The Scientific team of the School of Informatics of the Aristotle University of Thessaloniki has developed the fastest artificial intelligence processor in the world, which consists of photonic neurons that work with light instead of electricity.
The processor is based on a unique photonic neuron architecture and is capable of performing algebraic operations at very high speeds. In addition, a unique neural network training technique allows the algorithm to be optimally adapted to the light properties.
This combination led to the successful experimental demonstration of a photonic processor operating at speeds up to 50 GHz (more than 25 times faster than modern GPUs) and with 10 times lower power consumption.
Research from the School of Informatics of the Aristotle University of Thessaloniki has highlighted the significant speed and energy efficiency advantages of using light to perform multiplication and addition operations, allowing processors to operate many times faster than current NVIDIA GPUs and TPUs Google.
Research results have been published in Nature Communications, Journal of Lightwave Technology, IEEE Journal of Selected Topics on Quantum Electronics, SPIE Advanced Photonics and others.
They have also been published in international scientific journals and in the international press, such as ScienceX, PIC Magazine, Compound Semiconductor and Nanowerk.
In March 2023, some of the results were presented in three scientific papers at the world's largest optical fiber conference (OFC) in San Diego, USA.
The main researchers of the work were three PhD students, o Apostolos Tsakirides, The George Giamougiannis and Christos Pappas from the Informatics Department of the Aristotle University of Thessaloniki, supervised by him Professor Nikolaos Pleros of the same department.
These three doctoral students were selected by the International Conference Committee among seven finalists for the prestigious OFC Corning International Student Award, among 350 students from around the world. Thus, three of the seven finalists of the global competition were students of the Department of Informatics of AUTH.
The advantages of photonics technology in the field of information processing using neural networks is that data in the form of light and photonic integration technology can be used to implement deep neural networks, which attempt to “mimic” the functioning of the human brain and of human neurons in computing, and achieve faster and more efficient information processing.
This scientific field is today a rapidly developing technological field, which began to be intensively researched about five years ago, and the Informatics Department of the Aristotle University of Thessaloniki, together with two leading American universities, the Massachusetts Institute of Technology (MIT) and Princeton University.
Pioneers of KEDEK AUTH in this field are Wireless and Optical Systems and Networks Research Group of KEDEK AUTH (WinPhoS, http://winphos.web.auth.gr/) and the Computational Intelligence and Deep Learning Research Group of the Informatics Department of AUTH. It is based on interdisciplinary collaboration of the group (CIDL Research group, https://cidl.csd.auth.gr/), with the scientists Mr. Nikos Prelos in the upcoming years, while Anastasios Thefa of the Computer Science Department of AUTH, respectively.
Their work in the field of photonic neural networks for artificial intelligence applications has attracted significant funding from various European and national Horizon research programs, as well as from US startup Celestial AI, based in Silicon Valley, California.
The members of the AUTH Informatics Department research group that developed the fastest Artificial Intelligence processor. They can be seen from the left: standing Dr. Nikolaos Passalis, Asst. Prof. Nikolaos Pleros, Prof. Anastasios Tefas, Christos Pappas, Dr. Miltiadis Moralis-Pegios, Apostolos Tsakiridis, Georgios Giamougiannis and Manos Kirtas