Abstract
With the rapid progress made by industry and academia, quantum computers with dozens of qubits or even larger size are being realized. However, the fidelity of existing quantum computers often sharply decreases as the circuit depth increases. Thus, an ideal quantum circuit simulator on classical computers, especially on high-performance computers, is needed for benchmarking and validation. We design a large-scale simulator of universal random quantum circuits, often called 'quantum supremacy circuits', and implement it on Sunway TaihuLight. The simulator can be used to accomplish the following two tasks: 1) Computing a complete output state-vector; 2) Calculating one or a few amplitudes. We target the simulation of 49-qubit circuits. For task 1), we successfully simulate such a circuit of depth 39, and for task 2) we reach the 55-depth level. To the best of our knowledge, both of the simulation results reach the largest depth for 49-qubit quantum supremacy circuits.
Introduction
The quest for "quantum supremacy"—the point where a quantum device performs a task beyond the reach of any classical computer—requires a moving target. As quantum hardware improves, classical simulation techniques must also advance to provide a baseline for verification and benchmarking.
Simulating 49 qubits is a notorious "hard" threshold because storing a full state vector requires $2^{49}$ complex numbers. This translates to roughly 8 petabytes of memory (using double precision), which exceeds the RAM of even the world's largest supercomputers. To simulate circuits at this scale, researchers must find clever ways to trade off computation time for memory space, often by "cutting" the circuit into smaller, manageable pieces.
Methodology
The authors implemented a large-scale simulator on Sunway TaihuLight using several innovative strategies to overcome memory and communication bottlenecks.
Partitioning and Decomposition
Instead of using heuristic searches to decide how to split the quantum circuit, the authors used a Dynamic Programming (DP) algorithm. This algorithm finds the mathematically optimal partition scheme to minimize the total space and time complexity for a given circuit depth.
Furthermore, the team developed an Implicit Decomposition technique. When a 2-qubit gate (like a CZ gate) is "cut" to separate a circuit into two subsystems, it usually doubles the number of required paths to calculate. By exploiting the diagonal properties of the gates and their specific placement in 2D grid circuits, the authors decomposed 7 extra CZ gates without increasing memory overhead, effectively gaining "free" depth.
Hardware-Specific Optimizations
Sunway TaihuLight uses a unique many-core architecture (SW26010 processors). The team optimized single-node performance using:
- DMA (Direct Memory Access): To maximize memory bandwidth.
- Register Communication: Using ultra-fast on-chip communication buses to perform "qubit swaps" (matrix transpositions) between slave cores.
- Vectorization: Utilizing 256-bit instructions to process multiple amplitudes simultaneously.
- Communication Reduction: They optimized the final task of combining results by switching from standard MPI Gathers to a more efficient MPI Alltoall approach, resulting in a 57x speedup in network communication.
Key Results
The simulator was benchmarked using Universal Random Quantum Circuits, which are 2D grid-based circuits designed to generate high entanglement quickly, making them difficult for classical systems to simulate.
- Task 1 (Full State Vector): Successfully simulated a 49-qubit ($7 \times 7$) circuit at depth 39. This took 4.2 hours using 131,072 core groups (roughly 80% of Sunway TaihuLight).
- Task 2 (Single Amplitudes): Reached depth 55 for 49-qubit circuits.
- Comparison: For a 49-qubit circuit at depth 27, the Sunway simulator was over 29 times faster than previous results reported on IBM’s Blue Gene/Q.
- Efficiency: The simulation achieved a sustained performance of 4.3 - 4.92 PFlops. While this is only ~3.44% of the machine's peak, it is quite high for memory-bound quantum simulation tasks.
Discussion and Future Work
While this work pushes the boundaries of classical simulation, several challenges remain. The $2^{49}$ state vector remains a hard physical "memory wall"; simulating 50 or more qubits for a full state vector would require significantly more memory or a massive increase in computation time via path integral branching.
Additionally, the current simulator assumes ideal, noise-free quantum gates. Future work should incorporate noise models to better reflect the behavior of real-world NISQ (Noisy Intermediate-Scale Quantum) devices. The authors also propose exploring tensor network contractions as a potential way to further improve simulation efficiency for low-depth circuits.
Summary updated