Towards the end of 2020, the Jülich supercomputer JUWELS was ready: with its new booster module, it can perform 85 quadrillion computing operations per second (85 petaflops). That is equivalent to the computing power of more than 300,000 modern PCs. This massively expands simulation boundaries. JUWELS is also Europe’s strongest platform for the use of artificial intelligence (AI). By the time JUWELS was completed, it was not only the fastest supercomputer in Europe, but it was also the most energy-efficient system within the highest performance class.
The computer, developed by Forschungszentrum Jülich, the French-German company Atos and the Munich-based supercomputing specialist company ParTec together with the US manufacturer NVIDIA, is a powerful tool that Jülich scientists, along with partners from science and industry, use to answer complex research questions. A current example from the Covid-19 crisis is computer-supported drug development. The computing power of the booster makes it possible to simulate the processes taking place before, during and after the meeting of a potential active substance with a receptor or protein in a sufficiently realistic way (see p. 26).
In order to diagnose a neurodegenerative disease such as Alzheimer’s dementia, several test procedures are combined. Positron emission tomography (PET) can provide valuable information. Various biomarkers are available for this. For example, amyloid PET makes pathological amyloid deposits in the brain visible; 18F-fluorodeoxyglucose, or (18F-FDG)-PET, clears the way for evaluations of glucose metabolism and, thus, brain activity. Both methods complement each other.
However, there has been uncertainty about the optimal time to use these procedures as well as in what combination and order the PET biomarkers should be used. A panel of international experts from various disciplines, including Forschungszentrum Jülich and the University Hospital Cologne, has now turned to the available evidence and clinical expertise and developed an algorithm, which the experts present in the journal “Lancet Neurology”. They propose three main diagnostic pathways in which the biomarkers amyloid PET and 18F-DFG PET are applied at different positions in the sequence of diagnostic procedures, depending on the clinical presentation. In doing so, they also hope to stimulate further research on optimal diagnostic strategies.
A three-dimensional map of the brain that depicts the variability of brain structure with microscopic resolution – that is the Julich-Brain Atlas. A team of researchers from the Institute of Neuroscience and Medicine and from Heinrich Heine University Düsseldorf, led by Prof. Katrin Amunts, presented the atlas in the renowned journal “Science” in 2020. The atlas shows the different brain regions, which differ in the distribution and density of the total some-odd 86 billion nerve cells and in their function. So it is like a kind of “Google Earth” of the brain.
More than 24,000 wafer-thin brain slices from 23 brains in all were digitized, assembled on the computer in 3D and mapped by experts. Since no two brains are exactly alike, the Julich-Brain Atlas shows the size and location of the areas as probability maps..
Created in around 25 years of work, this is the most comprehensive digital map of the cellular brain architecture so far: it covers around 70 per cent of the cerebral cortex and deeper core regions. It is available to researchers worldwide through the new EBRAINS infrastructure (p. 46) as part of the European Human Brain Project. The atlas serves as an “interface” to connect information about the brain in a spatially precise way and thus to better understand the functioning of the brain and the mechanisms in diseases.
In the future, quantum computers could solve special tasks much faster and more efficiently than conventional supercomputers. Quantum information is fragile, however. This is why quantum computers must also be able to correct errors, such as when entire qubits – the carriers of quantum information – are lost.
Qubits are prone to errors induced by undesired environmental interactions. These errors accumulate during a quantum calculation and thus correcting them becomes a key requirement for reliable use of quantum computers. In the scientific journal “Nature”, a group of researchers from Jülich’s Peter Grünberg Institute and RWTH Aachen University, together with colleagues from the Universities of Innsbruck and Bologna, presented a method that enables quantum computers to continue computing even if they lose some qubits. For one thing, it enables an ion trap quantum computer to detect such errors and for another, to adapt to the loss of qubits in real time, all while maintaining the protection of fragile quantum information. The building blocks developed can also be used for other quantum computer architectures, the researchers emphasize.
Scientists from the Peter Grünberg Institute, together with colleagues from TU Berlin, have developed an artificial intelligence that can independently learn to grasp and move individual molecules.
Most materials consist of different molecules – similar to a Lego model composed of different building blocks. With a scanning electron microscope, atoms and molecules can be moved. However, the molecular building blocks of the nanoworld behave completely differently from macroscopic objects, each needing its own “instruction manual”. Targeted movement of molecules with the tip of the scanning electron microscope was previously only possible by hand, through trial and error.
With the help of a self-learning autonomous software control system, it has been possible to automate this process. In reinforcement learning, the software agent is not given a solution path, but success is rewarded and failure punished. This way, after completely random actions at the beginning, it learns rules over time about which movement is most promising in which situation, and gets better with each run.
This is the first time that artificial intelligence and nanotechnology have been successfully brought together. The method, published in the journal “Science Advances”, could pave the way for robotic, automated design of functional structures, such as quantum devices or innovative materials.
PHOTOS: Forschungszentrum Jülich/Wilhelm-Peter Schneider, Forschungszentrum Jülich/TRICKLABOR, De Visu/Shutterstock, Forschungszentrum Jülich/Katrin Amunts, Universität Innsbruck, SeitenPlan