Research Internships at HU BerlinOpen OpportunitiesA behavioral fish guidance structure employing a novel approach, based on turbulent eddies, was developed and tested in an ethohydraulic flume at VAW, ETH Zurich, for multiple fish species. Real-world applications of this guidance structure require understanding various operational challenges. This thesis aims to investigate flow field and operational aspects such as clogging probability by driftwood and organic fine materials of this novel guidance structure for different configurations using physical model experiments. - Water and Sanitary Engineering
- Master Thesis, Semester Project
| With over 14 million stroke cases annually, the global neurorehabilitation market presents a multi-billion-dollar opportunity for innovative solutions addressing motor recovery. The Rehabilitation Engineering Laboratory (RELab) at ETH Zurich is developing a revolutionary closed-loop neurorehabilitation device that leverages motion tracking and non-invasive brain stimulation to transform stroke rehabilitation. This project aims to develop a sophisticated financial model and a strategic business plan to propel the device to market leadership. The student will conduct market analysis, build financial projections, and craft a compelling business strategy, focusing on pricing, reimbursement, and investor engagement. By delivering investor-ready materials and a scalable commercialization plan, this work will position the device for rapid market entry and long-term success, offering the student a unique opportunity to blend business strategy, entrepreneurship, and healthcare innovation. - Finance Economics, Neurosciences, Rehabilitation Engineering, Small Business Management
- Bachelor Thesis, Internship, Master Thesis, Semester Project
| Bis 2050 soll die Wasserkraft in der Schweiz um rund 12 Prozent bezüglich der Jahreserzeugung ausgebaut werden. Die Ermittlung von geeigneten Gewässerstrecken unter Berücksichtigung des Gewässer- und Landschaftsschutzes ist eine Grundlage für diese Zielerreichung. Im Rahmen dieser Arbeit soll mittels einer GIS-Analyse das Wasserkraftpotenzial in ausgewählten Schweizer Bergkantonen bestimmt werden. - Water and Sanitary Engineering
- Master Thesis, Semester Project
| Article 6 of the Paris Agreement encourages international cooperation and allows high-income, high-polluting countries to meet their carbon reduction commitments affordably by funding carbon reducing activities in lower-income countries, claiming the carbon reductions for themselves1. Specifically, Switzerland, under its CO₂ Act and through the The Swiss Foundation for Climate Protection and Carbon Offset (KliK Foundation) aims to offset about 40 million tonnes of CO₂ by 2030—10% of its national emissions–with over half of these offsets occurring abroad.
One upcoming collaboration is with Malawi, which involves distributing 10,000 household biogas digesters to dairy farmers2. This project is expected to mitigate approximately 436,000 tons of carbon dioxide equivalent (CO2e) annually. The primary function of these digesters is to convert organic wastes, predominantly animal dung, into a methane-rich gas. Biogas can be used as a cooking fuel, replacing wood, which is often sourced from local forests, thus reducing carbon emissions from deforestation as well as from the burning of biomass.
The model being installed is the Sistema.bio biogas digester, a plastic bag digester with other components such as valves, pipes and stoves for cooking. These biogas parts are sourced globally—though the exact origins are uncertain—and then assembled in another country. They are shipped to Malawi as ready-to-assemble kits. Biogas has been used as a carbon offset technology for years, but the carbon footprint of the digesters are overlooked in carbon offset calculations. Sistema.bio, as well as other manufacturers, are implementing biogas projects in many countries worldwide using carbon financing, making it crucial to understand the carbon footprint of the installations to accurately estimate their carbon offset potential. Several life cycle assessment (LCA) studies of similar biogas digesters suggest that biogas plants need to operate for up to two years to offset their construction emissions3. This has significant implications for global carbon offsetting and trading and would mean that projects might not be offsetting as much carbon as calculated, and buyer countries, such as Switzerland, should do more to reduce their carbon footprints. It is currently unknown what the carbon footprint of Sistema.bio digesters are as well as other competitors with similar business models and how the carbon footprint compares to more traditional cement and brick digesters that can be built in the country of implementation. It is thus also unknown how the digester footprint might affect carbon reduction estimations of the projects. - Environmental Engineering, Mechanical and Industrial Engineering
- ETH for Development (ETH4D) (ETHZ), Master Thesis
| Three-dimensional (3D) reconstruction of dynamic scenes, particularly those involving humans interacting with their environment, remains a challenging problem in computer vision and graphics. Traditional volumetric and mesh-based methods can struggle with occlusions, complex geometries, and real-time performance. Recent advances in neural rendering—especially Gaussian splatting—offer promising alternatives by representing scenes as clouds of oriented 3D Gaussians (“splats”) that can be rendered extremely efficiently. The goal of this project is to develop a novel pipeline that leverages Gaussian splatting to reconstruct, analyze, and interpret human–scene interactions in 3D. - Computer Vision
- Master Thesis, Semester Project
| We will build a modular instrumentation stack that lets researchers amplify picoamp-level currents from next-generation ferroelectric synapses and interconnect multiple dies on a state-of-the-art parallel, FPGA-controlled setup.
Track 1 delivers low-leakage current-to-voltage converters with <10 pA input bias, bandwidth >1 MHz, and on-board digitisation.
Track 2 realizes a high-density daughter board linking two ferroelectric chips through impedance-controlled traces and an ArC 2 mezzanine connector, forming a plug-and-play 2.5-D IMC sandbox.
Together, the hardware will enable scalable, energy-aware evaluations of online learning algorithms on real devices. - Communications Technologies, Electrical and Electronic Engineering, Engineering/Technology Instrumentation, Nanotechnology, Processor Architectures
- Bachelor Thesis, ETH Zurich (ETHZ), Master Thesis, Semester Project
| Noisy signals, scarce labels, and black-box models hinder EEG-based BCIs for imagined speech and limb movement. We will tackle these issues with hybrid convolution-transformer networks—CTNet (github.com/snailpt/CTNet) and MSCFormer (github.com/snailpt/MSCFormer)—augmented by transfer learning and few-shot adaptation. Attention heat-maps and SHAP explanations (github.com/slundberg/shap) will expose which channels and time windows drive each decision. The work aligns with ViTFOX’s goal of low-power, explainable neuro-AI (vitfox.eu). - Engineering and Technology, Neural Networks, Genetic Alogrithms and Fuzzy Logic, Pattern Recognition, Signal Processing, Speech Recognition
- Collaboration, ETH Zurich (ETHZ), Master Thesis, Semester Project
| Silent-speech interfaces decode spoken content from muscle activity when no sound is produced. Surface electromyography (sEMG) offers a non-intrusive signal source but suffers from low SNR and large user variability. We propose a transformer-based pipeline that classifies silently mouthed words from multi-channel sEMG. We propose a rigorously benchmarked, transformer-based decoding pipeline that treats multi-channel sEMG as a structured time series and learns rich, long-range dependencies inaccessible to conventional CNN- or RNN-based systems. - Behavioural and Cognitive Sciences, Engineering and Technology, Intelligent Robotics, Neural Networks, Genetic Alogrithms and Fuzzy Logic, Pattern Recognition, Signal Processing
- ETH Zurich (ETHZ), Master Thesis, Semester Project
| Emerging ferroelectric resistive / capacitive devices can store analog synaptic weights and update them in situ, eliminating the memory-processor traffic that dominates AI energy budgets. We will build the first hardware-in-the-loop prototype that learns online with such devices, combining a mixed-precision supervised rule (Tiki-Taka) and a bio-inspired Hebbian rule. The demonstrator will show that continual learning on ferroelectric IMC arrays achieves software-class accuracy while slashing update energy, opening a path toward self-adapting edge hardware. - Artificial Intelligence and Signal and Image Processing, Electrical and Electronic Engineering, Memory Structures, Nanotechnology
- Bachelor Thesis, Collaboration, ETH Zurich (ETHZ), Master Thesis, Semester Project
| Neuromorphic computing calls for innovative memory elements that mimic brain-like efficiency. Ferroelectric capacitors (FeCAPs) based on hafnium zirconium oxide (HZO) have emerged as promising candidates due to their nonvolatile polarization states, high endurance, and low-power switching. However, to harness HZO FeCAPs in neuromorphic circuits, a robust device-to-circuit modeling framework is needed. This project addresses the gap by bridging detailed device physics and circuit-level simulation for HZO-based FeCAPs.
We propose to develop a comprehensive modeling flow that spans from TCAD device simulation to circuit integration. First, a physics-based TCAD model of the HZO FeCAP will be built and calibrated against experimental electrical data using Python-driven optimization, ensuring the model accurately captures ferroelectric behavior. Next, the calibrated device model will be distilled into a SPICE-compatible compact model (implemented in Verilog-A) for efficient circuit simulations. This compact model will enable designers to simulate neuromorphic architectures incorporating FeCAPs and assess performance impacts. Additionally, the influence of device geometry (such as ferroelectric layer thickness and capacitor area) on FeCAP behavior will be investigated to guide design optimizations. By leveraging TCAD, Verilog-A, and SPICE simulations in a unified flow, the project will produce a validated multi-scale model. This outcome is expected to accelerate the design-technology co-optimization of ferroelectric-based neuromorphic systems, providing valuable insight into how device-level engineering affects circuit-level functionality and efficiency. - Computer Hardware, Electrical and Electronic Engineering, Interdisciplinary Engineering, Materials Engineering
- Bachelor Thesis, Collaboration, ETH Zurich (ETHZ), Master Thesis, Semester Project
|
|
