Chapter 31Testing Photonic Chips
Measurement closes the loop: a design is only as good as the data that comes back from the chip.
- assemble a passive test setup: tunable laser, detectors, polarisation control, fibre array and temperature stage;
- describe automated probe stations and design-for-test rules;
- calibrate and reference measurements;
- test active devices and lasers (IV, LIV, spectra, SMSR, RIN, linewidth);
- measure high-speed behaviour (S-parameters, eye diagrams, BER) and run wafer-scale tests.
How do you trust a measured spectrum?
31.1The passive test setup
A passive photonic chip is characterised by its transmission spectrum. The standard setup has five parts:
- a tunable laser that sweeps across the band (for example 1500–1600 nm) in fine steps, a picometre or less, with its wavelength measured or triggered;
- polarisation control: a paddle or electronic polarisation controller, or polarisation-maintaining fibre, to put the light in the TE (or TM) mode the grating couplers accept (Chapter 15);
- fibre coupling to the chip: a fibre array over the grating couplers, or lensed fibres at the edge couplers, on motorised stages (Chapter 17);
- detectors: power meters or photodiodes synchronised with the sweep, with a dynamic range of 60 dB or more;
- a temperature-controlled stage, because a silicon ring drifts by 75 pm/K (Chapter 18) and a measurement without it drifts with the room.
31.2Automated testing and design for test
A chip with dozens of test structures, or a wafer with hundreds of dies, is measured by an automated probe station. It moves the fibre array from one structure to the next using the layout coordinates, aligns to each automatically (a search for first light, then optimisation, as in Chapter 17), runs the sweep the structure’s label asks for, and stores the result. The SiEPIC program publishes its probe-station software, SiEPIClab [25]. Automation only works if the layout follows design-for-test rules: couplers on the array pitch in a fixed orientation, machine-readable labels, alignment loopbacks, and electrical pads on a probe-card grid (Chapter 30).
31.3Calibration and de-embedding
What the detector sees is the device’s response multiplied by everything else on the path: the laser’s power against wavelength, the fibres and connectors, and above all the two grating couplers, each of which loses a few decibels and has a 30–40 nm wide passband (Chapter 15). Removing those is de-embedding:
- normalise to the laser’s own output (a reference detector);
- measure a loopback (two couplers joined by a short waveguide) on the same chip, and divide the device’s spectrum by it;
- or fit a smooth envelope to the device’s own off-resonance spectrum.
The loopback assumes that its couplers equal the device’s. Fabrication variation and slightly different fibre positions make that approximate, a source of a few tenths of a decibel of error in absolute loss. Relative measurements, such as resonance shapes and wavelengths and the slope of a cutback, are far more robust than absolute ones.
Try it. Sweep the virtual laser through a ring between two grating couplers; change the polarisation, the coupler reflections and the averaging, and switch to the loopback-calibrated view.
Three artefacts appear in real data and in the demo: fringes from reflections between the couplers (Chapter 15), which can be mistaken for weak resonances; a polarisation error, which lowers the whole spectrum and can add a TM resonance series; and noise near the band edges, where little light gets through.
31.4Testing active devices
Active devices add electrical probes, a needle or a probe card on the pads:
- Heaters: the IV curve gives the resistance; the resonance shift against power gives the tuning efficiency and \(P_\pi\) (Chapter 18).
- Modulators and junctions: the IV curve checks the diode; spectra at several reverse biases give the resonance shift or phase per volt, hence \(V_\pi L\) and the loss against bias (Chapter 19).
- Photodiodes: dark current against bias, and photocurrent against known input power, give the responsivity (Chapter 21).
31.5Testing lasers
A laser’s characterisation starts with the LIV sweep, output power and voltage against current, which gives the threshold, slope efficiency, series resistance and roll-over (Chapter 24), at several temperatures for \(T_0\). An optical spectrum analyser measures the spectrum and the side-mode suppression ratio (Chapter 26). An electrical spectrum analyser on a fast photodiode measures RIN, and a delayed self-heterodyne interferometer, or a beat note against a reference laser, the linewidth (Chapter 25). Optical isolation between the laser and the measurement is essential, or the instrument’s own reflections disturb what it measures.
31.6High-speed testing
Modulators and photodiodes are characterised in the frequency domain with a lightwave component analyser, a vector network analyser with a calibrated optical transmitter and receiver, which gives the electro-optic or opto-electronic S-parameters and the 3 dB bandwidth. In the time domain, a pattern generator drives the modulator and a sampling oscilloscope records the eye diagram (Chapter 20); an error detector counts errors to measure the bit-error ratio against received power, the sensitivity curve of Chapter 21. RF probes, cables and bias tees must be de-embedded like the grating couplers.
31.7Wafer-scale testing
The grating coupler’s great advantage is that the whole wafer can be tested before dicing (Chapter 15). Measuring the same structures on every die gives wafer maps of propagation loss, ring resonance wavelength, coupler efficiency and responsivity. These maps reveal the process variation of Chapter 29, screen out bad dies before packaging, and feed the statistics of Chapter 32.
The openEBL chips are measured on an automated probe station: a swept tunable laser, TE polarisation through a polarisation-maintaining fibre array at 127 µm pitch, a temperature-controlled stage, and a sweep of each labelled structure. Expect the raw spectra of Chapter 32: envelopes from the couplers, fringes, and the devices’ responses on top.
Reading
- [25] SiEPIC program. SiEPIClab: Python code for automated silicon photonics probe stations. https://github.com/SiEPIC/SiEPIClab
- [14] Mustafa Hammood, Sudip Shekhar and Lukas Chrostowski. “SiEPIC openEBL: remote silicon photonic design-fabrication-testing enabled by open-source process design kits and design tools.” 2025 IEEE Conference on Education and Training in Optics and Photonics (ETOP), Glasgow, 2025. doi:10.1109/ETOP64842.2025.11030682
- [9] Lukas Chrostowski et al. Silicon Photonics Design, Fabrication and Data Analysis (UBCx Phot1x), online course on edX. https://www.edx.org/learn/engineering/university-of-british-columbia-silicon-photonics-design-fabrication-and-data-ana
- [1] Lukas Chrostowski and Michael Hochberg. Silicon Photonics Design: From Devices to Systems. Cambridge University Press, 2015. https://www.cambridge.org/core/books/silicon-photonics-design/BF3CF13E8542BCE67FD2BBC7104ECEAB
- Passive setup: swept tunable laser, polarisation control, fibre array, synchronised detectors and a temperature-controlled stage.
- Automated probe stations need design-for-test layouts; SiEPIClab is open source.
- De-embed the couplers with a loopback or a fitted envelope; trust relative quantities more than absolute losses.
- Lasers: LIV, spectrum and SMSR, RIN and linewidth; high-speed devices: S-parameters, eyes and BER.
- Wafer-scale testing with grating couplers maps the process variation before dicing.