What is it about?
The interferometer is a fundamental building block of optical neural networks, however, its nonlinear response limits the accuracy. We demonstrated linearized interferometers that can provide 2 to 3 more bits of precision. The device can directly replace existing interferometers without adding complexity, and the corresponding optical neural network shows higher training accuracy.
Photo by Laura Ockel on Unsplash
Why is it important?
The conventional interferometer has a sinusoidal transfer function, which leads to inevitable dynamic phase errors. We introduced a superlinear phase change to compensate for the sublinear sinusoidal function, therefore achieving a much more linear response over its entire range.
Read the Original
This page is a summary of: Low-phase quantization error Mach–Zehnder interferometers for high-precision optical neural network training, APL Photonics, April 2023, American Institute of Physics, DOI: 10.1063/5.0146062.
You can read the full text:
Low-phase quantization error Mach–Zehnder interferometers for high-precision optical neural network training
A Mach–Zehnder interferometer is a basic building block for linear transformations that has been widely applied in optical neural networks. However, its sinusoidal transfer function leads to the inevitable dynamic phase quantization error, which is hard to eliminate through pre-calibration. Here, a strongly overcoupled ring is introduced to compensate for the phase change without adding perceptible loss. Two full-scale linearized Mach–Zehnder interferometers are proposed and experimentally validated to improve the bit precision from 4-bit to 6- and 7-bit, providing ∼3.5× to 6.1× lower phase quantization errors while maintaining the same scalability. The corresponding optical neural networks demonstrate higher training accuracy.
The following have contributed to this page