IEEE Circuits and Systems Magazine - Q3 2020 - 8

Smart City

eHealth

Environmental
Monitoring

Industry and
Business

Smart Home

Vehicular

Figure 3. TinyML ecosystem.

[25]. Besides, the reduced cost of end-devices will also foster their adoption in disadvantaged areas, which may help
to revamp local economy and business activities [26].
3) Vehicular Services
Thanks to the alliance between ML and MCUs, the vehicular market and, concretely, the Cooperative-Intelligent
Transportation Systems (C-ITS) will be enriched with a segment of vehicles that has been ignored so far, namely, personal mobility devices. Urban mobility trends are changing
during the last times due to the emergence of sustainable
and healthy vehicles such as shared bikes, electric mopeds, scooters, etc. However, current vehicular networks
just consider traditional road transportation means, i.e.,
cars, buses, trucks, etc., as their equipped On-Board Units
(OBUs) require great amount of power to operate. Although
some efforts have been done for connecting motorcycles to
C-ITS infrastructures [27], personal vehicles have not been
considered due to their novelty and inherent constraints.
However, coupling these devices with MCU-based OBUs
paves the way for their integration within the C-ITS and
smart-city ecosystems. Therefore, light vehicles will have
access to basic vehicular services such as driving safety,
device status monitoring, route planning, etc.
8

4) Industry 4.0
The potential impact of TinyML is
enormous in industrial and manufacturing sectors as they are experiencing a digitalization process
under the umbrella of the Industry
4.0 revolution [28]. The processing
tasks in these fields are highly heterogeneous in terms of computing
or memory requirements, which
causes that MCUs are not always capable of running certain tasks. However, the integration of ML-based
Decision Support Systems (DSSs)
within MCUs may permit them to
determine whether to assume a certain computation task or offloading
it onto upper processing layers such
as edge or cloud [29]. Besides, the
refinement of production processes, decision making, asset tracking,
etc. will be driven by the integration
of intelligence along the production
chain as well.

5) Smart Agriculture
and Farming
These sectors are also under a
deep digital transformation [30]. By
adopting ML-based smart monitoring and control, both of
them will increment the efficiency and health of crops and
animals, which will lead to greater revenues due to the improvement of production quality. Finally, common daily activities and home appliances will also reach the next level
with the support of ML-enabled devices. For example, biometric e-locks will be generalized as well as context-aware
domotic systems and wellbeing applications. Intelligent
pet trackers and feeders, smart garden-irrigation control,
or advanced safety & security home systems, are other
use cases that will be enabled thanks to the upcoming
-TinyML revolution [31].
III. Tinyml Frameworks
Given the great potential of TinyML to revolutionize multiple sectors, a number of libraries and tools for -easing
the implementation of ML algorithms on constrained platforms are available in the market. TinyML frameworks
can be categorized into three different families. The most
common approach relates to the conversion of already
trained models for adapting them to the restrictions of
MCUs. Thus, these tools usually take inference engines
generated by well-known ML libraries such as TensorFlow
[32], Scikit-Learn [33], or PyTorch [34], among -others,

IEEE CIRCUITS AND SYSTEMS MAGAZINE

THIRD QUARTER 2020

IEEE Circuits and Systems Magazine - Q3 2020

Table of Contents for the Digital Edition of IEEE Circuits and Systems Magazine - Q3 2020