The work involves the analysis of various hardware components of a System-on-Chip (SoC) from AMD’s Versal series in different mixed-criticality scenarios. Both the Network-on-Chip and the AI accelerator can be investigated.
Mixed-criticality systems are software applications in which individual tasks have different levels of importance and priority for the overall system. To develop models that accurately represent the behavior of the respective hardware components and enable reliable predictions, the timing behavior of individual tasks must be analyzed in detail.
- Research Project or Bachelor/Master Thesis
- Supervisor: M.Sc. Vincent Sprave
Today, most elaborate applications are embedded in a heterogeneous computing system, that is, the subtasks of the application are distributed along multiple different compute units, such as CPUs, GPUs, FPGAs, AI units and other specialized hardware. These complex heterogeneous systems can be found everywhere, ranging from high-performance computing clusters to energy-efficient edge devices such as smartphones. This research field offers a wide range of potential research topics for students interested in algorithmics, computer architecture, programming or heterogeneous system design.
- Research project or bachelor/master thesis
- Supervisor: Dr.-Ing. Martin Wilhelm
This thesis aim to exploit the existing AI Engine to accelerate near-sensor data processing in Ultra-Highy-Field-MRT Systems, allowing data reduction near sensor, and consequently power optimization, and meeting of real-time requirements, facilitating data post-processing.
The topic is suitable for Master's or motivated Bachelor's students.
- Bachelor/Master
- Supervisor: Dr.-Ing. Daniele Passaretti
This thesis explores existing accelerator architectures, such as GPUs, TPUs, FPGAs, and adaptive SoCs, for real-time UHF-MRI applications. The student will identify bottlenecks using models such as the Roofline Model, profile existing algorithms, and propose new parallel computing paths that exploit the selected hardware architecture to accelerate the assigned algorithm.
The topic is suitable for Master's or motivated Bachelor's students.
- Bachelor/Master
- Supervisor: Dr.-Ing. Daniele Passaretti
The RatatoskrM3D simulation framework, developed at our chair, enables the exploration of 3D Networks-on-Chip (NoCs) and their emerging challenges. Its focus lies on heterogeneous 3D NoCs, which introduce complex design aspects such as varying router counts and clock domains across layers. The project provides diverse research and development opportunities for students in the areas of computer architecture, hardware simulation, and system-level design. Students can contribute to extending the simulator, modeling new architectures, and participating in 3D NoC research.
- Research project or bachelor/master thesis
- Supervisor: M.Sc Max Tzschoppe
This thesis investigates RISC-V existing instruction set architectures, their different micro-architecture implementations, extending them for low-latency, data-intensive, real-time medical applications, such as the UHF-MRI application. The student identifies bottlenecks using models such as the roofline model, explore data formats customization and data processing exstensions like FFT unit, to identify a customization ISA for medical application.
The topic is suitable for Master's or motivated Bachelor's students.
- Bachelor/Master
- Supervisor: Dr.-Ing. Daniele Passaretti
Development of embedded vision algorithms. Required knowledge: C, High-Level synthesis, Image Processing, FPGA programming helpful.
The aim of this thesis is to examine existing solutions and propose new methods for Remote Direct Memory Access (RDMA) in heterogeneous computer architectures such as GPU, FPGA and TPU, targeting data-intensive applications with low latency, such as ultra-high-frequency magnetic resonance imaging.
The topic is suitable for Master's or motivated Bachelor's students.
- Bachelor/Master
- Supervisor: Dr.-Ing. Daniele Passaretti