Optoacoustic neural networks enabled by multi-frequency light-sound interactions

• Applicant: Professorin Dr. Birgit Stiller
• Subject Area: Optics, Quantum Optics and Physics of Atoms, Molecules and Plasmas
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 464294062

Nowadays complex problems – be it climate dynamics or economic processes – require the processing of big data and complex intelligent systems. Their backbone are artificial neural networks, that can be trained to solve the relevant questions. Major challenges are the capacity of data they can handle, long training times and their memory size. Artificial neural networks based on photonic connections can provide a solution, as an optical approach comes with a broad bandwidth, high processing speed and integrability into existing electronic chips. However, in reality optical neural networks are so far mostly based on a network of optical components such as couplers and phase shifters, which are fixed elements that act linearly and whose connections cannot be reconfigured.In this project we will use light-sound interactions for the implementation of an artificial neural network. The interaction of light and sound is based on a nonlinear optical effect – stimulated Brillouin scattering. This optoacoustic interaction has found applications in optical fiber sensing, narrow-linewidth lasers, microwave photonics and nonlinear optical signal processing. The latter includes delaying and storing optical information into sound waves in order to synchronize information streams and process data through optical control pulses. Within this project, we want to theoretically and experimentally explore the physics of multi-frequency light-sound interactions. With the gained knowledge, we aim to demonstrate a reconfigurable optoacoustic neural network in a single fiber and to show a proof-of-principle experiment for multi-frequency optoacoustic signal processing for its implementation into communication network architectures. The project will have an impact on how we design neural networks as it allows for a software-based solution – in contrast to hard-wiring the architecture. It has the potential to boost machine learning and artificial intelligence in terms of capacity and flexibility to solve new types of problems.

• DFG Programme: Research Grants