Xanadu Borealis

Overview

Xanadu Borealis is a photonic machine built around a clever trick: instead of creating 216 separate light sources, it generates a train of 216 pulses of squeezed light, one after another, and then makes them interact by sending them around loops of optical fiber with programmable delays. The result is a large entangled state of light, which is measured by detectors that count photons.

The machine performs one specific task, called Gaussian boson sampling. In June 2022 Xanadu reported in Nature that Borealis completed it in about 36 microseconds, while the best known classical algorithm would take more than 9,000 years on a leading supercomputer to produce a single exact sample. The machine was also placed on the cloud so that outside users could run and test it.

History and background

Xanadu was founded in Toronto in 2016 and developed photonic devices in a series of steps. Its earlier X8 and X12 chips, with up to 12 squeezed-state qubits, were available through the cloud. Borealis scaled the approach to 216 modes by switching from a chip with many physical paths to a time-multiplexed design in which pulses travel through the same optical components in sequence. The work was done in collaboration with the National Institute of Standards and Technology, which supplied superconducting photon-number-resolving detectors.

After the Nature paper, other groups, including those behind Chinese Jiuzhang experiments, tested the limits of the advantage claim, and some proposed classical algorithms that sample from related distributions faster than expected. Xanadu itself moved on to Aurora, a modular and error-corrected architecture announced in 2025, and today positions Borealis as an important proof of principle.

How it works

A pulsed laser pumps a nonlinear crystal or waveguide, which produces squeezed light: pulses whose quantum noise has been reshaped so that one property is more precisely defined than normal. Borealis generates 216 of these pulses in sequence. They then pass through three loops of fiber with different lengths, in which each pulse can be combined with its neighbors, with pulses 6 positions away, and with pulses 36 positions away. Programmable phase shifters and beam splitters set the strength of each interaction, which lets the user choose the interferometer's overall transformation.

At the end, the pulses are split across 16 detection channels, and photon-number-resolving detectors count how many photons arrive in each time bin. A single sample is a list of 216 integers. Because the interferometer is large and the input states are quantum, the probability of each possible list depends on a mathematical quantity called a hafnian, which is extremely hard to compute exactly for large matrices. That hardness is the basis for the advantage claim.

Loss is the main enemy. Every optical element absorbs a little light, and the Borealis experiment is validated by comparing the measured distribution with several classical models, including ones that simulate noise, to check that the output does depend on genuine quantum interference. This statistical validation is typical of sampling-based advantage claims.

Connectivity

Programmable three-dimensional entanglement structure created with three delay loops (couplings between modes 1, 6, and 36 time bins apart)

Specifications in detail

Key features

• 216 squeezed-light modes in a programmable interferometer
• Three-dimensional connectivity from three time-delay loops
• Photon-number-resolving detection with events of up to 219 photons
• Sampled in about 36 microseconds, according to the 2022 Nature paper
• Opened to outside users through Xanadu Cloud and Amazon Braket

Limitations

• Gaussian boson sampling is a specialized task, not a general-purpose computation
• Photon loss limits the accuracy of large circuits
• Later classical algorithms and approximations have narrowed the claimed gap for related experiments
• Not a universal, fault-tolerant quantum computer

Timeline

What it is used for

• Demonstrating quantum advantage in sampling and testing classical simulation methods
• Research on graph problems and molecular vibrational spectra, which can be mapped to Gaussian boson sampling
• A teaching example of continuous-variable quantum computing

How it compares

• Compared with Google Sycamore, Borealis is a different kind of advantage experiment: sampling light instead of sampling qubit circuits.
• Compared with gate-based machines, Borealis is a special-purpose sampler and cannot run Shor's or Grover's algorithms.
• Compared with Xanadu's Aurora, Borealis is a proof of advantage while Aurora targets fault tolerance.

Technical notes

Borealis does not use qubits in the usual sense. It uses squeezed states of light, which are continuous-variable systems, and its output is a list of photon counts for each of 216 time bins. Xanadu's estimate of 9,000 years refers to the time for exact classical sampling with the best known algorithms on a leading supercomputer. Classical approximate methods can sample faster at lower fidelity, so the claim is that Borealis matches the target distribution at a cost classical methods struggle to meet.

Why it matters

Borealis is the clearest example of a photonic quantum advantage experiment that outsiders could actually use. It broadened the range of physical platforms that have shown beyond-classical sampling beyond superconducting chips.

It also shows how much specifics matter. Gaussian boson sampling is a hard problem for classical computers, but it is also a narrow one, and the real test for photonics is whether it can reach error-corrected, general-purpose computing.

Frequently asked questions

Keep exploring

• Company profile: Xanadu Quantum Technologies
• Related processor: Google Sycamore
• Related processor: D-Wave Advantage2
• Related processor: Pasqal Fresnel (Orion generation)

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