For decades, the central promise of quantum error correction (QEC) has rested on a conditional: if physical error rates fall below a critical threshold, then adding more qubits should exponentially suppress logical errors rather than compound them. In December 2024, a Google Quantum AI team reported crossing that threshold on their Willow superconducting processor, presenting a distance-7 surface code operating at 0.143% error per correction cycle. The result is significant, but the distance between a laboratory demonstration and a fault-tolerant quantum computer remains substantial.
The Threshold Problem in Context
Quantum error correction encodes a single logical qubit across many physical qubits, using syndrome measurements to detect and correct errors without destroying quantum information. The surface code—the most studied QEC architecture—arranges qubits on a two-dimensional lattice and tolerates relatively high individual qubit error rates compared to other codes. However, the approach only works when physical error rates sit below a critical threshold, typically estimated near 1% for the standard depolarizing noise model.
Prior experiments had demonstrated surface code elements, but none had convincingly shown that increasing the code distance—the number of qubits encoding one logical qubit—actually reduces the logical error rate. The theoretical prediction is exponential suppression characterized by a factor Lambda: each increment in code distance should multiply the error suppression by Lambda. Below threshold, Lambda exceeds 1; above threshold, adding qubits makes things worse.
What Willow Achieved
The Google team (Acharya et al., 2024) operated two surface code memories on the Willow processor. The headline result is a distance-7 code using 101 physical qubits, achieving a logical error rate of 0.143% ± 0.003% per cycle of error correction. Critically, when the team compared performance between their distance-5 and distance-7 codes, they measured Lambda = 2.14 ± 0.02—meaning each two-step increase in code distance roughly halved the logical error rate.
The logical memory also exceeded breakeven: the encoded logical qubit survived 2.4 times longer than the best individual physical qubit on the chip. This is a meaningful milestone because it demonstrates that the collective encoding genuinely outperforms its components, rather than merely redistributing errors.
A second notable achievement was real-time decoding. The distance-5 code operated with a real-time decoder achieving 63-microsecond average latency at cycle times of 1.1 microseconds, sustained over a million correction cycles. Real-time decoding is essential for any practical quantum computation, since classical post-processing delays would negate the advantage of error correction during active computation.
Critical Assessment: What the Numbers Do Not Say
Several aspects of this work warrant careful interpretation.
Lambda of 2.14 is modest. While any Lambda above 1 confirms below-threshold operation, practical fault-tolerant quantum computing requires logical error rates many orders of magnitude below current levels—typically 10^-10 or better for useful algorithms. Achieving such rates with Lambda = 2.14 would require extremely large code distances, translating to millions of physical qubits. The relationship between current Lambda values and what might be achieved with larger, more optimized chips remains an open extrapolation.
Correlated errors impose limits. The team ran repetition codes up to distance 29 and found that logical performance was ultimately limited by rare correlated error events occurring approximately once per hour (roughly 3 x 10^9 cycles). These events, which simultaneously affect multiple qubits, fall outside the independent error model on which threshold estimates are based. How these correlations scale with chip size and code distance is not yet well understood.
The gap between memory and computation is wide. Willow demonstrated a quantum memory—maintaining a stored logical state—not a fault-tolerant computation. Executing logical gates between surface-code qubits, particularly the non-Clifford gates required for universal quantum computing, introduces additional error channels and architectural complexity through techniques like magic state distillation. Memory experiments, while necessary, test a simpler operational regime.
Resource overhead remains daunting. Even optimistic extrapolations from these results suggest that running algorithms like Shor's factoring or quantum chemistry simulations at useful scales will require physical qubit counts in the millions, well beyond the 105-qubit Willow chip. The engineering challenges of scaling superconducting qubit systems—wiring, cooling, cross-talk mitigation—grow nonlinearly with qubit count.
Broader Implications for the Field
The Willow results do confirm a core theoretical prediction of QEC: below-threshold operation with genuine error suppression is physically achievable. This lends empirical support to the entire fault-tolerant quantum computing program, which has operated largely on theoretical guarantees since Shor's 1996 threshold theorem.
The demonstration also creates a concrete performance benchmark. Other platforms—trapped ions, neutral atoms, photonic systems—can now be compared against a clearly defined target: Lambda > 2 at distance 7 with real-time decoding. Competition among hardware platforms is likely to accelerate as a result.
However, the quantum computing community should resist interpreting this as evidence that fault-tolerant quantum advantage is imminent. The path from Lambda = 2.14 at distance 7 to the error rates needed for practical quantum algorithms spans multiple orders of magnitude in both qubit quality and qubit count.
Open Questions
- Can Lambda be significantly improved through better qubit fabrication, or does it represent a fundamental characteristic of superconducting platforms at this stage?
- How do correlated error events scale as chip sizes increase, and can they be mitigated through architectural innovations?
- What is the realistic timeline for demonstrating fault-tolerant logical gates (not just memory) below threshold?
- Will competing QEC codes—such as the recently proposed quantum LDPC codes—offer better scaling characteristics than the surface code?
Closing Reflection
Google's Willow result is a genuine scientific milestone: the first clear demonstration that quantum error correction works as theory predicted, with error rates suppressed by increasing code distance. Yet milestones are not destinations. The engineering, physics, and systems challenges that remain between this proof-of-principle and practical fault-tolerant quantum computing are formidable. The field has crossed a threshold in the laboratory; crossing it at scale is the work of the coming decades.