Grating Coupler Fiber Alignment: Finding First Light

Scope

First light on a surface grating coupler is found by setting the fiber 15–20 μm above the chip, raster-scanning a 40 × 40 μm window in 5 μm steps, and then optimizing polarization, lateral position and height in turn. This article describes the procedure for active alignment of a single-mode optical fiber to a surface grating coupler on a silicon photonic integrated circuit (PIC). Coverage is restricted to the search algorithm used to obtain initial coupling and to the diagnostic checks performed when coupling cannot be established. Passive alignment to V-grooves and to edge couplers is outside scope.

Geometry

A surface grating coupler diffracts light between an in-plane slab waveguide mode and a near-vertical free-space mode. For a uniform one-dimensional grating of period $\mathrm{\Lambda }$ etched into a waveguide of effective index $n_{\text{eff}}$ and operating at free-space wavelength ${\lambda }_0$, the central output angle in the upper cladding of index $n_c$ is given by the first-order grating equation

$$\sin \theta =\frac{n_{\text{eff}}}{n_c}-\frac{{\lambda }_0}{n_c\mathrm{\Lambda }}\mathrm{.}$$

For a standard silicon-on-insulator (SOI) 220 nm waveguide operating at ${\lambda }_0=1550$ nm with TE polarization, the effective index of the grating region is about 2.7 for a 70 nm etch (a first estimate, the average of the etched and unetched slab sections), and periods of about 615 to 630 nm put the beam 10 to 13.5° from vertical in air, where the angle follows from $\sin {\theta }_{\text{air}}=n_c\sin \theta$ (grating coupler). Surface gratings are typically detuned a few degrees off normal to suppress second-order back-reflection into the waveguide.

Coupling tolerances

Standard SOI surface grating couplers have a mode field diameter of approximately 10 μm at the optimal fiber height of 15–25 μm above the chip surface. Coupling efficiency falls off as a Gaussian function of lateral offset with characteristic width determined by mode overlap between the grating mode and the fiber mode; the lateral 1 dB tolerance is approximately $\pm 2.5$ μm for a mode matched to the fiber's (the loss at $\pm 1$ μm is about 0.2 dB). The 1 dB bandwidth in wavelength is typically 30–40 nm centered on the design wavelength. The 1 dB tolerance in fiber polar angle, computed for a beam matched to the fiber mode, is about $\pm {2.6}^{\circ }$ at a fixed wavelength (mode mismatch loss); changing the angle also moves the peak wavelength, by about 8 nm per degree.

For peak-coupling values, uniform SOI grating couplers achieve 25–50% peak efficiency per coupler (3–6 dB insertion loss). Apodized designs with a buried metal reflector or distributed Bragg reflector below the grating reach 60–80% (1–2 dB). Silicon nitride grating couplers operating at the same wavelengths typically achieve 25–40% efficiency.

Equipment

A minimal active alignment setup requires:

Polarization control is required because most grating couplers are designed for a single polarization (typically TE); the input state of polarization (SOP) at the grating must be matched to the design polarization. A linear state 10° away from it costs only 0.13 dB, but one close to TM, more than about 70° away, costs 10 dB or more.

Procedure

1. Pre-position via top-down vision

With the fibers retracted to at least 200 μm above the chip surface, position each fiber tip laterally over its target grating coupler by reference to the top-down microscope. At this stage, lateral position is established to within approximately ±10 μm.

2. Establish fiber height

Lower each fiber toward the chip in 5 μm increments while observing the microscope focus. Contact with the chip surface produces a focus shift; back off by 15–20 μm from the contact point. A height error mostly moves the beam footprint sideways, by about 0.18 μm per micrometer of height for a fiber tilted 10°: a 30 μm error costs about 4.7 dB at a fixed lateral position but only about 0.3 dB once the lateral position is re-optimized, and the raster in step 3 still finds the signal for height errors up to roughly 60 μm.

For setups with side-view cameras, fiber height can be set directly by visual measurement of the air gap.

3. Lateral coarse search

Execute a 2D raster scan in the chip plane over a 40 × 40 μm window centered on the eyeball-aligned position, with 5 μm step size. The detector should remain on a fixed sensitive range (typical: $-50$ to $-30$ dBm) throughout the scan; autoranging is disabled.

The expected signal at the coarse-search peak is approximately 20 dB below the optimized peak, due to combined sub-optimal lateral position, sub-optimal height, and unoptimized polarization.

4. Polarization optimization

At the coarse-search peak, rotate the polarization controller paddles to maximize received power. For TE-designed grating couplers, the SOP-dependent contrast is typically 20–25 dB.

5. Lateral fine search

Repeat the 2D scan over a 10 × 10 μm window with 1 μm step size, centered on the coarse-search peak.

6. Height optimization

At the fine-search peak, sweep fiber height in 2 μm steps over $\pm 20$ μm of the current position. The optimum is a smooth maximum.

7. Iterate

Repeat steps 5–6 until peak position converges (typically 2–3 iterations). Convergence is required because the lateral and height optima are coupled through the fiber tilt: at 10°, each micrometer of height moves the beam about 0.18 μm sideways across the grating.

8. Angle optimization

If the positioner permits, sweep the fiber polar angle over $\pm 2^{\circ }$ of the design angle. Many production setups omit this step by fixing the fiber holder at the design angle in hardware.

Validation

After alignment, coupling can be validated against expected losses. For a loopback test structure consisting of two grating couplers connected by a length $L$ of waveguide, the measured insertion loss is

$${\text{IL}}_{\text{loopback}}=2\cdot {\text{IL}}_{\text{coupler}}+\alpha \cdot L,$$

where $\alpha$ is the waveguide propagation loss (typical SOI: 1–3 dB/cm; silicon nitride: $<0.1$ dB/cm). For an uncalibrated alignment of a uniform SOI grating coupler with $L<1$ cm, expected loopback insertion loss is 8–13 dB. Measured loss more than 5 dB worse than this range indicates a process or alignment issue rather than nominal device performance.

Common failure modes

The following account for the majority of alignment failures, in approximate descending order of incidence.

Wavelength outside coupler bandwidth. Grating couplers have 1 dB bandwidths of 30–40 nm. Testing a 1550 nm coupler at 1310 nm produces $>$ 20 dB of additional loss and may not yield detectable signal. The design wavelength must be confirmed against the PDK or layout file before alignment.

Incorrect input polarization. Most grating couplers are TE-only; TM input produces 20+ dB of additional loss. Polarization must be optimized after coarse position is established, not before.

Fiber height error. The next most common no-signal cause. Touch-down to chip followed by 15–20 μm retraction is reliable; visual height setting alone is unreliable above ±15 μm error.

Detector autoranging during raster. Autoranging meters may miss transient peaks during fast scans. The detector range must be fixed at a sensitive setting for the duration of the search.

Light launched into the wrong fiber. In multi-port setups, the input optical path may not reach the active launch fiber. Verification with a known-good reference path is performed before alignment.

Chip not in vacuum contact. Thermal drift and vibration of a loose chip cause coupling instability that masquerades as alignment error during long sweeps.

Damaged or contaminated grating coupler. Visible damage or particulate contamination on the grating, confirmed by microscope inspection at high magnification, can produce permanent excess loss. This is the last hypothesis to evaluate, not the first.

References

For grating coupler design theory, see Cheng et al. (2020) and Marchetti et al. (2019). For the broader context of silicon photonics characterization workflows, see Bogaerts and Chrostowski (2018). For practical alignment automation and search algorithms in commercial probe stations, see the FormFactor application note on silicon photonics wafer-level testing.