ETP's participation in the BMFTR's ErUM-Data action plan

The DEEP research network is coordinated by Prof. Dr. Torben Ferber at the Karlsruhe Institute of Technology (KIT) and develops new solutions for ultra-fast real-time AI. At KIT, the Institute for Experimental Particle Physics (ETP) and the Institute for Information Processing Technology (ITIV) under the direction of Prof. Dr.-Ing. Dr. h. c. Jürgen Becker are involved. The goal is to use the developed technologies in large-scale research facilities and make them available for science and industry in the long term. 

The focus is on embedded parallel processors and next-generation system-on-a-chip hardware, in particular the AMD Versal architecture. A new and strategically important development is the close cooperation with SiMa.ai Germany (Stuttgart) as an industrial partner, which strengthens the transfer of state-of-the-art AI technologies from research to industrial applications and contributes to the development of sustainable technological expertise in Germany.

Other partners in the consortium include the Technical University of Hamburg, the University of Freiburg, and the University of Bonn. The solutions developed will be tested under real conditions at Belle II, LHCb, AMBER, ESS, European XFEL, and PETRA III/IV.

Associated partners in DEEP are SICK AG (Waldkirch) and the German Electron Synchrotron DESY (Hamburg).

The BRAID research consortium, coordinated by Prof. Dr. Jan Kieseler at KIT, is developing novel AI methods for data-driven reconstruction in particle, hadron, and astroparticle physics. The goal is to create a detector-independent (“agnostic”) framework that efficiently processes complex sensor data using graph and transformer networks while respecting physical laws.

Adaptive dimensionality reduction and open, standardized datasets enable cross-domain applications ranging from basic research to medical technology. Other partners include RPTU Kaiserslautern-Landau, TU Dortmund, and the GSI Helmholtz Center for Heavy Ion Research. The consortium strengthens the bridge between physics and computer science and contributes to sustainable, open, and energy-efficient AI-supported science.

The associated partner is inomed Medizintechnik GmbH (Emmendingen).

The SUSFECIT research consortium, coordinated at the University of Freiburg, is investigating new concepts for energy-efficient data center management in close collaboration with GridKa at the Scientific Computing Center (SCC). The focus is on developing a “breathing data center” that dynamically adapts computing capacities to the availability of renewable energies.

Intelligent orchestration and optimization of CPU clock frequencies are intended to significantly reduce energy consumption and the carbon footprint. The developments, which are based on the COBalD/TARDIS resource meta-scheduler, are being incorporated into international infrastructures such as the Worldwide LHC Computing Grid, thereby making an important contribution to the sustainable digitization of science.

Other partners in the consortium are RWTH Aachen University, the University of Bonn, DESY, the University of Freiburg, the University of Göttingen, and the Öko-Institut e.V.

Associated partners are CERN (Geneva) and, at KIT, the Institute for Astroparticle Physics (IAP), the Institute for Accelerator Physics and Technology (IBPT), and the Scientific Computing Center (SCC).

The FASTER joint project, coordinated at the University of Siegen, aims to develop autonomous and unattended experiments in synchrotron and neutron research in order to accelerate scientific findings and use beam time more efficiently. Results will be made available via the ErUM Data Hub to promote scientific collaboration. At the same time, an industrial partnership is being sought to make the developed technologies usable in other areas as well. The project is taking the first steps toward autonomous experiments in synchrotron and neutron research in order to increase research efficiency and data quality. 

The Laboratory for Applications of Synchrotron Radiation in Accelerator Physics and the Institute for Accelerator Physics and Technology (IBPT) are contributing to the project and cooperating with the consortium partners with machine learning algorithms, their optimization, FPGA-based data analysis and control, as well as fast control systems for accelerators and their transmission.

In addition to the University of Siegen and KIT, the partners in the consortium include Eberhard Karls University of Tübingen, Christian Albrecht University of Kiel, Helmholtz-Zentrum hereon GmbH, the German Electron Synchrotron (DESY), and the company TXproducts UG in Hamburg.

The joint project ErUM-IFT2, coordinated at the Max Planck Institute for Astrophysics in Garching, builds on the success of the previously funded ErUM-IFT project and deals with the application of methods of information field theory in the context of large-scale research experiments. The project covers a broad range of topics from astroparticle and particle physics to radio astronomy and condensed matter physics.

At the Institute for Astroparticle Physics (IAP) at KIT, the focus is on developing high-precision models for the signal propagation of radio waves in Antarctic ice by combining simulations with the CORSIKA 8 code with methods of information field theory. Such ice models form the basis for the search for high-energy neutrinos via their radio emissions in dense media in detectors such as IceCube-Gen2 and the Radio Neutrino Observatory Greenland (RNO-G).

In addition to the Max Planck Institute for Astrophysics, partners in the consortium include the German Electron Synchrotron (DESY), Forschungszentrum Jülich, Friedrich-Alexander University Erlangen-Nuremberg, Helmholtz-Zentrum hereon, RWTH Aachen University, Augsburg University of Applied Sciences, Technical University of Munich, Bielefeld University, and the University of Hamburg. 

Erium GmbH is an associated partner.

The joint project MorphoSphere, coordinated by the Laboratory for Synchrotron Radiation Applications (LAS) at KIT, has the vision of establishing intelligent, interactive data storage and analysis repositories. As a pilot application, 3D imaging is considered an important analysis method in many areas of application. Currently, petabytes of raw data are being generated here, which represent a data bottleneck in the analysis process and slow down scientific progress in many areas of application. To overcome this bottleneck, large data-generating devices, data and computing centers, and scientific communities have joined forces.

KCETA scientists from the Institute for Process Data Processing and Electronics (IPE) and the Scientific Computing Center (SCC) are implementing distributed big data management in the MorphoSphere project, which virtually connects storage systems with each other. They are implementing efficient data organization and data reduction and developing new approaches for distributed machine learning that aim to overcome the problem of large data sets for a single graphics processor by developing domain-parallel neural networks.

In addition to KIT, the partners in the network include Heidelberg University and the German Electron Synchrotron (DESY). Twenty associated national and international partners, consisting of research institutions, synchrotron radiation sources, and museums, complete the network.