Bordersearch: Efficient Characterization of Automotive Electronic Systems Through Machine Learning

Auteurs

• Markus Dobler

Paramètres

• Informatique, programmation

En savoir plus sur le livre

The complexity of modern electrical and electronic systems is steadily increasing, particularly in the automotive industry, driven by advanced driver assistance systems and electric vehicles. To ensure safety and reliability, thorough verification of chips and devices is essential, especially with new regulations like ISO 26262. This necessitates testing chips within their application context, which complicates the investigation of potential failures due to the increased number of parameters involved. Traditional methodologies struggle with this complexity, as they require an impractical number of tests for reliable results. To address this challenge, a new methodology called Bordersearch is introduced. It analyzes systems through functional tests yielding binary responses—“pass” or “fail.” By integrating various machine learning techniques, Bordersearch efficiently handles systems with numerous parameters. It employs a robust classifier to predict the boundary between pass and fail regions and assesses the certainty of these predictions across the parameter space. An adaptive sampling strategy focuses testing near the border, maximizing information gained from each test while minimizing the number of samples needed. Additionally, Bordersearch automates the detection of relevant parameters and offers innovative visualization methods for understanding multi-dimensional interactions. This makes experimental results accessible

Achat du livre

• Année de publication: 2018