IEEE Circuits and Systems Magazine - Q3 2020 - 5

Abstract
The TinyML paradigm proposes to integrate Machine Learning
(ML)-based mechanisms within small objects powered by Microcontroller Units (MCUs). This paves the way for the development of novel applications and services that do not need the
omnipresent processing support from the cloud, which is power
consuming and involves data security and privacy risks. In this
work, a comprehensive review of the novel TinyML ecosystem is provided. The related challenges and opportunities are
identified and the potential services that will be enabled by the
development of truly smart frugal objects are discussed. As a
main contribution of this paper, a detailed survey of the available
-TinyML frameworks for integrating ML algorithms within MCUs
is provided. Besides, aiming at illustrating the given discussion,
a real case study is presented. Concretely, we propose a multiRadio Access Network (RAT) architecture for smart frugal objects. The issue of selecting the most adequate communication
interface for sending sporadic messages considering both the
status of the device and the characteristics of the data to be
sent is addressed. To this end, several TinyML frameworks are
evaluated and the performances of a number of ML algorithms
embedded in an Arduino Uno board are analyzed. The attained
results reveal the validity of the TinyML approach, which successfully enables the integration of techniques such as Neural
Networks (NNs), Support Vector Machine (SVM), decision trees,
or Random Forest (RF) in frugal objects with constrained hardware resources. The outcomes also show promising results in
terms of algorithm's accuracy and computation performance.

I. Introduction
achine Learning (ML) has revolutionized our
way of understanding the world by enabling the
inference of valuable information and knowledge
from huge amount of data, so far invisible to human eye.
Data are produced by our digital activity and grow at almost unlimited rates [1], hence calling for large pools of
resources in terms of processing power and storage for
their handling and analysis. In this regard, research and
innovation advances during the last years have been
notable in multiple areas of study [2]-[5]. These efforts
have focused on obtaining a better response from the
wide range of available ML techniques, e.g., Deep Learning (DL), Neural Networks (NNs), Reinforcement Learning
(RL), clustering, etc., by exploiting the potential offered
by state-of-the-art processors, data-centers, and supercomputers [6]. However, a non-negligible segment of
processing entities has received scarce attention due to
their inherent computing constraints: Microcontroller
Units (MCUs). This tiny processing entities are usually
embedded in daily-use appliances such as vehicles, medical devices, personal gadgets, etc., but they may also constitute the core processing unit of small frugal objects [7].

Digital Object Identifier 10.1109/MCAS.2020.3005467
Date of current version: 12 August 2020

Frugal objects are independent "things" specialized
in perform concrete operations that can be programmed
with software rather than requiring specific electronics for each task, so they can be massively produced at
inexpensive cost. These devices incorporate enough
processing and memory resources, sensors, and connectivity capabilities to perform the required operation,
while being highly energy efficient. A concise definition
of types of MCU-based devices according to their hardware resources is given in [8].
The penetration of frugal objects in our lives due to
the emergence of the Internet of Things (IoT) is skyrocketing, thus provisioning them with cognitive skills opens
a plethora of opportunities for developing a widespread
web of collective intelligence (Fig. 1) [9]. Different paradigms such as fog or edge computing are fueling this
hyper-connected distributed scheme supported by the
new wave of communication technologies, which permit
coordinating end-devices efforts to place intelligence
close to the users [10]. Nevertheless, given the power
constraints of MCUs, reducing data transmissions is crucial for their sustainability as this type of operations are
more power demanding than computational tasks [11].
In this line, much data captured by sensors are
wasted because of transporting costs, bandwidth limitations, or power constraints. For those reasons, performing on-device ML processing is a buzz trend today. The
TinyML community [12] is boosting the integration of ML
within MCU-based smart frugal objects. In fact, forecasts
predict that the global edge computing market will reach
1.12 trillion dollars by 2023 [13] and some strong companies such as Ericsson are already offering TinyML-as-aService solutions [14]. In this -paper, we provide a comprehensive overview of current challenges and opportunities

Figure 1. Interconnected smart services.

Ramon Sanchez-Iborra and Antonio F. Skarmeta are with the University of Murcia, Murcia, Spain, (ramonsanchez@um.es; skarmeta@um.es).
THIRD QUARTER 2020

IEEE CIRCUITS AND SYSTEMS MAGAZINE

IEEE Circuits and Systems Magazine - Q3 2020

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