QESEM Error Mitigation on IBM's 156-Qubit Heron Shows Quantum Advantage Fugaku Couldn't Match

Forum topic · 小凯 · 2026-08-25

Summary

On July 30, 2026, BlueQubit, Qedma, IBM, and Japan's RIKEN jointly reported a quantum advantage result using Qedma's QESEM error-mitigation software on IBM's 156-qubit Heron processor. The experiment simulated a 74-qubit two-dimensional Floquet Ising model, producing results that classical computers could not reproduce within the allotted time. RIKEN attempted classical verification on the Fugaku supercomputer, consuming over 500,000 CPU core-hours without achieving convergence, reflecting the exponential scaling of classical simulation cost at this qubit count. The result was listed on the Quantum Advantage Tracker, reportedly unmatched by classical methods for eight months. Notably, Quantinuum independently reproduced the same result signature on its trapped-ion quantum computer, providing cross-platform validation from a different hardware modality. The work highlights error mitigation rather than error correction: instead of using extra qubits for redundancy, software post-processing statistically reconstructs signals from noisy data, extracting usable compute from NISQ-era devices. This suggests error-mitigation software stacks may deliver verifiable, independently reproducible quantum advantage before fault-tolerant quantum computing arrives.

Event

On July 30, 2026, BlueQubit, Qedma, IBM, and Japan's RIKEN jointly announced a quantum advantage result: using Qedma's QESEM error-mitigation software on IBM's 156-qubit Heron processor, they simulated a 74-qubit two-dimensional Floquet Ising model and obtained results that classical computers could not replicate within the prescribed time. The result was listed on the Quantum Advantage Tracker and reportedly remained unmatched by classical methods for eight months.

The key point is error mitigation rather than error correction: instead of relying on extra qubits to correct errors, software performs statistical reconstruction on noisy data, recovering the signal from the noise.

Classical Verification: Even a Supercomputer Fell Short

RIKEN exhausted over 500,000 CPU core-hours on the Fugaku supercomputer (Arm-based, formerly the world's fastest) without getting the classical simulation to converge. This is not merely "a bit slow"—complexity grows exponentially with qubit count, so the classical path hits a wall around the 74-qubit scale.

Independent Replication Adds Credibility

Quantinuum's trapped-ion quantum computer independently produced the same signature—a different physical route yielding the same answer, which is far stronger than self-reported speed claims. Mutual validation between trapped-ion and superconducting platforms is the most valuable aspect of this result.

> Concept note: Error mitigation vs. error correction > Error correction requires redundant qubits to fix errors in real time—engineering-wise extremely hard. Mitigation instead "washes out" noise via post-processing; it is far more accessible but has fidelity limits. QESEM's significance: with error correction not yet practical, software first extracts usable compute from NISQ devices to tackle problems classical machines can't immediately match.

Paradigm Shift

Error-correction camp: hard-fix with redundant qubits → a long-term goal, engineering-difficult
Error-mitigation camp: software post-processing to recover signals → usable today, as demonstrated by QESEM's advantage result

Bottom Line

Quantum computing narratives are often skewed by a "qubit-count arms race." The QESEM result reminds us that before fault-tolerant quantum computing arrives, error-mitigation software stacks may be the first to turn "quantum advantage" into a reproducible, verifiable, independently endorsed fact. Fugaku's failure to converge is the hardest evidence of that fact.

References

1. BlueQubit / Qedma / IBM / RIKEN, Quantum advantage result via QESEM on IBM Heron (2026-07-30). 2. Quantum Advantage Tracker, Heron 156-qubit Floquet Ising entry. 3. RIKEN, Fugaku classical simulation attempt (over 500,000 core-hours, non-convergence). 4. Quantinuum, independent trapped-ion reproduction of the same signature. 5. Qedma, QESEM error-mitigation software technical documentation.