Master's Thesis: Advanced AI Accelerators

Three-dimensional integration can help AI accelerators overcome memory bottlenecks, but it can also make heat management more difficult. This thesis at Huawei Sweden R&D will develop a simulation framework to assess power use and thermal behaviour in 3D-integrated computing systems using conventional and emerging memory technologies, including advanced SRAM and MRAM.

The assignment

• Build a simulation framework that connects architectural simulators such as gem5 with power estimation models and thermal tools such as HotSpot.
• Use the framework to evaluate how different memory technologies affect power consumption, performance and temperature distribution in an AI system-on-chip.
• Investigate the effects of reinforcement-based task scheduling and workload characteristics on these outcomes.

Who we are looking for

• Master’s student with a background in Computer Engineering or Electrical Engineering.
• Basic understanding of computer architecture and programming in Python/C++.
• Strong spoken and written English.
• Ability to work independently and collaborate with others, with careful attention to quality and detail.
• Experience with gem5, HotSpot or McPAT, and knowledge of memory technologies and 3D integration, are advantageous.

What we offer

• An opportunity to apply technical skills to an industry-relevant research problem.
• An innovation-focused R&D environment with international engineering collaboration.

Practical details

Planned start: spring 2027. Location: Kista.