Self-driving Ultrafast Nanophotonics
Published:
Self-Driving Nanophotonics
We are developing nanophotonic systems that utilize autonomous experimentation to manipulate femtosecond pulses. This involves exploring large parameter spaces to optimize light-matter interactions and uncover physical principles.
Our platform, AutoSciLab, integrates machine learning techniques—such as variational autoencoders, active learning, and neural network equation learners—to close the design–fabrication–characterization loop. This allows for the efficient extraction of physical insights from high-dimensional experimental data.
Key Developments
Ultrafast Spin Steering ― We have demonstrated chiral metasurfaces capable of independently rotating left- and right-handed circularly-polarized beams by >30° within 140 femtoseconds. This enables sub-picosecond manipulation of ultrafast quantum states and spin-encoded information.
Physics-Aware Optical Computing ― Our photonic Ising machines utilize free-space diffraction to solve optimization problems. By harnessing Fourier optics, we address 2-million-spin optimization problems at kilohertz rates.
Chip-Scale Femtosecond Diagnostics ― We have developed a passive photonic streak camera by encoding quadratic phase profiles into metasurfaces. This device deflects 300-femtosecond pulses by 15° to capture ultrafast dynamics without complex electronics.
Autonomous Discovery ― AutoSciLab has been used to rediscover physical laws and uncover new structure–property relationships in nanophotonic emission, demonstrating the capability of AI to aid in physical understanding.
Impact
Our work aims to enable photonic processors capable of real-time sensing, ultrafast communication, and efficient computation. This represents a shift towards programmable photonic systems.