< retour aux séminaires

Marcello TRAIOLA - Postdoctoral researcher, Lyon Institute of Technology (INL), École Centrale de Lyon, France

Theme: Approximate Computing forReliability and Testing: a JourneyTowards Next-generation Energy-efficient Safety-critical Systems
Date: December 3rd, 2020 - 10:00 - 12:00, Internal videoconference


In 2019, Marcello TRAIOLA earned a Ph.D. degree in Computer Engineering from the Montpellier University, in France.
In 2016, he earned a Master’s degree in Computer Engineering- summa cum laude -from the Italian University “Federico II” of Naples, Italy.
From October 2019 to January 2020, he was in the USA at the semiconductor company MediaTek USA Inc., as intern R&D engineer.
Since February 2020, he is at the Lyon Institute of Technology (École Centrale de Lyon), in France, as post-doctoral researcher.
He actively researches one merging computing paradigms with focus on design, test and reliability.


In the last decades, Approximate Computing (AxC) has become increasingly studied as a next-generation computing paradigm. AxC takes advantage of the intrinsic error-tolerance of some applications and systems to achieve high benefits in terms of efficiency. AxC carefully introduces some inaccuracy – that will be intrinsically tolerated – to increase the computing systems' efficiency (in terms of performance, area and power consumption).
A plethora of AxC techniques have been proposed at different levels, from software to hardware. Such techniques have always been applied to inherent error-tolerant systems, such as deep-learning-based systems, e.g. Deep Neural Networks (DNN) and Convolutional Neural Networks (CNN).
However, in the last years, these systems have been increasingly employed in safety-critical domains, as autonomous driving and health. In such scenarios, catastrophic events can occur, as environmental harm, loss or damage to equipment, serious injury to people or even death. Therefore, complying with reliability standards is crucial to avoid incurring critical consequences for people, environment, and business.
In this context, challenges and obstacles to the application of AxC techniques to safety-critical systems must be thoroughly investigated.
In this perspective, this research project proposes to comprehensively study all the aspects related to the application of AxC techniques to next-generation advanced computing systems operating in safety-critical scenarios.
In the first part of the project, studies on the approximation impact on reliability and testability requirements will be carried out. Methodologies and techniques to assess the impact of AxC on safety-critical systems will be produced. As a consequence, enhanced approximation techniques complying with reliability and testing requirements will be designed.
Integrated hardware-software architectures for safety-critical applications will be realized and implemented on development platforms for experimental purposes.
The second part of the project will be devoted to empower the widespread use of approximation techniques for safety-critical systems. Specifically, integrated design solutions will be studied to produce approximate yet reliable and easy-to-test systems. Such systems will be highly efficient and also perfectly suitable in safety-critical scenarios.
Moreover, the issue of trust will be addressed: on one side, verification techniques will be designed to prove approximate systems trustworthy; on the other side, the design of diagnostic techniques will help to deeply understand failures that can possibly happen inoperating conditions.
The ultimate objective is to pave the way to finally perceive the opportunities provided by approximation techniques and no longer consider them as a threat.