Machine learning internship: explain neural network models by visualization

Introduction

In-Device Metrology (IDM) aims at inspecting the quality of semiconductor patterns printed in production wafers. This is achieved by using data-driven mathematical models that relate physical parameters to optical signals. The quality of such models is critical to ensure accuracy of the measured parameters. Understanding how optical signals encode physical information enables a better understanding and can help further improve model performance.

Your Assignment

Based on measurement data, a model is trained to map high-dimensional optical response to physical parameters of a semiconductor stack. When the mapping is linear, it is easy to visualize and show which part of the signal has a stronger contribution to physical parameters. However, with complex models such as neural networks, it is not a trivial task to see which parts of the signal contribute more to a certain physical parameter. The goal of this internship is to build a software tool to visualize and explain neural network models.

Tasks

• Investigate best practices for explainable neural networks such as Saliency maps and tailor their implementation for IDM data.
• Given two trained neural networks, trained with data measured on different sized targets on the wafer, analyze their most contributing features. Find out if overfitting is a problem for certain sized targets.
• Given one trained CNN model, apply the same analysis as above and report which features are important and how convolutional kernels are trained to extract them.

Your Profile

• You are a (HBO) bachelor student or a master student in Computer Science, Applied Mathematics, Applied Physics or a related field looking for an apprenticeship or a graduation internship.
• You have skills in machine learning and are familiar with machine learning tools.
• You have affinity with physics, metrology and mathematical understanding.
• You have a can-do mentality and the ability to work independently and collaboratively in a team environment.
• You have effective communication skills and are fluent in English both verbally and in writing.

This is a (HBO) bachelor or master apprentice or graduation internship for a minimum duration of 3 months, for 4 to 5 days per week (at least 3 days onsite for full-time). The start date of this internship is flexible.

Other Requirements

• You are enrolled at an educational institute for the entire duration of the internship.
• You are located in the Netherlands to perform your internship. If you are currently living/studying outside of the Netherlands, your CV/motivation letter should include the willingness to relocate.
• If you are a non-EU citizen studying in the Netherlands, your university must be willing to sign the documents relevant for doing an internship (i.e., Nuffic agreement).
• This position requires access to controlled technology, as defined in the United States Export Administration Regulations (15 C.F.R. § 730, et seq.). Qualified candidates must be legally authorized to access such controlled technology prior to beginning work. Business demands may require ASML to proceed with candidates who are immediately eligible to access controlled technology.

About ASML

ASML is a leading supplier of lithography equipment, used by the world’s top chipmakers to print microchips that are increasingly powerful, fast and energy efficient. Working at ASML is inspiring, no matter what field you're in. We push the boundaries of technology: if it’s moving the world forward, chances are, we’re behind it. We’re a global team of more than 32,000 people from 122 different nationalities and counting. Headquartered in Europe’s top tech hub, the Brainport Eindhoven region in the Netherlands, our operations are spread across Europe, Asia and the US.