Xanadu: Borealis and Photonic Quantum Advantage

Photonic Quantum Computing

Most quantum computers use superconducting qubits (IBM, Google) or trapped ions (IonQ, Quantinuum). Xanadu takes a fundamentally different approach: encoding quantum information in light.

Advantages of photonic quantum computing:

• Operates at room temperature (no dilution refrigerator)
• Photons naturally resist some forms of decoherence
• Integrates with existing fiber optic infrastructure for quantum networking
• Gaussian operations are deterministic; the challenge is non-Gaussian gates

Gaussian Boson Sampling (GBS)

GBS is the photonic equivalent of random circuit sampling. The task: send squeezed light through a programmable optical interferometer and record the photon-number detection outcomes.

Simulating the GBS output distribution classically requires computing the permanent of a large matrix - a problem believed to be classically hard (#P-hard). As the number of modes and photons grows, classical simulation time grows exponentially.

Borealis

Borealis has 216 programmable squeezed modes connected by a series of beamsplitters and delay loops. The programmable elements are:

• Squeezing amplitude per mode
• Beamsplitter angle at each coupling
• Phase shift per loop

This gives full programmability - unlike earlier GBS demonstrations that used fixed optical circuits.

import strawberryfields as sf
from strawberryfields import ops
import numpy as np

# A simplified 4-mode GBS program
prog = sf.Program(4)

with prog.context as q:
    # Squeezed inputs
    ops.Sgate(1.0) | q[0]
    ops.Sgate(1.0) | q[1]
    ops.Sgate(1.0) | q[2]
    ops.Sgate(1.0) | q[3]

    # Interferometer: beamsplitters + phase shifts
    ops.BSgate(np.pi/4, 0) | (q[0], q[1])
    ops.BSgate(np.pi/4, 0) | (q[2], q[3])
    ops.BSgate(np.pi/4, 0) | (q[1], q[2])
    ops.Rgate(np.pi/3) | q[0]

    # Photon-number detection
    ops.MeasureFock() | q

# Run on Xanadu Cloud hardware
eng = sf.RemoteEngine("borealis")
result = eng.run(prog, shots=1000)
print(result.samples)

The Advantage Demonstration

The Nature paper measured Borealis running GBS with up to 219 photons detected across 216 modes. For a representative problem instance:

• Borealis: 36 microseconds per sample
• Best classical algorithm (Metropolis sampling): estimated 9000 years for equivalent problem

The comparison used state-of-the-art classical algorithms, not naive simulation. Xanadu and collaborators developed new classical GBS algorithms specifically to provide a rigorous baseline.

Applications of GBS

Beyond the demonstration, GBS has proposed applications:

Molecular vibronic spectra: The mathematical structure of GBS matches the Franck-Condon factors governing molecular vibrations. Xanadu’s team used GBS to simulate the vibronic spectrum of formic acid.

Graph problems: GBS samples from distributions over graph matchings, connecting to problems in graph theory, drug discovery, and network analysis.

Quantum machine learning: GBS output distributions can serve as features for kernel-based classification.

# Using GBS for molecular vibronic spectra
from strawberryfields.apps import vibronic

# Formic acid molecular data
freq, Udisplace, alpha = vibronic.sample.formic_acid_data()

# Generate GBS program for vibronic spectra
prog = vibronic.sample.gbs_params(freq, Udisplace, alpha, n_mean=2)

Accessing Borealis

Borealis is available via Xanadu Cloud using Strawberry Fields:

pip install strawberryfields
sf configure --token YOUR_API_TOKEN

eng = sf.RemoteEngine("borealis")
result = eng.run(prog, shots=100)

Academic access is available for researchers. The hardware is genuinely different from superconducting systems and offers unique access to continuous-variable quantum computing.

Learn more: Strawberry Fields Reference

Sources

• Madsen et al., Quantum computational advantage with a programmable photonic processor, Nature 2022