RIA Novosti: An Experimental Prototype of an Analog Photonic Computing System Will Be Assembled in Russia

The assembly of an experimental prorotype of an analog photonic computing system that processes information hundreds of times faster than modern digital neural networks based on traditional semiconductor computers will begin in August and finish by the end of this year, Roman Skidanov, a participant in the project, professor of the Department of Technical Cybernetics at Samara University, Doctor of Physical and Mathematical Sciences, told RIA Novosti.

"The implementation of our project is going according to plan; the creation of an experimental sample of a photonic processor is currently in the stage of body assembly. At the moment, all the main elements of the experimental sample have been manufactured, and this month, in August, we will begin assembly," Skidanov said.

"It was decided to use another diode-type laser in the experimental sample: it is more compact and has less coherence, which should improve the characteristics of the processor. Future tests and experiments will show how much better. It is planned to complete the assembly and conduct tests by the end of 2024," he added.

A demonstration sample of the processor was created by specialists of Samara University as part of the research program of the National Center for Physics and Mathematics (NCPhM), with support of the Rosatom State Corporation. The processor operates on a new photonic component base in which information is transmitted by light particles (photons), and not by electrons, as in conventional computers.

The sample of photonic processor was created for the photonic computer of Megascience Class to have been implemented by 2030 in the NCPhM. According to the project, the computer’s performance is expected to be record-breaking and reach 10²¹ operations per second. Such a Megascience Class device will allow solving applied tasks for processing big data arrays and obtaining fundamental results in artificial intelligence and machine learning. The specialized processor even now allows recognizing huge arrays of data in large video streams.

The analog photonic computing system makes it possible to analyze and recognize objects hundreds of times faster than modern digital neural networks based on traditional semiconductor computers. This is especially important for the operational analysis of so-called hyperspectral data, which initially represent significant amounts of information. As was reported earlier, during the first experiments the demo sample showed recognition reliability of 93.75%. Recognition accuracy and reliability of the experimental sample should increase, due to selection of components with improved specifications. It was also reported that the prototype of the installation is planned to be ready in 2025.

Source: ria.ru