Christoph Lenzen, group leader at the Max Planck Institute for Computer Science in Saarbrücken, received an ERC proof-of-concept grant. This additional funding to an already granted ERC Grant will only be awarded if, in addition to the scientific excellence of a research object, its commercial exploitation becomes possible. For the next 18 months Lenzen will demonstrate the applicability of his research results on the "theory of reliable hardware" on a silicon chip.
Usually, a certain degree of safety is built-in at the design stage in all technical artifacts, from experience that usually never all eventualities in use can be considered from the outset. In the case of pressure vessels, the walls are designed to be stronger to absorb fluctuations, in electrical installations fuses are built-in to prevent overloads. In the European low voltage network (230 / 400 Volt) the actual voltage may only deviate 10% from the nominal value; electrical appliances must be designed for this. Microprocessors also have such protection in their design, e.g. for a short-term drop of the supply voltage. This must not lead to catastrophic consequences, especially in critical applications such as automotive control, medical technology, etc. Up to now, a limitation of the clock rate below the actually possible value has been used as a safety measure from the outset. Logically, this also reduces the efficiency of the circuit; calculations take longer or more space is used on the substrate. The dependency on the supply voltage is high: even a 1% drop reduces the effective clock rate and thus the computing speed. With increasing integration density this dependency tends to become even stronger.
This is where the approach, Lenzen has been pursuing for about ten years now, comes into play, and which has been supported by an ERC Starting Grant for three years. A hardware design that makes it possible to intercept difficulties, such as failures or even partial damage within the integrated circuits, in such a way that the overall system continues to function. "Imagine that you pull a screwdriver over your processor, which should still function smoothly under continuous bombardment with X-rays afterwards. That is of course exaggerated, but not so far from the truth," the researcher explains with a smile.
The new project focuses on the electrical side of the chip. The researchers around the Saarbrücken computer scientist want to reduce the influence of voltage fluctuations by adjusting the frequency of the clock. This will be achieved by a purely digital circuit technology that can react quickly enough. It is capable of absorbing both very steep and deep voltage drops and ensuring the functionality of the circuit during operation.
"A special feature of our approach is that we formally prove the correctness of our approach. This means that we not only verify experimentally whether the system behaves as expected, but also use mathematical proofs to show that the correct functioning of the system is guaranteed at all times," says Lenzen. The development of the mentioned computer chip for the practical demonstration of the results will be done together with the group of Prof. Milos Krstic (University of Potsdam/IHP).
Webpage Christoph Lenzen
Webpage Milos Krstic