Scientists from the Karlsruhe Institute of Technology have launched practical research on a new class of high-performance computing platform based on a photonic processor. The system uses the developments of the German company Q.ANT and relies on light signals instead of traditional electric current in transistors. Instead of completely replacing the familiar microelectronics, the new device functions as a specialized optical accelerator within a hybrid architecture, working in conjunction with conventional central and graphics processors.
The principle of operation relies on the conversion of digital data into light wave parameters, such as phase and amplitude. The process is controlled by an integrated photonic circuit with optical waveguides and interferometers, where the light beams overlap and perform the necessary mathematical operations directly during their passage. This physical interference is extremely effective in matrix calculations, which form the basis of neural networks and artificial intelligence algorithms.
The main advantage of the technology is expressed in the parallel processing of data sets and the drastic reduction in the need for energy-intensive information transfer between memory and computing cores. According to initial estimates, the photonic chip can provide up to 30 times higher energy efficiency in specific calculations compared to standard silicon components. Tests in a real environment will yet determine the optimal distribution of tasks between optical and electronic modules, and in the future it is planned to add nonlinear light transformations.