Silicon Photonics: Tricks and Tweaks for Wafer and Chip-Level Optical Test
Photonic integrated circuit (PIC) and silicon photonics technologies are being used to manufacture devices for optical communications at higher volumes with lower costs, energy consumption, and size. This is now especially driven by the rapidly increasing needs of data centers. These technologies also offer the means to realize new functionality with high-level integration of electronics and optics.
Like in the semiconductor electronics industry, appropriate test and measurement at an early stage, like wafer-level test, is valuable to avoid the high cost of processing and packaging substandard devices that will fail final test as well as for control and diagnostics of the wafer production process. Typically, this involves parametric tests to characterize the material and structuring quality, like measuring sheet resistance and capacitance electrically and attenuation and responsivity optically. Optical testing generally also includes the dependence on wavelength and on the alignment of optical polarization with the waveguide structures. For transponder devices like photodetectors and modulators, the RF-frequency dependence is also important for characterizing the bandwidth of the devices.
The Challenges of Polarization Alignment
Adding optical measurements to wafer-level test requires optical probes to couple light to or from the wafer. Optical fiber cable is the usual way to connect to the instruments.
Standard single-mode fiber (SMF) has a nominal core diameter of 9 μm, which is significantly larger than typical waveguide dimensions on the semiconductor wafer, where the higher refractive index results in shorter wavelengths and stronger confinement of the intensity profile. Also, the waveguides typically have a rectangular cross-section, while the fiber core is round. Therefore, an adapting structure is needed to provide matched coupling between the probe fiber and the wafer waveguides. For example, coupling into the surface of a wafer to waveguides running parallel to the wafer surface is achieved with coupler grating structures that can match the beam profiles and provide the refraction needed to change the direction of the light as shown in Figure 1 below. Tapered waveguides can be used for edge-coupling into chips after dicing the wafer. For packaging, the chips can be attached to an interposer, generally a passive planar structure that provides accessible connections to the smaller chip.
One way to assure alignment is using polarization-maintaining fiber (PMF), which has structure to define optical axes so that when the input light is exactly aligned with either axis the output light will also be aligned with that axis. This can be used if the laser source provides the output aligned in PMF and the alignment of the probe fiber axis with the wafer axis can be assured. It can be a good approach but is limited in flexibility. Such probes can only be easily used for measuring with that one state of polarization but not for determining polarization dependence. And any intermediate instruments or components, like an optical attenuator or switch or coupler must also be a polarization-maintaining device. Every connection to PMF is also subject to some misalignment that increases the crosstalk between the modes and degrades the measurements.
Another approach uses SMF probes with a polarization controller before the probe. The output signal from the device under test (DUT), either electrical or optical, is optimized by adjusting the polarization controller. In this way, when the laser source is stable, the polarization at the DUT can be aligned.
Unfortunately, the polarization at the output of the polarization controller and the SMF probe generally varies with the wavelength of the light So if the device parameter (like attenuation or responsivity) should be determined over a wide wavelength range it is usually necessary to repeat the optimization step at several wavelengths over the desired range.
This prevents getting good measurements of a particular axis by simply scanning the wavelength of the tunable laser while sampling the DUT output. Typical results show an optimum signal at the wavelength used for alignment and (falsely) degraded performance as the wavelength is increasingly far from this point. An example of this is shown in Figure 2 (blue curve) for a measurement of a polarization filter.
The severity of degradation depends on the wavelength dependence and on the amount of birefringence in the polarization controller, fibers and other instruments in the optical path. Breaking the sweep into shorter separately optimized segments to avoid this requires much longer test duration.
Read more to learn about an improved approach, especially for obtaining wavelength-dependent data.