The scientific team of the Department of Informatics of the Aristotle University of Thessaloniki created the world’s fastest AI processor, which consists of photonic neurons and works with light instead of electricity.
The processor is based on an original photonic neuron architecture, which allows the implementation of algebraic operations at very high speeds, while exploiting original Neural Network Training techniques for optimal adaptation of algorithms to the specific characteristics of light.
This combination led to the experimental demonstration of a photonic processor operating at speeds up to 50GHz (>25 times faster than the speed of modern GPU processors) while enabling 10 times lower power consumption.
The research of the Department of Informatics of the Aristotle University of Thessaloniki highlights the significant speed and energy efficiency advantages that the use of light in the implementation of multiplication and addition operations provides, and allows processors to operate at speeds many times faster than today’s corresponding GPUs of NVIDIA and Google’s TPUs.
The results of the research have been published in international scientific journals such as Nature Communications, Journal of Lightwave Technology, IEEE Journal of Selected Topics on Quantum Electronics and SPIE Advanced Photonics as well as in the international press, such as ScienceX, PIC Magazine, Compound Semiconductor, Nanowerk etc.
In March 2023, part of the results were presented, through three scientific papers, at the world’s largest Optical Fiber Communication (OFC) Conference in San Diego, USA. The three main researchers of the papers were the PhD candidates of the Department of Informatics of the Aristotle University, Apostolos Tsakyridis, Georgios Yamougianis and Christos Pappas, supervised by Professor Nikolaos Pleros. The three PhD students were selected among the seven finalists, by the International Conference Committee, from among 350 students from all over the world, for the prestigious international OFC Corning Student Award. Thus, three of the seven finalists of the global competition were students of the Department of Informatics of AUTh.
The advantages of photonic technology in the field of information processing using neural networks are that this technolgy utilizes data in the form of light and photonic integration technologies to implement deep neural networks, which attempt to “mimic” the functioning of the human brain and human neurons in the computer field, with the goal of faster and more efficient information processing.
This scientific field is nowadays a rapidly developing technological field, which started to be intensively researched about five years ago, with the Department of Informatics of the Aristotle University, being among the three pioneers, together with two of the leading American universities, the Massachusetts Institute of Technology (MIT) and Princeton University.
The pioneering research of the Aristotle University in this field was based on the interdisciplinary collaboration of the Wireless and Photonic Systems and Networks (WinPhoS) research group of the Center for Interdisciplinary Research and Innovation of AUTh and the Computational Intelligence and Deep Learning Research Group of the Department of Informatics of the Aristotle University, with scientific leaders Prof. Professor Nikos Pleros and Professor Anastasios Tefas.
Their research in the field of photonic neural networks for AI applications has attracted significant funding from numerous European Horizon and national research programmes, as well as from the US start-up company Celestial AI, based in Silicon Valley, California.