Superconducting
First to claim quantum supremacy in 2019. Willow (2024) demonstrated below-threshold quantum error correction, a foundational milestone for fault-tolerant quantum computing.
About Google Quantum AI
Google's quantum computing effort began with the Xmon qubit architecture and the Bristlecone 72-qubit processor, before culminating in the 2019 announcement that Sycamore had performed a specific sampling task in 200 seconds that Google estimated would take a classical supercomputer 10,000 years. The result, published in Nature, is the most-cited quantum supremacy demonstration to date, though classical simulation algorithms have since closed some of that gap.
In December 2024, Google published results from Willow, a 105-qubit processor, demonstrating that surface code logical error rates decrease as the code distance grows past a threshold. This is the first experimental confirmation of below-threshold error correction, meaning adding more physical qubits to the error correcting code actively improves, rather than just changes, the logical qubit fidelity. It is a landmark result for the long-term viability of fault-tolerant quantum computing.
The primary SDK for Google's hardware is Cirq, an open-source Python framework for writing, simulating, and running quantum circuits. Cirq is designed around the specific gate sets and connectivity of Google's hardware, with native support for the fSim gate family that appears on Sycamore-architecture processors. Alongside Cirq, OpenFermion handles quantum chemistry, TensorFlow Quantum handles quantum machine learning, and ReCirq provides research-grade circuit tools used in Google's published experiments.
Hardware Specifications
| Specification | Value |
|---|---|
| Current systems | Willow (105q, 2024), Sycamore (53q, 2019), Bristlecone (72q, research) |
| Qubit technology | Superconducting transmon qubits (Xmon design) |
| Two-qubit gate fidelity | ~99.7% on Willow |
| Readout fidelity | 99%+ |
| Connectivity | 2D grid (nearest-neighbor) |
| Error correction | Below-threshold surface code demonstrated (Willow, 2024) |
| Benchmarking method | Cross-entropy benchmarking (XEB) |
| Cloud access | Google Cloud research program (application required) |
| Primary SDK | Cirq (open source, Python) |
| Supported SDKs | Cirq, OpenFermion, TensorFlow Quantum, ReCirq |
105 qubits
Google's 2024 flagship processor and the site of the below-threshold error correction milestone. Willow also demonstrated computational speed on a random circuit sampling benchmark that would take classical supercomputers an estimated 10 septillion years. Current focus for Google's research collaboration program.
53 qubits
The processor behind Google's 2019 quantum supremacy claim. Sycamore uses a 2D grid of Xmon transmon qubits with tunable couplers. It served as the primary research platform for time crystal demonstrations (2021) and many foundational algorithm experiments before Willow.
72 qubits
A 2018 research processor that established Google's path toward quantum supremacy. Bristlecone used a rectangular grid layout and demonstrated the scalability of Google's Xmon qubit fabrication process. It is a research predecessor, not an active production system.
Use Cases
Willow's 2024 demonstration that surface code error rates decrease as the code grows past a threshold is the defining result for the platform. Researchers focused on QEC will find no more relevant hardware.
Cross-entropy benchmarking (XEB) was developed by Google to certify that Sycamore's output was classically hard to simulate. Google hardware remains the benchmark for supremacy-style circuit sampling experiments.
OpenFermion maps fermionic molecular Hamiltonians to qubit operators for Cirq. The combination is Google's primary toolchain for variational quantum eigensolver experiments in computational chemistry.
Google's fSim gate is natively implemented on Sycamore-family hardware. Circuits that use fSim directly avoid decomposition overhead, making the hardware efficient for fermionic simulation problems.
TensorFlow Quantum integrates Cirq circuits into TensorFlow's differentiable programming model. Researchers can define hybrid quantum-classical models and train them with standard TF optimizers.
Google demonstrated a discrete time crystal on Sycamore in 2021, showing the platform's strength for studying non-equilibrium quantum phases of matter that are difficult to access classically.
Getting Started
pip install cirq
Cirq installs with all simulators included. No API key is needed to run circuits against the local simulator. For quantum chemistry, also install openfermion (which now includes the former openfermion-cirq integration). For QML, install tensorflow-quantum.
import cirq
# Define two qubits on a grid
q0, q1 = cirq.LineQubit.range(2)
# Build a Bell state circuit
circuit = cirq.Circuit([
cirq.H(q0),
cirq.CNOT(q0, q1),
cirq.measure(q0, q1, key='result'),
])
print(circuit)
# Simulate locally
simulator = cirq.Simulator()
result = simulator.run(circuit, repetitions=1000)
print(result.histogram(key='result'))
Cirq circuits use explicit qubit objects (GridQubit, LineQubit, NamedQubit). GridQubit matches Google's 2D hardware connectivity; LineQubit is convenient for linear experiments.
import cirq
import numpy as np
q0, q1 = cirq.GridQubit(0, 0), cirq.GridQubit(0, 1)
# fSim is Google's native 2-qubit gate (theta, phi parameterized)
fsim = cirq.FSimGate(theta=np.pi/4, phi=np.pi/6)
circuit = cirq.Circuit([
fsim(q0, q1),
cirq.measure(q0, q1, key='m'),
])
print(circuit)The fSim gate is native to Sycamore-family hardware. Writing circuits directly with fSim avoids decomposition overhead that occurs when using CNOT on Google hardware. See the Cirq reference for gate decomposition details.
import cirq
q0, q1 = cirq.LineQubit.range(2)
# Build a noise model approximating Sycamore-class hardware
noise = cirq.ConstantQubitNoiseModel(cirq.depolarize(p=0.005))
circuit = cirq.Circuit([cirq.H(q0), cirq.CNOT(q0, q1), cirq.measure(q0, q1, key='m')])
noisy_simulator = cirq.DensityMatrixSimulator(noise=noise)
result = noisy_simulator.run(circuit, repetitions=500)
print(result.histogram(key='m'))
Depolarizing noise with p=0.005 approximates Sycamore two-qubit gate error rates. Use noise modeling during algorithm development before applying for hardware access.
Google Quantum AI hardware access is not available via a public queue. Researchers can apply through the Google Quantum AI research program. Applications are reviewed based on scientific merit, research goals, and alignment with Google's quantum computing roadmap. Students and academic groups with strong proposals are encouraged to apply.
Access
Google does not offer a public pay-per-use hardware queue. Access to real QPUs is through the research collaboration program. However, free simulation via Cirq is available to anyone.
Application-based (free for approved researchers)
Qualified academic and research teams can apply for direct QPU access via the Google Quantum AI research program. Access is granted based on scientific merit and research alignment. No public queue exists.
View details →Research program / limited preview
Google Cloud provides access to quantum hardware through a limited preview program integrated with Google Cloud infrastructure. Availability is restricted; interested teams should apply via Google Cloud.
View details →Free local simulation
Cirq's built-in simulators run locally on any machine at no cost. The noise model simulators can approximate Sycamore-class hardware behavior, and statevector simulation is practical up to roughly 30 qubits on a standard workstation.
View details →Free cloud simulation
Google provides free Colab notebooks with Cirq pre-installed. All tutorials and algorithm examples on quantumai.google run in Colab against local simulators, with no hardware cost or account setup required.
View details →Learning Resources