New technical note AGE-PN-001: reproducible benchmark for WDM stability See evidence
LAB 01 · Photonic circuit laboratory

Interactive photonic circuit

A lab bench for dual-rail photonic circuits: assemble the circuit part by part, watch phasors add at each coupler and solve six challenges where identifying and compensating coherent errors is exactly what AGE does.

Assembly bench

Dual-rail circuit

SOI 220 nm waveguide at 1550 nm (typical values, approx.): n_eff ≈ 2.44, n_g ≈ 4.2, dn/dT ≈ 1.8×10⁻⁴ K⁻¹.
Transfer matrices

The circuit is a product

Phasors and outputs

Amplitude and phase at the cut

Interference and visibility

Fringe explorer

Lab challenges

Six challenges, from basic physics to AGE logic

Each challenge loads its circuit on the bench; parts marked fixed cannot be touched, like an already-fabricated chip. The tags show which AGE step it rehearses: identify, compensate or track.

Why it works

Seven results behind the lab

REF = established result in the literature; VER = verified numerically in this lab; HEUR = heuristic reading for AGE, not a proved result.

1

The directional couplerREF

MDC = √(1−L) · [[ √(1−η), i√η ], [ i√η, √(1−η) ]]

η is the power fraction that crosses over. The i on the off-diagonal is mandatory: without that 90° shift the matrix would not be unitary and energy would not be conserved.

2

Unitarity = energy conservationREF

U†U = I  ⟺  |a1|² + |a2|² is conserved

With loss, U stops being unitary: its singular values drop below 1 and measure how much light is lost for the best and worst input. The matrix panel shows them live.

3

The Mach-Zehnder interferometerREF

UMZI = MDC · diag(eiφ, 1) · MDC  ⟹  P1 = sin²(φ/2),  P2 = cos²(φ/2)

With 50:50 couplers, the phase difference between arms decides which output the light goes to. It is a switch and, at intermediate φ, a tunable splitter.

4

Visibility and unequal lossVER

V = (Pmax − Pmin)/(Pmax + Pmin) = 2√T / (1 + T),   T = 10−L/10

For a 50:50 MZI with a loss of L dB in one arm. Inverted, the formula turns a measured visibility into the hidden loss: 3 dB gives V = 0.943, and the lab reproduces it to machine precision.

5

Imperfect couplers limit extinctionVER

Pbar,min = (√((1−η1)(1−η2)) − √(η1η2))²,   zero only if η1 + η2 = 1

With two 42 % couplers the bar output never drops below 0.0256: 15.9 dB extinction, however the phase is tuned. For both outputs to switch you need η₁ = η₂ = 0.5. With one added tunable coupler the bench tops out at 21.9 dB. With two, one before and one after and both at 50:50, the assembly switches fully again: this is Miller’s (2015) compensation with redundant tuning.

6

Two photons: the Hong-Ou-Mandel effectREF

P11 = |U11U22 + U12U21|² = (1 − 2η)² for a lossless coupler

Two identical photons entering one per port of a 50:50 always leave together: coincidences vanish. It is the matrix permanent, not the determinant, and it is the most direct test that two temporal modes are indistinguishable.

7

Why this is AGEHEUR

Ureal = Uideal · Ecoherente  →  identify E, compensate with tunable parts, track its drift

A coupler error or a thermal drift is not random noise: they are coherent, deterministic and therefore identifiable errors. Any U(2) is synthesised with a tunable coupler and one phase (Reck 1994, Clements 2016), and imperfect components can be compensated with redundant tuning (Miller 2015). This lab rehearses that logic in 2×2; it does not reproduce the AGE-T algorithm.

Core tests against the analytic laws

Auditor's findings

What this lab does not model

  • Monochromatic light and a single mode per waveguide; no coupler dispersion with λ.
  • No reflections or cavities: forward propagation only.
  • Temperature acts only through the length difference ΔL; a different thermal arm is not modelled.
  • SOI indices are approximate typical values, not a specific PDK.
  • For two photons, identical temporal modes unless compared with distinguishable photons.
  • No detection noise or finite statistics: powers are exact.
  • Random challenges (loss detective) draw the hidden loss on each attempt; the sealed record stores the circuit, not the solution.

Save and share the circuit

The current circuit as JSON. Edit it, or paste someone else's and load it.

Reproducible, sealed record

Current circuit, conditions and results with their SHA-256 fingerprint, to document a measurement or pre-register a prediction.