Artificial intelligence (AI) mimics human skills but works faster. Neural networks power AI by copying how brain neurons connect. Neural networks need an activation function to work. This function adds nonlinearity.

Researchers at the Max Planck Institute, Leibniz University Hannover, and MIT have developed a new activation function using light and sound waves. This photonic activation function works with optical neural networks. Optical neural networks use light, not electricity, to process data. Light-based systems, called photonics, save energy and work faster than electronic ones.

Advantages of photonics in AI systems

Photonics offers high bandwidth, which means it handles lots of data quickly. It also uses high-dimensional symbols to encode information efficiently. These features speed up communication systems like the internet. Photonic systems process data in parallel, meaning they handle multiple tasks at once. They connect easily to optical fiber networks. When scaled up, photonics uses less energy for complex problems. Scientists explore photonics to make AI more efficient.

The new photonic activation function uses sound waves to control light. This method, called optoacoustics, keeps data in the light form. It avoids converting data to electronic signals, which saves energy. The function changes light signals based on their intensity. This nonlinearity helps neural networks solve hard tasks. The researchers show it can act like common functions such as sigmoid or ReLU. These functions transform data in neural networks.

The researchers have tested this in optical fibers and photonic chips. Sound waves help tune the activation function for different tasks. This setup allows parallel computing, where multiple calculations happen at once. It also keeps the light signal clear and efficient.

This discovery marks a step toward energy-efficient AI. Scaling up photonic systems remains a challenge. However, the new function shows promise for future optical neural networks. It could make AI faster and less power-hungry in the long term.

The researchers have described the methods and results of this study in a paper published in Nanophotonics.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.