Optimizing High-Fidelity Readout Circuit Using Reinforcement Learning Methods

We seek to automate the tuning of high-fidelity readout and gate control in quantum dot (QD) systems using reinforcement learning. We use Elzerman's readout technique to couple a qubit's spin state with a current signal. In practice, this signal must be disentangled from noise and drift using a circuit board, whose precise configuration and parameters must be finely-tuned in order to maximize readout fidelity. The tuning process can be done manually in real-time, but it must be automated to scale up to more general, complex QD systems, which are necessary for the construction of large-scale quantum computers. To that end, we are studying how to use reinforcement learning techniques to automate this process reliably and efficiently.