Abstract
Achieving high-performance computation on quantum systems presents a formidable challenge that necessitates bridging the capabilities between quantum hardware and classical computing resources. This study introduces an innovative distribution-aware Quantum-Classical-Quantum (QCQ) architecture, which integrates cutting-edge quantum software framework works with high-performance classical computing resources to address challenges in quantum simulation for materials and condensed matter physics. At the heart of this architecture is the seamless integration of VQE algorithms running on QPUs for efficient quantum state preparation, Tensor Network states, and QCNNs for classifying quantum states on classical hardware. For benchmarking quantum simulators, the QCQ architecture utilizes the cuQuantum SDK to leverage multi-GPU acceleration, integrated with PennyLane's Lightning plugin, demonstrating up to tenfold increases in computational speed for complex phase transition classification tasks compared to traditional CPU-based methods. This significant acceleration enables models such as the transverse field Ising and XXZ systems to accurately predict phase transitions with a 99.5% accuracy. The architecture's ability to distribute computation between QPUs and classical resources addresses critical bottlenecks in Quantum-HPC, paving the way for scalable quantum simulation.
Introduction
In the current "Noisy Intermediate-Scale Quantum" (NISQ) era, quantum machine learning (QML) faces significant hurdles, primarily the qubit bottleneck and the data/communication bottleneck, often referred to as the Quantum Interconnect Bottleneck (QIB). To address these, this research introduces QSandwich, a hybrid "Quantum-Classical-Quantum" (QCQ) architecture. By strategically distributing tasks between classical GPUs and quantum simulators, the workflow reduces the number of qubits required for complex classification tasks in materials science and condensed matter physics.
The QSandwich Architecture
The core innovation is the QSandwich (QCQ) architecture, which processes data in three distinct stages:
- Quantum State Preparation (VQE): The process begins with a Variational Quantum Eigensolver (VQE). Instead of standard templates, the authors utilize a "checkerboard" tensor network ansatz to efficiently approximate the ground states of quantum systems like the Ising model.
- Classical Feature Extraction (CNN): This "filling" of the sandwich converts quantum states into classical data for a 1D Convolutional Neural Network. The CNN performs feature selection and pooling, drastically reducing data dimensionality.
- Quantum Classification (QCNN): Refined features are re-encoded into a final Quantum Convolutional Neural Network (QCNN) to classify the system's phase (e.g., "ordered" vs. "disordered").
To optimize this workflow, the team employed the cuQuantum SDK with PennyLane’s Lightning plugin, enabling simulations across multiple NVIDIA A100 GPUs using NVLink for high-speed communication. The use of a checkerboard tensor network ansatz was critical for efficient state preparation, allowing the model to approximate complex ground states with fewer parameters.
Benchmarking and Results
The architecture was evaluated using the Transverse Field Ising Model (TFIM) and the Antiferromagnetic XXZ Spin Chain Model. Key results include:
- High Accuracy: The model achieved 99.5% test accuracy for the Ising model and 94.6% for the more complex XXZ model.
- Performance Scaling: Multi-GPU acceleration provided a 10x speedup compared to CPU-based methods, with near-linear scaling from one to four NVIDIA A100 GPUs.
- Data Efficiency: By recording circuit parameters instead of full quantum state information, data size for a 10-qubit circuit was reduced from 1,024 floats to just 100, significantly easing the communication bottleneck.
Conclusion
The QSandwich framework demonstrates that Quantum-HPC Middleware is essential for near-term quantum utility. By using GPUs to handle heavy simulation and data processing, quantum resources can be focused on specialized tasks. Future work will integrate the architecture with the CUDA Quantum platform to further enhance flexibility and speed for multi-class classification problems in materials science.
Summary updated