Large-scale experiments in fundamental research require more and more computing and storage resources. In order to gain further scientific insight in the future, physicists from several research institutions have now joined forces in a project funded by the Federal Ministry of Education and Research (BMBF) to develop innovative digital processing methods.
The participating researchers contribute their diverse experience and knowledge in the fields of distributed computing infrastructures and algorithm development to the project.
Within the next three years, the joined project will develop and test new computing systems. One promising approach is the use of virtualization technologies to tap previously inaccessible resources. The scientists are also thinking about the use of new processor architectures, which are used, for example, in graphics cards and promise better energy efficiency (Green IT). The researchers see an important pillar in the development of improved algorithms and the use of artificial intelligence (AI) for Big Data analyses. Innovative methods of "machine learning" will play an important role here.
"The huge amounts of data are a great challenge for us. Innovative digital methods will be indispensable in the future if fundamental research is to advance decisively," said network coordinator Prof. Thomas Kuhr. However, it is not only physical research that faces the digital challenge. "Sooner or later, other scientific disciplines will also need powerful computing environments and will benefit from the new competences," Kuhr is certain. The joint project offers the participating young scientists an excellent opportunity to acquire comprehensive knowledge in new computing technologies. This means they are well prepared to fill leading positions in science or business in order to drive digital change forward.
The contribution of the ETAP group within this network is the replacement of existing algorithms for the real time detection and reconstruction of physics objects by deep neural networks. The group is targeting a processing rate of up to 40 MHz and a maximum response time of less than a millionth of a second. This will be sufficient to cover a large range of applications within the network. Field Programmable Gate Arrays (FPGAs) are the ideal choice.
The challenge is to optimally adapt the neural networks to the architecture of FPGAs, and the first trigger stage of the ATLAS experiment provides an ideal test environment for this.