Quantum Hardware Database
IQM Garnet
IQM Garnet is a 20-qubit superconducting quantum processor with a square-lattice layout, a median CZ gate fidelity of 99.51%, and genuine 20-qubit entanglement demonstrated in a GHZ-state test. It is available on Amazon Braket and IQM's own cloud and has been installed at customer sites.
Qubit Count
20
Qubit Type
Superconducting (flux-tunable transmon qubits with tunable couplers)
Architecture
Square lattice (rotated 45 degrees, forming an effective 5-by-5 grid) with tunable couplers
Released
2023 to 2024 (benchmarks February 2024; cloud access from May 2024)
Gate Error (2Q)
Median two-qubit (CZ) fidelity of 99.51% across 30 pairs, with the best pair at 99.8%; single-qubit fidelity about 99.9%
Coherence Time
Median T1 of about 40 microseconds and median T2* of about 25 microseconds (2024 benchmark)
Operating Temp
Below 20 millikelvin inside a dilution refrigerator
Cloud Access
Yes — Amazon Braket and IQM Resonance; also deployed on-premises
Company
IQM
Overview
IQM Garnet is the 20-qubit processor at the heart of IQM's current product range for research and education. Its qubits sit on a square lattice, which matches the layout used for surface-code error correction, and each pair of neighbors is linked by a tunable coupler that can turn their interaction on and off. IQM describes it as the representative of its qubit crystal family, the same technology used in its larger on-premises Radiance systems.
Garnet is available in several ways. Users can run programs on it through Amazon Braket or IQM's own Resonance cloud, and customers such as national laboratories and universities have bought systems built around it.
History and background
IQM was founded in 2018 as a spin-off from Aalto University and VTT in Finland. It built several generations of smaller processors before announcing Garnet. In February 2024 the company published benchmarks for the 20-qubit computer, including the median CZ gate fidelity of 99.51% and a 20-qubit GHZ state, and in May 2024 Garnet became available on Amazon Braket, which was the first time IQM computers were offered through a self-service cloud.
Garnet's architecture was designed to scale. IQM followed it with Emerald, a 54-qubit processor on the same architecture with 99.5% median CZ fidelity, and plans to scale the square-lattice design to 150 qubits. In 2024 IQM also reported a 99.9% two-qubit gate fidelity on a planar transmon in a research device, and it has developed a Star topology that uses a central resonator to link qubits.
How it works
Garnet's qubits are flux-tunable transmons, superconducting circuits whose frequency can be adjusted with a magnetic flux. They sit on a square lattice rotated by 45 degrees, forming an effective 5 by 5 grid with 20 qubits in the occupied sites. Between each pair of neighboring qubits is a tunable coupler, itself a flux-tunable transmon. When the coupler is set to one frequency, the qubits are decoupled, and when it is moved, the qubits interact.
The native gates are a phased X rotation (PRX) for single qubits and a controlled-Z (CZ) for pairs. A CZ takes about 20 to 40 nanoseconds and is carried out by moving qubit frequencies toward an avoided crossing for a short time. The tunable-coupler design allows fast gates while keeping idle crosstalk low, and IQM reports a CZ with fidelity of 99.81% in a research version of the architecture.
A system built around Garnet includes IQM's in-house control electronics, a dilution refrigerator, and software for calibration and job management. IQM publishes benchmark data, including randomized benchmarking, quantum volume, and GHZ fidelity, so users can compare the system with others on a consistent basis.
Connectivity
Square lattice: each qubit connects to up to four neighbors through tunable couplers (30 qubit pairs)
Specifications in detail
| Item | Detail |
|---|---|
| Native gates | PRX single-qubit rotation and CZ two-qubit gate |
| Gate time | About 20 to 40 nanoseconds |
| Median T1 / T2* | About 40 microseconds / 25 microseconds |
| Readout | High-fidelity readout with median error in the low single-digit percent range |
| Larger sibling | IQM Emerald (54 qubits, same architecture) |
| Product link | Garnet's qubit crystal is the basis of IQM Radiance on-premises systems |
Key features
- 20 qubits in a square lattice with tunable couplers
- Median two-qubit CZ fidelity of 99.51% across 30 pairs; best pair 99.8%
- Median single-qubit fidelity about 99.9%
- 20-qubit GHZ state fidelity above 0.5, confirming genuine multi-qubit entanglement (about 0.62 with readout error mitigation)
- Available on Amazon Braket since May 2024
Limitations
- 20 qubits is small compared with IBM or Google chips
- Coherence times of around 40 microseconds limit circuit depth
- Raw GHZ fidelity falls below 0.5 for more than 12 qubits without readout error mitigation
- Tunable-coupler control adds calibration complexity
Timeline
| When | What happened |
|---|---|
| 2018 | IQM is founded in Espoo, Finland. |
| February 2024 | IQM publishes 20-qubit benchmarks including 99.51% median CZ fidelity. |
| May 2024 | Garnet becomes available on Amazon Braket. |
| 2024 | IQM demonstrates 99.9% two-qubit fidelity on a research planar transmon and introduces the Star topology. |
| 2025 to 2026 | Larger Radiance systems built on the same crystal technology are delivered to supercomputing centers. |
What it is used for
- Teaching and training in quantum programming
- Testing algorithms with surface-code-compatible connectivity
- Hybrid HPC research when combined with supercomputers
- Benchmarking and calibration research
How it compares
- Compared with IBM Heron (156 qubits), Garnet has far fewer qubits and a similar two-qubit fidelity range, with an open, published benchmark set.
- Compared with the RIKEN-Fujitsu 64-qubit computer, it is a smaller European counterpart built with in-house electronics.
- Compared with Rigetti Novera (9 qubits), it is more than twice as large, with a comparable approach to tunable couplers.
Technical notes
A GHZ state is a maximally entangled state of all qubits. If its measured fidelity is above 0.5, the system must contain genuine multi-qubit entanglement across all qubits. IQM reported a fidelity above 0.5 for 20 qubits after readout error mitigation, with a value of about 0.62. It is a useful, if demanding, system-level test, because every error in every gate and measurement lowers the score.
Why it matters
Garnet is a good example of a mid-sized processor that is useful for learning and for early research, and its published benchmarks show how fidelity and entanglement tests are done in practice. It also anchors IQM's strategy of selling on-premises systems that connect with supercomputers.
For readers comparing machines, Garnet is a reminder that fidelity numbers should come with context: median versus best-pair values, the pairs and dates measured, and whether readout error mitigation was applied.
Frequently asked questions
How many qubits does IQM Garnet have?
20 superconducting qubits on a square lattice.
Can I use IQM Garnet?
Yes. It has been available through Amazon Braket since May 2024 and through IQM's Resonance cloud. Pricing and availability change.
What is a GHZ state?
A GHZ state is an entangled state of all qubits in the system. A measured fidelity above 0.5 proves genuine multi-qubit entanglement, and IQM reported this for all 20 qubits.
What is the difference between Garnet and Emerald?
Emerald has 54 qubits and uses the same square-lattice, tunable-coupler architecture, with a similar median CZ fidelity of about 99.5%.
What is a tunable coupler?
A small circuit between two qubits whose frequency can be adjusted to switch their interaction on or off, which reduces unwanted crosstalk.
Keep exploring
- Company profile: IQM
- Related processor: IBM Heron (r1, r2, and r3)
- Related processor: Rigetti Novera QPU
- Related processor: RIKEN-Fujitsu 64-Qubit Superconducting Quantum Computer
- Related processor: Origin Wukong
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.