Open Positions

The Extreme Light Infrastructure – Nuclear Physics (ELI-NP) is a premier European large-scale research facility located near Bucharest, Romania, and part of a major international initiative in high-power laser science and advanced nuclear physics. A central research focus at ELI-NP is the development of innovative high-energy gamma-ray sources based on inverse Compton scattering, enabling a new class of precision experiments in nuclear and fundamental physics.

Building on extensive experience with the Munich Compact Light Source — a globally recognized inverse Compton source for high-brilliance X-ray radiation — this postdoctoral research project aims at the advanced development, experimental implementation, and performance optimization of key components of an inverse Compton gamma-ray source in close collaboration with ELI-NP. The project encompasses the design and realization of high-power optical resonators in the green wavelength regime, comprehensive experimental characterization, and dedicated measurement campaigns at the ELI-NP facility in Bucharest.

The position offers the opportunity to conduct independent, high-impact research at the interface of laser physics, accelerator science, and high-energy photon sources, while contributing directly to the advancement of a world-class international research infrastructure.

A strong background in laser physics, accelerator physics, X-ray or photon science is highly desirable.

For further information, please contact: franz.pfeiffer(at)tum.de

The Munich Compact Light Source (MuCLS) is a laboratory-scale synchrotron radiation facility located at the Munich Institute of Biomedical Engineering (MIBE) in Garching and operated by the Chair of Biomedical Physics. Its X-ray source uses the principle of inverse Compton scattering to produce a partially coherent, quasi-monochromatic and high-flux X-ray beam [1].

Among these techniques, propagation-based phase-contrast imaging (PBI) is a very versatile single-shot phase-sensitive X-ray imaging method. On the one hand, this enables studying dynamic processes in specimens of biomedical interest, such as in vivo small-animal X-ray imaging of airways or lungs [2]. On the other hand, rapid micro-CTs of centimeter-scale samples at resolutions better than 10 μm become feasible.

The current PBI framework at the MuCLS employs the standard Paganin single-material single-distance phase retrieval. Within the framework of this thesis, more sophisticated phase-retrieval techniques, e.g. employing convolutional neuronal networks or the Fokker-Plank equation, will be developed and implemented. Furthermore, the duration of the micro-CT scans should be reduced by exploiting (both classical and AI-based) methods for denoising and sparse sampling. The improved PBI framework will be used to measure biomedical specimens, e.g., cochleas.

Character of the thesis: theory, programming & data processing (60%), experimental (40%)

Experience in X-ray or phase-contrast imaging as well as Python programming is desirable.

For further information, please contact Prof. Franz Pfeiffer (franz.pfeiffer(at)tum.de).

[1] Günther et al. 2020. https://dx.doi.org/10.1107/s1600577520008309
[2] Gradl et al. 2018. https://dx.doi.org/10.1038/s41598-018-24763-8

The Munich Compact Light Source (MuCLS) is a laboratory-scale synchrotron radiation facility located at the Munich Institute of Biomedical Engineering (MIBE) in Garching and operated by the Chair of Biomedical Physics. Its X-ray source uses the principle of inverse Compton scattering to produce a partially coherent, quasi-monochromatic and high-flux X-ray beam [1].

X-ray absorption spectroscopy (XAS) provides element-specific information about the chemical surroundings of an atom and, therefore, has a wide range of applications in different research fields. The XAS setup implemented at MuCLS [2] has successfully been applied in different catalysis research projects in collaboration with the TUM Department of Chemistry [3].

This PhD project will focus on further developing the energy-dispersive XAS setup at MuCLS and its applications. Currently, the setup employs a flat crystal in Laue geometry to generate a spatial energy gradient (see [1] for details). This geometry also results in a polychromatic line focus, which in the past had been inaccessible for experiments due to geometrical constraints. With a recent major remodeling of the experimental area of the MuCLS, these constraints have been removed. The PhD project is expected to make use of this increased flexibility by designing, implementing, and conducting experiments with the polychromatic focus. Furthermore, dynamical bending of the crystal shall be investigated to control both the bandwidth and the focusing of the transmitted beam. Exploration of alternative setup geometries (through simulations and experiments) and combination with other techniques like, e.g., X-ray fluorescence spectroscopy, is desirable.

Character of the thesis: simulation & data processing (50%), experimental (including setup design) (50%).

Experience in Python programming and CAD is desirable. Previous experience with XAS is considered an asset but not a requirement.

For further information, please contact Prof. Franz Pfeiffer (franz.pfeiffer(at)tum.de).

[1] B. Günther et al. 2020. https://doi.org/10.1107/S1600577520008309
[2] J. Huang et al. 2020. https://doi.org/10.1038/s41598-020-65225-4
[3] J. Huang et al. 2021. https://doi.org/10.1039/D1JA00274K

The Extreme Light Infrastructure – Nuclear Physics (ELI-NP) is a leading European large-scale research facility located near Bucharest, Romania, and part of one of the most ambitious international initiatives in high-power laser science and nuclear physics. One of ELI-NP’s key research directions is the development of novel high-energy gamma-ray sources based on inverse Compton scattering for cutting-edge nuclear physics experiments.

Building on our long-standing expertise with the Munich Compact Light Source — a world-leading inverse Compton source for brilliant X-ray radiation — this PhD project focuses on the development and experimental realization of the core components of an inverse Compton gamma source in close collaboration with ELI-NP. The work includes the design, setup, optimization, and detailed characterization of a high-performance optical resonator operating in the green wavelength range, as well as extensive experimental campaigns at the ELI-NP research facility in Bucharest.

This doctoral project offers a unique opportunity to work at the forefront of modern laser physics and accelerator-based photon sources, contributing directly to an internationally visible flagship research infrastructure.

Background knowledge in laser physics, X-ray science, or accelerator physics is advantageous but not mandatory.

For further information, please contact franz.pfeiffer(at)tum.de

Grating-based X-ray dark-field imaging provides additional structural information beyond conventional X-ray attenuation and is a promising technology for preclinical lung imaging and research on pulmonary diseases.

Within our group, we are developing a novel small-animal dark-field CT research platform. The aim of this Master’s thesis is the development and implementation of a dedicated sample positioning and monitoring system for mice and its integration into the existing scanner setup.

The thesis may include the mechanical design and construction of the sample bed, including controlled positioning in x-, y-, and z-direction, as well as the development of a mechanism for simple and reproducible mounting and removal of the sample bed from the scanner. Further tasks may comprise the selection and integration of suitable actuators, sensors, and electronics, as well as the implementation of a monitoring system for the mice, including respiratory monitoring and camera integration.

Depending on the scope of the project, simulation, manufacturing, assembly, commissioning, and experimental testing of the developed components can also be part of the thesis.

Character of the thesis: Mechanical engineering, mechatronics, electronics, prototyping, and system integration. Experience with CAD is desirable; knowledge of simulation, electronics, control systems, or Python is beneficial.

To apply, please send a short motivation letter, CV, and other relevant information to Prof. Franz Pfeiffer, franz.pfeiffer(at)tum.de.

Permanent magnet arrays are a key component of compact ultra-low-field MRI systems. For accurate image reconstruction, the spatial magnetic field distribution and its inhomogeneities must be known with high precision.

This MSc project will focus on the automated three-dimensional characterisation of magnetic fields generated by permanent magnet arrays using a Hall-probe sensor and a motorised measurement setup. The resulting field maps will be analysed to quantify magnetic field inhomogeneities and provide accurate input data for MRI image reconstruction.

The work will include the development and operation of an automated 3D measurement setup, Hall-probe-based field measurements, data processing and visualisation, and comparison with magnetic field simulations.

A strong background in physics, electrical engineering, or related fields is desirable. Experience with experimental setups, programming, automation, electromagnetism, or data analysis would be an advantage.

For further information, please contact franz.pfeiffer(at)tum.de.

Ultra-low-field magnetic resonance imaging (ULF-MRI) is an emerging imaging technique that allows magnetic resonance measurements to be performed at magnetic field strengths several orders of magnitude lower than those used in conventional clinical MRI systems. Permanent magnet arrays are a key technology for enabling compact and low-cost ULF-MRI systems. Achieving a sufficiently homogeneous magnetic field, however, requires precise knowledge of the magnetic properties of the individual magnets used in such arrays.

This MSc project will focus on the automated experimental characterisation of several hundred to several thousand individual permanent magnets. The aim is to develop and implement a measurement setup that can efficiently determine relevant magnetic properties of each magnet, such as magnetic field strength, magnetic moment, and variations in magnetisation magnitude and orientation.

By accounting for the measured properties of each individual magnet rather than assuming ideal and identical magnets, the project aims to identify optimised magnet configurations that produce a more homogeneous overall magnetic field for ULF-MRI applications.

The work will include the design and implementation of an automated measurement setup, experimental magnetic field measurements, data acquisition and analysis. Depending on the candidate’s interests, the project may also include mechanical design, automation and control, or validation of the optimised magnet array using experimental field mapping.

A strong background in physics, electrical engineering, mechanical engineering, or a related field is desirable. Experience with experimental setups, automation, programming, electromagnetism, or magnetic field simulations would be an advantage.

For further information, please contact franz.pfeiffer(at)tum.de.

Ultra-low-field magnetic resonance imaging (ULF-MRI) is an emerging imaging technique that allows magnetic resonance measurements to be taken at magnetic field strengths that are several orders of magnitude lower than those of conventional clinical MRI systems. By combining innovative permanent magnet arrays with advanced signal acquisition and AI-based reconstruction methods, ULF-MRI has the potential to facilitate the development of compact, low-cost MRI systems that are accessible for use in biomedical and industrial applications.

This MSc project will focus on developing and characterising permanent magnet arrays for ultra-low-field MRI applications. The work will include magnetic field simulations, setting up experiments and characterising magnet configurations, and investigating field homogeneity and imaging performance. Depending on the candidate’s interests, the project may also encompass MR hardware development, signal processing and image reconstruction. This work will be conducted within an interdisciplinary research environment at the intersection of medical physics, medicine, and imaging science.

A strong background in physics, electrical engineering, or related fields is desirable. Experience with electromagnetism, magnetic field simulations or MR physics would be an advantage.

For further information, please contact franz.pfeiffer(at)tum.de.

Ultra-low-field magnetic resonance imaging (ULF-MRI) is an emerging imaging technique that allows magnetic resonance measurements to be taken at magnetic field strengths that are several orders of magnitude lower than those of conventional clinical MRI systems. Although ULF-MRI has the potential to provide compact, low-cost and accessible imaging systems, the significantly reduced signal levels present major challenges for image quality and reconstruction.

This MSc project focuses on developing advanced AI-based image processing methods for ULF-MRI, particularly denoising and image reconstruction using state-of-the-art deep learning approaches. The project will investigate modern convolutional neural network (CNN) architectures, as well as emerging implicit neural representation (INR) methods, to improve image quality, suppress noise and enhance spatial resolution in highly undersampled or low signal-to-noise ratio MRI data. Depending on the candidate’s interests, the work may include the implementation and training of neural networks, simulation studies and the evaluation of experimental MR datasets. The project will be carried within an interdisciplinary research environment at the intersection of medical imaging, machine learning, and computational physics.

A strong background in physics, computer science, electrical engineering, or related fields is desirable. Experience with Python, deep learning frameworks, or image processing would be an advantage.

For further information, please contact franz.pfeiffer(at)tum.de.

The Extreme Light Infrastructure – Nuclear Physics (ELI-NP) is a premier European large-scale research facility located near Bucharest, Romania, and part of a major international initiative in high-power laser science and advanced nuclear physics. A central research focus at ELI-NP is the development of innovative high-energy gamma-ray sources based on inverse Compton scattering, enabling a new class of precision experiments in nuclear and fundamental physics.

Building on extensive experience with the Munich Compact Light Source — a globally recognized inverse Compton source for high-brilliance X-ray radiation — this M.Sc. project aims at experimental implementation of key components of an inverse Compton gamma-ray source, namely a high-power optical resonators in the green wavelength regime. While the main part of the master thesis will be done in our laboratories at TUM, within the project there is the possibility to participate in dedicated measurement campaigns at the ELI-NP facility in Bucharest.

Some strong background in laser physics or optics is desirable.

For further information, please contact: franz.pfeiffer(at)tum.de

X-ray absorption spectroscopy (XAS) provides element-specific information about the chemical surroundings of an atom and, therefore, has a wide range of applications in different research fields.

Within this project, the energy-dispersive XAS setup at the MuCLS, our inhouse compact synchrotron, is re-commissioned after the upgrade of its experimental area and the data acquired with this setup is going to be processed and analysed.

Data analysis will be performed with support by an experienced synchrotron XAS team. Since spectroscopy data taken at the MuCLS and at energy-dispersive XAS synchrotron beamlines are processed the same, data already acquired at this beamline will be analysed within the project as well.

Character of the thesis: experimental (25%), data analysis & processing (75%),

Experience in Python programming is desirable.

For further information, please contact Dr. Martin Dierolf (martin.dierolf(at)tum.de), Dr. Benedikt Günther (benedikt.guenther(at)tum.de), or Prof. Franz Pfeiffer (franz.pfeiffer(at)tum.de).

Industrial computed tomography (CT) has become a key technology for the non-destructive characterization of lithium-ion batteries over the recent years. As demand for higher inspection throughput and advanced defect analysis increases, detector performance increasingly limits achievable scan speed, image quality, and information content. Recent advances in photon-counting detector (PCD) technology have introduced the possibility of direct photon discrimination and energy-resolved X-ray imaging while using conventional polychromatic X-ray sources. Unlike conventional energy-integrating detectors, PCDs can classify individual photons according to their energy, enabling spectral imaging and opening new opportunities for material differentiation and characterization. While these capabilities are expected to enable improved material characterization and spectral CT workflows, their practical advantages and limitations in industrial battery inspection remain insufficiently investigated.

This thesis aims to experimentally investigate the performance of photon-counting detectors in comparison with conventional energy-integrating CMOS and TFT detector technologies. Automotive battery cells will be examined using both 2D radiography and 3D computed tomography under representative industrial imaging conditions.

Key research questions include:

• How do PCDs compare to state-of-the-art CMOS and TFT detectors with respect to image quality and defect detectability?
• What imaging parameters maximize the performance of photon-counting detectors in battery applications?
• To what extent can photon-counting detectors improve material differentiation within lithium-ion battery cells?
• What trade-offs exist between imaging speed, spectral information, and reconstruction quality?

The thesis combines experimental and independent work, image analysis, and detector characterization and contributes to the assessment of next-generation detector technologies for industrial CT.

A strong background in physics, X-ray imaging and python programming skills are desirable.

For further information, please contact: lisa.marie.petzold@tum.de or benjaminraphael.berger@tum.de

Latent diffusion models have recently emerged as a powerful framework for generating realistic medical images, enabling the synthesis of anatomically plausible data while operating in a compact latent representation. Such models offer a promising strategy for mitigating the scarcity of annotated medical imaging datasets, particularly for rare diseases and specialised imaging protocols. We have successfully implemented a latent diffusion framework for generating pulmonary emphysema CT scans and benchmarked it on downstream machine learning tasks. 

The goal of this project is to investigate latent diffusion models for generating medical images beyond pulmonary emphysema. The work will explore the synthesis of various pathologies and anatomical locations using computed tomography data, and potentially extend the approach to other modalities such as magnetic resonance imaging (MRI). Model performance will be evaluated both through image quality metrics and through their utility in downstream applications, such as image classification performance.

Character of thesis work: Machine Learning (70%), Medical Image Analysis (20%), Image Processing (10%).

Experience in X-ray imaging and Python programming are desirable.

For more information, please contact: Theresa Hiu [theresa.hiu(at)tum.de], Dr. Florian Schaff [florian.schaff(at)tum.de], and Prof. Franz Pfeiffer [franz.pfeiffer(at)tum.de].
 

X-ray dark-field radiography has proven significant diagnostic potential for the detection of pulmonary diseases in the first clinical prototype. Studies have shown that radiation dose can be substantially reduced while maintaining diagnostic performance [1], motivating the development of advanced image reconstruction and processing techniques for low-dose imaging. Continuous neural representations have emerged as a powerful framework for solving imaging inverse problems by directly incorporating the underlying physics into the reconstruction process, recently demonstrated for the human-scale dark-field CT prototype [2].

The goal of this project is to investigate physics-guided representations, in particular implicit neural representations (INRs) [3], for enhanced dark-field radiography. Possible research directions include neural phase retrieval using a Talbot-Lau forward model, continuous fusion of overlapping slot-scanning measurements, and improved noise suppression in low-dose acquisitions.

Character of thesis work: Machine learning (40%), X-ray physics (40%), Image processing (20%).

Experience in X-ray imaging and Python programming are desirable.

For more information, please contact: Daniel Frey [daniel.frey(at)tum.de], Lennard Kaster [lennard.kaster(at)tum.de], Dr. Florian Schaff [florian.schaff(at)tum.de], and Prof. Franz Pfeiffer [franz.pfeiffer(at)tum.de].

References:
[1] Bast et al. (2026). Ultra-Niedrigdosis Dunkelfeld-Radiographie für die Erkennung von Lungenemphysem. RöFo.
[2] Frey et al. (2026). SONAR: A Physics-constrained Neural Representation for X-ray Dark-field CT. MIDL.
[3] Sitzmann et al. (2020). Implicit Neural Representations with Periodic Activation Functions. NeurIPS.

Limited-angle computed tomography (CT) arises in many practical imaging scenarios where full angular coverage is not feasible. Examples include digital breast tomosynthesis, dental and interventional cone-beam CT, and on-site non-destructive testing using portable scanners. Fast acquisition times in helical CT lead on average to decreased angular sampling per slice, posing a particular challenge for achieving clinically feasible breath-holding times in human-scale dark-field CT [1].

The goal of this project is to investigate continuous scene representations as an alternative to conventional voxel-based reconstruction for limited-angle CT. In particular, implicit neural representations (INRs) [2] and 3D Gaussian Splatting [3] will be explored for existing attenuation and dark-field CT data acquired with existing experimental systems.

Character of thesis work: Machine learning (40%), CT reconstruction (40%), Image processing (20%).

Experience in X-ray imaging and Python programming are desirable.

For more information, please contact: Daniel Frey [daniel.frey(at)tum.de], Dr. Florian Schaff [florian.schaff(at)tum.de], and Prof. Franz Pfeiffer [franz.pfeiffer(at)tum.de].

References:
[1] Viermetz et al. (2022). Dark-field computed tomography reaches the human scale. PNAS.
[2] Sitzmann et al. (2020). Implicit Neural Representations with Periodic Activation Functions. NeurIPS.
[3] Zha et al. (2024). R$^2$-Gaussian: Rectifying Radiative Gaussian Splatting for Tomographic Reconstruction. NeurIPS.

Ring artifacts are a common issue in X-ray computed tomography (CT), caused by variations in detector pixel sensitivity. These artifacts manifest as circular patterns in reconstructed images and vary in shape and intensity. While traditional post-reconstruction image processing techniques are frequently employed to mitigate them, they often fall short under challenging conditions.

This project aims to develop machine learning–based methods for the effective removal of ring artifacts in CT images. Tasks include simulating ring artifacts on existing micro-CT datasets, training machine learning models for artifact correction, and benchmarking their performance against classical image processing techniques.

Character of thesis work: Numerical Simulation (33%), Image Processing (33%), Machine Learning (33%)

Experience in X-ray imaging and Python programming are desirable.

For more information, please contact: Sebastian Peterhansl [sebastian.peterhansl (at)tum.de], Dr. Florian Schaff [florian.schaff(at)tum.de], and Prof. Franz Pfeiffer [franz.pfeiffer(at)tum.de].

Noise in X-ray imaging degrades image quality and can obscure fine structural details. It is primarily caused by shot noise due to a limited number of photons, common in low-dose or rapid imaging, and electronic noise from detector hardware. A distinct characteristic in X-ray imaging with flat-panel detectors is the spatial correlation of shot noise across neighboring pixels. Traditional denoising techniques, often developed for uncorrelated noise, typically do not account for this.

The aim of this project is to develop machine learning–based methods to denoise X-ray images and enhance image quality. The work involves simulating realistic noise in existing micro-CT datasets, training machine learning models for noise removal, and benchmarking their performance against classical image processing techniques.

Character of thesis work: Numerical Simulation (33%), Image Processing (33%), Machine Learning (33%)

Experience in X-ray imaging and Python programming are desirable.

For more information, please contact: Sebastian Peterhansl [sebastian.peterhansl (at)tum.de], Dr. Florian Schaff [florian.schaff(at)tum.de], and Prof. Franz Pfeiffer [franz.pfeiffer(at)tum.de].

Computed Tomography (CT) image quality and quantitative accuracy strongly depend on the precise knowledge of the imaging geometry. Small inaccuracies in geometric parameters, such as the source-to-detector distance, source-to-object distance, detector orientation, or rotational alignment, can lead to reconstruction artefacts, image distortions, and reduced spatial resolution. 

The aim of this project is to develop methods for automatic geometry parameter estimation in Computed Tomography systems from raw data alone. The work involves modelling and simulating geometric misalignments, designing algorithms to estimate calibration parameters directly from projection data, and evaluating their impact on image reconstruction quality. Machine learning–based approaches may be explored and benchmarked against classical calibration and optimisation techniques to assess their accuracy, robustness, and computational efficiency

CT reconstruction (40%), Image processing (40%), Machine learning (20%)

Experience in X-ray imaging and Python programming are desirable.

For more information, please contact: Zijin Huang [zj.huang (at)tum.de], Dr. Florian Schaff [florian.schaff(at)tum.de], and Prof. Franz Pfeiffer [franz.pfeiffer(at)tum.de].

This project seeks to model comprehensive financial market dynamics through the innovative integration of physics-inspired functions and artificial intelligence techniques. Long-term market growth is represented using an exponential function with linear–quadratic parameters, enabling the capture of both sustained trends and acceleration or deceleration effects over time. Sudden market downturns are realistically depicted through negative exponential signal pulses, incorporating parameters such as decay time and bonding potential to reflect market resilience. In addition, speculative phases are modelled using specialised analytical functions designed to identify and predict the emergence of market bubbles.Long-term cyclical fluctuations are analyzed and modeled using techniques such as Fourier Series or wavelet decomposition to capture underlying periodicities and trends.

The methodology further involves separating the low frequency from noise using Empirical Mode Decomposition (EMD). The overall framework is validated through extensive backtesting on historical financial data.

Strong proficiency in Python is essential for this project, along with a strong interest in machine learning—particularly CNNs—and financial markets.

For more information, please contact: Prof. Franz Pfeiffer (franz.pfeiffer(at)tum.de)

This project aims to model and predict comprehensive financial market dynamics through the systematic application of Hidden Markov Models (HMMs). The approach represents market behaviour as a sequence of latent states corresponding to different market regimes, such as growth phases, high-volatility periods, speculative bubbles, and crash scenarios. Transitions between these hidden states are governed by probabilistic rules, enabling the realistic capture of regime shifts and temporal dependencies in financial time series.

Observable market indicators, including price movements (to start with), volatility measures, and trading volumes (later), are used to infer the underlying market states and to estimate transition probabilities. By learning these patterns from historical data, the model can identify prevailing market conditions and forecast likely future developments. The methodology allows for both short-term dynamics and long-term structural changes to be incorporated within a unified probabilistic framework.

The predictive performance is evaluated through rigorous backtesting using historical market data, assessing the model’s ability to detect regime changes and anticipate major market movements.

Essential requirements include solid Python programming skills, along with a strong interest in statistical modelling, machine learning, and financial markets.

For more information, please contact: franz.pfeiffer(at)tum.de

This project applies advanced deep learning techniques, specifically one-dimensional Convolutional Neural Networks, to model and forecast complex financial market dynamics. Financial time series, including price movements, returns, volatility indicators, and trading volumes, are interpreted as sequential signals, allowing the network to automatically extract meaningful temporal features to predict future probabilities for price changes, incl. Probabilities for abnormal market behaviour and sudden crash events. 

Through stacked convolutional layers with diverse kernel sizes, the model learns to capture both short-term market fluctuations and longer-term structural patterns. These learned features support accurate trend prediction, identification of regime shifts, and detection of critical market events such as crashes or speculative surges. The network architecture is optimised to efficiently process large financial datasets while maintaining strong generalisation capabilities.

The workflow encompasses data preparation, network training and tuning, and comprehensive validation using historical market data. Predictive performance is assessed through rigorous backtesting and relevant quantitative metrics.

Proficiency in Python is required, along with a strong interest in deep learning methods, time series modelling, and financial markets.

For more information, please contact: franz.pfeiffer(at)tum.de