Microprocessors for exascale scaling, hybridization of deep learning with physics, unstructured meshes, and the combination of AI and electromagnetism. The modeling and digital awards presented on June 14 at the Polytechnic School at the end of the Teratec forum testify to an excellent track record.
For this seventh issue, the editors of Industrie & Technologies and Usine Nouvelle, accompanied by representatives from Teratec, selected 3 nominees for each of the 3 trophy categories – Startup, Innovation and Code Design – among the many submissions received.
Then a jury of professionals in the field attended the candidates’ presentations on June 9 during the webinar and determined the winners. At the same time, readers of Industrie & Technologies and Usine Nouvelle were able to vote in favor of the Grand Prix du public.
The Startup Trophy was presented by Christine Menashe, Manager of the CEA Very Large Computing Center, SiPearl, represented by Founding President Philippe Naughton, for the Rhea project.
Established in January 2020, French startup SiPearl wants to make Brussels’ ambition to develop a European microprocessor capable of bringing exascale supercomputing to life, i.e. 1 billion billion calculations per second, a reality. A strategic area that affects multiple applications.
This processor, dubbed Rhea, will be 80% based on ARM Zeus cores and will also include Risc-V cores for security management. The choice of ARM should allow Rhea to reduce power consumption by a factor of 2.
SiPearl, launched in February 2020 with 6.2 million euros in European funds, is moving forward with great speed: expanding partnerships with Intel, Nvidia, HPE, Graphcore…, SiPearl surpassed one hundred employees in May and just received investment from the EIC (European Innovation Council) in the amount of 15 million euros and 2.5 million grants.
Cerfacs and Safran Aircraft Engines received the Innovation Prize from Jean-Yves Berthoud, Director of the INRIA Center in Saclay-Ile-de-France, to Gabriel Staffelbach, Cerfacs Researcher (European Center for Research and Excellence in Scientific Computing).
Cerfacs was awarded for developing code that uses unstructured meshes for particle-in-cell plasma simulation in an industrial configuration. More precisely, it is the plasma generated by Hall effect thrusters for satellites.
This type of innovative electric propulsion will become widespread with the deployment of mega constellations of satellites. However, the physics of these plasmas is still poorly understood and forces manufacturers to launch costly material testing campaigns.
The need for modeling comes up against the inability of liquid methods to accurately represent the kinetic effects characteristic of these plasmas. Hence the use of the particle-in-cell method, which, being resource-intensive, nevertheless remained limited to academic configurations discretized on structured grids.
The award in the Codesign category was given to Daniel Vervarde, President of Teratec, IRT SystemX and RTE, represented in particular by Muad Yagubi, IRT Project Manager, for the HSA project, hybridization simulation training.
This project aims to overcome certain limitations of the classical physical numerical simulation applied in the industrial world. First, the computational costs for complex systems. Data-based learning methods such as machine learning have shown good results in this regard, but have their limitations, in particular, less control over the behavior of the model, which may be unacceptable for industrial use.
A hybrid of physical modeling and machine learning is a widely studied way to get the best of both worlds. To speed up its use in industrial applications, SystemX and RTE have developed a test environment called “Industrial Learning Physical Simulation” that has been tested in the case of electrical networks but intended for use in other areas.
Audience Choice Grand Prix
Startup Altrans won the Public Grand Prix for its automated cable inspection solution called Translocator. Its purpose is to detect fault precursors in cables in order to reduce breakdowns and the cost of inspections.
The solution is based on emitting at one point at one end of the cable a special waveform that will interact with faults. This wave also interacts with other elements or phenomena in the cables, artificial intelligence software is used to identify signals coming from faults. They can be characterized and localized.
The Translocator is being tested in France and China in cooperation with Enedis, EDF and Shanghai Electric.
Selected for you