Tracking Quantum Dot Spectral Wandering in Real-time with Neuromorphic Silicon Photonics
Loading...
Authors
Grace, Adam
Date
2025-05-02
Type
thesis
Language
eng
Keyword
Quantum dots , Spectral Wandering , Neuromorphic Computing , Silicon Photonics
Alternative Title
Abstract
The field of quantum information science continues to grow as new ways to represent, process, and transmit information emerge. While hardware advances have supported this progress, further developments are needed to enable scalable quantum information systems. Quantum photonics presents a promising hardware platform that benefits from low decoherence, high-speed transmission, and on-chip integration with silicon photonics. Semiconductor quantum dots are attractive single-photon emitters for this platform, but they suffer from random fluctuations in the frequency of emitted photons — a phenomenon known as spectral wandering. Addressing this issue is an important step towards realizing practical quantum computing and communication systems.
In this thesis, we present a novel method for tracking quantum dot spectral wandering in real time. Typical resonance fluorescence scans are too slow to capture the relevant noise timescales. Instead, our approach reduces the scan to just two points and uses deep learning to infer the central emission frequency. By modeling spectral wandering as an Ornstein-Uhlenbeck process, we evaluate these two-step scans in simulation and show that simple feedforward neural networks can generate accurate detuning estimates. Notably, we derive the optimal measurement duration that balances shot noise with spectral wandering noise under various experimental conditions.
We propose that neuromorphic silicon photonics is well-suited to implement these deep learning models for detuning estimation. This hardware platform emulates neural networks in the analog domain and offers ultrafast latency for real-time signal processing. To address scalability challenges, we reduce the model size at little cost to performance. We further support this approach by experimentally characterizing a silicon photonic integrated circuit that implements a basic three-neuron model using microring weight banks and microring modulators. Strong optical weight bit precision values near 9.2 are reported, with an achievable range of ±0.8. We also observed reversed sigmoid-like microring modulator transfer functions, offering atypical but viable nonlinear activation functions. Despite these constraints on precision and flexibility, we see a limited impact on estimation accuracy in simulation, supporting the feasibility of neuromorphic silicon photonics for the real-time tracking of quantum dot spectral wandering.