| Supervisor | CVLab, Mathieu Salzmann and Andrew Price |
| Type of Project | Master project, 1 student |
| Duration | 1 Semester |
Recommended
This project is suitable for a student interested in one or more of the following topics: the space sector, hands on experiments, machine learning, electromagnetic radiation, and/or edge deployment.
Context
CVLab and the Space Centre are currently engaged in research supporting deploying machine learning algorithms on edge devices in-orbit. This includes power constraints, bandwidth constraints, explainable AI, and improving fault-tolerance of machine learning algorithms.
With funding secured from ESA and the Swiss Space Office, CVLab and the Space Centre will perform some radiation experiments at the PSI Proton Irradiation Facility. A machine learning algorithm will be deployed to a Nvidia Jetson Orin NX. A proton beam will irradiate the jetson to cause Single Event Events (SEEs). When exposed in such a manner, an SEE can cause bit flips in the machine learning algorithm resulting in reduced performance or outright failure.
Project scope
The student will support the proton irradiation experiments and validation against simulated results. The student will prepare practice runs of the radiation testing and then perform the radiation testing at the Paul Scherrer Institute. The student will log and analyze the results.
Expected outcomes
- Completion of proton radiation experiments on Nvidia Jetson Orin NX
- Validation against simulation results (provided by Andrew)
- Thesis / report
List of tasks and tentative planning
- Familiarize self with literature; effect of radiation on hardware.
- Familiarize self with hardware.
- Perform radiation experiment test runs.
- Perform radiation experiment at the Paul Scherrer Institute
- Analyze the fault propagation through the devices.
- Comparison of experimental results to simulated results.
- Report writing
Requirements
- Excellent python programming; C is an asset
- Experience with edge devices or deployment is an asset
- Independent methodical worker, ability to troubleshoot
- Good communication skills in English
Type of Work
20% Theory, 50% Experiments, 20% Analysis, 10% Documentation
Contact
Andrew Price, [email protected]
References
-
Sources of Single Event Effects in the NVIDIA Xavier SoC Family under Proton Irradiation https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9897236
-
Proton Evaluation of Single Event Effects in the NVIDIA GPU Orin SoM: Understanding Radiation Vulnerabilities Beyond the SoC https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=10616076
-
Using ensemble learning to improve radiation tolerance of CNNs in space applications https://zenodo.org/records/13885495
