Quantum Pulse Control with Spiking Neural Networks

An implementation for learning spiking-neural-network (SNN)–based control pulses for superconducting qubits.

---

🚀 Project Overview

Superconducting qubits require precisely shaped microwave pulses for high-fidelity gates, and need to react to realtime changes.

Our pipeline:

1. Build the transmon qubit model (Hamiltonian, dissipation)
2. Generate pulse samples from an SNN
3. Drive the simulation with those samples
4. Compute gate fidelity and use the loss for training
5. Backpropagate through time (BPTT) to update the SNN

---

📂 Repository Structure

demo                          # Demo for 0 to 1 state
├── quantum_snn_pulse_demo.py
└── writeup.txt
src
├── utils.py                  # Utilities for logging and contexts
├── run_quantum_pulse.py      # Demo for |0> -> |1> pulse
└── quantum_pulse_control/    # SNN model module
    ├── simulator.py           # Quantum simulator (differentiable & non-differentiable)
    ├── model.py               # Feedforward SNN pulse generator
    ├── trainer.py             # Training loop with GRAPE-style loss
    └── evaluator.py           # Evaluation and visualization

Usage

As a module:

from quantum_pulse_control import train_snn_controller, evaluate_and_visualize

model, history = train_snn_controller(num_epochs=300, batch_size=8, lr=1e-4)
evaluate_and_visualize(model)

Components

• SingleQubitSimulator: Simulates single qubit evolution with I/Q control pulses
• FeedforwardSNN_PulseGenerator: 4-layer spiking neural network that outputs analog control signals
• train_snn_controller: GRAPE-inspired training with fidelity, smoothness, bandwidth, and energy losses
• evaluate_and_visualize: Tests trained model and generates plots

---

🛠️ Setup & Installation

0. Install uv
1. Clone
2. Setup virtual environment: uv sync
3. Run the respective main file

---