JuliaCon 2025

ASML is the leading edge company production photo-lithography machines used to produce the most advanced chips in the world.

Currently our machines produce around 300 wafer per hour, each wafer contains somewhere between 100 and 10,000 chip dies that are exposed with a scanning technique. This is achieved by exposing groups of dies at a time somewhere between 100 and 300 exposures on a wafer.

This roughly means that a wafer is exposed every 12 seconds and an exposure completed in at most 40 ms.

In order to produce sub-nanometer precision wafers, we need to run complex correction models for each exposure. Besides that those models run in parallel they are still very constrained in time and have to complete in <30 ms per exposure.

To keep up with the pace of technology improvements we need to increase complexity of the algorithms and quickly adapt to the demanding requirements of semi-conductor manufacturers.

As a pilot we are using Julia as a language for a few of such algorithms.

That poses few challenges:
* how to reach the desired performance (fast and consistent)?
* how to deploy algorithms compiled ahead of time?
* how to connect the algorithms with other algorithms and facilities of the machine?

We employ the new ahead of time compiler juliac (presented by Gabriel and Jeff on JuliaCon 2024) which will be released in Julia 1.12.
However, using juliac poses its own challenges. This talk will be a summary of several issues that we encountered and how we solved them or still struggling with.