Making Sense of Quantum Performance - A Look Inside Google's Willow Chip (1)

Hi everyone,

Let’s be honest: keeping up with quantum computing can at times feel like trying to drink from a firehose. Every few weeks there’s a new chip, a new benchmark, an updated roadmap, and a flood of numbers. How do you figure out what’s actually a big leap forward? How do you interpret the metrics that are typically listed or mentioned in headlines?

That’s what we’re doing today. And we’ll do it through an example: taking a close look at one of the most advanced quantum processors to date - Google’s Willow chip - and breaking down its performance stats, one by one.

The Foundation: Qubits and Coherence

A quantum processor’s potential starts with its fundamental hardware. For a quantum chip, the key factors are its physical qubits, how long they can maintain their quantum state, and how they are connected.

Qubits and Connections: Willow packs 105 superconducting qubits onto its chip, each linked to about three or four neighbors through dedicated elements known as couplers. Both the qubits and these couplers are flux-tunable, so by applying magnetic flux through dedicated on-chip lines, their frequencies can be adjusted.

Staying Quantum (Coherence): 

A qubit is only useful as long as it can maintain its delicate quantum state. This lifetime is called coherence and is measured by two different “clocks”:

T₁ (the energy clock) measures how long an excited qubit (in state ∣1⟩) takes to lose its energy and relax back to the ground state (∣0⟩).

T₂ (the phase clock) is more subtle, and often the bigger challenge for quantum computation. It measures how long a qubit can preserve a well-defined relationship between the ∣0⟩ and ∣1⟩ states before random fluctuations cause that relationship to drift.

On Willow, the chip built for running algorithms achieves T₁ times close to 100 microseconds (µs) across all qubits, a strong result at this scale. The chip designed for error correction comes in lower at 68 µs, a deliberate trade-off to accommodate the dense wiring and controls required for those experiments.

Putting the Qubits to Work: Precision and Speed

So, we have good qubits that can maintain their quantum state. The next step is to perform operations on them. In a quantum computer, these fundamental operations are called "gates." The quality and speed of these gates determine the real-world capability of the processor.

First, let's talk about Gate Fidelity: This is a direct measure of how accurately each gate operation is performed. A score of 100% would mean a perfect, error-free operation every time. In reality, tiny imperfections in control signals and environmental noise always introduce some errors. Minimizing this error rate is one of the central challenges in building and operating quantum processors.

On the chip optimized for random circuit sampling, Willow's performance in this area is state-of-the-art:

• Single-Qubit Gates, which operate on one qubit at a time, have an average error rate of just 0.036%. This corresponds to 99.964% fidelity.

• Two-Qubit Gates, which are more complex and essential for creating entanglement between qubits, have error rates as low as 0.14% (99.86% fidelity).

It is vital to understand that these numbers are measured while the entire chip is operating simultaneously. Testing a gate in isolation can produce artificially optimistic results. These figures reflect the processor's ability to maintain precision amidst the background activity of a full-scale computation, which is a much more difficult and meaningful achievement.

Beyond the Gates: Measurement and Reset

Just as important as performing gate operations accurately is reading out qubit states with high fidelity. Even a perfectly executed algorithm is useless if the result can’t be measured reliably. This is where measurement fidelity comes in. It quantifies how accurately the processor can determine whether a qubit is in the ground state (∣0⟩) or the excited state (∣1⟩).

Measurement doesn’t only happen at the end of a computation. In protocols like quantum error correction, measurements must be performed mid-circuit and repeated many times without disturbing the rest of the system. For this repetitive scheme on the dedicated QEC chip, Willow reports an error rate of about 0.77%.

Once a qubit has been measured, it must be quickly and reliably prepared for reuse. This is handled by reset. A simple reset takes a qubit from ∣1⟩ back to ∣0⟩, but real superconducting qubits are not perfect two-level systems: they can occupy higher energy states (∣2⟩, ∣3⟩, etc.). If a qubit “leaks” into one of these non-computational states, it will no longer respond correctly to standard gates, effectively removing it from the computation.

To address this, Willow uses advanced reset protocols:

• Multi-level reset: Actively cools a qubit from any excited state, whether ∣1⟩ or higher, back to the ground state (∣0⟩).

• Targeted leakage removal: Efficiently returns a qubit that has leaked specifically to the ∣2⟩ state back into the computational subspace.

From Components to System Performance…

The metrics we’ve detailed - qubit count, coherence times, and operational fidelities - are the universal language of quantum hardware. They form a foundational scorecard agreed upon by the entire community, allowing us to assess the quality of the basic building blocks of any quantum processor.

But as crucial as they are, these numbers only tell half the story. They measure the parts, not the system. The real test is how those parts perform together in a large, coordinated computation - an area with fewer standards, where the rules and benchmarks are far less clear.

Next time, we’ll step into exactly that space and look at how Google Quantum AI evaluates Willow’s full-system performance.

Talk to you soon,

References

[1] Blog Post: ‘Meet Willow, our state-of-the-art quantum chip’

[2] "Quantum error correction below the surface code threshold." Nature 638, no. 8052 (2025): 920-926.