Fine-Grained Mapping of AI Training & Inference Workloads Across the Full Memory Hierarchy

Understanding this behavior modern AI training and inference workloads is crucial for designing optimized memory hierarchies and minimizing data-movement bottlenecks. In this internship, the student will develop a fine-grained workload-to-memory mapping framework that characterizes how representative AI workloads interact with the entire memory stack. This includes profiling layer-level and operator-level footprints, traffic patterns, reuse distances, and temporal/spatial locality. By integrating these insights into a system-level simulation environment, the student will evaluate bottlenecks, identify optimization opportunities, and generate actionable guidelines for future memory hierarchy design targeting large-scale AI systems.

Skills to stand out:

• Solid understanding of memory subsystem
• Familiarity with AI training and inference workload characteristics
• Strong programming skills in C++ or Python
• Experience with performance modelling techniques
• Exposure to system-level simulation tools or benchmarking frameworks is a plus (Ramulator, DRAMSys, DRAMSim, Gem5, ...)

Type of internship: Master internship, PhD internship

Duration: 6-9 months

Required educational background: Computer Science, Electrotechnics/Electrical Engineering, IT, Nanoscience & Nanotechnology

Supervising scientist(s): For further information or for application, please contact Khakim Akhunov (<e-post slettet av sikkerhetsgrunner>)

The reference code for this position is 2026-INT-045. Mention this reference code in your application.

Imec allowance will be provided for students studying at a non-Belgian university.

Applications should include the following information:

• resume
• motivation
• current study

Incomplete applications will not be considered.