Neutral-atom hardware built for reproducibility
Strontium-88 atoms in optical tweezers, Rydberg entanglement, and an AI calibration layer that closes the loop between every gate operation.
Programmable 2D optical tweezer arrays
Each trap is an individually addressable qubit site. A spatial light modulator defines the trap geometry before each run, allowing circuit topology to be tuned per experiment.
- Trap depth ~1 mK
- Trap frequency (radial) 150-180 kHz
- Site-to-site spacing 3.0 micron (programmable)
- Loading efficiency >96% per site
- Atom species Strontium-88 (88Sr)
Rydberg blockade gates
When one atom is excited to a Rydberg state, the strong dipole-dipole interaction shifts the energy of neighboring atoms, preventing their simultaneous excitation. This blockade enables high-fidelity two-qubit CZ and CNOT gates without physical contact between qubits.
Measured via randomized benchmarking on our development configuration.
Fast gate operation relative to coherence time enables deep circuits.
Rydberg interaction range exceeds typical array spacing, enabling long-range connectivity without atom transport.
Closed-loop correction at every scale
Drift is continuous. Calibration must be too. The AI layer runs simultaneously with experiments, monitoring dozens of parameters and making micro-adjustments that are invisible to the experiment controller.
Trap frequencies and qubit transition frequencies are sampled thousands of times per second using non-destructive probe pulses interleaved with circuit operations.
A lightweight recurrent network trained on the system's own drift history predicts parameter evolution and pre-empts corrections before fidelity degrades measurably.
Each atom's trap is corrected independently. Spatially correlated drift (e.g., from a pointing fluctuation) is distinguished from uncorrelated single-site noise and handled with the appropriate response.
Early-access system parameters
| Parameter | Value | Notes |
|---|---|---|
| Qubit count (current config) | 100+ | Scalable by SLM pattern update |
| Single-qubit gate fidelity | 99.8% | Internal randomized benchmarking |
| Two-qubit gate (CZ) fidelity | 99.2% | Rydberg blockade; internal benchmarking |
| T1 coherence time | >4 s | Ground state population lifetime |
| T2 (Ramsey) coherence time | >1 s | With dynamical decoupling |
| Gate time (CZ) | 100-200 ns | Depends on inter-site spacing |
| Readout fidelity | 99.5% | Fluorescence imaging, per-qubit |
| Loading efficiency | >96% | Per-site, stochastic Zeeman loading |
| Calibration downtime | <5 min/session | vs. 2-4 hours without AI layer |
Specifications reflect development configuration as of mid-2026. Early-access partners receive updated parameter sheets with each system release.
Where we are going
First operational tweezer array with AI calibration loop running continuously. 50-qubit configuration with 99%+ single-qubit fidelity.
Array scaled to 100+ qubits. First research partners onboarded. Two-qubit fidelity benchmarks published.
Array density increase through improved SLM optics. REST API for remote experiment submission. Continued calibration layer tuning at scale.
First fault-tolerant logical qubits on the platform using surface code protocols adapted for the tweezer array geometry and long coherence times.
Request early access to the platform
Hardware access is currently by application for research groups and labs. Contact us to discuss your use case.