New technical note AGE-PN-001: reproducible benchmark for WDM stability See evidence
Photonic control software

Stability for photonic systems that never stand still.

AGE turns temporal characterisation logs into measurable diagnostics and compensation paths tailored to each platform, PDK and version.

Free, bounded offline evaluation No new chip · no initial integration Explore
Measure before integration

Drift is not always noise. Sometimes it is structure the controller does not yet use.

When devices share a substrate, package and environment, part of their dynamics is shared too. AGE identifies that common and local structure to estimate what improvement is actually achievable against the existing baseline.

We begin with data the team already owns. If no robust advantage appears on a held-out time segment, the answer is not to integrate. If it does, we engineer the next phase around the real monitor, actuator and platform constraints.

See the method
Open package availableTechnical note + code + example data
Nominal improvement1.80×reproducible simulation
SISO result1.06×negative result preserved
LicenceMITauditable code
What we do

A software layer between telemetry and actuation.

One principle—learning temporal structure—applied with an explicit maturity level for every capability.

Evidence before promises

The first product is a technical decision.

The open benchmark shows where gain appears and, more importantly, where it does not: a better single-ring filter buys only 6%. Cross-channel information is what changes the nominal result.

reproducible simulation 1.80× relative RMS error reduction in the nominal scenario

16 channels · 8 hours · 60 runs. This is simulation evidence, not a performance guarantee for a real platform.

Free demonstration

One log. One baseline. One useful answer.

The entry demonstration lets a foundry or photonics team assess AGE without changing hardware, transferring unnecessary IP or committing to integration in advance.

Review data requirements
  1. 01

    Scope

    Fix the signal, period, baseline, metric and failure criterion before running.

  2. 02

    Evaluate

    Identify on one time block and score on another; the owner can keep the data on their machine.

  3. 03

    Decide

    Quantitative result, uncertainty, observed limitations and go / no-go recommendation.

Visible maturity

Demonstrated work and pending work do not share the same label.

Available now

  • Reproducible nominal benchmark.
  • Offline evaluation on multichannel logs.
  • Comparison against an agreed baseline.
  • Sealed-forecast protocol.

Requires platform validation

  • Platform-specific improvement factor.
  • Real-time closed-loop operation.
  • Final monitor and actuator interface.
  • Experimental temporal-mode compensation.
Frequently asked

Before sharing data.

First decision

Discover whether your data already contains a better control strategy.

Describe the platform and available log. Do not send confidential data in the first contact.