"Solving puzzles is one of my favorite things to do," says Christoph Lenzen. The 39-year-old has been a senior scientist at CISPA since the beginning of July. Together with his team, he is currently developing a mathematical approach to model circuits in processors in such a way that errors do not throw them out of sync, thus avoiding performance losses. The EU has been funding this project since 2017 and will continue to do so until 2022 with an ERC Starting Grant totalling 1.5 million euros. The mathematician and theoretical computer scientist is also conducting research on designing circuits in such a way that they can react flexibly to potential voltage drops. For this project, Lenzen was also able to secure an ERC Proof-of-Concept Grant of 150,000 euros. This funding is awarded in addition to the Starting Grant and gives scientists the opportunity to test whether their research results can also be translated into marketable products.
Processors have experienced a real performance explosion in recent decades. This is mainly due to the continuous reduction in the size of hardware components. More than a billion transistors can now be installed on a single processor chip. These switch electrical currents on and off within the processor, ultimately enabling the various computing operations to be carried out. However, as the number of installed components increases, so does the number and variety of potential errors and safety problems. This is the case in the complex circuit system of a processor and increases when many computers are networked and work together in so-called distributed systems. Guaranteeing security and freedom from errors in such constantly growing systems with many components is a major task for research.
Distributed systems have been used everywhere since their development in the 1980s. The most prominent example is the Internet. Several comparatively cheap computers are networked, thus allowing many processes to be executed simultaneously and quickly. Lenzen has been researching the theory of such systems since his days as a doctoral student at ETH Zurich, where he earned his doctorate from 2007 to 2011. One of the problems that has particularly concerned Lenzen since then is what is known as clock synchronization in distributed systems. "It is important for different applications to have a common notion of time," says Lenzen. He is working with a team at ETH Zurich to create a secure time base for a modernized Internet. The existing Internet architecture is being completely revised for this purpose.
There are some parallels between clock synchronization in distributed systems and in processors. However in processors, timing must be even more precise by a million times. Why is that? The processor is the brain of a computer. It processes numerous instructions from various computer programs every second. Clock signals in the form of electrical pulses coordinate these processes by initiating the next computing step. If the difference between the clocks in a computer network becomes too great, for example because errors at individual points lead to delays, problems arise in the joint solution of tasks. Built-in buffers compensate for such failures, but the computers must be constantly synchronized to guarantee smooth cooperation. Delays in this synchronization process occur repeatedly and slow down the computer. Lenzen is therefore working on new error-tolerant clocking methods that guarantee smooth cooperation even in growing systems." More or less as a byproduct of the work already done, a circuit design emerged that can compensate for voltage drops without leading to synchronization delays." His research team estimates that up to a 5 percent performance increase for processors is possible with this design. Lenzen hopes to find industry partners for this project so the technology can be successfully brought to market.
"I'll be occupied with these projects for a while longer, but I'm already very much looking forward to working with my new colleagues from CISPA," says Lenzen. The 39-year-old, a native of Kleve near the German-Dutch border, already knows the Saarland quite well. Before moving to CISPA, he was a research group leader at the Max Planck Institute for Computer Science in Saarbrücken for seven years. Lenzen's other research interests are also in the area of distributed systems theory and include consensus or distributed graph problems.
translated by: Tobias Ebelshäuser