Skip to content

Benchmarks Overview

This page summarises the performance benchmarks and validation milestones for the Q-Memory photonic platform across its development phases.

Phase 0 is a structured engineering experiment designed to answer specific pass/fail questions before committing to full-scale fabrication. It is not a performance benchmark — it is a component qualification run.

TestMetricPass Threshold
Waveguide propagation lossLoss per unit length< 2 dB/cm (Phase 0 target)
Phase element responsePower for π phase shift< 25 mW
Phase element speedResponse time< 10 µs
Beam splitter accuracyExtinction ratio> 25 dB
Optical coupling efficiencyInsertion loss per facet< 2 dB
Two-photon interferenceHong-Ou-Mandel visibility> 90%
Mesh programmabilityDecomposition error< 1%
Phase stabilityDrift rate< 5 mrad/minute

If these pass, Phase 1 is approved. Failures drive specific design changes rather than project cancellation.

Based on published literature from comparable silicon nitride photonic platforms:

ComponentTarget PerformanceLiterature Basis
Waveguide propagation loss< 0.5 dB/m (Phase 1+)Demonstrated in silicon nitride platforms
Beam splitter insertion loss< 1 mdB per 50:50 splitDirectional coupler literature
Waveguide crossing loss~1–2 mdB per crossingSilicon nitride crossing demonstrations
Edge coupler insertion loss~50–100 mdB per facetInverse taper demonstrations
Photon detection efficiency> 98% (cryogenic detectors)Superconducting nanowire literature
  • Entangled pair generation rate: Determined by pump power and ring resonator quality factor; targets MHz-scale rates on-chip
  • Photon indistinguishability: > 99% required for long-distance QKD; target > 90% for Phase 1 demonstration
  • Telecom wavelength compatibility: 1550 nm C-band — directly compatible with standard fibre infrastructure
  • Generation rate: Limited by detector dead time; target > 100 Mbit/s certified random bits
  • Certification: Quantifiable against quantum optics models — not pseudo-random
  • Classical hardness threshold: Estimated to require ~50 photons across ~100 modes for practical quantum advantage
  • Phase 1 target: Demonstrate interference at 16–32 photon scale; validate sampling fidelity

The optical matrix-vector multiplication performance scales with chip size and detector speed.

Matrix SizeOptical Propagation TimeDetector ReadoutTotal Latency
4 × 4 (Phase 0)< 1 ns~10 ns~10 ns
64 × 64 (Phase 1)< 5 ns~10 ns~15 ns
256 × 256 (Phase 2)< 20 ns~10 ns~30 ns

For comparison, a conventional GPU performing a 256 × 256 matrix multiplication is typically limited by memory bandwidth — reading the weight matrix at ~100 ns–10 µs depending on caching.

A key benchmark for the photonic platform’s AI acceleration mode is the power saved by using non-volatile optical memory versus continuous phase control:

Network size (optical elements)Continuous thermal phase controlNon-volatile optical memory
64 elements~1 W0 W
256 elements~4 W0 W
1024 elements~15 W0 W

This saving is structural — the non-volatile optical memory holds the weight matrix configuration with zero ongoing power. The saving scales linearly with network size.

Fault-tolerant photonic quantum computing requires keeping total photon loss below approximately 2.7% per photon path (based on published fault-tolerance thresholds for fusion-based quantum computing architectures).

Current best-demonstrated systems in research literature are estimated at approximately 3.2% per photon — just above the threshold. The Phase 1 design targets closing this gap.

Loss sourceEstimated contribution
Photon source coupling into waveguide~0.7%
Chip edge coupler~1.2%
Waveguide routing (10 cm path)~0.01%
Beam splitter elements~0.1% per element
Detector coupling and efficiency~0.5–1.0%
Total (estimated)~2.5–3.5%

Achieving < 2.7% requires careful management of every element in the path and is the primary engineering constraint for fault-tolerant operation.