Artificial intelligence (AI) is seen as a key technology with fields of application in a wide variety of areas in society. However, researching, developing and, in particular, using AI systems presents enormous challenges for the computing power and storage capacity needed to process large data volumes. These are generated for example in internet applications such as the Internet of Things and broadband services such as HD video on demand and social media. Traditional electronic hardware is no longer able to meet this challenge. A new research alliance headed by Dr. Wolfram Pernice, a professor at the Institute of Physics at the University of Münster, is developing fast, energy-efficient optical hardware alternatives. The alliance is now to receive almost six million euros for this research, over four years, from the European Commission, as part of the FET Proactive (Horizon 2020) funding line. The research teams involved include those from the University of Exeter (UK) and École polytechnique fédérale de Lausanne (EPFL, Switzerland).
"Our modern electronic technologies are fast approaching their limit, from a physics point of view," says Wolfram Pernice. "We need completely new methods for processing the enormous data volumes which are necessary for AI applications."
In the project, the PHOENICS consortium plans to use new types of materials to create the photonic neuromorphic processors. Another aim is to develop new methods of significantly increasing computing power.
The project is based on previous work done by Wolfram Pernice’s group. A few weeks ago, for example, the team published a study in "Nature" in which it presented a hardware accelerator for so-called matrix multiplications. These multipliers handle the main processing load within neuromorphic networks. The researchers had combined the photonic structures with phase change materials (PCMs) to create very fast and energy-efficient photonic processors. PCMs are normally used in optical data storage with DVDs or Blu-Ray discs. In the processor which the team described, this enables the matrix elements to be stored and preserved without any energy input being needed. The light source which the physicists used was a chip-based frequency comb. Such a light source provides different optical wavelengths which, independently of one another, are processed in the same system. This enables parallel data processing to be carried out.
EU Commission’s "FET Proactive" funding line
FET Proactive provides funding - thematically focused - for revolutionary, multidisciplinary technological research as a response to social and industrial challenges. The aim is to mature novel research themes in technology and to open up and develop the research landscapes necessary for this. The idea is to enable ambitious topics to be included when the relevant research communities are structured and set up - as well as when industrial research agendas are developed. FET Proactive is part of the EU’s "Horizon 2020" Framework Programme for Research and Innovation.