We implement a multi-color laser engine with silicon nitride photonic integrated circuit technology, that combines four fluorophore excitation wavelengths (405 nm, 488 nm, 561 nm, 640 nm) and splits them with variable attenuation among two output fibers used for different microscope imaging modalities. With the help of photonic integrated circuit technology, the volume of the multi-color laser engine’s optics is reduced by two orders of magnitude compared to its commercially available discrete optics counterpart. Light multiplexing is implemented by means of a directional coupler based device and variable optical attenuation as well as fiber switching with thermally actuated Mach-Zehnder interferometers. Total insertion losses from lasers to output fibers are in the order of 6 dB at 488 nm, 561 nm, and 640 nm. Higher insertion losses at 405 nm can be further improved on. In addition to the system level results, spectrally resolved performance has been characterized for each of the developed devices.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Visible wavelength photonic integrated circuits (PICs) have proven to be a versatile technology effectively addressing the needs of a range of sensing and diagnostics applications in the life sciences, such as Raman sensing , fluorescent sensing , or refractive index sensing . Due to its wide transparency range from the ultra-violet (UV) to the mid-infrared (MIR)  and its compatibility with standard complementary-metal-oxide semiconductor (CMOS) technology [5,6], the silicon nitride (SiN) integrated waveguide platform [2,5–7] is a prime candidate for the implementation of visible range PICs. In comparison with silicon-on-insulator (SOI) platforms, the smaller refractive index contrast of silica-cladded SiN waveguides reduces the achievable routing density, but also results in much reduced scattering losses . In addition to low confinement SiN platforms (using thin SiN layers), with induced losses as low as 0.5 dB/m in the near-infrared (NIR) C-band , stress-controlled fabrication methods of high confinement SiN waveguides (using thicker SiN layers), enabling much denser routing than germanium-doped silica photonic lightwave circuits with similar low losses , have been demonstrated with losses as low as 5 dB/m . Furthermore, low waveguide background luminescence in SiN waveguides has enabled enhanced sensitivity waveguide-based sensors . Lastly, fabrication with CMOS manufacturing might enable low cost lab-on-a-chip consumables if fabricated in high enough volume.
Laser stimulation of fluorescent markers (fluorophores) is the underlying mechanism for a number of biophotonics analytical techniques, such as fluorescent confocal microscopy , fluorescent light sheet microscopy , flow cytometry  and optogenetics [16,17], among others. Multi-color laser engines (MLEs) provide simultaneous or sequential sourcing of multiple wavelengths allowing stimulation of commonly used fluorophores. For some imaging modalities such as fluorescent recovery after photobleaching (FRAP) or superresolution microscopy, required optical power levels can vary significantly depending on whether they are used for bleaching or excitation. Moreover, multiple fiber outputs allow switching between different imaging and excitation modes, for example between confocal and flood illumination. Digital intensity modulation at the output of MLEs in the kHz to MHz range presents itself as another crucial functionality of these instruments typically required for confocal fluorescent microscopy techniques. Millisecond switching times, respectively kHz modulation, are typically sufficient for applications such as FRAP, even though faster switching times can be advantageous for some of the more advanced imaging techniques .
The primary motivations for the reported work are miniaturing, reducing cost and increasing the robustness of such MLEs with help of SiN PICs. Conventional MLEs are assembled from a number of discrete devices such as acousto-optic tunable filters (AOTFs), dichroic mirrors, and fiber switches, that are all replaced by devices integrated on the PIC. While lasers are implemented with laser diodes that can be directly current switched, on-chip variable attenuation remains essential as driving laser diodes close to threshold, to source the smallest targeted power levels, would also result in much increased noise. The compactness afforded by the PIC reduces free-space beam paths. At the same time, since the low confinement on-chip waveguides have been engineered to have similar mode sizes as the single mode fibers used at visible wavelengths, alignment requirements at the PICs’ edge couplers are not increased compared to the requirements associated to fiber coupling in discrete optics solutions. This results in the placement accuracy of coupling lenses to be relaxed, so that active beam steering after assembly is no longer required. To enable this PIC functionality, a number of challenges had however to be overcome arising from the wide spectral range being addressed, covering the entire visible spectrum from 405 to 640 nm. For example, multiplexers had to be developed that are adaptable to the entire visible range. The chip-to-fiber interface at the chip output also has to support reasonable insertion losses (ILs) over the entire wavelength range of operation.
To implement the PIC, we chose the TriPleX SiN platform [19,20] that has been further adapted by LioniX International to cover the visible wavelength range , with a transparency window from 405 nm to 2.35 µm, as limited by absorption from the SiO2 cladding at long wavelengths. Since silicon-rich films absorb light below the nominal bandgap of stochiometric SiN , the waveguide core is formed by a high quality Si3N4 film deposited by low-pressure chemical vapor deposition (LPCVD), in order to extend the transparency window as far as possible at its short wavelength bound, resulting in waveguide losses in the range of 0.3 to 0.4 dB/cm at 405 nm. A single, stoichiometric SiN thin film is deposited onto an oxidized silicon wafer with 8 μm thick SiO2, fully etched, and overclad with an additional 8 μm SiO2 (the single strip geometry ). Low mode confinement of the thin (∼26 nm) waveguide cores enables the engineering of the mode profile at the edge couplers to match that of standard visible wavelength optical fibers and thus to reduce the coupling losses. However, the low modal confinement also constrains bending radii and the integration density. However, since the PIC size was also constrained here by the placement of the laser diodes and coupling lenses, this turned out not to be the primary constraint and the PIC could be further miniaturized. While high-speed phase tuners relying on the application of stress with piezoelectric materials have also been investigated in the TriPleX platform , here we rely on thermal phase tuners with kHz modulation speeds  sufficient for most of the targeted applications.
Previously, we have investigated edge coupling to these SiN PICs, both with direct butt-coupling to a fiber as well as free-space coupling from laser diodes with help of an interposed lens system , as well as a concept for visible wavelength combiners (WLC)  and a simpler, single output fiber version of the MLE . Here, we report for the first time the implementation of the complete MLE with integrated fiber switch and variable optical attenuation. Detailed and systematic spectrally resolved characterization of the individual devices and of the entire system is given, with good agreement to modeling results. Section 2 describes the architecture of a conventional MLE and of the PIC based MLE concept, including the PIC layout and the module assembly. Section 3 covers the overall characteristics of the photonic integrated system chip in terms of transmission loss, spectral characterization and switching times. In Section 4, the simulation and characterization of the utilized building blocks such as input/output edge couplers (ECs), Mach-Zehnder interferometer based switches, that can be gradually tuned to set the power at a given output fiber, and WLC stages are presented.
2. Dual output SiN-PIC-based MLE concept
The architecture of a typical multi-color laser engine  is depicted in Fig. 1. Four wavelengths in the visible range, 405, 488, 561, and 640 nm, are combined into a single beam with dichroic mirrors, that is then modulated with an AOTF and split between two fibers with a fiber switch. Fully autonomous adaptive beam steering maintains alignment over drifting temperatures. Since laser diodes at 561 nm are not readily available, a diode pumped solid state laser (DPSS) or a more compact frequency doubled diode laser (FDDL) are used to supply this wavelength.
The schematic of the proposed PIC-based MLE is shown in Fig. 2. The PIC comprises a number of integrated components implementing the MLE’s functionality. The lasers are coupled with a pair of lenses to the ECs of the PIC. Thermally tunable Mach-Zehnder interferometers (MZIs) with a dual waveguide output split the power for each wavelength between the two output fibers. Each MZI comprises two thermal phase tuners, with one of them embedded in each of its branches. The complementary MZI outputs are each combined with their other wavelength counterparts by a cascade of directional coupler based WLCs. Finally, light is out-coupled with further ECs from the PIC to the attached fiber array (FA). While grating couplers have also been implemented in low confinement SiN platforms , due to their limited optical bandwidth, they would not be suitable as an output coupler for the present architecture in which wavelengths spanning over the entire visible range have to be supported, even though substantial progress has been made in that regard . In its overall functionality, the PIC can be seen to be a wavelength selective switch such as is also required in communications at different wavelengths, typically implemented with arrayed waveguide gratings (AWG) or Echelle gratings due to the requirement for small channel spacing . The directional coupler based WLCs used here, while not enabling such closely packed channels, have the advantage of being straightforwardly extendable to the entire visible wavelength range and beyond.
The layout of the SiN PIC is shown in Fig. 3. Free space coupling via a pair of convex focusing and collimating lenses is implemented at the input of the PIC. For the three wavelengths 405, 488, and 640 nm, laser diodes are directly used as light sources. The light at 561 nm is supplied by a compact FDDL sourced with a 1122 nm laser diode [Fig. 4(a)].
At the output of the PIC, a FA is attached with a process ensuring a direct mechanical contact between PIC and FA , without interposed epoxy, which is essential for supporting short wavelengths such as 405 nm that lead to rapid degradation of organic materials. The fiber loop labeled close to the output port in Fig. 3 was used to align the FA, by aligning two auxiliary fibers embedded in the FA to the loop and measuring the transmission. The 405 nm input EC is also protected with an endcap . High power densities at this near-UV wavelength lead to photo-induced degradation of the input facet, with the optical tweezer effect accelerating the accumulation of carbon contaminants, which are subsequently deteriorated by the high energy photons . An endcap as implemented here allows focusing the light inside the material, leading to a reduction of the power density at the air interface by several orders of magnitude, significantly improving its long-term reliability . With endcap, power levels at 405 nm and at 532 nm, close to the 561 nm used here, have been applied with respectively a couple of 100 mW and 3 W, without noticeable degradation of the coupling efficiencies. More systematic reliability testing is under way.
Unused WLC drop ports are terminated by being routed to the edge of the chip. They are combined beforehand with low reflection Y-junctions to facilitate routing (while light cannot be combined without losses from independent waveguides into a unique single mode waveguide, non-supported modes are primarily radiated out without generating noticeable back-reflections). Due to the optical modes being highly delocalized from the SiN core, back-reflections at the edges of the chip are determined by the SiO2-air interface and in the order of -14 dB.
The ordering of the ports at the input of the chip are constrained by bending radii: Since the longest wavelengths, 561 and 640 nm, require the largest radii, their input ports are chosen to be the first and last, facilitating the implementation of large, low-loss bends (see section 4.1).
The mounted module, with dimensions 56 mm by 89.5 mm by 15 mm, has a volume that is about 2 orders of magnitude smaller than currently commercially available equivalents. While it does not yet contain the required electronics, the size of current commercial modules is limited by the discrete optics so that a very sizeable volume reduction is also expected for a fully self-contained module once the electronics have been miniaturized.
3. System chip measurement
Prior to reporting the details of the individual devices used in the PIC, we describe here the overall performance of the PIC and of the assembled module. Light with transverse magnetic (TM) polarization, for which the PIC has been designed, is injected via a cleaved fiber through the nominal output ports of the chip and collected with free-space optics (microscope objective and large area detector) at its nominal input ports. Due to Lorentz reciprocity, recorded transfer functions correspond to those also obtained during normal chip operation. The fiber is a PANDA type polarization maintaining fiber with a core diameter adapted to maintain single mode operation in the wavelength range of interest (Nufern PM-S405-XP). It has a 1/e2 mode field diameter (MFD) ranging from 3.3 µm at 405 nm to 4.6 µm at 630 nm. Since it is also used to populate the FA used for module assembly, the ILs at the PIC-to-fiber interface also correspond to those expected for the full module. Additional ILs occurring in the free-space coupling between the lasers and the PIC are not taken into account in this experiment and are characterized at the module level. Direct measurement of the collimated laser light without interposed PIC is used for normalization of the recorded raw data and enables the assessment of the PIC induced ILs due to mode mismatch at the fiber interface and internal losses of on-chip devices.
Two sets of measurements are reported here. In a first set of measurements, light was injected with a fixed wavelength corresponding to one of the four wavelengths used during normal module operation, while the setpoints of the switches were varied. In a second experiment, the light source was replaced by a supercontinuum source followed by a tunable notch filter, used to characterize the overall spectral response of the PIC.
Results for the first characterization experiment are shown in Fig. 5, for which the voltage applied to the MZIs / switches was varied between 0 to 15 V for either of the two thermal phase tuners. It is apparent that light is split between the two fibers with ILs of 6 ± 1 dB at 488, 561, and 640 nm, corresponding to a slight reduction in performance compared to a previously reported single output fiber version of the chip, for which 5 dB ILs at maximum switch transmission was obtained consistently for these three channels . Additional losses are attributed to the imbalance of the 2-by-2 MMIs used at the MZI outputs in this PIC, as well as to waveguide crossings (that could not be implemented with 90o angles throughout). The increased ILs at 405 nm (12 ± 1 dB) are due to an increased mode mismatch at the fiber interfaces. As explained in Section 4.2, these could be reduced by 4 dB by resizing the waveguide width at the corresponding EC, at the price of 1-2 dB extra losses for the other 3 wavelengths, equalizing ILs to ∼8 dB for all. As shown below in this section, further 1-3 dB improvement can also be obtained by retargeting the spectral response of the devices. When assembling the PIC inside the module, an additional average ∼1.5 dB ILs are observed across the four wavelengths, due to ILs occurring during free-space in-coupling of the beams at the input of the PIC .
An extinction ratio above 10 dB is obtained for all devices. Confocal scanning fluorescence microscopy is a typical application that requires a high dynamic range. While taking images requires power levels as low as 10 to 100 μW, bleaching of the samples requires up to 10 mW from the laser sources. Without external attenuation, this dynamic range of 1:1000 can only be reached by running the laser diodes below threshold. This results in high noise levels and highly detrimental reduction of the imaging quality. Combining the 10 dB external attenuation achieved here with a factor 10 reduction in laser diode output power enables 20 dB dynamic range (OD2 attenuation) while maintaining low noise laser diode operation. An external attenuation of at least 20 dB would be ideal here, which could be reached by reducing the imbalance of the 2-by-2 MMIs used in the switches, by retargeting the device lengths (see subsection 4.3). Stray light might also play a role here as a better, but still limited extinction in the order of 16 dB was observed in a single output fiber version of the PIC only utilizing nominally balanced 1-by-2 MMIs , however we have not been able to precisely quantify it yet.
Spectral characterization of the system chip is shown in Fig. 6, in which each curve corresponds to the transmission between one input port and a common output port. Due to the low power of the supercontinuum source used here around 405 nm, the spectrally resolved transfer function between the 405 nm input port and the fiber coupled output port could not be reliably resolved and is not shown. As in the previous data, recorded power levels are normalized relative to the input power, which varies with wavelength due to the characteristics of the supercontinuum source. Moreover, the plotted data is normalized relative to the peak transmission, since ILs at a fixed wavelength can be more reliably extracted from the previous dataset in which the switch setpoint was swept.
It can be seen that the spectral responses are shifted towards longer wavelengths by 12, 23, and 35 nm, respectively for the target wavelengths 488, 561, and 640 nm. As shown in the following, this is mainly due to the WLCs that were suboptimally sized due to a systematic bias between simulations and experiments (Section 4.4), which could straightforwardly be taken into account in a further chip iteration (the bias proved to be systematic across fabrication runs and to correspond to a slight deviation from the expected SiN film thickness, initially assumed to be at 22 nm and later corrected to 26 nm). The 3-dB passband of the transfer functions is also mostly limited by the WLCs and was extracted to be 48, 58, and 53 nm for the three optical paths with target wavelengths 488, 561 and 640 nm. As mentioned earlier, retargeting of the spectral responses would allow recovering 1, 1.5 and 3 dB, respectively.
An overall comparison between the measured PIC transmission for each of the wavelengths and fiber ports and predictions from power budgets derived from device level characteristics reported in the following can be found in Table 1. Overall, a relatively good consistency is obtained. ∼1 dB ILs for MMIs are estimated from experimental data as an average over device type (MMIs in 1-by-2 and 2-by-2 configurations, as well as targeted towards different wavelengths). The relatively high WLC ILs can be compensated by retargeting them, as also seen in the spectral transfer functions in Fig. 6.
4. Device level characterization
4.1 Waveguides, bends, and crossings
We performed a mode sweep analysis with different waveguide widths and wavelengths using a finite element mode solver with a 10 by 10 µm simulation domain bounded by perfectly matched layer boundary conditions (Fig. 7), based on which the single mode cutoff waveguide widths were estimated as 1.4, 1.8, 2.4, and 2.7 µm for the 405, 488, 561, and 640 nm wavelengths, respectively.
For routing inside the chip, waveguide widths of 1.2, 1.5, 1.6, and 2.2 µm were selected, slightly below the single mode cutoff conditions. Waveguides after combiners are sized according to the largest wavelength, in order to maintain sufficient confinement for all. However, this also leads the waveguides to be multimode for the shortest wavelengths, in particular 405 nm, after the downstream combiner stages, so that additional losses may also result as a consequence of mode conversion in bends.
We calculated bending and crossing losses occurring during routing in the chip for the nominal waveguide widths. At a fixed radius, losses for a 90-degree bend grow with wavelength, as expected [Fig. 8(a)]. A detailed analysis, however, showed these to be primarily due to mode mismatch between straight and bent waveguide sections, as opposed to single-mode bending losses inside the bent waveguides, and could thus have been improved with parabolic bends. The bending radii in the final system layout are chosen to be 1.3, 1.7, 5, and 7 mm in order to maintain the losses per bend in the order of 0.01 dB. In retrospect, this criterion was very conservatively set given the other losses in the system, and bending radii could have been reduced to 0.5, 1, 2.2, and 3 mm, while maintaining bending losses below 0.1 dB.
As seen in Fig. 3, a substantial number of waveguide crossings proved to be necessary in order to implement a compact chip. Figure 8(b) shows the results of three-dimensional (3D) finite-difference time domain (FDTD) simulations, in which the angle between crossing waveguides was swept. As expected, best results are obtained for 90 degree crossing angles, with losses of ∼0.2, ∼0.15, ∼0.1, and ∼0.07 dB at wavelengths of 405, 488, 561, and 640 nm, respectively. Cross-talk levels also drop from below -35 to below -47 dB as the crossing angle is increased from 30o to 90o. Overall, due to the low bending and crossing losses, as well as due to the low losses of the TriPleX platform, we expect measured losses to be dominated by mode mismatch at ECs, as well as internal losses arising at the MMIs used inside the switches as well as in the WLCs. Resizing of the WLCs in particular is expected to lead to significant improvements (Section 4.4).
4.2 Optical interfacing
As explained in Section 3, inside the module, light is free-space coupled at the input of the PIC and fiber coupled at its output. In order to optimize mode matching, the width of the waveguide is tapered at the ECs.
We implemented a series of test structures to extract the PIC-to-fiber interface losses as a function of EC waveguide width for each of the four wavelengths. As seen in Fig. 9, for the 488, 561, and 640 nm wavelengths, the smallest interface losses were obtained for 1 µm wide ECs, the smallest that could be fabricated based on the design rules of the process. They are in the order of 1 to 2 dB and increase as the wavelength is reduced, as the minimum waveguide width then becomes more constraining. These are inverse tapers for which the mode is close to circular and is expanded in both directions, due to the waveguide width being substantially below the single mode cutoff condition. Losses for the 405 nm wavelength (8 dB) are high for the 1 μm EC waveguide width, so that here a direct taper for which the waveguide width is increased at the edge of the chip proved to be optimal. For 405 nm, best results are obtained for a ∼3.5 μm EC waveguide width. The remaining 4 dB ILs are then primarily due to the large anisotropy of the waveguide mode, that is expanded in one direction only. It should also be noted that these ILs were all recorded using the selected visible-wavelength single-mode fiber, for which the MFD also varies with wavelength, which allowed us to maintain low insertion losses across 488 to 640 nm for a single EC geometry (1 μm width). Below 488 nm, however, the MFD of the fiber and of the EC trend quite differently, leading to the high losses at 405 nm for the inverse taper.
As also explained in Section 3, since all the multiplexed wavelengths are coupled out at the same EC with a single waveguide width, a trade-off had to be struck in regards to the wavelength dependent ILs. In the system chip we opted for a 1 μm EC, resulting in optimum results in the 488 to 640 nm wavelength range (1-2 dB ILs), but also high insertion losses at 405 nm (8 dB). Alternatively, we could have chosen a 3.5 μm output EC width, with moderate ILs for all wavelengths (2-4 dB). Since the requirement for low coupling losses is primarily driven by the need to ensure enough output power can be reached to service relevant microscopy applications, the presently implemented configuration appears optimum in view of the power sourcing capabilities of laser diodes and required power levels at different wavelengths. Indeed, typical 405 nm laser diodes can routinely source in the order of 300 mW, while longer wavelength laser diodes, e.g. at 515 mW, typically source only 100 mW. The FDDL used here to supply the 561 nm wavelength generates a maximum 50 mW. Moreover, the required power levels at 488 nm and 561 nm are typically higher than at 405 nm for common fluorescence microscopy applications, so that penalizing the 405 nm ILs to improve the ILs at the other wavelengths appears to be the right choice here.
Since ECs are wavelength specific at the input of the PIC, we could use separate EC widths there. These were chosen close to the optimum required for fiber coupling in order to also allow fiber connectorization, and are 4.25 μm for 405 nm, 1 μm for 488 and 561 nm, and 1.1 μm for 640 nm. Far field diffraction angles were obtained by applying a far-field transformation to the mode profiles and also experimentally determined with a beam profiler (Table 2). As already mentioned, a pair of commercial collimating/focusing convex lenses was then selected to map them to the laser emission.
Dynamic optical power regulation is not only an essential function in communication systems, for example for equalizing the optical intensity of different wavelength division multiplexed (WDM) channels , but also plays a central role in life science applications, for example to prevent damaging biological or medical samples at excessive light intensities  or to switch between fluorophore bleaching and excitation modes. Moreover, when using external attenuation, the sourced laser power can remain at the fixed level, increasing the stability of the total system .
For the PIC utilized in this work, we developed MZI based switches with dual waveguide output, also serving for variable power splitting between the optical paths routed to the two output fibers. As shown in Fig. 10, the switch consists of a 1-by-2 MMI used as a 3-dB power splitter [Fig. 10(a)], two MZI-arms consisting in waveguides with thermal phase tuners, and a 2-by-2 MMI used as a power combiner / splitter at the output [Fig. 10(c)]. The MMIs have tapered waveguides at their in- and outputs and have been individually optimized for the target wavelength of each switch.
The design parameters of the 1 × 2 and 2 × 2 MMIs are summarized in Table 3.
Test structure measurements show excess insertion losses below 1.5 dB for both the 1 × 2 and 2 × 2 MMIs for all targeted wavelengths. As expected based on the symmetry of the devices, the 1 × 2 MMIs also feature a negligible output-power splitting-ratio imbalance. However, in the order of 20% power imbalance, defined as the deviation of the ratio of output powers from unity, is observed at the outputs of the 2 × 2 MMI devices. It is expected to result from the deviation of the SiN film thickness from the designed for 22 nm target (see below). This sensitivity on fabrication tolerances could be reduced by reoptimizing the width of the access waveguides . The imbalance is expected to be the main cause of the finite switch extinction seen in Fig. 5 (even though a higher imbalance would be required to fully account for it).
The spectral transfer functions of the wavelength optimized MMIs has been recorded on test structures, as previously with a supercontinuum light source followed by a tunable optical filter. They are shown in Figs. 11 and 12, respectively for the 1-by-2 and 2-by-2 devices. The data is normalized relative to reference measurements done on straight waveguides, so that wavelength dependent coupling losses are taken into account, and the curves subsequently normalized to the maximum observed optical power transmission. Interruptions in the plotted datasets correspond to spectral regions where the supercontinuum source was instable and this de-embedding procedure thus not reliable. Moreover, the low power output of the supercontinuum source at short wavelengths prevented reliable characterization below 400 nm. Experimental data is overlaid with simulation results assuming a film thickness of 26 nm, which results in the best overlay for all devices (as is also the case for the WLCs, see Section 4.4).
The 1 × 2 MMIs have a relatively large 3-dB passband, which is 229 nm, 446 nm, 642 nm, and 673 nm, respectively for devices having their center wavelength at 405, 488, 561 and 640 nm. The MZI passbands are thus limited by the narrow passbands of the 2 × 2 MMIs, which are ∼70, ∼160, ∼180 and ∼185 nm.
The 1500 μm long thermal phase shifters are formed by 20 μm wide chromium heaters deposited on top of the 8 μm thick upper SiO2 cladding layer, see section III-D of  for more details on their fabrication. Thermal cross-talk between the phase shifters is reduced by placing them 80 µm apart. They feature a resistance of 400 ± 50 Ω. Variation of their impedance is partially attributed to the metal interconnects to the pad frame, which have varying length. The electrical power required to achieve a 2π phase shift (P2π) is 270, 307, 375, and 450 mW, at 405, 488, 561, and 640 nm, respectively. The wavelength dependency of the phase shift efficiency results from the wavelength dependency of the materials’ thermo-optic coefficients. Moreover, the response time of the phase tuners is measured by applying a pulse shaped signal. The measured 1/e response time of 100 µs corresponds to a cutoff frequency of 1.6 kHz.
4.4 Wavelength combiner (WLC)
A multiplexing scheme adapted to the low confinement platform used here is based on cascaded symmetric directional couplers and allows combining wavelengths over the entire visible range. Rather than requiring similar waveguide properties over this wide wavelength range, as an AWG or Echelle grating would, it exploits instead the widely differing waveguide properties as wavelengths are spanned over a large range. Each of the three cascaded wavelength combiners (WLC), as represented in Fig. 13(a), has a waveguide width designed to provide high confinement, and thus low evanescent coupling, for the shorter wavelengths, and reduced confinement for the longer wavelength, that, combined with the right coupling length, results in near 100% coupling. Thus, the shorter wavelengths stay in the waveguide in which they were injected and the longer wavelength is coupled over to it, so that they all exit the device through the same output port. These stages successively add 488 to 405 nm, 561 nm to the other two, and 640 nm to the previous three. The naming of the devices is based on the wavelength added by the combiner to the bus waveguide, which in each of these is the next longer wavelength.
The wavelength selectivity of directional or contra-directional couplers has been used in the past to implement wavelength multiplexers and filters , including in the SiN platform . In forward directional couplers, the difference between the beating length of two multiplexed wavelengths has been leveraged. If the length of the device is sized to be an even number of half beating lengths for one wavelength and an odd number of half beating lengths for the other wavelength, one wavelength is transmitted straight through and the other coupled over to the opposite waveguide. Consequently, the device can be used as a two wavelength multiplexer or demultiplexer. However, it becomes generally difficult to handle more than two wavelengths with this method, for example add a wavelength (ADD port) to several other wavelengths already contained in a main bus (IN port), as needed here, as a length would have to be found that corresponds, e.g., to an even number of beating lengths for all wavelengths injected in the IN port and an odd number of beating lengths for the wavelength injected in the ADD port. A three wavelength demultiplexer can be implemented with a directional coupler by adding a grating to the device . Two wavelengths are demultiplexed with the aforementioned feed-forward configuration, while the third wavelength is reflected back by the grating, coupled over to the opposite waveguide, and exits the device via its fourth port. Since the coupling length for this third wavelength is determined by the position of the grating, it can be freely chosen without interdependencies with the forward coupling length of the other two wavelengths. In principle, the wavelength separation in contra-directional configuration could be extended to a larger number of wavelengths by inscribing several, in some cases superimposed gratings in the device , each selecting a wavelength and reflecting it back at a different effective reflection point selected for obtaining the required coupling length. Grating based approaches would, however, result in a number of challenges in the low index contrast platform we are using, since they would result in very long and narrowband devices and would impose a much smaller critical dimension on the lithography than available in the utilized process. Here, we rely only on the varying light confinement in our waveguide geometry over the visible range. Due to the low confinement of the utilized platform, the mode size and thus the resulting beating length are very sensitive to the wavelength. As a consequence, the WLCs can be engineered to have a much longer beating length for all the shorter wavelengths already preexisting in the main bus compared to the beating length of the added longer wavelength. Consequently, the coupling length needs to be chosen only according to the beating length of the added wavelength, while the performance in regards to the transmission of the other wavelengths is relatively insensitive to the chosen device length. Thus, this device configuration can be straightforwardly extended to a number of targeted wavelengths. As a consequence of the chosen multiplexing mechanism, the minimum wavelength resolution is however limited, as discussed further in the following.
Figure 13(b) exemplarily shows the lowest order supermode profiles of WLC488, the stage adding the 488 nm wavelength to the 405 nm bus, at 405 and 488 nm. The vanishing mode overlap at 405 nm (upper panel) and the much increased mode overlap at 488 nm (lower panel) can be readily seen. Waveguide width and coupler gap are optimized in simulations based on the eigenmode expansion method in order to maximize the coupling of the longer wavelength, while keeping the coupling of the shorter wavelength at a minimum. Top-view field profiles, recorded at the mid-height of the waveguide slab, are shown for WLC488 for both the shorter and longer wavelength in Fig. 13(c) (405 nm, upper panel, and 488 nm, bottom panel).
Table 4 summarizes the design parameters for the three WLC stages.
The WLC stages have been individually characterized with breakout structures following the previously described methodology. In addition to the nominal devices implemented on the system chip, combiners with different coupling lengths were investigated as test structures. Figure 14 compares simulations to experimental data for WLCs of varying coupling length, wherein a 26 nm SiN layer thickness also resulted in the best overlay, confirming the fabrication bias already identified with the MMIs. A cross-coupling ≥ 80% is obtained for straight coupler lengths of 1535, 3020, and 6114 µm for WLC 488 nm, WLC 561 nm, and WLC 640 nm, respectively. These optimum lengths are larger than those implemented on the system chip, as simulations were not experimentally calibrated in the initial design phase and coupling strengths had been overestimated. As shown in Fig. 14, the maximum experimentally recorded coupling strength for the added wavelengths stays below 100% (particularly so for 561 nm in the WLC561 combiner), indicating that there might be some small phase mismatch in the fabricated devices, the cause of which we have however not been able to identify.
The spectral response of the WLCs, with the sizing implemented in the system chip, was also measured [Fig. 15(a)]. The recorded 3-dB bandwidths were ∼52, ∼55, and ∼ 60 nm for the add-ports of WLC488, WLC561, and WLC640, respectively. The passbands of each of these devices are shifted to longer wavelengths by about 20 nm, at which the coupling strength is higher, due to their undersized lengths, resulting in insertion losses below 2 dB at the targeted wavelengths. It is apparent that the insertion losses recorded from test structures are smaller than those seen at the system chip level (see Table 1), as a consequence of a larger off-centering of the passbands in the latter. This might be partially due to process bias, as the test structures and the system chip were obtained from separate runs. However, we attribute it primarily in differences in the S-bend routing before and after the straight directional coupler sections leading to a longer effective coupling length in the case of the test structures.
Figure 15(b) shows the corresponding simulations over the entire wavelength range from 400 nm to 650 nm. From this, it is also apparent how closely the wavelengths can be stacked while maintaining a maximum amount of losses and cross-talk. The density of wavelength stacking is fundamentally limited by the fact that the shorter wavelengths, that are not supposed to be cross-coupled between waveguides by the device, have a much suppressed coupling due to a better waveguide confinement, but not a perfectly suppressed one. For example, it can be seen that the WLC adding the 488 nm wavelength to the 405 nm wavelength attenuates the latter by 4% (the 4% cross-coupling level is shown by the horizontal grey line), or, conversely, results in 4% cross-talk of the 405 nm channel onto the 488 nm channel if used as a demultiplexer. A similar 4% attenuation / cross-talk is seen for WLC561 at 488 nm and WLC640 at 561 nm. The minimum stacking is thus in the order of 80 nm for this 4% tolerance, which is fine for the application pursued here (in which it only results in acceptable losses) but is evidently much larger than what can be achieved with AWGs or Echelle gratings.
We demonstrated a miniaturized dual-output-fiber PIC-based multi-color laser engine implemented in the TriPleX silicon nitride technology with an optimized layer thickness of ∼26 nm, operating in the visible spectrum and aimed at fluorescent detection and imaging in the life-sciences. The PIC combines, variably attenuates and variably splits the four wavelengths 405, 488, 561, and 640 nm, thus covering the full visible range. We designed cascaded wavelength combiners as well as a PIC-to-fiber interface servicing the entire required wavelength range. Integrated dynamic on-chip power adjustment and fiber switch functionality are accomplished by using MZI based switches optimized for each specific color. Device-level characteristics are reported for all constituting components. Insertion losses of the complete system chip, including those arising from mode mismatch at the PIC to fiber interface, are 6 ± 1 dB for the three longer wavelengths (488, 561, and 640 nm) and 12 ± 1 dB at 405 nm. These additional losses are mostly due to mode mismatch at the PIC-to-fiber interface. A different output taper configuration would lead to losses in the order of 8 dB for all four wavelengths, but is expected to be suboptimum in view of wavelength dependent power levels required to service targeted fluorescent microscopy applications. Further improvements can be obtained by retargeting the on-chip devices based on the SiN layer thickness fitted from experimental results.
Horizon 2020 Framework Programme (688519).
TOPTICA Photonics AG, LioniX Int. BV and Miltenyi Biotec BV have a commercial interest in the developed technology.
1. A. Dhakal, P. C. Wuytens, F. Peyskens, K. Jans, N. L. Thomas, and R. Baets, “Nanophotonic waveguide enhanced Raman spectroscopy of biological submonolayers,” ACS Photonics 3(11), 2141–2149 (2016). [CrossRef]
2. P. Muñoz, G. Micó, L. A. Bru, D. Pastor, D. Pérez, J. D. Doménech, J. Fernández, R. Baños, B. Gargallo, R. Alemany, A. M. Sánchez, J. M. Cirera, R. Mas, and C. Domínguez, “Silicon nitride photonic integration platforms for visible, near-infrared and mid-infrared applications,” Sensors 17(9), 2088 (2017). [CrossRef]
3. D. Duval and L. M. Lechuga, “Breakthroughs in photonics 2012: 2012 breakthroughs in lab-on-a-chip and optical biosensors,” IEEE Photonics J. 5(2), 0700906 (2013). [CrossRef]
4. R. Soref, “Mid-infrared photonics in silicon and germanium,” Nat. Photonics 4(8), 495–497 (2010). [CrossRef]
5. N. Daldosso, M. Melchiorri, F. Riboli, M. Girardini, G. Pucker, M. Crivellari, P. Bellutti, A. Lui, and L. Pavesi, “Comparison Among Various Si3N4 Waveguide Geometries Grown Within a CMOS Fabrication Pilot Line,” J. Lightwave Technol. 22(7), 1734–1740 (2004). [CrossRef]
6. S. Romero-García, F. Merget, F. Zhong, H. Finkelstein, and J. Witzens, “Silicon nitride CMOS-compatible platform for integrated photonics applications at visible wavelengths,” Opt. Express 21(12), 14036–14046 (2013). [CrossRef]
7. D. J. Blumenthal, R. Heideman, D. Geuzebroek, A. Leinse, and C. Roeloffzen, “Silicon Nitride in Silicon Photonics,” Proc. IEEE 106(12), 2209–2231 (2018). [CrossRef]
8. R. Baets, A. Z. Subramanian, S. Clemmen, B. Kuyken, P. Bienstman, N. Le Thomas, G. Roelkens, D. Van Thourhout, P. Helin, and S. Severi, “Silicon Photonics: silicon nitride versus silicon-on-insulator,” in Optical Fiber Communication Conference, OSA Technical Digest (online) (Optical Society of America, 2016).
9. D. T. Spencer, J. F. Bauters, M. J. R. Heck, and J. E. Bowers, “Integrated waveguide coupled Si3N4 resonators in the ultrahigh-Q regime,” Optica 1(3), 153–157 (2014). [CrossRef]
10. Y. Hibino, “Recent Advances in High-Density and Large-Scale AWG Multi/Demultiplexers With Higher Index-Contrast Silica-Based PLCs,” IEEE J. Sel. Top. Quantum Electron. 8(6), 1090–1101 (2002). [CrossRef]
11. M. H. P. Pfeiffer, A. Kordts, V. Brasch, M. Zervas, M. Geiselmann, J. D. Jost, and T. J. Kippenberg, “Photonic Damascene process for integrated high-Q microresonator based nonlinear photonics,” Optica 3(1), 20–25 (2016). [CrossRef]
12. A. Dhakal, P. C. Wuytens, A. Raza, N. L. Thomas, and R. Baets, “Silicon nitride background in nanophotonic waveguide enhanced Raman spectroscopy,” Materials 10(2), 140 (2017). [CrossRef]
13. M. Lang, “Lasers for Confocal Microscopy: All Colors of the Rainbow,” Imag. & Microsc. 2, 2011 (2011).
14. B.-C. Chen, W. R. Legant, K. Wang, L. Shao, D. E. Milkie, M. W. Davidson, C. Janetopoulos, X. S. Wu, J. A. Hammer III, Z. Liu, B. P. English, Y. Mimori-Kiyosue, D. P. Romero, A. T. Ritter, J. Lippincott-Schwartz, L. Fritz-Laylin, R. D. Mullins, D. M. Mitchell, J. N. Bembenek, A.-C. Reymann, R. Böhme, S. W. Grill, J. T. Wang, G. Seydoux, U. S. Tulu, D. P. Kiehart, and E. Betzig, “Lattice light-sheet microscopy: Imaging molecules to embryos at high spatiotemporal resolution,” Science 346(6208), 1257998 (2014). [CrossRef]
15. S. P. Perfetto, P. K. Chattopadhyay, and M. Roederer, “Seventeen-colour flow cytometry: unravelling the immune system,” Nat. Rev. Immunol. 4(8), 648–655 (2004). [CrossRef]
16. W. D. Sacher, X. Luo, Y. Yang, F.-D. Chen, T. Lordello, J. C. C. Mak, X. Liu, T. Hu, T. Xue, P. G.-Q. Lo, M. L. Roukes, and J. K. S. Poon, “Visible-light silicon nitride waveguide devices and implantable neurophotonic probes on thinned 200 mm silicon wafers,” Opt. Express 27(26), 37400–37418 (2019). [CrossRef]
17. A. Mohanty, Q. Li, M. A. Tadayon, S. P. Roberts, G. R. Bhatt, E. Shim, X. Ji, J. Cardenas, S. A. Miller, A. Kepecs, and M. Lipson, “Reconfigurable nanophotonic silicon probes for sub-millisecond deep-brain optical stimulation,” Nat. Biomed. Eng. 4(2), 223–231 (2020). [CrossRef]
18. J. R. Moffitt, C. Osseforth, and J. Michaelis, “Time-gating improves the spatial resolution of STED microscopy,” Opt. Express 19(5), 4242–4254 (2011). [CrossRef]
19. K. Wörhoff, R. G. Heideman, A. Leinse, and M. Hoekman, “TriPleX: a versatile dielectric photonic platform,” Adv. Opt. Technol. 4(2), 189–207 (2015). [CrossRef]
20. A. Leinse, R. G. Heideman, M. Hoekman, F. Schreuder, F. Falke, C. G. H. Roeloffzen, L. Zhuang, M. Burla, D. Marpaung, D. H. Geuzebroek, R. Dekker, E. J. Klein, P. W. L. van Dijk, and R. M. Oldenbeuving, “TriPleX waveguide platform: low-loss technology over a wide wavelength range,” Proc. SPIE 8767, 87670E (2013). [CrossRef]
21. D. Geuzebroek, R. Dekker, E. Klein, and J. V. Kerkhof, “Photonic integrated circuits for visible light and near infrared: controlling transport and properties of light,” Sens. Actuators, B 223, 952–956 (2016). [CrossRef]
22. A. Gorin, A. Jaouad, E. Grondin, V. Aimez, and P. Charette, “Fabrication of silicon nitride waveguides for visible-light using PECVD: a study of the effect of plasma frequency on optical properties,” Opt. Express 16(18), 13509–13516 (2008). [CrossRef]
23. N. Hosseini, R. Dekker, M. Hoekman, M. Dekkers, J. Bos, A. Leinse, and R. Heideman, “Stress-optic modulator in TriPleX platform using a piezoelectric lead zirconate titanate (PZT) thin film,” Opt. Express 23(11), 14018–14026 (2015). [CrossRef]
24. D. Pérez, J. Fernández, R. Baños, J. D. Doménech, A. M. Sánchez, J. M. Cirera, R. Mas, J. Sánchez, S. Durán, E. Pardo, C. Domínguez, D. Pastor, J. Capmany, and P. Muñoz, “Thermal tuners on a Silicon Nitride platform,” arXiv:1604.02958 (2016).
25. S. Romero-García, T. Klos, E. Klein, J. Leuermann, D. Geuzebroek, J. V. Kerkhof, M. Büscher, J. Krieg, P. Leisching, and J. Witzens, “Photonic integrated circuits for multi-color laser engines,” Proc. SPIE 10108, 101080Z (2017). [CrossRef]
26. S. Romero-García, J. Leuermann, F. Merget, T. Klos, S. Koch, P. Leisching, E. Klein, D. Geuzebroek, R. Dekker, J. Van Kerkhof, J. Krieg, M. Büscher, and J. Witzens, “Silicon nitride photonic integrated circuits for multi-color optical engines with application in flow cytometry,” Proc. IEEE Photon. Soc. Summ. Top. Meet. (2017).
27. A. T. Mashayekh, T. Klos, S. Koch, F. Merget, D. Geuzebroek, E. Klein, T. Veenstra, M. Büscher, P. Leisching, and J. Witzens, “Miniaturized PIC multi-color laser engines for the life sciences,” Proc. SPIE 10922, 109221U (2019). [CrossRef]
28. iChrome Datasheet. https://www.toptica.com/products/multi-laser-engines/ichrome-mle/
29. S. Romero-García, F. Merget, F. Zhong, H. Finkelstein, and J. Witzens, “Visible wavelength silicon nitride focusing grating coupler with AlCu/TiN reflector,” Opt. Lett. 38(14), 2521–2523 (2013). [CrossRef]
30. J. C. C. Mak, Q. Wilmart, S. Olivier, S. Menezo, and J. K. S. Poon, “Silicon nitride-on-silicon bi-layer grating couplers designed by a global optimization method,” Opt. Express 26(10), 13656–13665 (2018). [CrossRef]
31. C. R. Doerr, L. L. Buhl, L. Chen, and N. Dupuis, “Monolithic Flexible-Grid 1 ( 2 Wavelength-Selective Switch in Silicon Photonics,” J. Lightwave Technol. 30(4), 473–478 (2012). [CrossRef]
32. D. Geuzebroek, R. Dekker, and P. van Dijk, “Photonics Packaging Made Visible,” Opt. Photonik 12(5), 34–38 (2017). [CrossRef]
33. C. C. Kim, Y. Choi, Y. H. Jang, M. K. Kang, Minho Joo, and M. S. Noh, “Degradation modes of high-power InGaN/GaN laser diodes on low-defect GaN substrates,” Proc. SPIE 6894, 68940O (2008). [CrossRef]
34. A. Knigge, C. Knothe, U. Oechsner, and G. Federau, “Fibers with End Caps,” Physics’ Best2–5 (2017).
36. M. Lenzi, S. Tebaldini, D. D. Mola, S. Brunazzi, and L. Cibinetto, “Power control in the photonic domain based on integrated arrays of optical variable attenuators in glass-on-silicon technology,” IEEE J. Sel. Top. Quantum Electron. 5(5), 1289–1297 (1999). [CrossRef]
37. M. T. Hill, X. J. M. Leijtens, G. D. Khoe, and M. K. Smit, “Optimizing imbalance and loss in 2×2 3-dB multimode interference couplers via access waveguide width,” J. Lightwave Technol. 21(10), 2305–2313 (2003). [CrossRef]
38. C. G. H. Roeloffzen, M. Hoekman, E. J. Klein, L. S. Wevers, R. Bernardus Timens, D. Marchenko, D. Geskus, R. Dekker, A. Alippi, R. Grootjans, A. van Rees, R. M. Oldenbeuving, J. P. Epping, R. G. Heideman, K. Wörhoff, A. Leinse, D. Geuzebroek, E. Schreuder, P. W. L. van Dijk, I. Visscher, C. Taddei, Y. Fan, C. Taballione, Y. Liu, D. Marpaung, L. Zhuang, M. Benelajla, and K.-J. Boller, “Low-Loss Si3N4 TriPleX Optical Waveguides: Technology and Applications Overview,” IEEE J. Sel. Top. Quantum Electron. 24(4), 1–21 (2018). [CrossRef]
39. J. H. Song, J. H. Lim, R. K. Kim, K. S. Lee, K.-Y. Kim, J. Cho, D. Han, S. Jung, Y. Oh, and D.-H. Jang, “Bragg Grating-Assisted WDM Filter for Integrated Optical Triplexer Transceivers,” IEEE Photonics Technol. Lett. 17(12), 2607–2609 (2005). [CrossRef]
40. X. Nie, N. Turk, Y. Li, Z. Liu, and R. Baets, “High extinction ratio on-chip pump-rejection filter on cascaded grating-assisted contra-directional couplers in silicon nitride rib waveguides,” Opt. Lett. 44(9), 2310–2313 (2019). [CrossRef]
41. X. Zhao, Y. Wang, Q. Huang, and J. Xia, “Two-mode contra-directional coupler based on superposed grating,” Opt. Express 25(3), 2654–2665 (2017). [CrossRef]