The growing integration of AI services demands significant improvements in the energy efficiency and performance of AI accelerators. To achieve these goals, the AI architectural simulation frameworks play a critical role in enabling the design and optimization of next-generation accelerators.
A state-of-the-art simulation framework must accurately predict architectural behavior and memory subsystem interactions while providing reliable performance estimates. These capabilities are essential for guiding design decisions and ensuring that accelerators meet stringent performance targets. However, existing simulators exhibit limited validation against diverse real hardware platforms under the emerging AI workloads. This lack of correlation introduces uncertainty and limits confidence in simulation results, making design space exploration for future accelerators both complex and time-consuming.
Closing this gap requires a systematic approach to benchmarking, validation, and enhancement of simulation tools. By improving accuracy and fidelity, the simulation ecosystem can become a powerful enabler for innovation in AI hardware design.
As part of this internship, you will:
Ideal candidate profile:
Master's degree: Master of Science, Master of Engineering Technology, Master of Engineering Science
Required educational background: Computer Science, Electrotechnics/Electrical Engineering
Duration: 12 months (NNFC internship only)
For more information or application, please contact the supervising scientists Tommaso Marinelli (< e-mail verwijderd om veiligheidsredenen >), Lev Mukhanov (< e-mail verwijderd om veiligheidsredenen >) and Konstantinos Tovletoglou (< e-mail verwijderd om veiligheidsredenen >).
A state-of-the-art simulation framework must accurately predict architectural behavior and memory subsystem interactions while providing reliable performance estimates. These capabilities are essential for guiding design decisions and ensuring that accelerators meet stringent performance targets. However, existing simulators exhibit limited validation against diverse real hardware platforms under the emerging AI workloads. This lack of correlation introduces uncertainty and limits confidence in simulation results, making design space exploration for future accelerators both complex and time-consuming.
Closing this gap requires a systematic approach to benchmarking, validation, and enhancement of simulation tools. By improving accuracy and fidelity, the simulation ecosystem can become a powerful enabler for innovation in AI hardware design.
As part of this internship, you will:
- Correlate the performance of the target AI accelerator simulator with real platforms when running existing AI workloads.
- Implement and integrate emerging AI inference and training workloads into the simulator.
- Investigate inaccuracies in the simulator at both the architecture and memory subsystem levels.
- Summarize the identified inaccuracies across various AI workloads.
- Propose simulator design improvements to increase the accuracy of the performance estimates.
- Design and develop an automated design exploration engine for the simulator.
Ideal candidate profile:
- MSc student in Computer Science, Electrical Engineering, or a related program.
- Strong understanding of computer hardware architecture, memory systems, and AI accelerators.
- Familiarity with neural network architectures, especially LLMs, and knowledge of emerging paradigms such as multi-agent systems.
- Proficiency in C/C++ and Python programming languages.
- Experience with performance analysis tools or simulation frameworks is a plus.
- Self-starter with the ability to work independently and think critically.
- Available for a 1-year internship and eligible to work in Belgium.
- Strong written and verbal English communication skills.
Master's degree: Master of Science, Master of Engineering Technology, Master of Engineering Science
Required educational background: Computer Science, Electrotechnics/Electrical Engineering
Duration: 12 months (NNFC internship only)
For more information or application, please contact the supervising scientists Tommaso Marinelli (< e-mail verwijderd om veiligheidsredenen >), Lev Mukhanov (< e-mail verwijderd om veiligheidsredenen >) and Konstantinos Tovletoglou (< e-mail verwijderd om veiligheidsredenen >).
