Perlmutter Supercomputer Sets new benchmark in Quantum chip
The Perlmutter supercomputer simulates quantum devices in unparalleled detail. The Perlmutter Supercomputer
Effective quantum hardware development has improved because to the Perlmutter supercomputer at the National Energy Research Scientific Computing Centre (NERSC). Perlmutter has been hired to simulate large-scale quantum circuits and quantum microchips in unprecedented detail to increase quantum chip performance.
These conventional simulations aim to provide broad, high-fidelity analysis that accelerates quantum software and hardware development, comparison, and refinement. These extensive classical computations are essential for quantum hardware design and verification, notwithstanding their cost.
A First-Time Physical Simulation of the Microchip
A superconducting quantum microprocessor created by UC Berkeley's Quantum Nanoelectronics Laboratory and Berkeley Lab's Advanced Quantum Testbed (AQT) was simulated in groundbreaking simulation. Simulation of device assembly and functioning at the full-wave physical level was the key goal. Its realism in material, layout, and wiring distinguishes it from other simulations that treat the quantum device as a “black box”. The simulation, which included behaviour over four orders of magnitude, was complicated. Time-domain PDEs for entire device electromagnetic (Maxwell) systems were solved by researchers. Using circuit parts with nonlinear behaviour, the physics model allowed researchers to observe the chip's spectrum and transient responses. This time-domain strategy is crucial because it avoids the transient and nonlinear effects that many frequency-domain or reduced models miss. Simulated object was a 0.3 mm thick, 10 mm square multi-layered chip. To accurately capture electromagnetic wave propagation, the researchers discretised the chip into 11 billion grid cells, including one-micron etchings. To manage this detail, massive processing was needed. The simulation employed over 7,000 of Perlmutter's 7,168 NVIDIA GPUs in a day. The team used the full system to conduct over a million time steps in under seven hours. Researchers tested three circuit designs in one day at this speed. Device-level, time-domain simulations at this scale provide quantum engineers with performance information. It captures nonlinear interactions and realistic noise coupling. This feature considerably improves crosstalk and gate fidelity forecasts compared to simplified or frequency-domain models. By accurately recreating large parts of the hardware, researchers can find hidden resonances, emergent failure modes, and spectral coupling issues that smaller models may miss.
Digging into Quantum Circuits and Algorithms
QIS@Perlmutter, which simulates quantum circuits and algorithms on classical hardware, relies on Perlmutter. Researchers are developing and testing high-performance quantum simulation software and other approaches on the platform. These software development initiatives prioritise GPU-native optimisations using CUDA-Q, NVSHMEM for inter-GPU communication, and TensorCores. Q-Gear transforms Qiskit instructions into CUDA-Q for optimum GPU utilisation on Perlmutter, and TANQ-Sim is a TensorCores-accelerated density-matrix simulator. These optimisations ensure strong scaling and nearly 100% GPU use. Simulated systems solve key quantum development issues. Simulated evolution of the Quantum Approximate Optimisation Algorithm (QAOA) indicated that it frequently enters a phase of high entanglement. Some classical algorithms struggle to recreate QAOA due to this entanglement. Quantum Systems Dynamics has also been studied using Perlmutter time-evolution simulations of the transverse-field Ising model (TFIM), a 2D spin lattice with systems scaled up to 40 qubits. Simulations with 20 to 30 qubits were 600x faster with a single A100 GPU instead of a multi-threaded CPU. Circuit Cutting/Knitting is a novel Perlmutter approach. This method divides large quantum circuits into smaller subcircuits that can be put on small quantum devices. Perlmutter's classical computing power reassembles the data. HPE and NVIDIA's 40-qubit simulations, which took 24 minutes and used 1,024 A100 GPUs, proved this strategy. Also important is noise modelling, which simulates decoherence and noise in quantum systems to assist researchers reduce errors in real-world quantum gear. Enhancing Quantum Co-Design and Error Reduction Time-domain, device-level simulations with near-device realism allow quantum chip designers to properly benchmark their designs. This capacity improves gate error, spectrum, and crosstalk forecasts before expensive fabrication. Co-design efforts benefit from these simulations. Designers may carefully test layout combinations, device geometry, and control pulse forms to reduce noise and resonances. The Perlmutter supercomputer enables rigorous, high-fidelity verification and design iteration, speeding up the research and optimisation needed to build reliable quantum computing systems.
