Google Sycamore

Overview

Google Sycamore is one of the most famous quantum processors ever built. In October 2019, Google researchers published a paper in Nature titled 'Quantum supremacy using a programmable superconducting processor', reporting that a 53-qubit chip had completed a specific random-circuit sampling task in about 200 seconds. They estimated that the same task would take the world's best supercomputer about 10,000 years.

The result was celebrated and criticized in equal measure. It showed that a programmable quantum chip could perform a computation that was extremely hard to simulate classically, but the task had no practical use, and later classical algorithms reduced the gap considerably. Today Sycamore is best understood as the experiment that proved superconducting chips could be controlled at this scale, rather than as a lasting quantum advantage.

History and background

Google's quantum effort moved from annealing to gate-based hardware in 2014, when it recruited John Martinis and his superconducting-qubit group from the University of California, Santa Barbara. The team first built a 72-qubit chip called Bristlecone in 2018, which did not reach the target performance, and then switched to a design with tunable couplers that became Sycamore.

The 2019 paper described a 54-qubit device in which one qubit was faulty, leaving 53 usable qubits and 86 couplers. IBM published a response arguing that the task could be done classically in days with a smarter use of storage, and further simulation work, including tensor-network methods run on large GPU clusters, reported times of hours or less. In 2023 Google followed up with experiments on a 70-qubit Sycamore-class device and with a surface-code error-correction demonstration. By 2024 the company had moved on to Willow, which delivers much better qubit quality.

How it works

Sycamore's qubits are transmons, superconducting circuits whose two lowest energy levels represent 0 and 1. Each qubit has a frequency that can be tuned with a magnetic flux pulse, and neighboring qubits are linked by tunable couplers. When a coupler is switched on for a short time, the two qubits swap energy in a way that implements a two-qubit gate; when it is off, the qubits are nearly isolated. This design reduces unwanted interactions between idle qubits.

In the supremacy experiment, Google ran random circuits: sequences of random single-qubit gates alternating with fixed two-qubit gates. After 20 cycles, the chip's output is a bitstring, and the probability of each bitstring follows a distribution that is difficult to compute classically for 53 qubits. Google compared the experiment's outputs against classical simulation on smaller instances using a statistic called cross-entropy benchmarking, and then extrapolated the classical cost for the full circuit.

Fidelity matters because the entire experiment ran with a total circuit fidelity of only about 0.2%. In other words, the signal was small but statistically detectable over millions of samples. That fragility is why later Google work shifted to error correction, where each added qubit is meant to improve reliability instead of adding noise.

Connectivity

2D grid: each qubit connects to its four nearest neighbors through tunable couplers (86 couplers in the 53-qubit device)

Specifications in detail

Key features

• 53 working qubits (54 fabricated, one non-functional) on a 2D grid
• Tunable couplers between neighbors that allow fast, high-fidelity two-qubit gates
• Random circuit sampling of 20 cycles in about 200 seconds
• First widely discussed experimental claim of a beyond-classical computation

Limitations

• The sampling task has no known practical application
• Classical algorithms later cut the estimated simulation time from 10,000 years to days, hours, or less
• Qubit coherence and gate error are well behind Google's later Willow chip
• Not available for public cloud access

Timeline

What it is used for

• A benchmark of beyond-classical sampling, not a practical application
• A platform for early error-correction experiments and for studying noise
• A reference point for later claims, including Willow's random circuit sampling result

How it compares

• Compared with Willow, Sycamore has about half the qubits, noticeably higher error, and shorter coherence times.
• Compared with IBM Eagle (127 qubits, 2021), Sycamore has fewer qubits but a more compact grid layout and tunable couplers.
• Compared with Quantinuum H2, Sycamore has much faster gates but lower gate fidelity and only nearest-neighbor connectivity.

Technical notes

The Sycamore experiment is a case study in how advantage claims evolve. IBM argued within days that a better classical method using disk storage could finish the task in about 2.5 days. Later tensor-network simulations on GPU clusters reported even shorter times. Google's own later experiments moved to larger circuits and to error-correction benchmarks, partly because the supremacy task proved a moving target.

Why it matters

Sycamore marked the first time a large technology company claimed to have crossed a clear line between quantum and classical computation. Even though the specific claim was revised, the experiment pushed the field to define better benchmarks and to take classical simulation seriously.

It also illustrates a lesson that applies to every advantage claim: a result is only as strong as the best classical algorithm available when it is checked. Anyone reading about quantum supremacy or advantage should ask what the classical baseline was, and who has tried to beat it.

Frequently asked questions

Keep exploring

• Company profile: Google (Google Quantum AI)
• Related processor: Google Willow
• Related processor: IBM Eagle
• Related processor: Quantinuum H2

This profile is written for educational purposes and summarizes company announcements, published papers, and vendor specification sheets. Last reviewed: September 30, 2026. Figures marked as company-reported have not all been independently verified, and benchmarks differ between vendors, so compare them with care.

← Back to Hardware Database