Wafer-Level Optical Testing of Photonic Chips: Setup, Alignment and Test Flow
How silicon photonic wafers are tested optically before dicing: the optical probe station, fiber probes over grating couplers, per-die alignment and its share of test time, reference structures, wafer maps of loss and resonance, probe calibration, and the failure modes of an automated flow.
Scope
This article describes the optical testing of photonic integrated circuits on whole, undiced wafers: the probe-station hardware, how the fiber probes reach the devices, how alignment is automated from die to die, which structures are measured and what the resulting wafer maps show, how the optical probes are calibrated, and where automated flows go wrong. It complements the wafer-level test entry, which sets out why testing before dicing pays, and the die-level test entry for what must wait until after dicing. Single-chip alignment in the lab is covered in Grating Coupler Fiber Alignment, and the wider coupling picture in Fiber-to-Chip Coupling for Silicon Photonics.
The optical probe station
| Subsystem | Function | Notes |
|---|---|---|
| Wafer chuck and prober | Holds the wafer by vacuum and steps it die to die | Vacuum chuck, usually temperature-controlled; 200 mm or 300 mm |
| Optical probes | Single fibers or fiber arrays held above the wafer surface at the grating couplers' design angle | Typically 8° to 12° from vertical; motorized in X, Y, Z and angle, with fine piezo stages for alignment |
| Electrical probes | DC needles or probe cards for heaters, photodiodes and modulator bias; RF probes for high-speed tests | Placed so their shadows and bodies clear the fibers |
| Instruments | Swept tunable laser and synchronized power meters, source-measure units, optionally an RF network analyzer | Swept measurements capture a full spectrum per structure in one pass |
| Vision | Top-down camera for die registration and fiber placement; a side camera for fiber height | Pattern recognition locates alignment marks on each die |
| Control software | Executes the test plan, runs alignment, logs data against die coordinates | Produces the wafer maps |
The optical access is almost always through grating couplers, since edge couplers need a facet that does not exist until the wafer is diced. The trade between the two, and the common pattern of grating test taps on dies whose product I/O is edge-coupled, is discussed in Edge Coupler vs Grating Coupler.
Test flow
1. Load and register the wafer
The wafer is loaded, and the vision system finds alignment marks on a few dies to fix the wafer's position and rotation relative to the stage. From these marks and the die map, the stage can move to any die within a few microns.
2. Calibrate the probes
Before the first die, the loss of the optical path outside the chip is measured: the fiber probes, their connectors and any switches. A common method is a fiber-to-fiber reference taken with the probes joined through a known path, or measurement of a reference structure on a known die whose loss has been established. The reference is repeated at intervals during the wafer, because fiber tips collect debris and drift is otherwise indistinguishable from process variation.
3. Align on the first die
The first die receives a full search as in the single-chip procedure: fiber height set, a raster or spiral search over a window of tens of microns, polarization set, fine search and height optimization. The resulting fiber position relative to the die's alignment marks is stored.
4. Step and re-peak
On each following die, the stage moves the wafer by the die pitch, the fibers are brought to the stored position, and only a short fine search is run to recover the peak. The fine search is usually on a loopback structure with both fibers, and the coordinates found are then applied to the other structures on the die by their layout offsets.
5. Measure the structures
Each structure on the test plan is measured, usually by sweeping the laser across the band and recording transmitted power at every wavelength, together with any electrical measurements. The data are stored against the die coordinates.
6. Map and bin
After the wafer, results are assembled into wafer maps and each die is binned against its pass criteria. The known-good-die list goes forward to dicing and packaging.
Where the time goes
Test time per wafer is the number of dies times the time per die, and the time per die is alignment plus measurement plus stepping. As an illustration, a die with one re-peak of 10 s, five structures each swept over 100 nm at 10 nm/s (10 s each) and a 2 s step spends 62 s per die, of which the sweeps are 50 s; across 100 dies that is about 1.7 h. Where many structures are measured per die, the sweeps dominate; where few are measured, alignment does, and faster search algorithms or fiber arrays that reach several structures at one placement make the difference.
Test structures and what they reveal
| Structure | Measured quantity | What the wafer map shows |
|---|---|---|
| Loopback (two couplers and a short waveguide) | Twice the coupler loss, and its spectrum | Coupler efficiency and center wavelength across the wafer; the first sign of etch-depth or film-thickness variation |
| Waveguide spirals of several lengths | Propagation loss by cutback | Sidewall roughness and material loss by wafer position |
| Ring resonators | Resonance wavelength, Q factor, free spectral range | Effective-index variation; resonance maps are among the most sensitive probes of waveguide width and thickness |
| Mach-Zehnder interferometers and directional couplers | Extinction, splitting ratio | Coupling-gap and width variation |
| Photodiodes | Dark current, responsivity | Germanium and contact process health |
| Heaters and modulators | Resistance, tuning efficiency, and with RF probes the and bandwidth | Doping and metal process |
Coupler loss taken from loopbacks is subtracted from the other structures' insertion loss, as described in De-Embedding Coupler Loss. The coupler's own wafer map is therefore needed before any other loss map can be read.
Common failure modes
Drift mistaken for process variation. A slowly contaminating fiber tip or a drifting laser makes the loss map show a gradient in the order the dies were tested rather than across the wafer. Periodic reference measurements, and testing in an order that does not follow a single wafer direction, separate the two.
Chuck temperature. Ring resonances and other interferometric structures shift with temperature (see thermo-optic coefficients); an unregulated or unsettled chuck adds a spurious resonance map.
Polarization drift. With standard single-mode fiber probes, the polarization at the grating changes as the fibers move and cool; polarization-maintaining fiber probes or a re-optimization step per die avoid this.
Stored coordinates that do not transfer. Wafer bow and die-to-die placement errors move the coupler positions by more than the fine search window, and the re-peak finds a side lobe or nothing. The first-die search is then repeated on that die.
Probe collisions. Fiber tips at 15 to 20 μm above the surface have little margin against wafer bow, particles and electrical probe bodies; touch-down detection and conservative approach heights prevent broken probes and scratched dies.
References: L. Carroll et al., "Photonic packaging: transforming silicon photonic integrated circuits into photonic devices," Applied Sciences 6, 426 (2016); R. Marchetti, C. Lacava, L. Carroll, K. Gradkowski and P. Minzioni, "Coupling strategies for silicon photonics integrated chips," Photonics Research 7, 201 (2019); L. Chrostowski and M. Hochberg, Silicon Photonics Design: From Devices to Systems (Cambridge University Press, 2015), chapters on testing and on fabrication variability.