GPU-based noise simulation for Clifford quantum circuits via cuQuantum SDK

The development of high-quality quantum error correcting codes often requires efficient classical simulations of quantum circuits. To this end, specialized algorithms restricted to only Clifford quantum gates are most appropriate, as the codes often rely on these types of gates. The simulators that use such algorithms are often called stabilizer simulators. In this work, we present techniques for GPU-accelerated stabilizer simulation with cuQuantum SDK. While most stabilizer simulators allow drawing measurements from a noise-less quantum circuit, our library focuses on simulating the effect of quantum noise channels on the state. This workload is useful for evaluating performance of error correcting codes and generating data for error syndrome decoders and other AI based decoders. The most efficient CPU simulator for this workload is the Stim library, which simulates only the Pauli frames instead of the full stabilizer tableau. With our APIs and kernels, users can accelerate this simulator, and others, to deliver up to 150 times improvement in sampling rate on a large-distance code compared to single-thread CPU.