1. Introduction

Driven by the growing need for reduced power consumption and greater bandwidth to meet the demands of exponentially increasing data transfer volumes [1], there has been a notable shift towards research and development of compact and high-performance photonic integrated circuits (PICs). PICs hold great potential for a wide range of applications, including optical communication, quantum computing, and sensor technology to mention a few [2,3]. While significant advancements have been made in the development of silicon waveguide-based photonic components and circuits using CMOS foundries [4,5], there are few fabrication services available [6] that offer integration of on-chip light sources with silicon for both prototyping and high volume mass production. This is primarily due to the indirect bandgap nature of silicon, which makes it an unsuitable gain material for building an efficient laser source and thus ineffective in utilizing the traditional fabrication platform. Semiconductor lasers based on III-V gain materials have historically proven to be ideal laser sources due to their direct band gap properties. Various approaches to integrating III-V gain devices with silicon photonics have been explored, including monolithic [7], heterogeneous [8,9], and hybrid integration [10,11]. However, each of these methods has its own set of challenges in terms of reproducibility, scalability, efficient coupling, and laser performance as compared and summarized in these reviews [2,12]. In monolithic integration, the III-V material stacks are directly grown on the silicon wafer using epitaxy and then patterned [3]. However, this approach lacks flexibility in device design and hinders laser performance testing before integration. Pre-characterization of the laser is also not possible in heterogeneous integration, which involves a complex fabrication process of wafer bonding, with additional issues such as the lattice mismatch of different materials and the thermal budget [10]. Efficient alignment and coupling are issues in hybrid integration, where laser dies are, in essence, picked and placed in line with the input coupling region of the photonic circuits.

An emerging technological solution to the challenges of various laser integration methods is photonic wire bonding (PWB) [10,13–17], where a femto-second pulsed laser three-dimensionally (3D) prints a low-loss polymer waveguide. The system, utilizing image detection and automated trajectory planning, ensures precise and efficient fabrication of 3D-printed waveguides, eliminating the need for active alignment techniques seen in some hybrid integration approaches [18]. Notably, PWB exhibits lower sensitivity to alignment, allowing for less critical placement of the laser near the coupling region and achieving reproducible insertion loss as low as 0.73$\pm$ 0.15 dB measured from experiments conducted on a hundred PWBs [16]. The alignment tolerance for PWB can be up to 30 $\mathrm{\mu}$m in the x-y plane while maintaining distances between bond interfaces (along z) in the few-hundred-micrometer range [15,19–21]; this is two to three orders of magnitude greater than techniques like flip-chip bonding, which demands sub-micron precision from all three axes [22]. The technology thus has the advantage of low loss coupling between diverse optical facets including optical fibers, surface-emitting lasers, and edge-emitting lasers to name a few [10,14,15,23]. This technique can be used to integrate optical devices operating at different wavelengths, with minor adjustments to the PWB writing recipe, which means that a single silicon photonics process can be compatible with a wide range of wavelengths. A broad range of wavelengths is more difficult to achieve using heterogenous or monolithic integration since it requires extensive growth and integration optimization, with fabrication processes that are specific to each wavelength. The printing process itself is also rapid, with reported times ranging from 30 seconds to 5 minutes for each PWB depending on the bond volume [13,16] without any process optimization. The dominant factors contributing to these times are the exposure speed governed by the galvanometer scanners and the settling time of the piezo-actuator used to control the objective’s movement along the axis. Through the advancement of tool mechanics in commercial platforms [13,24], by incorporating high-speed galvanometers with full capacity (5000 lines/s for a line length of 40 $\mathrm{\mu}$m) as well as harnessing the continuous movement capabilities of the piezo actuators, the writing time is anticipated to be reduced to well below 30 seconds even for larger PWBs [16]. The photonic wire bonds have demonstrated good mechanical stability with uncladded samples surviving rigorous water rinse during the development process, manual handling with tweezers, blow-drying by nitrogen gun, and drop tests from 1 m height [13,16,17]. Stable performance has been observed in various technically relevant conditions such as multiple temperature cycles ranging from -40 $^{\circ }$C to 85 $^{\circ }$C and damped heat conditions of 85 $^{\circ }$C at 85% humidity [13,16]. The PWBs are shown to survive power as high as 19 dBm on power tolerance tests on silicon chips; these experiments have been only limited by the damage of the silicon waveguides due to non linear absorption indicating the threshold maximum power of the bonds is higher [16]. A recent study also demonstrates the PWB’s robustness in cryogenic conditions [21]. Experimentally verified low absorption of the resist used for PWB in O-band and C-band indicates negligible temperature rise due to photothermal absorption during operation in these wavelength ranges [16]. Given that known good chips can be integrated with this process, the technology has the additional advantage of being more reliable than the heterogeneous or monolithic integration where pre-characterization of the laser is not possible and more critical process control development is required to ensure high yield. Similar to existing pick-and-place machines, PWB tools can be automated for every step starting from dispensing the resist to 3D printing and development, offering the potential for mass production [20,25–28]; thereby lowering integration cost.

In this paper, as illustrated by Fig. 1 (a) - (b), we present the successful on-chip hybrid integration of a commercially available distributed-feedback (DFB) laser with multiple passive circuits fabricated through a multi-project run via SiEPICfab [29]. The integration process involved the formation of a deep trench in the substrate [30] to achieve height matching between the laser facet and the silicon device layer. The laser was reflow soldered into the designated pocket and connected to the waveguides through PWB techniques. In addition, this work explores the utilization of a swept frequency laser (SFL) system to characterize the PICs, whereby the laser is thermally tuned through self-heating by varying its bias current, resulting in a wavelength change [31]. In high-speed electrical current modulation applications, this effect is called frequency chirp and is considered a nonlinear and undesirable parasitic effect [32,33]. However, in various applications such as LIDAR [34], 3D imaging [35], optical biosensing, and spectroscopy [36], the chirping effect has been exploited, and linearized through feedback mechanisms to achieve the desired applications. In this project, we investigated the low-speed direct modulation of DFB lasers by sweeping the injection current in a range of values above the laser’s threshold current. This allowed us to characterize their nonlinear behavior through curve fitting of experimentally acquired sweep data. The analysis facilitated the determination of the current-wavelength relationship and the generation of a calibration curve for wavelength/frequency-swept measurements. By employing this calibration fit model, we characterized the optical circuits on the chip using the SFL measurement method.

 

Fig. 1. (a) Three-dimensional chip schematics after the full assembly. The DFB laser was die bonded using tin gold preforms onto the deep trench. Photonic wire bonding was then performed to connect the laser output facet to the silicon surface coupler. (b) I. Microscope image depicting the DFB laser and photonic circuit interface after the assembly (top view). II. Post measurement SEM image of the on-chip DFB laser photonic wire bonded to the surface coupler. A roughness is observed around the middle of the PWB (marked with the black ellipsoid). The PWB is also abruptly bonded from the edge of the surface coupler instead of from the top (side view). (c) Fabrication flow diagram showing all the steps prior to assembly: I. The silicon substrate with handle wafer, 3.5 $\mu$m of buried oxide, and 220 nm of silicon waveguide layer. II. 550 nm of ZEP 520A, a positive electron beam resist was deposited through spin coating. III. Photoresist pattern after E-Beam exposure and the resist development. IV. Silicon pattern after ICP etching of the exposed silicon layer and resist strip. The passive silicon waveguide device layer was formed. V. A positive photoresist AZP4620 was spin-coated. VI. Photoresist pattern after exposure through maskless aligner and photoresist development. VII. A deep trench for the laser, created after the deep reactive ion etching into the silicon handle wafer. An approximate 80 $\mathrm{\mu}$m deep trench was formed to accommodate the DFB laser. VIII. 10 nm of titanium as an adhesion layer followed by 200 nm of gold was deposited via electron beam evaporation. IX. Final pattern after lift-off with the deep trench and the device layer. The chip was now ready for assembly (die bonding and PWB). The drawings (a and c) are not to scale.

Download Full Size | PDF

This paper highlights the potential of an SFL as a cost-effective measurement system, offering an alternative to expensive benchtop lasers by utilizing commercially available low-cost DFB laser dies. To the best of our knowledge, this is the first demonstration of an on-chip laser and photonic circuit fabrication flow using PWBs, in combination with on-chip SFL measurements.

2. Methods

2.1 Photonic circuit design

To test the performance of the integrated DFB laser as an SFL, arrays of Fabry-Perot (FP) cavities using waveguide Bragg gratings (WBG) were designed. The goal was to design each of the circuits such that the output resonance peaks could be characterized using the integrated 1310 nm DFB laser by sweeping the laser injection current and utilizing its self-heating characteristics. A preliminary characterization of the die laser revealed that the injection current sweep span was approximately 0.55 nm. Hence, the free spectral range (FSR) of the resonance peaks of the FPs in each design was required to be sufficiently small, preferably less than 0.2 nm, to facilitate the detection of multiple resonant peaks. The layout consisted of a variety of WBG-based FP cavities, connected to the single laser through a Y-splitter tree. The design and layout aggregation was carried out using a Python script in the open source mask design tool, KLayout [37], SiEPIC tools library [38] and SiEPIC Ebeam ZEP process design kit (PDK) [39]. Here, the silicon layer height, h, was 220 nm, and the width, w, was 370 nm in order to confine the single quasi-TE optical mode at the O-band operating wavelength. The simulations for the waveguides and photonic circuits were conducted using Ansys- Lumerical MODE, FDTD (Ansys, Inc.), and Transfer Matrix Method (TMM) [40] in MATLAB or Python. The wavelength of operation for the O-band resonator was between 1270 nm and 1330 nm, with an approximate center at 1310 nm. It should be noted that considering the fabrication etch bias of our process [39], the width used during the simulations was 335 nm.

Figure 2 (a) shows the schematic of a typical FP cavity with WBG as the distributed Bragg reflectors on each side. A variety of cavities were designed with different corrugation widths, w, Bragg period, $\Lambda$, number of periods, N, and the cavity length, L, with the goal of creating a reflection and transmission response of multiple peaks having high-quality factors in a span of about 0.55 nm and a central wavelength of around 1310 nm. To achieve this, the FP cavity length, L was chosen to be at least 1000 nm, based on Eq. (1), where $\Delta$$\lambda$_FSR was FSR in nm, $\lambda$_c was the central wavelength in nm and n_g was the group index of approximately 4.8. The particular FP cavity chosen to be analyzed in this paper had a corrugation width of 72 nm, a cavity length of approximately 1200 nm, a period of 290 nm, and a grating number of 125. Each FP can be tested in both transmission and reflection mode, where the reflection is measured by inserting a Y-splitter before the resonator [41] as shown in Fig. 2 (b).

 

Fig. 2. (a) The schematic of a WBG-based FP cavity. (b) The concept schematic of the FP cavity in the layout with transmission and reflection port. The input is from the on-chip laser after passing the Y-splitter tree. (c) The overall photonic circuit layout with all the devices. (d) A zoomed-in version of 3 circuits in one group. Each group of circuits can be measured by coupling GCs through an 8-channel FA and off-chip detectors. Both reflection and transmission spectra of the Bragg resonator are simultaneously measured.

Download Full Size | PDF

For the purpose of laser characterization, we also included circuits without any Bragg gratings, i.e., simple waveguides designed to measure the laser output. Figure 2 (c) and (d) display the full design layout of the chip, with a magnified version of a group of three different circuits consisting of a loop and two FP cavities that would be measured together through grating couplers (GCs) connected to the detectors via 8-channel fiber array (FA). The layout also included GCs positioned in the top corners of the chip to ensure coupling of the FA with the benchtop laser prior to on-chip laser measurements through the GCs. The alignment check was also conducted during the on-chip measurement for each group of circuits through channel 1 and channel 8 using the loopbacks. Considering that the output of the laser in PIC underwent splitting eight times through Y-branches and had loss incurred through the GC output, we estimated a significant total loss, IL_total, of 51.5 dB. Equation (2) shows how this was calculated. Based on prior photonic wire bonded laser to fiber characterization [15], the insertion loss through PWB, IL_PWB, was estimated to be 3 dB for the wavelength of 1310 nm. The loss of each Y-splitter was estimated to be 4 dB, where 3 dB was the characteristic splitting loss, IL_splitter, of the Y-branch, and 1 dB was the estimated excess loss, EL_splitter, due to scattering and other fabrication uncertainties. This excess loss emerged due to the non-optimized Y-splitter designs for this specific fabrication platform. The insertion loss for the GCs, IL_GC, was approximated to be 15 dB, as confirmed by the historical data of the GC-GC test structures in this platform. Finally, about 1.5 dB of propagation loss due to routing, IL_routing, was added (assuming propagation loss of 3 dB/cm and routing length of 5 mm).

2.2 Fabrication and assembly

The fabrication process, Fig. 1 (c), involved a combination of electron beam (E-beam) and optical lithography techniques. Specifically, E-beam lithography was used to pattern the silicon device layer, while optical lithography was employed to create the trench in which the laser was placed.

The process started by patterning a 15 mm by 15 mm chip with a positive tone photoresist ZEP 520A (Zeon Specialty Materials Inc., CA, USA) using E-beam lithography (JEOL JBX-8100FS system, JEOL, Tokyo, Japan) [42]. Silicon was removed from exposed areas through inductively coupled plasma etching (PlasmaPro 100, Oxford Cobra Instruments, UK). Using photolithography (MLA-150, Heidelberg Instruments, Germany) the trench area was patterned and then deep etched through the Bosch process (Rapier$^{\textrm {TM}}$ DRIE, SPTS Technologies Ltd., UK). This was followed by metalization through lift-off. Step-by-step details of the fabrication flow are provided in Supplement 1.

For assembly, the chip was sub-diced into 5 mm x 5 mm pieces using the dicing saw and surface-treated with 20 sccm of oxygen plasma in the plasma cleaner for 30 seconds. Using a thermal adhesive (8349 TFM, MG Chemicals, Canada), it was mounted on a rectangular aluminum shim. An O-band DFB laser (MACOM, MA, USA) was die bonded into the deep trench area using tin-gold (Sn-Au) preforms with a pick and place tool. The photonic wire bonder (Vanguard SONATA 1000, Vanguard Automation GmbH, Germany) was used to connect the silicon surface taper on the chip to the laser facet interface. To fabricate the PWB, a negative-tone photoresist (VanCore A, Vanguard Automation GmbH, Germany) was initially deposited by manual drop-casting at the target location. The tool employed image analysis to detect the facets of the laser and the silicon surface coupler, generating a model of the PWB. A femtosecond laser in the tool exposed the photoresist via two-photon absorption with a recipe tailored to the speed and exposure specified for the model [14]. The writing time (including stage movement, alignment, wire bond path calculations, and exposure) was approximately 1 minute. Afterward, the unexposed resist was removed in a wet development process by soaking in PGMEA for 20 minutes, followed by a 5-minute dip in isopropanol and nitrogen drying. It should be noted that the PWB is typically preferred as the final step in the assembly process. Figure 1 (a), (b-I), and b(-II) depict a three-dimensional schematic, a post-PWB microscope image, and a scanning electron microscope (SEM) image of the integrated laser after the optical measurements, respectively.

2.3 Measurement setup

Photonic circuit characterization was conducted using a Maple Leaf SD-100 optical probe station. The on-chip DFB laser was probed using microneedles connected to a laser diode controller (LDC) (LDC501, Stanford Research Systems, USA). Laser spectral characterization was performed using an optical spectrum analyzer (OSA) (AQ6317B, Ando Electric Co., Japan).

For the circuit characterization using SFL, both the transmission and reflection output channels were connected to the O-band mainframe (8164B, Keysight Technologies, USA) photodetectors. The on-chip laser was swept using the LDC, Fig. 3 (a-I). To compare the on-chip DFB laser measurement with an off-chip standard, the benchtop O-band laser was connected through the FA to the GC at the reflection port, and the transmission channel output of the circuit was measured by the mainframe detector, coupled to the respective GCs through the FA, Fig. 3 (a-II). A microscope image of the corresponding measurement setup of this on-chip laser and PIC with the FA and needle probes is shown in Fig. 3 (b).

 

Fig. 3. (a) Different configurations of the measurement setups: I. The PIC with FP cavity where output is measured through GCs by SFL method by the on-chip DFB laser. II. The PIC output is measured by the benchtop laser input through the reflection port. III. A DFB laser PWB to fiber for general characterization through photodetector without photonic circuit integration. (b) Top view of the experimental setup for current tuning on PIC. The electrodes of the laser are connected using the needle probes. The fiber array also in the camera view was aligned to the grating couplers. (c) Top view of the configuration of DFB laser PWB to fiber. The annotated black lines represent needle probes.

Download Full Size | PDF

As a control experiment, we also characterized another die of the MACOM DFB laser which was photonic wire bonded to a fiber; a comparable assembly to [15] but without the surface taper. The output of this DFB laser PWB to fiber was connected to a power meter (HP 81531A, Agilent Technologies, Inc., USA), Fig. 3 (a-III). Figure 3 (c) is the optical micrograph of this configuration with black annotated lines as needle probes. More details of the setup are available in Supplement 1.

2.4 Figures of merit

To evaluate the integrated laser as well as the photonic circuit performance, several figures of merit (FOM) were used.

First, we conducted measurements for the laser characterization to get a power-current-voltage (L-I-V) curve for the DFB lasers. This was used to compare the performance of the two DFB lasers assembled differently, i.e., the on-chip DFB laser photonic wire bonded to PIC and the DFB laser PWB to the fiber. Next, by utilizing the characteristic wavelength tunability of the semiconductor laser through current sweeps in a range above the threshold, we checked if the tunable range of approximately 0.55 nm was maintained by the on-chip laser with reference to the preliminary in-house characterization of these laser dies prior to the integration. This aspect of tunability was utilized to perform multiple sweeps and compute a current-wavelength calibration fit curve for the on-chip laser. This fit equation in turn was used to characterize the PICs. More details about this method are provided in Sections 3.2 and 3.3.

To evaluate the PICs, the output resonance peaks were measured, and their quality factor, Q, was calculated from the experimental data using Eq. (3). Here, λ_r is the resonance wavelength and Δλ_FWHM is the full width at half maximum of the peak. A Lorentzian fit function [43] was used to curve fit the peaks for the quality factor calculation.

Another FOM used for evaluation of the PICs was the free spectral range, FSR, which was the distance between the resonator peaks as could be calculated from the output spectrum data.

Since we used both benchtop laser and on-chip DFB laser to measure the PICs, the FOMs of the resonators as characterized by these two systems were also compared to evaluate the on-chip measurement system.

3. Results and discussion

3.1 Characterization of on-chip integrated laser

To characterize the on-chip laser, we conducted experiments on one of the structures in the circuit design that does not have an FP cavity. We compared these findings with that of an assembly consisting of a DFB laser photonic wire bonded to an optical fiber, Fig. 3 (a-III) and (c). The characteristic power-current-voltage (L-I-V) curves for both of the lasers are shown in Fig. 4. The DFB laser PWB to fiber output shows a maximum output of approximately 5 mW, Fig. 4 (a) as expected from such an assembly where the laser had a single output through the fiber. In contrast, the output of the on-chip DFB laser was significantly lower, Fig. 4 (b). This was expected from the estimates of Eq. (2), where the power of the on-chip DFB laser was split eight times and had output through GCs. The resulting LI curve, being closer to the noise floor of the mainframe detector, was therefore not as smooth as that of the DFB laser PWB to fiber.

 

Fig. 4. Characteristic L-I-V curves of the lasers measured in two different setups, through PWB to fiber (red) and through GCs in PIC (black). The stage temperature for all the measurements was set to 25°C (a) The L-I curve of the DFB laser PWB to fiber shows a threshold of approximately 7 mA with power in the mW range. (b) The L-I curve for PIC integrated DFB laser shows a threshold current of around 10 mA with output power significantly lower (less than a nW range) measured through GC. (c) The V-I curves for the two setups show the maximum operating voltage of 1.6 V at the operating current of I_th + 20 mA. This matches the range provided by the manufacturer.

Download Full Size | PDF

As shown by the L-I curve in Fig. 4(b) and the output spectrum in Fig. 5 (a), the output power of the laser in the PIC was measured to be approximately -65 dBm to -70 dBm at the operating conditions of 35 mA and 25°C. On a dB scale, assuming the input power from the DFB laser was 0 dBm, this gave us an approximate loss of 65 dB to 70 dB. This implied that there were additional losses of approximately 15 dB to 20 dB that were unaccounted for in our initial estimate of 51.5 dB, Eq. (2). One source of this loss could be from the splitters, which were not optimized nor independently characterized, so the estimated loss through them could have been underestimated in our initial calculation. Another uncertainty could be the insertion loss of the PWB itself. In several of our previous in-house assemblies of the DFB laser PWB to fiber, we noticed a variation in output power from assembly to assembly, which sometimes would be as low as -10 dBm due to high insertion loss. This was attributed to improper alignment, which was more challenging with an FA or single fiber. The on-chip laser assembly to the surface taper is comparatively simpler but a relatively recent process flow in our facility. Additionally, no protective cladding layer on top of the PWB bond was used which made it vulnerable to external factors such as dust accumulation and thus more prone to loss. This was confirmed by a microscope image taken after measurement completion and compared to an image taken before measurement, see Supplement 1 (Fig. S1). Unlike the optical microscope image taken right after assembly, Fig. 1 (c), we observed some anomaly on the PWB surface around the middle. This was also captured in the SEM micrograph, Fig. 1 (b-II). The side view in the SEM also showed how the bond sharply fell near the edge of the trench instead of over the surface of the inverted tapers as detailed in Supplement 1 (Fig. S2). Both of these phenomena could potentially increase loss through surface scattering and bending [13]. Because the PIC design did not integrate a tap to measure the laser power before the splitter tree and the PIC, we are unable to isolate the laser integration losses from the photonic circuit losses. The measured excess loss thus likely originates from both of these factors. Optimized Y-branch designs, use of PWB cladding, and improved imaging for the detection of the facets for the PWB could minimize the losses through the laser integrated circuit in the future.

 

Fig. 5. Current-wavelength calibration fit of the on-chip DFB laser for swept-frequency measurements. (a) Representative optical output spectra of the DFB laser at various injection currents at a constant stage temperature of 25°C. Each output shows the optical spectrum at the GC output of the photonic loopback structure measured through the optical spectrum analyzer (OSA). (b) Calibration plot of central emission wavelength as a function of injection current. A wavelength tuning range of approximately 0.42 nm is accessible by sweeping the current over the range of 25 mA to 50 mA. Data were collected over 163 replicate current sweeps over two days, with OSA measurements at each current value to demonstrate repeatability. We fit the experimental data using a third-order polynomial fit function to characterize the nonlinear wavelength-current relationship. Fit equation= ax³+ bx²+cx+d, where a=-1.676e-6 (-1.985e-06, -1.367e-06), b=0.00031 (0.00028, 0.00034), c=0.001 (-7.673e-5, 0.002), d= 1308.35 nm (1308.34 nm, 1308.36 nm), Norm of residuals= 0.24 nm, R²=0.999. The 95% confidence interval represents the range of uncertainty of the fit function.

Download Full Size | PDF

Despite the challenge of having a very low signal-to-noise ratio depicted by the L-I curve, the V-I curve of the lasers on both chips was comparable and within the range of specification provided by the manufacturer, Fig. 4 (c). The L-I curves also showed the threshold current, I_th, to be 7 mA for the DFB laser PWB to fiber while it was about 10 mA for the on-chip laser, Fig. 4 (a-b). This would provide an operating current of 27 mA and 30 mA, (i.e., I_th + 20 mA, by the manufacturer recommendations) for the two lasers respectively. Nevertheless, these values were within the maximum expected I_th of 15 mA at the operating temperature of 25°C, with the voltage of 1.6 V at the operating current as specified by the manufacturer. We also measured the optical spectra for both of the lasers using OSA for currents between 20 mA to 45 mA, see Supplement 1 (Fig. S3). Measurement of the spectral width at the point 3 dB lower than the maximum peak for each current showed a consistent 20 pm for all the peaks (the measurement is limited by the resolution bandwidth of the OSA).

3.2 Formation of current-wavelength calibration model

To establish a calibration model for the SFL measurements, we measured the current-wavelength relationship of the on-chip laser, again using a structure with no cavity. To ensure the laser output was in the linear region of the diode, while not crossing the maximum current limit (I_th + 40 mA) as recommended by the manufacturer, we performed the injection current sweep in the range between 20 mA to 50 mA with 5 mA increments and decrements. The output was measured through an OSA, and the spectrum is shown in Fig. 5 (a). A total of 163 replicates of the above measurement were carried out over 2 days with 61 sweeps on day 1 and 102 sweeps on day 2. The peak wavelengths for each input current for all replicates were plotted. An estimated 0.55 nm of wavelength change was observed for the current range of 20 mA to 50 mA. As expected from literature [32,34], the frequency or the wavelength of the semiconductor laser showed non-linearity due to several factors including the active medium, refractive index change, and reflections [44]. From the measurement data, we developed a fit model of the wavelength versus current relationship using a third-order polynomial curve-fitting. Figure 5 (b) depicts the calibrated model with the raw data and 95% confidence interval. This curve fit was used as a calibration function to convert current to wavelength for subsequent SFL measurements. It should be mentioned that at the curve fitting stage, we removed the peaks detected at 20 mA as they proved to be unstable during continuous sweeps for the replicates rendering inappropriate fits. The fit data thus portray the sweep between 25 mA to 50 mA reducing the wavelength tuning range to approximately 0.42 nm.

3.3 Comparison of PIC outputs between on-chip and benchtop laser measurements

For on-chip laser measurements, the injection current of the laser was swept from 20 mA to 50 mA using the LDC with 2000 data points providing a 0.015 mA step size. The output was again measured through the mainframe detector. To characterize the WBG-based FP cavities in the PIC, we also used a benchtop laser to sweep through the O-band spectrum. The output was measured through the mainframe detector. The main configurations of the setups for these measurements of on-chip and benchtop lasers are shown in Fig. 3 (a-I) and (a-II), respectively.

While plotting the findings from the on-chip laser measurements, the current axis was first translated to wavelength using the non-linear fit equation of the calibration model. The prediction intervals were calculated using MATLAB function [45] for simultaneous observation bounds as a measure of the expected uncertainty in the use of the calibration model. This method provided the 95% confidence interval for the fitted output wavelengths during the on-chip SFL measurement. The corresponding output power in this range, as also observed through the optical spectrum in Fig. 5 (a), becomes more stable at a similar higher magnitude value after 30 mA compared to a lower current. Since it provided a higher signal-to-noise ratio, we calculated the quality factor of the peak corresponding to the operating current 35 mA (approximately 1308.7 nm) via Lorentzian fitting. The data from the benchtop laser was also plotted over the same calculated wavelength range that the calibration model provided and the quality factor of the peak in a similar location was calculated. Figure 6 (a-c), shows the overlaid plots of the on-chip and benchtop laser measurements with the prediction interval and the corresponding peaks with Lorentzian fits to calculate the quality factors.

 

Fig. 6. a) The output of a Bragg grating resonator with on-chip SFL and off-chip benchtop Keysight O-band tunable laser. The injection current parameters for the on-chip laser measurement were translated to wavelength using the fit model described in Fig. 5 (b). The output power was normalized to the maximum and minimum within the on-chip laser full sweep range for each case. b) Quality factor of approximately 51,500 calculated from on-chip laser measurement, with a peak corresponding to 35 mA, i.e., around 1308.7 nm, using the Lorentz fit function [43]. Fit model used is y(x) = P1/((x - P2)² + P3) + C where the fit parameters are P1=1.499e-4, P2=1308.67 P3=1.6103e-4 C=-0.0534 and goodness of fit, R²=0.95. c) Quality factor of about 52,500 calculated from off-chip laser measurements, with a peak corresponding to around 1308.7 nm by same fit Lorentz fit model [43] with fit parameters P1=1.5310e-4, P2=1308.69 P3=1.5520e-4 C=-0.0136 and goodness of fit, R²=0.99.

Download Full Size | PDF

We observed a difference in FSR between the spectra measured using the two laser interrogation methods. While peaks measured through the benchtop had an FSR of 120 pm, the on-chip ones had an FSR of 160 pm (c.f. simulated FSR of 110 pm). We hypothesized that the discrepancy in FSRs between the two measurements resulted from thermal crosstalk during on-chip laser sweeps, leading to a red shift in the resonator peaks. To verify this, we conducted two experiments with the additional goal of quantifying the temperature change due to the on-chip laser current sweeps. For both experiments, we measured the resonant peaks of the circuits using the benchtop laser sweep. For the first experiment, we increased the on-chip DFB laser current from 0 mA to 50 mA with 1 mA steps. For the second experiment, we increased the stage temperature from 0°C to 0.5°C with 0.1°C step. We found the on-chip laser current increment from 0 mA to 50 mA caused an approximate 40 pm red-shift of the resonance peak of the PIC, matching the observed FSR discrepancy between the two measurement setups. This peak shift corresponded to a temperature rise of about 0.45 $^{\circ }$C of the stage as depicted in Supplement 1 (Fig. S4). These data support the hypothesis that during the on-chip measurements, the DFB laser heated up the resonators causing the wavelength shift of the resonance peaks. Future work will explore modeling solutions to compensate for the red-shift due to thermal crosstalk. Improved thermal stability strategies, such as using a thinned-down SOI chip or incorporating thermal isolation regions through trenches, are also worth exploring.

Despite the differences in FSR, calculations showed a comparable quality factor of about 51,500 for the peak measured through the calibrated SFL system and about 52,500 for the peak measured through the benchtop laser demonstrating the possibility of using this approach of measurement.

3.4 Comparison with simulations of Bragg gratings

During our investigation of the Bragg grating cavities employing benchtop laser measurements across the full spectrum, a notable blue shift of the stop band was observed relative to the on-chip DFB laser’s operational wavelength. This shift aligned with the simulation model displayed in Fig. 7 (a) and (b). The peaks detected within the operational region of the on-chip DFB laser sweep were from segments of lower Fabry-Perot cavity reflectivity. Additionally, the FSRs between the simulation and measurements were also different, with simulations showing 110 pm, and benchtop measurements showing 120 pm, Fig. 7 (c). We hypothesized this shift to be attributed to the fabrication uncertainties which would affect the effective index and subsequently the parameters in the TMM model. To validate this assumption we took the SEM image, shown in Fig. 7 (d) where we found the corrugations of the Bragg gratings to have more rounded edges. Additionally, the SF₆ and CF₄ gas mixture in the ICP etcher also affected the SiO₂ BOX layer leading to an over etch of approximately 100 nm. This can be observed in the tilted side-view SEM image by the darker region below the brighter device layer. By integrating these considerations into waveguide modeling, modifications were introduced to the effective index expression and the coupling constant (kappa value) in the TMM model. These adaptations resulted in the stop band location and FSR value to align with the benchtop measurement outcomes, Fig. 7 (c).

 

Fig. 7. (a) The full spectrum of the FP cavity measured by benchtop laser showed the cavity location or the stop band below 1290 nm. (b) The initial and revised simulation models showing the overall spectrum. (c) The FSRs of the different simulations and the benchtop measurement. (d) 45$^{\circ }$ tilted SEM of the Bragg gratings showing the over etch into the BOX layer and more rounded corrugations.

Download Full Size | PDF

3.5 Future work

While the measurements have proven successful, it is essential to assess the system’s stability over time for practical applications. Subsequent efforts can involve replacing needle probes with electrical wire bonding for laser electrode connections during measurements to mitigate the impact of mechanical vibrations in the output signal. Introducing a protective cladding layer on the PWB can shield against external factors, such as dust accumulation during long-term stability analysis [10,16]. The demonstrated process in this paper with the reduced assembly size, can also be implemented at wafer-scale. As an example, for a 200 mm wafer containing a die of size 5x5 mm, there would be 1029 dies [46]; with each die containing one laser, the write time per wafer (including image detection and exposure) would be approximately seventeen hours with the recipe we used in this project. Further investigation on wafer scale fabrication of the PWB technology can be conducted henceforth. In this study, we used lasers with low power and the system did not have any heaters to introduce any other form of temperature rise except for the laser self-heating through injection current. For applications or any procedures requiring higher temperatures, more investigation on the tolerance of heat on PWB [13,16,21] can be carried out as it is expected to have a maximum threshold due to the material properties of the polymer; causing it to deform at high temperatures for example, and thus changing its optical characteristics.

4. Conclusion

We have successfully demonstrated a complete fabrication process flow for the hybrid integration of low-cost DFB lasers with silicon photonic circuits through photonic wire bonding. Arrays of WBG-based FP cavities were designed and fabricated in the SiEPICfab UBC in-house process flow, with deep trenches for laser placement followed by photonic wire bonding from laser to waveguide coupling regions. By employing a calibrated SFL method, we analyzed the photonic circuits after characterizing the photonic wire bonded lasers. We also identified that thermal crosstalk affects the resonators from the temperature rise of the on-chip laser during the current sweep. Despite this, successful measurement of the output spectrum was possible through this method, with detected peaks having comparable quality factors to that measured with the benchtop laser. The difference between measurements and the initial simulation of the FP resonator could also be resolved by taking into account the process biases during fabrication. In addition to exploring ways to mitigate the thermal crosstalk effect by additional trench regions in design and an enhanced calibration model that might capture the temperature effect, future work will involve an investigation into the system’s stability over time with more robust packaging with electrical wirebonding and PWB cladding.