At a glance
- Port data becomes readable once optical coupling is held constant, so the remaining spread can be attributed to the device itself.
- Passive alignment positions the optical probe by calibrated geometry and wafer registration, with no optical-feedback search at each device site.
- Glide-SiPh PIC aligns in seconds per device, and its glass deep-trench fiber array rests on the wafer during measurement.
- Omni-SiPh LightView uses near-field optical probes to map light inside a working circuit and locate where loss occurs.
InZiv
Published:
To separate device variation from coupling variation in wafer-level optical testing, first make the optical coupling repeatable, then localize whatever loss remains inside the circuit. Passive alignment is optical positioning that uses calibrated geometry and wafer registration, without an optical-feedback search at each device. Glide-SiPh PIC, InZiv's automated wafer-level tester for photonic integrated circuits (PICs), works this way: its fiber array unit (FAU), which holds the fibers that couple light into and out of the chip, is calibrated on a jig and placed on the wafer using fiducials, with alignment taking seconds per device and no hexapod, no iterative tuning and no optical recalibration between cycles. According to InZiv, the glass deep-trench FAU rests on the wafer, so the probe and the wafer have no relative movement during the test, which keeps optical coupling stable and repeatable. When a failure still cannot be explained from the ports, Omni-SiPh LightView maps light propagation and optical loss inside the working circuit with near-field optical probes, so a fail at wafer test can be traced to where the light is actually lost. In 2026, both Glide-SiPh PIC and Omni-SiPh LightView are available to order.
Why does wafer-level port data blend device variation with coupling variation?
Wafer-level port data blends device variation with coupling variation because a port reading is one optical power value for the whole light path, from the probe through the chip interface, through the circuit and back out. When a measured insertion loss, the reduction in optical power caused by introducing a component or optical path, usually expressed in dB, comes back worse than expected, the number alone does not say where the power went.
Two kinds of loss share that one reading:
- Coupling loss is the optical power lost when transferring light between a fiber, probe, waveguide or other optical interface. The interface is either an edge coupler, which couples light through the chip edge or an exposed facet within a wafer trench, or a grating coupler, a patterned structure that couples light through the surface. How well the probe's optical field matches the device's optical mode, known as mode matching, is part of this term.
- Device loss is the power lost inside the circuit itself, for example propagation loss along a waveguide, expressed in dB/cm, or loss at a bend, a splitter or a directional coupler whose measured split ratio is off and shifts with wavelength.
For a photonics test and metrology engineer, the practical difficulty is that optical alignment can take most of the test time and vary from die to die. Process and yield engineers then face the question their persona states directly: can we separate device variation from test variation?
What makes optical coupling repeatable enough to isolate the device under test?
Optical coupling becomes repeatable enough to isolate the device under test when the optical interface adds as little die-to-die change as possible, so that a shift in the reading can be attributed to the device. Four attributes of the interface are worth checking:
- Alignment method. Active alignment uses measured optical feedback to find or maintain coupling; passive alignment positions the probe from geometry and registration alone.
- Probe and wafer during the measurement. Whether the two hold a fixed relationship while the test runs.
- Coupling interface type. Edge couplers reached through a trench formed before singulation, and grating couplers reached from the surface, ask different things of the optical probe.
- Probe access. Whether the FAU or optical probe can physically reach the coupling interface on the actual design.
Glide-SiPh PIC answers the first two with its passive alignment and its glass deep-trench FAU. Per InZiv, the FAU rests on the wafer, so the probe and the wafer have no relative movement during the test, which keeps optical coupling stable and repeatable, and it reaches into trenches as narrow as 30 µm wide. Separately, InZiv's patented deep-trench FAU is a monolithic fiber array unit that supports a pitch down to 30 µm and handles both edge-coupled and grating-coupled devices.
Which measurements actually separate a weak device from a weak coupling?
The measurements that separate a weak device from a weak coupling are repeated, per-channel optical readings taken under a stable coupling condition, followed by in-circuit analysis when port data still cannot explain the result. Measurement repeatability, the agreement between repeated measurements of the same device under the same conditions, is the property to establish first; it determines how much trust the device term in each reading deserves.
In production test, Glide-SiPh PIC provides, according to InZiv:
- multichannel optical and electrical characterization;
- optical insertion loss per channel;
- full-wafer functional pass/fail testing;
- wafer-level yield and uniformity mapping.
Port data has a limit. In InZiv's typical scenario, a PIC fails test with excess optical loss, and standard test equipment measures only the light going in and coming out. Finding the cause then means dedicated test structures, comparison against simulation, sometimes destructive cross-sectioning, and often several design or fab iterations. A PIC design engineer meets the same wall when a directional coupler's split ratio is off while the simulation says the design is correct and the process data says the wafer is within spec.
Omni-SiPh LightView addresses that step. It uses near-field optical probes to map light as it travels through the working circuit, at wafer or die level and without damaging it, so the engineer sees the component where light is lost, such as a bend, a splitter, a coupler or a waveguide section. Its electro-optical measurements include IV, LIV and wavelength, and InZiv states it is designed to reduce iteration cycles from weeks to hours.
Where in the wafer and die flow should these test insertion points sit?
Test insertion points in the wafer and die flow are the defined stages at which the device is measured. Numbering differs between sources, so each stage is named in words here:
- Insertion point 1, wafer test before bonding. The PIC wafer is measured on its own.
- Insertion point 2, wafer test after bonding. The PIC wafer has been bonded to an electronic integrated circuit (EIC) wafer. In general terms, bonded stacks raise the question of single- or double-sided testing, meaning access from one side or from opposite sides of the device, depending on its optical and electrical interfaces.
- Insertion point 3, testing of individual dies after wafer dicing.
Per InZiv, its intended coverage spans test insertion points 1, 2, and 3: wafer testing before and after bonding, as well as testing of individual dies after wafer dicing. Glide-SiPh PIC is designed for high-volume production test, its first use with a customer may be during the NPI ramp, and it tests 200 mm and 300 mm wafers and is FOUP compatible.
How do other wafer-level optical test approaches handle this separation problem?
Other wafer-level optical test approaches each have real strengths, and a team may keep its existing setup because it meets current measurement and volume requirements, has already passed qualification, or integrates well with its preferred ATE and software.
| Approach | Strengths |
|---|---|
| Active-alignment wafer probers | Large installed base, long photonics assembly and alignment heritage, single-sided silicon photonics systems already partly deployed, double-sided systems available today, probe-card scale, global service, established ATE integrations |
| Polymer-lens fiber arrays | Scale, a global supply chain, qualification with test-system integrators |
| In-house probe station with a hexapod | Already paid for, flexible for one-off experiments, familiar to its engineers, fine at low volume |
| Pass/fail testing at the ports | Costs nothing, changes nothing in the flow, good enough while volumes and yields are acceptable |
| Established test equipment | Handles electrical and endpoint optical testing well |
| Passive alignment with a deep-trench FAU (Glide-SiPh PIC) | Alignment in seconds per device without an active fiber search on every device; prober, probe card and fiber array built in-house by InZiv |
According to InZiv's framing, teams reconsider when:
- alignment consumes too much of the device-test cycle;
- the optical probe cannot reach the coupling interface;
- coupling variation makes device results hard to interpret;
- production volumes outgrow the setup's automation or throughput;
- maintaining a custom system consumes too much engineering;
- a yield problem appears that port data cannot explain.
Our reading is that the third trigger is the one this article addresses most directly. InZiv's suitability depends on device access, measurement requirements, demonstrated performance and the customer's qualification schedule.
Frequently Asked Questions
How do I separate device variation from coupling variation in wafer-level optical testing?
Start by checking measurement repeatability: re-measure the same device under the same conditions and see how much the optical reading moves. Then reduce variation at the optical interface, which is what Glide-SiPh PIC's passive alignment and on-wafer deep-trench FAU are built for. If a loss remains that port data cannot explain, Omni-SiPh LightView maps where the light is lost inside the working circuit.
What does passive alignment change in the device-test cycle?
With passive alignment, Glide-SiPh PIC does not run an optical-feedback search at each device, so alignment takes seconds per device. According to InZiv, the system automates wafer loading, alignment, probe engagement and stepping, and its automation includes wafer handling (EFEM) and probe-card replacement. The system is available to order.
Which system sorts wafers and which one shows where the light is lost?
InZiv splits the two jobs across its silicon photonics systems, both available to order.
| System | Job | What it produces |
|---|---|---|
| Glide-SiPh PIC | Production testing: high-throughput pass/fail sorting at volume | Per-channel optical insertion loss, functional pass/fail, wafer-level yield and uniformity maps |
| Omni-SiPh LightView | Analysis in process development and NPI, including design validation, debugging yield problems and offline failure analysis | Near-field maps of light propagation and optical loss inside a working circuit, plus IV, LIV and wavelength |
Omni-SiPh LightView's probe options include near-field, deep-trench and angled probes, with custom probe designs for specific customer architectures.
What if my edge couplers sit in trenches too narrow for my current fiber array?
New designs can put edge couplers in trenches too narrow for the current fiber array. InZiv states that the Glide-SiPh PIC glass deep-trench FAU reaches into trenches as narrow as 30 µm wide, and that its proprietary glass fiber is durable and flexible and does not damage probes or the wafer.
Which test stages does InZiv cover, and what falls outside that scope?
According to InZiv, its intended coverage spans test insertion points 1, 2, and 3: wafer testing before bonding, wafer testing after bonding, and testing of individual dies after wafer dicing. InZiv's integrated scope is the prober, the electrical probe card and the optical FAU; the customer's existing automated test equipment (ATE) comes from the ATE supplier as part of the test configuration. Testing of the finished co-packaged module, and very low-volume labs that only need input and output power measurements, are outside InZiv's fit today.
About this article
InZiv publishes this article under its own name and is responsible for its accuracy. Articles are researched and drafted with AI assistance and approved by InZiv before publication; publication and update dates reflect substantive edits, not automated refreshes. Last updated: 2026-11-01