New paper shows the use of photonic structures for AI

As we enter the next chapter of the digital age, data traffic grows exponentially. To further enhance artificial intelligence and machine learning, computers need the ability to process large amounts of data as quickly and efficiently as possible.

Conventional computer methods are not the task, but researchers have seen the light literally in the search for a solution.

Light-based processors, called photonic processors, enable computers to perform complex calculations at incredible speeds. New research was published in the journal this week Earth explores the potential of photonic processors for artificial intelligence applications. The results show for the first time that these devices can process information quickly and in parallel, something that today’s electronic chips cannot do.

“Neural networks ‘learn’ by capturing large sets of data and recognizing patterns through a series of algorithms,” said Nathan Youngblood, assistant professor of electrical and computer engineering at the University of Pittsburgh Swanson School of Engineering and co-author, explain. “With this new processor, it can perform multiple calculations simultaneously, with different optical wavelengths for each calculation. The challenge we wanted to address is integration: how can we perform calculations using light in a scalable and efficient way?”

The fast, efficient processing that researchers have been searching for is ideal for applications such as self-driving vehicles, which need to process the data they observe from various inputs as quickly as possible. Photonic processors can also support applications in cloud computing, medical imaging and more.

“With light-based processors for speeding up tasks in the field of machine learning, complex mathematical tasks can be processed at high speeds and throughputs,” said senior co-author Wolfram Pernice at the University of Münster. “It’s much faster than regular chips that rely on electronic data transfer, such as graphics cards or specialized hardware such as Tensor Processing Unit (TPUs).”

The research was conducted by an international team of researchers, including Pitt, the University of Münster in Germany, the universities of Oxford and Exeter in England, the École Polytechnique Fédérale (EPFL) in Lausanne, Switzerland, and the IBM Research Laboratory in Zurich.

The researchers combined phase change material – the storage material used on DVDs, for example – and photonic structures to store data in a non-volatile way without requiring a continuous energy supply. This study is also the first to combine these optical memory cells with a disk-based frequency comb as a light source, enabling them to simultaneously calculate at 16 different wavelengths.

In the paper, the researchers used the technology to create a convolutional neural network that would recognize handwritten numbers. They found that the method allowed data rates and computer densities never seen before.

“The contraction operation between input data and one or more filters – which can be, for example, an edge of a photograph – can very well be transferred to our matrix architecture,” said Johannes Feldmann, a graduate student at the University of Münster. lead author of the study. “By extracting light for signal transmission, the processor can perform parallel data processing by means of wavelength multiplexing, which results in a higher computing density and many matrix multiplications are performed in just one time. Unlike traditional electronics, which usually operate in low GHz optical modulation speeds can be achieved at speeds up to 50 to 100 GHz. ”

The article, “Parallel Convolution Processing Using an Integrated Photonic Tensor Core”, was published in Earth and co-author of Johannes Feldmann, Nathan Youngblood, Maxim Karpov, Helge Gehring, Xuan Li, Maik Stappers, Manuel Le Gallo, Xin Fu, Anton Lukashchuk, Arslan Raja, Junqiu Liu, David Wright, Abu Sebastian, Tobias Kippenberg, Wolfram Pernice , and Harish Bhaskaran.