UDiTQC: U-Net-Style Diffusion Transformer for Quantum Circuit Synthesis

TL;DR

UDiTQC advances quantum circuit synthesis by fusing a U-Net-style diffusion architecture with a Diffusion Transformer, enabling efficient multi-scale feature extraction and robust global-context modeling. The framework encodes quantum circuits as tensor-based representations and uses conditional diffusion with SRV or gate-set labels, alongside a unitary encoder for compilation tasks, to outperform prior baselines like GenQC. Through the UDiT architecture and its circuit-specific extension UDiTQC, the approach achieves superior entanglement generation and unitary compilation accuracy while enabling masking and editing to enforce physical constraints. This work both improves practical quantum circuit design and contributes to diffusion-model architecture by integrating residual, asymmetric, and transformer-based elements for scalable generative modeling.

Abstract

Quantum computing is a transformative technology with wide-ranging applications, and efficient quantum circuit generation is crucial for unlocking its full potential. Current diffusion model approaches based on U-Net architectures, while promising, encounter challenges related to computational efficiency and modeling global context. To address these issues, we propose UDiT,a novel U-Net-style Diffusion Transformer architecture, which combines U-Net's strengths in multi-scale feature extraction with the Transformer's ability to model global context. We demonstrate the framework's effectiveness on two tasks: entanglement generation and unitary compilation, where UDiTQC consistently outperforms existing methods. Additionally, our framework supports tasks such as masking and editing circuits to meet specific physical property requirements. This dual advancement, improving quantum circuit synthesis and refining generative model architectures, marks a significant milestone in the convergence of quantum computing and machine learning research.

Figures (8)

• Figure 1: The U-Net-Style Diffusion Transformer (UDiT) architecture: (a) The overall structure of UDiT is similar to DiT, but replaces the connections between multiple DiT blocks with U-Net-style layers. (b) The specific form of the U-Net-style DiT layers. (c) Details of the DiT block, where each DiT layer is composed of N DiT blocks. (d) Residual connections between DiT layers.
• Figure 2: The pipeline of the UDiTQC framework: (a) Quantum circuit are encoded as two-dimentional tensors, then embedding into high-dimensional continuous tensor space. (b) Conditional embedding in UDiTQC, with SRV or compilation gate sets embedded through class labels; unitary encoder is trained with UDiT to create the encoding of an input unitary. (c-d) Schematic representation of the diffusion model training process and posterior inference from the trained model.
• Figure 3: Model accuracy vs. the number of entangled qubits for circuits with different numbers of qubits. The accuracy represents the average generation accuracy for SRVs corresponding to each number of entangled qubits.
• Figure 4: Confusion matrix comparing the input and generated SRVs for 5-qubit circuits. The SRVs are grouped by the number of entangled qubits for clarity.
• Figure 5: Masking and editing circuits. (a) Masking: The layout of a quantum processor can be embedded as a mask, preventing the model from placing gates at specific locations in the input tensor (represented by the white area). (b) Editing: Portions of the circuit can be fixed to specific gates before generation, such as to incorporate an initial quantum state, ensuring that the desired quantum computation is performed on this state.