Sample fabrication and mounting

In this paragraph the fabrication of the toroidal microresonators is outlined and our method of sample handling and cleaning is introduced. The microfabrication process has been described before elsewhere [107, 108] , but nevertheless we will highlight some particular aspects here. To fabricate high-Q toroidal microresonator, we start out with commercial silicon wafers with a 2µm oxide layer (Silicon Valley Microelectronics).3 _{Briefly, we use optical lithography to define disks of varying}

diameters (80−120µm) and HF etching to remove excess silicon oxide (cf. Figure

2.3(a)-(c) ). In the next step the glass disks are undercut using sulfur hexafluoride. In a process developed by Emanuel Gavartin, we then use a dicing saw to cut groves that define the chips with the samples on the wafer (cf. the rectangular pattern in Figure2.3). This way the wafer is still suited to undergo a final automated cleaning step in hot sulfuric acid in the cleanroom.

Each predefined chip has a size of 4 mm×12 mm and carries 20 disks. To separate a single chips from the wafer it is sufficient to scribe along the grove with a razor blade or scalpel and then carefully cleave the wafer using a pair of wafer tweezers that are placed next to each other on each side of the grove. Compared to the conventional cleaving method involving a diamond scribe, the dicing method allows us the obtain well defined cleaves and samples of uniform size. The latter is particularly important, as the coupling chamber of the FCD and the sample transfer stage are laid out for a specific sample size. Moreover the method increases overall cleanliness, because of the additional cleaning step on wafer scale and because residues from scribing and cleaving remain inside the grove.

The toroidal resonator shape is obtained from the microdisk in a final laser reflow step. To this end, a CO2 laser (Synrad 48-series, 10 W), which is strongly absorbed

by silica, is focused on the disk to melt the glass. The silicon pillar provides a heat sink, such that the temperature is highest at the rim of the disk, where the liquid SiO2 forms a ring due to surface tension. The ring shrinks towards the center, while

gaining in thickness, until the cold from the pillar terminates the process and sets 3_{Oxide layers of 1µm thickness have been tested , but the optical quality and reproducibility}

30 2. The experimental setup

(a) (b) (c)

2 μm silica layer on silicon wafer

Silica pads on silicon wafer after lithography, HF-etching

Free standing silica discs after

XeF2dry etching

Figure 2.3.: The sample preparation at a glance. (a) Starting point of the process is a commercial 4” silicon wafer with a 2µm oxide layer (e.g., from Silicon Valley Microelectronics). (b) The positive structures are defined using standard pho- tolithographic methods; the excess SiO2 is removed with a hydrofluoric acid wet etch. Typically disks with diameters from 80µm to 120µm are defined. (c) Next the silicon is selectively etched using either xenon difluoride of SF6. The undercut is adjusted via the gas pressure and the number of etching cycles. The pictures at the bottom of the figure show the final wafer. The samples have been pre-diced, such that the wafer can still undergo a final cleaning step. The insets shows a magnified view of a single chip, carrying 20 microdisks and an identification/alignment structure.

the major toroid radius. There is controversy, whether a fast reflow, using a short (10−100 ms) laser pulse, or a slow reflow, by gradually increasing the laser power over a few seconds, lead to better results. Here, we found that the second method provides better control over the optical quality. However it is highly recommended that the laser power remains constant at 60%−70%, to avoid power fluctuations. To this end we work at a constant pulse width modulation and use an adjustable attenuator (ULO optics, CO2mpact attenuator with enhanced Brewster windows) to control the power.4

For the use with the inverted coupling setup, we designed a support that consists of a steel rod (_∅= 2 mm) attached to a PMMA block (20 mm×8 mm×2 mm), which is glued to a nickel plate (cf. Figure2.4). The support is mounted in the coupling setup via magnets and can easily be handled with grooved tweezers. Figures 2.4

4_{The power level of the 10W Synrad CO}

2laser that we used for reflow is regulated via pulse with

modulation (PWM). Here we set the PWM rate to the highest level (20 kHz). It is possible to control the power output fully via the PWM; however at the low power level required for laser reflow, which is typically around 10−15% of the maximum power, we observed that the intensity fluctuates and the focal spot drifts. In particular the latter issue affects the repeatability. Moreover we found that the attenuator, which is based on a rotatable pair of Brewster plates, does not cause any visual displacement of the focal spot.

2.1 Coupling setup 31

a

e

b

c

d

Figure 2.4.: Sample handling. The chip with the micro-resonators is glued to a support with a nickel base plate that can be attached to a magnetic holder. Nickel was chosen because of its insensitivity against a humid environment. Using grooved tweezers, the sample can be manipulated safely. A schematic view of the mount with a chip is shown in panel (c). The photograph (b) shows how the chip is aligned on a glass stage while the support is positioned using a translation stage (not in the picture) and while the glue is still wet. A groove prevents the microresonators from being damaged. (a) The adhesive is cured using a UV lamp. (d) Two samples are attached to a glass slide with matgnets and plunged in a beaker with cleaning solution. (e) A close up view shows that the sample is only wetted from the side that carries the resonators. Hence the potentially aggressive agent does not attack the glue.

(a) and (b) show how the chip is glued to the support with UV adhesive (Thorlabs, NOA-81) and how the chip is aligned on a stage assembled from microscopy slides. For sample handling and wet chemical processing, the support is attached to a glass slide with magnets (cf. Figure 2.4 (c) ). This way the sample can be transferred between different cleaning and rinsing solutions without risking contamination, e.g., from tweezers.

Between two measurements the FCD is rinsed with DI water. Flushing with 4 ml of DI water has shown to be sufficient to visually remove any trace of a colorant (before a white background, cf. Figure2.1 (d) ); we flush with 30 ml to remove any residue of an analyte in between measurements. A sample that was used for a measurement can be recycled without significant degradation of its optical quality. To this end, the sample is first placed in a 65 : 25 : 1 solution of chloroform, methanol, and DI water for 20 minutes and afterward rinsed with DI water. In a second step we use SC-1 solution (5 : 1 : 1 solution of DI water, 30% hydrogen peroxide, and ammonium hydroxide at 70◦C) for 10 to 20 minutes and rinse again with DI water. During the cleaning and preparation procedure the sample holder is attached to a microscope

32 2. The experimental setup

slide, which we put on top of the beaker that contains the solution. It is an advantage of our sample mount, that the fluid level of the cleaning solution can be adjusted to a level such that the chip with the samples is only wetted from the side carrying the resonators. This avoids contamination and prevents that the UV adhesive (which is used to mount the chip) is attacked by the agent. Moreover the samples can be quickly transferred between two solutions. Following this procedure we were able to re-use a sample up to ten times. After an initial linewidth degradation from 5 MHz to 8 MHz no further loss of quality was observed. However, each recycling cycle shifts the resonance towards higher frequency by a few GHz. We believe that the SC-1 cleaning step takes off some SiO2 material. Indeed a radius reduction corresponding

to one Si−O bond length shifts the frequency by roughly 3 GHz. The effect limits the total lifetime of a sample, because the resonance eventually shifts out of the laser tuning range.