Unlocking scalable photonic quantum computing with AI-enhanced circuit design

Abstract: We used an AI-enhanced, physics-grounded numerical optimization to automatically design photonic circuits that generate 4-5 qubit graph states, the building blocks of photonic fusion-based quantum computing, with up to 7.5 times higher success than standard baselines, and the first-ever photonic circuits that generate 5-qubit graph states.

Leading with AI for quantum hardware

Photonic quantum computing has long promised scalability and fault tolerance, but progress has been limited by the difficulty of generating the complex resource states that underpin this approach.

Now, we have taken a major step forward. Using AI-enhanced optimization grounded in physics, our team discovered the first known photonic circuits that generate 5-qubit graph states with high success rates, achieving up to 7.5 times greater performance than existing baselines. These graph states are essential components for fusion-based photonic quantum computing, and this breakthrough points to a new way forward in scaling these architectures.

Read on to explore how design automation is opening new frontiers in photonic quantum computing.

Applying AI tools to photonic quantum computer design

In recent research, we asked: Can we apply tools developed in AI to photonic circuit design, specifically to resource-state generation (RSG) for Fusion-Based Quantum Computing (FBQC)?

FBQC achieves fault-tolerant quantum computation by repeatedly applying fusion operations to smaller graph states, created by RSGs. The bottleneck in FBQC is the low success probability of the RSGs: if each attempt rarely succeeds, the entire fusion process slows or fails, the overhead balloons, and scalability is blocked.

Our AI-enhanced approach numerically discovers full photonic optical circuits end-to-end, using a physics-faithful simulator. Rather than combining hand-crafted submodules, the design is fully automated and guided by gradient information across a vast search space. 
Our automated pipeline is split into two stages that exploit tools from AI, including GPU acceleration, to achieve significant speed-ups:

• Stage 1 uses automated differentiation techniques to efficiently explore the entire photonic circuit configuration space to find high-fidelity, high-success solutions.‍
• Stage 2 applies sparsity-promoting regularization to prune unnecessary elements, reducing circuit complexity while preserving performance.

A new path to scalable photonic quantum computing

By applying an AI-enhanced, physics-faithful optimization loop to photonic circuit design, we were able to automatically discover circuits that dramatically improve performance. Our central result  produced RSG circuits that achieve:

• Up to 4.7 times higher single-shot success for 4-qubit graph states
• Up to 7.5 times higher single-shot success for 5-qubit graph states
• The first known circuits for several 4- and  5-qubit graph states
• Much simpler designs: typically 50–80% fewer beamsplitters, reducing optical depth and cumulative loss

The outcome: We can directly create larger building blocks (4- and 5-qubit clusters instead of 3-qubit trios), reducing the number of fusion steps required to assemble large clusters.

This work is part of our broader strategy to integrate AI throughout the quantum stack from autonomous calibration and noise suppression to quantum circuit design. By applying the same AI infrastructure software that’s been validated in quantum sensing and computing, we’ve shown that design automation can meaningfully accelerate quantum hardware development.

For the broader quantum industry, this marks a shift in what’s possible. AI is no longer just powering quantum algorithms; it is improving the performance and usability of the hardware itself.

Read the technical manuscript for more information and get in touch to discuss how we can help you scale photonic quantum computing.