Optimizing machine-learning process using light-based processors

A chip-based “frequency comb”, as a light source.

The world is generating exponentially increasing amounts of data that need to be processed quickly and efficiently. Highly parallelized, fast, and scalable hardware is therefore becoming progressively more important. All this places a heavy burden on the ability of current computer processors to keep up with demand.

An international team of scientists developed a photonic processor using light rays inside silicon chips to process information much faster than conventional electronic chips. These photonic processors have surpassed conventional electronic chips by processing information much more rapidly and in parallel during experiments.

Scientists developed hardware accelerators for so-called matric-vector multiplications, which are the foundation of neural networks, which are utilized for machine-learning algorithms

Since various light wavelengths (colors) don’t interfere with one another, the scientists could utilize different frequencies of light for parallel calculations. Yet, to do this, they used another creative innovation, created at EPFL, a chip-based “frequency comb,” as a light source.

Professor Tobias Kippenberg at EPFL, one of the study’s leads, said, “Our study is the first to apply frequency combs in the field of artificial neural networks. The frequency comb provides various optical wavelengths that are processed independently of one another in the same photonic chip.”

Senior co-author Wolfram Pernice at Münster University, one of the professors who led the research, said, “Light-based processors for speeding up tasks in the field of machine learning enable complex mathematical tasks to be processed at high speeds and throughputs. This is much faster than conventional chips which rely on electronic data transfer, such as graphic cards or specialized hardware like TPU’s (Tensor Processing Unit).”

Scientists tested their photonic chips on a neural network that recognizes hand-written numbers.

Johannes Feldmann, now based at the University of Oxford Department of Materials, said, “The convolution operation between input data and one or more filters – which can identify edges in an image, for example, are well suited to our matrix architecture. Exploiting wavelength multiplexing permits higher data rates and computing densities, i.e., operations per area of processer, not previously attained.”

David Wright at the University of Exeter, who leads the EU project FunComp, which funded the work, said“This work is a real showcase of European collaborative research. While every research group involved is world-leading in their way, it was bringing all these parts together that made this work truly possible.”

This Light-based processor has far-reaching applications: higher simultaneous (and energy-saving) processing of data in artificial intelligence, more extensive neural networks for more accurate forecasts and more precise data analysis, large amounts of clinical data for diagnoses, enhancing rapid evaluation of sensor data in self-driving vehicles, and expanding cloud computing infrastructures with more storage space, computing power, and applications software.

Journal Reference:
  1. J. Feldmann, N. Youngblood et al. Parallel convolution processing using an integrated photonic tensor core. Nature 06 January 2021. DOI: 10.1038/s41586-020-03070-1

See stories of the future in your inbox every morning.