Artificial intelligence (AI) has become a part of our daily lives, helping us with tasks like language processing and data summarization. However, as AI continues to evolve, it requires new solutions to increase processing speed and reduce energy consumption. Optical neural networks, which use light instead of electric signals, have the potential to handle large amounts of data quickly and efficiently. Until now, most optical neural networks have relied on fixed components and steady devices, limiting their reconfigurability.
A team of researchers from the Max Planck Institute for the Science of Light and the Massachusetts Institute of Technology has found a way to create reconfigurable building blocks for optical neural networks using sound waves. Their approach uses hair-thin optical fibers, already widely used for fast internet connections, to create traveling sound waves that can manipulate the computational steps of an optical neural network.
The Optoacoustic REcurrent Operator (OREO)
The researchers have developed the Optoacoustic REcurrent Operator (OREO), which uses sound waves to implement a recurrent operator, a technology commonly used in recurrent neural networks. Recurrent operators allow the linking of computational steps, providing context for each calculation. This is important for understanding human language, where the order of words can change the meaning of a sentence.
Traditional neural networks struggle to capture context because they require access to memory. Recurrent neural networks overcome this challenge by using internal memory to capture contextual information. While this is straightforward to implement digitally, it is challenging in optics and has relied on artificial cavities to provide memory. OREO offers a solution by harnessing the intrinsic properties of an optical waveguide without the need for an artificial reservoir or newly fabricated structures.
The Potential of Sound Waves in Optical Neural Networks
One of the key advantages of OREO is that it is entirely optically controlled, making it programmable on a pulse-by-pulse basis. The researchers have used this feature to implement a recurrent dropout optically for the first time, a regulation technique previously only used to boost the performance of digital recurrent neural networks. OREO has demonstrated its ability to process context by distinguishing up to 27 different patterns.
“The all-optical control of OREO is a powerful feature. Especially the possibility to program the system on a pulse-by-pulse basis gives several additional degrees of freedom. Using sound waves for photonic machine learning is disrupting the status quo and I am very eager to see how the field will evolve in the future,” says Steven Becker, a doctoral student in the Stiller Lab.
In the future, using sound waves for optical neural networks could unlock a new class of optical neuromorphic computing that can be reconfigured spontaneously and allow large-scale in-memory computing in the present telecommunication network. On-chip implementations of optical neural networks can also benefit from this approach, as it can be implemented in photonic waveguides without additional electronic controls.
“Photonic machine learning might hold huge potential for parallel processing of information and energy-efficient operations. Adding acoustic waves can contribute to this endeavor with an all-optically-controlled and easy-to-operate tool-kit,” says Dr. Birgit Stiller, head of the Quantum Optoacoustics Research Group.
