Silicon nitride (Si₃N₄) ring resonators are critical for efficient and compact on chip optical routing [1–3], frequency combs [4–7], and high precision sensing [8–11], however the intrinsically high film stress of silicon nitride has limited both the optical confinement and quality factor (Q) of ring resonators. Whereas the silicon and silicon dioxide platforms generally suffer from high losses or delocalized optical modes, the Si₃N₄ platform provides advantages of both high confinement and high Q. High Q disks have also been demonstrated in Si₃N₄ [12] but disks have larger mode volumes and are challenging to dispersion engineer for nonlinear applications. Si₃N₄ is also a deposited material, which enables seamless integration with other material platforms. However, the high film stress of Si₃N₄ prevents thick (>400 nm) films of high optical quality from being deposited; catastrophic cracking occurs, severely limiting device yield. In principle low stress nitride films can be grown thicker with plasma enhanced chemical vapor deposition (PECVD) and modified low pressure chemical vapor deposition (LPCVD) processes, but these deposition chemistries yield films with stronger material absorption caused by dangling H and O bonds with the Si and N in the films.

Thick films, limited to date by stress, would enable high confinement and high Q. Thick films lead to smaller optical mode overlap with the boundaries of the waveguides (responsible for scattering losses) leading to lower losses. This can be observed in Fig. 1 where we show the mode profile for a mode confined in two different waveguide thicknesses. Figure 1(a) shows the mode confined in a waveguide with a traditional thickness of 400 nm limited to date by stress. One can see that the mode overlaps significantly with the boundaries of the waveguide. Figure 1(b) shows the mode confined in a waveguide with a much higher thickness of 910 nm. One can see that very little of the mode overlaps with the boundary of the waveguide. Note that increasing the size of the waveguide increases the number of modes supported; however these higher order modes are only weakly excited.

 

Fig. 1 Transverse electric (TE) mode simulations at 1550 nm wavelength for the (a) 400 nm x 1800 nm and (b) 910 nm x 1800 nm waveguides with 71% and 93% modal confinement, respectively. Less of the optical field interacts with the waveguide boundaries for the taller waveguide (b).

Download Full Size | PDF

Previous high Q Si₃N₄ ring resonators have circumvented film stress issues by exploiting either highly delocalized optical modes with extremely thin films [13] or highly confined modes within film stress limits [14]. High Q ring resonators based on extremely thin Si₃N₄ films can avoid film stress issues, but they suffer from highly delocalized optical modes, requiring millimeter-scale bending radii and up to 15 µm of silicon oxide cladding. In addition, these resonators only support the transverse electric (TE) mode which prevents integration with devices that support the transverse magnetic (TM) mode. High confinement ring resonators based on thicker films can be achieved using high temperature deposition and anneal to relieve film stress, but films greater than 750 nm in thickness remain challenging.

In order to overcome the stress limitations of Si₃N₄, we strategically place mechanical trenches to isolate the photonic devices from propagating cracks. Physical shocks near the edge of the wafer, which occur often during handling of the wafer, can provide enough energy to induce cracking of the stressed film. Once initiated, these cracks originating from the edge of the wafer propagate continuously in a uniform stress field, terminating only once they encounter crack resistance at the edge of the wafer or at another crack boundary. We introduce trenches around our devices that terminate cracks before they can spread to our device region (shown in Fig. 2). A single trench does not guarantee crack termination however [15]. Overstressed films store energy in the enhanced acceleration of the crack and the extended penetration into the substrate. With this stored energy, cracks can overcome the crack resistance of a single trench and continue propagation. To increase crack resistance, we create between two and five parallel trenches to ensure crack termination.

 

Fig. 2 Microscope images in the dark field showing crack propagation terminating at a trench created with a diamond scribe. The film to the left of these trenches is crack free.

Download Full Size | PDF

We define a series of trenches around the edge of the wafer before deposition of the Si₃N₄ film to prevent crack propagation. Beginning with a thermally oxidized 4 inch silicon wafer, we lightly scribe a series of lines into the silicon oxide surface to define a 5 cm by 5 cm rectangular region in the center of the wafer. This rectangular region will be the crack free region in which we pattern our photonic devices; elsewhere, the film will crack. Note that for ease of processing and throughput we define trenches with a diamond scribe, but in principle one can define more sophisticated trenches with photolithography and etching. Etching may actually improve the crack resistance of these trenches because of the increased roughness on the etched surfaces [15]. In addition, photolithography could be used to define trenches only at the wafer periphery to enlarge the area of the crack free region. Following trench definition, we proceed with device fabrication as described in [14]. After deposition of 910 nm of Si₃N₄, in steps of 400 nm and 510 nm, we pattern devices with electron beam lithography using ma-N 2405 resist, post exposure bake for 5 minutes at 115°C, and etch in an inductively coupled plasma reactive ion etcher (ICP RIE) using CHF₃/O₂ chemistry. After stripping the resist, we anneal devices at 1200°C in a nitrogen atmosphere for 3 hours. We clad devices with 250 nm of high temperature silicon dioxide (HTO) deposited at 800°C followed by 2 µm deposition of silicon dioxide using plasma enhanced chemical vapor deposition (PECVD). The fabricated device is shown in Fig. 3. Note that our trench definition method is not specific to this process; it can be applied to any highly stressed film in which cracks are initiated near the edge of the substrate.

 

Fig. 3 Scanning electron microscope (SEM) image of the resonator.

Download Full Size | PDF

We measure an intrinsic quality factor of 7 million, the highest quality factor reported to date for high confinement Si₃N₄ ring resonators; this quality factor corresponds to an ultra-low propagation loss of 4.2 dB/m. The ring resonator measured has a radius of 115 µm, a coupling gap of 680 nm, and a cross section of 910 nm tall by 1800 nm wide. We couple a tunable laser light source, transmitted through a polarization controller, into the inverse nanotaper of our device using a lensed fiber. We collect the output of the ring resonator through another inverse nanotaper and collimating lens. After passing the output through a polarizer, we monitor the output on a photodetector. In order to measure a single resonance, we finely scan the laser frequency by applying a triangular-wave voltage signal to the piezoelectric transducer of the laser, while monitoring the photodetector signal on an oscilloscope. We calibrate the voltage-frequency conversion with respect to a free space bowtie cavity. We observe some higher order mode resonances (shown in Fig. 4), but these resonances have very low extinction, suggesting that most of the optical power is in the fundamental mode. Minimal higher order mode excitation is expected because we design our waveguides to support few modes: five TE modes and four TM modes. From the single resonance scan normalized to the maximum value, we measure a resonance linewidth of 36 MHz and resonant transmission of 26% (shown in Fig. 4), corresponding to an intrinsic quality factor of 7 million for TE polarization. For TM polarization we measure a Q of 4 million. The Q for TE is higher because the TE mode has higher optical confinement and effective index, so the mode suffers less scattering and bending loss.

 

Fig. 4 (a) Transmission spectrum with the finely scanned resonance outlined in red. (b) Resonance with 36 MHz linewidth corresponding to an intrinsic Q of 7 million.

Download Full Size | PDF

We calculate the propagation loss within the ring to be 4.2 dB/m and we estimate the absorption loss to be 71% of this loss, suggesting that the Q can be further improved with increased optical confinement. We calculate propagation loss using the relation [16]

We demonstrate a high quality factor of 7 million in a high confinement Si₃N₄ ring resonator using crack resistant trenches to overcome stress limitations of thick Si₃N₄ films. Our high Q devices herald advances in low loss optical routing, low power threshold nonlinear optics, and high sensitivity sensors. We have also overcome film stress limitations for Si₃N₄, revealing a new design space for integrated optics and microelectromechanical (MEMS) devices that has been unexplored to date.