Photonic Circuits Become Trainable as AI Hardware
A physics-based in-situ learning method enables post-fabrication training of circuits by directly measuring optical gradients.
Researchers have enabled photonic integrated circuits to be trained even after manufacturing by developing a new physics-based in-situ learning method that directly measures optical gradients.
New Method in Photonic Circuits
Thanks to the developed physics-based in-situ learning method, photonic integrated circuits can be successfully trained even after manufacturing.
This approach allows different circuit topologies to be flexibly processed and optimized.
Direct Optical Gradient Measurement
The new technology is based on the principle of directly measuring optical gradients, thereby enhancing hardware capabilities.
In this way, complex computational processes can be performed much more efficiently at the optical level.
Advanced Hardware Capabilities
The hardware in question successfully performs operations such as matrix design and multi-wavelength computing.
Additionally, the system has the capacity to perform fast meta-learning without requiring a full digital model.