Intern (not graduate)
Assignment:
Master internship: Distribution classification by using supervised and unsupervised ML
Background
Checking whether an IC (Integrated Circuit) complies with its specifications after the manufacturing process is an essential task to guarantee the quality of devices shipped to the customer. As a result, at the end of the production line, we perform electrical measurements that covers the area of the chip. These measurements all require limits, to be able to judge the chip’s performance. With roughly 30,000 electrical measurements, this requires the same number of limits. These need to be set and adjusted manually, an automated process is not possible here, as not all measurements show a normal distribution.
Objective
We aim to use both supervised and unsupervised ML techniques on the measurement data, to enable the automation of limit tuning. By using robust techniques (not outlier sensitive), statistical info can be obtained per distribution, and they can be classified based on this info.
This Master internship explores the topic of distribution classification in an industrial testing environment and its limitations. The focus will be on exploring ML techniques which can identify distributions, without being vulnerable to the presence of outliers. Production and qualification data will be available for the interested student to perform the following tasks:
- Explore the different robust ML techniques
- Explore the usage of supervised and unsupervised ML techniques, and the best fit for this use-case
- Apply a combination of ML techniques on electrical measurements
Distribution classification Description
Distribution classification is the process of identifying the distribution that best fits a dataset, whether it's discrete or continuous. This classification is essential because distributions have different shapes, and behaviors, which influence how data should be interpreted.
Understanding the data’s distribution helps in applying correct statistical methods and set meaningful limits. For example, if your data follows a normal distribution, you can use standard deviation-based limits (e.g., ±3σ) to detect outliers or monitor process stability. If the data is skewed or has a tail, using normal-based limits could lead to incorrect conclusions.
NXP Introduction
NXP Semiconductors enables a smarter, safer, and more sustainable world through innovation. As the world leader in secure connectivity solutions for embedded applications, NXP is pushing boundaries in the automotive, industrial & IoT, mobile, and communication infrastructure markets.
For more information, please visit our website https://www.nxp.com
More information about NXP in the Netherlands...
#LI-0d06
