In the digital age, data traffic is growing at an exponential rate. The demands on computing power for appli­cations in arti­ficial intelli­gence such as pattern and speech recognition in particular, or for self-driving vehicles, often exceeds the capa­cities of conven­tional computer processors. Working together with an inter­national team, researchers at the University of Münster are developing new approaches and process archi­tectures which can cope with these tasks extremely efficient. Together with researchers at the Univer­sities of Oxford, Exeter, Pittsburgh, École Poly­technique Fédérale (EPFL) and IBM Research Europe they have now shown that photonic processors, with which data is processed by means of light, can process information much more rapidly and in parallel – something electronic chips are incapable of doing.

Light-based processors for speeding up tasks in the field of machine learning enable complex mathe­matical tasks to be processed at enormously fast speeds. Conventional chips such as graphic cards or specialized hardware like Google's TPU (Tensor Processing Unit) are based on electronic data transfer and are much slower. The team of researchers led by Wolfram Pernice from the Institute of Physics and the Center for Soft Nano­science at the University of Münster implemented a hardware acce­lerator for matrix multi­plications, which represent the main processing load in the computation of neural networks. Neural networks are a series of algorithms which simulate the human brain. This is helpful, for example, for classifying objects in images and for speech recognition. 

The researchers combined the photonic structures with phase-change materials (PCMs) as energy-efficient storage elements. PCMs are usually used with DVDs or BluRay discs in optical data storage. In the new processor this makes it possible to store and preserve the matrix elements without the need for an energy supply. To carry out matrix multi­plications on multiple data sets in parallel, the physicists used a chip-based frequency comb as a light source. A frequency comb provides a variety of optical wavelengths which are processed inde­pendently of one another in the same photonic chip. As a result, this enables highly parallel data processing by calculating on all wave­lengths simul­taneously. “Our study is the first one to apply frequency combs in the field of arti­ficially neural networks,” says Wolfram Pernice.

In the experiment the physicists used a convo­lutional neural network for the recog­nition of handwritten numbers. These networks are a concept in the field of machine learning inspired by biological processes. They are used primarily in the processing of image or audio data, as they currently achieve the highest accuracies of classification. “The convo­lutional operation between input data and one or more filters – which can be a highlighting of edges in a photo, for example – can be transferred very well to our matrix archi­tecture,” explains Johannes Feldmann. “Exploiting light for signal transference enables the processor to perform parallel data processing through wavelength multi­plexing, which leads to a higher computing density and many matrix multi­plications being carried out in just one timestep. In contrast to tradi­tional electronics, which usually work in the low GHz range, optical modu­lation speeds can be achieved with speeds up to the 50 to 100 GHz range.” This means that the process permits data rates and computing densities, i.e. operations per area of processor, never previously attained.

The results have a wide range of applications. In the field of arti­ficial intelli­gence, for example, more data can be processed simultaneously while saving energy. The use of larger neural networks allows more accurate, and hitherto unattainable, forecasts and more precise data analysis. For example, photonic processors support the evaluation of large quantities of data in medical diagnoses, for instance in high-reso­lution 3D data produced in special imaging methods. Further appli­cations are in the fields of self-driving vehicles, which depend on fast, rapid evaluation of sensor data, and of IT infra­structures such as cloud computing which provide storage space, computing power or appli­cations software. (Source: U. Muenster)

Reference: J. Feldmann et al.: Parallel convolutional processing using an integrated photonic tensor core, Nature 589, 52 (2021); DOI: 10.1038/s41586-020-03070-1

Links: Responsive Nanosystems, Institute of Physics, University of Münster, Münster, Germany • Laboratory of Photonics and Quantum Measurements, Swiss Federal Institute of Technology Lausanne (EPFL), Lausanne, Switzerland