN
2072-666X (ISSN-L)
OpenAlex
W4282025104
Cited by
224
References
37
Citations by year
2026: 67 · 2025: 71 · 2024: 56 · 2023: 26 · 2022: 4
FWCI
17.41
Type
journal article
Language
en
Open access
Locations
  • Micromachines (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
  • Physical Sciences › Computer Science › Information Systems › Internet of Things and AI
Keywords
TinyML, edge computing, Internet of Things, ultra-low power electronics, microcontroller, deep learning inference, on-device learning, resource-constrained devices, IoT datasets

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: TinyML: Enabling of Inference Deep Learning Models on Ultra-Low-Power IoT Edge Devices for AI Applications. https://app.cochat.ai/papers/10.3390/mi13060851
The paper
Alajlan, N. N., & Ibrahim, D. M. (2022). TinyML: Enabling of Inference Deep Learning Models on Ultra-Low-Power IoT Edge Devices for AI Applications. Micromachines. https://doi.org/10.3390/mi13060851

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.