IBM Nighthawk r2 Completes Quantum Sampling in 19 Seconds

A quantum computing milestone has been reached using IBM’s Nighthawk r2 processor, which completed a complex random-circuit sampling experiment in just 19 seconds. The task involved manipulating 61 qubits through 36 rounds of operations, generating one million results that researchers estimate would require approximately 110 years for the Frontier supercomputer to replicate with comparable fidelity.

The Random-Circuit Sampling Benchmark

This experiment serves as a stress test rather than a practical solution, designed to determine if a quantum machine can produce patterns difficult for conventional computers to reproduce. Researchers applied 918 two-qubit interactions across the 61 qubits, scrambling them through randomly chosen operations before recording the resulting strings of zeros and ones.

While any laptop can generate random binary strings, the challenge lies in replicating the specific probability pattern created by quantum entanglement. The study, led by Tigran Sedrakyan of BlueQubit, a San Francisco-based quantum software company, highlights this distinction between simple randomness and structured quantum output.

Verifying Results Without Classical Calculation

Validating the quantum computer’s work presented a significant hurdle: calculating the full correct answer classically is too difficult. To address this, the team employed two distinct verification methods. First, they created “patched” versions of the circuit by cutting connections to split the system into three or four smaller groups. These smaller pieces were simple enough for conventional simulation, providing a reference answer to estimate the fidelity retained by the full circuit.

The second method utilized mirror circuits, where the quantum processor performed a sequence of operations followed immediately by the exact reverse sequence. In an ideal scenario, the second half should undo the first, returning qubits to their starting state. Deviations from this baseline indicated accumulated noise and error, allowing assessment without requiring a classical computer to calculate the full quantum output.

At 36 cycles, the surviving quantum signal had a fidelity of about 0.23 percent. Although low in absolute terms, this represents a measurable trace of the ideal quantum pattern. The authors note that 36 cycles hit a critical sweet spot where classical simulation became extremely expensive while the quantum signal remained detectable.

Hardware Speed and Reset Mechanisms

The rapid data collection was enabled by improvements in IBM’s hardware. While Nighthawk r2 features 120 programmable qubits, its primary advantage over previous Heron processors lies in its reset mechanism. Quantum experiments are intensely repetitive, requiring the processor to run, measure, and reset thousands of times.

Nighthawk r2 uses a dedicated mechanism to actively drain energy from each qubit, reducing idle time between runs to as little as one microsecond. IBM states the processor can execute more than 100,000 circuits per second, compared to roughly 4,000 per second on Heron. Consequently, a batch of executions that would occupy Heron for about 25 seconds passes through Nighthawk r2 in roughly one second.

Contextualizing the Computational Advantage

The claim of computational advantage remains subject to evolving classical algorithms. Previous demonstrations, such as Google’s 2019 Sycamore experiment, initially suggested tasks would take supercomputers thousands of years, but subsequent research using China’s Sunway supercomputer reduced that estimate to minutes.

The new study acknowledges that the 110-year figure assumes a particular tensor-network simulation and unlimited memory. Better algorithms and approximation strategies could significantly shrink this number. Nevertheless, the experiment marks a shift: a quantum task appearing extraordinarily expensive to reproduce classically was executed in seconds on hardware accessible via IBM’s cloud platform, allowing external researchers to verify the findings themselves.

The results have been described in a preprint on arXiv and have not yet undergone peer review.