Packaging challenges: Case studies in packaging engineering

Transcription

1 Packaging challenges: Case studies in packaging engineering Ronald C. Stearns and Scott Trask Advanced Process Technology Center Newport Corporation 1791 Deere Av., Irvine, CA, USA ABSTRACT: A series of three case studies are presented to illustrate the technical challenges associated in the design and manufacturing of active photonic component packages. After reviewing the general steps of process realization, we will follow through three packaging engineering studies. 1. DWDM Source Laser Package: A package containing 980-nm pump laser, Distributed Bragg Reflector, ITU grid output laser, and associated optical components was optimized. This included optimization of the alignment of the output fiber plus two internal focusing lenses and a beam sampler. The package was designed for compensation of post-weld shifts, typically on the order of 3 to 5 µm, in order to attain actual alignment tolerances of less than 1 µm. 2. Single-Mode Laser Coaxial Package: We examine tolerance requirements in single-mode laser coaxial packaging while attempting a manufacturing cost reduction through passive alignment and attachment. We consider packaging options that depend upon which components are passively attached and estimate the acceptable tolerance in each option Fast Active Alignment in Photonics Device Packaging: We examine active alignment options for fast alignment. In particular, we look at the package design for a 980-nm pump laser. We look at 3D fast alignment data and the results of the connectorized coaxial transmitter CASE STUDY #1: DWDM SOURCE LASER 1.1 Design Concept: The challenge to develop a unique hybrid package design for Northstar Photonics presented an opportunity to explore integration assembly problems. The design is for a DWDM source laser package. The final product was to deliver a fiber-coupled laser output with a wavelength selected from the ITU-T DWDM grid. In the particular case here, the selected output is ITU Channel 53 at nm. As shown in Figure 1, a 980-nm pump laser is coupled to an ErYb:glass waveguide with a distributed Bragg reflector (DBR). The DBR acts as the wavelength-selective element. The output of the waveguide is coupled to an output fiber. There is also a beam sampler at the output to direct a small portion of the light to monitor laser power, plus an optical isolator to prevent feedback to the source laser. The waveguide was coated to maximize the path length for absorption of 980-nm pump laser light and to minimize crosstalk of 980-nm light into the 1535-nm output beam. Side 1 (input) of the waveguide was coated for anti-reflection at 980 nm and high reflection at 1535 nm, while Side 2 (output) was coated for high reflection at 980 nm at anti-reflection at 1535 nm. Figure 1: DWDM Laser Source Package 1.2 Design Challenges:

2 The laser was die bonded onto its TEC. The DBR was then also bonded onto its own mount and long TEC that ran the length of the waveguide. Both of these subassemblies were then soldered into the package before the focusing lens was positioned. Because of this, the positioning of the focusing lens became highly critical. There was no margin for error because neither the laser nor the DBR could be adjusted to compensate for any errors in the alignment of the focusing lens. The lens weld clip (see Figure 2) is fixed to the base plate of the package. The lens holder was then positioned and aligned. When the alignment had been optimized, using Lens Holder coupling efficiency to monitor the position, the lens holder was then welded into place in the weld clip using fillet welds at three to five points. After the weld, there will be a shift in the alignment due to post-weld shift (PWS). Depending upon the quality of the design and the manufacturing processes, this shift can be anywhere in the range of 2 to 20 µm. The goal is to reduce this to 0.5 µm, but this is a major challenge, one that was addressed in this design study. The ideal situation is one where the gap between the weld clip and the lens holder is zero. There are two things t hat affect PWS: (1) The design and tolerance of the weld clip and lens holder; (2) The amount of energy used in the weld. The design of the lens holder and weld clip and their manufacturing tolerances contribute directly to the degree of PWS. The ideal situation is one in which the gap is zero. The larger the gap, the more space that must be filled by the weld, and the larger will be the shift after the weld. The trade-off results from the fact that alignment of the lens holder requires some space for adjustment to optimize the alignment. Against that is the need to have the adjustability as small as possible to minimize the PWS. The second trade-off is the weld energy vs. the PWS. As the amount of energy used in making the welds goes up, the amount of PWS increases dramatically. However, greater energy also produces a stronger bond. So, the trade-off is weld strength vs. PWS. The requirement is that one must use just enough energy to create an adequately strong bond. More than this minimum amount will increase the PWS. Using only this minimum amount will also minimize the PWS. The choice of materials used in welding is very important. First, it is necessary that no phosphorus in used anywhere in the welding process. Even if nickel-plated, the use of phosphorus leads directly to cracking of the weld. Good material choices include Kovar, Invar, nickel, and 304L (low carbon, corrosion-resistant) steel. Bad choices include 303 and 304 steels and copper copper-tungsten alloys. Although copper is used in some applications, it is very difficult to weld in miniaturized packages such we have here. At the output of the package, the critical alignment is that of the output fiber. The focusing lens at the end of the waveguide and DBR can be passively aligned, since correct positioning of the output fiber can compensate gross misalignment. As was the case for the weld of the lens in position to couple from the laser to the waveguide, so also with the welding of output fiber the key is the minimization of PWS. Again, the ideal is for the gap to be zero. In the case of this output coupler, the weld clip is best made from nickel or dead-soft Kovar, as shown schematically in Figure 3. The initial PWS was in the range of 2 to 8 µm. The soft weld clip allows it to be deformable. Therefore, manual adjustment through plastic deformation allows for compensation of the PWS. 1.3 Process Description: Fillet Weld (3-5 plcs.) Lens Weld Clip Figure 2: Focusing Lens Subassembly Kovar or nickel weld clip Fiber output ferrule Figure 3. Schematic of output fiber ferrule in its weld clip. After the problems presented by these technical details had been solved, the final process proceeded as described here. First, the 980-nm-pump subassembly, which included the 980-nm laser diode chip, its TEC, the collimating lens, and the submount, was attached to the base plate. Electrical contact was made to the laser diode and photodiode. The optical bench was gripped with special tooling. The laser diode was turned on. The collimating lens was prepared and soldered into the lens holder. The laser diode was turned on and the collimating lens was loaded. The collimating lens was aligned using edge reflection as the alignment criterion. This was soldered into place once it was aligned. Photonics 2004 Full Text 2

3 The optical bench was positioned onto the base plate, and coarse x, y, z vision-based alignment was done. The alignment was made by aiming the collimated output from the laser through the open snout of the package (where the output fiber would be fed through in the final assembly) and verifying that one had obtained a well collimated beam with good alignment to through the output snout. The optical bench was then soldered to the base plate. Next, the TEC for the waveguide chip was soldered to its copper base plate. The waveguide chip was then soldered to the base plate. This subassembly was then visually aligned and soldered into the package. The first focusing lens (the lens coupling light from the 980-nm diode into the waveguide) was then aligned. This is a high precision alignment and attachment process. Laser welding is the process of choice to meet the tight tolerances. The feedback for this alignment is the output optical power at 1535 nm. This optical power is picked up by a golden fiber. [The golden fiber is a temporary output fiber, which may be a multimode fiber to optimize light gathering. It is used only to gather enough light to determine the relative maximum of light out of the waveguide.] The focusing lens was soldered into its holder. The load holder was then loaded into a custom gripper. The lens was aligned so that the collimated beam from the optical bench was focused into the waveguide. The weld clip was soldered to the base plate and the lens holder was laser welded into the clip. The important process guideline is that the PWS must be minimized in to order to optimize the coupling efficiency. Once the focusing lens is welded into place, neither the laser nor the waveguide nor the focusing lens may be adjusted. The optical isolator was aligned and soldered into place. Since this is a bulk-optics component that does not add significant displacement or angle to the beam, a coarse alignment in the range of 100 to 200 µm is sufficient. Finally, the output fiber was aligned and laser welded into position, as shown in Figure 4. The output fiber ferrule was gripped using pneumatic tweezers. Since all other components in the package were now fixed in place, the alignment of the fiber and PWS shift were key. Control of PWS by proper laser welding technique and mechanical adjustment resulted in optimized output coupling. In the final steps, the assembly was soldered into place and the laser diode and photodiode were wire bonded. The fiber feedthrough was soldered and the package was hermetically sealed using a seam sealer. Fiber X, Y, Z Beam Sampler Focusing Lens X, Y, Z Figure 4: Final assembly of output end of package. CASE STUDY #2: SINGLE-MODE LASER COAXIAL PACKAGE: The use of TO cans contributes to reducing the cost of packaging optoelectronic devices. Virtually all active diodes, including LEDs, VCSELs, FP lasers, DFB lasers, and photodiodes, can be mounted on TO cans. Even though TO packages are regarded as one of the most economical solutions, the package itself is still very complex and includes considerable high cost material and parts, primarily because of the tight optical tolerance requirements. In addition, a rapidly growing desire for 10 Gbit capability on the TO header imposes a huge pressure on packaging. A leap to a novel package scheme will be necessary to reduce the cost dramatically while improving the signal handling capability. Until then, a series of small steps of improvement will continue to drive the packaging cost down. 2.1 Review of a Typical Coaxial Packaging Process for a TO Device TO-46 and TO-56 cans are among the most popular types of TO cans for packaging optoelectronic devices. In the TO-46 package there is a step in the header so that the cap is passively placed on, or snapped onto it. Low cost, multimode laser diodes and photodetectors have been traditionally packaged in the TO-46 packages. On the other hand, the TO-56 header does not have such a snap-on feature. Single mode lasers have been packaged in the TO-56 packages because the lensed cap can be actively aligned. Figure 5 illustrates t he difference between the two types of headers. Photonics 2004 Full Text 3

4 The package design starts with characterization of the laser diode chip to define the optical train. Due to the high diffraction of the emitted beam, it is customary to use a cap with either a ball or an aspherical lens. By performing an optical simulation, one can determine the design requirements of the optical and mechanical components to meet the target specifications. Once the lens is selected, the optical characterization of the lens with the chip is performed experimentally to verify the simulation results and to finalize the package design. Optical train length, coupling efficiency, fiber alignment tolerances, mechanical tolerances of the cap and the header, cap height, lens positional tolerance, and height of the pedestal, if that is to be used, must all be experimentally verified. The laser diode chip and the monitor photodiode are eutectically soldered using a die bonder. For an edgeemitting laser, the header is typically placed horizontally and the chip is bonded at the tip of the pedestal. If a surface-emitting laser is to be used, it can be placed directly on the surface of the header, or on top of the pedestal. Wire bonding follows once all the components are in position for electrical connection. Next is the metal can attachment, once the test results show that all the components work properly. Either a projection welding or a laser welding method can be applied to seal the cap. The final step is to attach the pigtail fiber or the receptacle. Figure 6 shows the components to be used to laser weld the TO can in the coaxial manner. The situation is identical when the receptacle is to be welded. The ferruled fiber or the receptacle is inserted in a z-sleeve and then attached together by welding along the length of the sleeve. The sleeve and the laser housing are necessary in order to provide a freedom to position the fiber with respect to the laser source. The pigtailing is completed after the sleeve and the TO can housing are attached together. Figure 6. A schematic of a typical pigtail packaging component. From left, the components shown above are, a ferruled fiber, a z-sleeve, a laser housing, and a TO package, 2.2 Opportunities for Cost Reduction Even though the TO package is one of the cheapest packaging options, any possibility in further reduction of production cost must be explored. An obvious starting point is to look for a possibility of replacing an active alignment process with a passive process. An active alignment scheme not only consumes a longer process time but also requires a complex measurement system and a feedback algorithm. A successful passive alignment of the cap requires keeping the optical axis as collinear as possible to the mechanical axis. Deviation from that collinearity introduces excessive aberration in the optical system and thus reduces coupling, which is an expensive and painful trade-off in automated alignment. One option uses a TO-46 can instead of a TO-56 can, and removes a step of actively aligning the cap to the optical axis. This approach has been traditionally used for packaging multimode lasers or photodiodes, whose beam profiles are much wider than those of single mode lasers. For single-mode fibers, the die placement accuracy and the diametrical tolerance of the cap and the header must be controlled extremely tightly. Among other recent developments is the effort to combine the lens and the fiber receptacle structure by using plastic or metal molding technology. This drastically reduces the number of components used in the package as well as the material cost. The technology is at this time limited to multimode laser packaging, but it may soon also be applied to single mode lasers. The manufacturing tolerances are much tighter than the traditional approach because the lens and the fiber position are tied together. 2.3 Measurements Figure 5. The TO-56 (left) and the TO-46 (right) header viewed from top. The TO-56 package typically has a pedestal so that an edge-emitting laser can be packaged and its top is flat. The TO-46 has a snap-on feature so that the cap can be placed passively. Photonics 2004 Full Text 4

5 The optimized coupling is achieved by aligning the laser source, the lens on the cap, and the pigtail fiber along the optical axis. Figure 7 is an example of a lateral beam profile of a typical Fabry -Perot edge emitting laser operated at 1310 nm, packaged in a TO-56 package with a 1.27 mm diameter sapphire ball lens. The clear aperture of the lens is specified at 0.70 mm. An 8-degree angle cleaved single mode fiber was used to couple light. The globally optimal coupling position was found when the distance between the chip and the lensed cap was around 0.45 mm, and between the second surface of the lens and the fiber tip was 2.4 mm. The FW90%M (full width at 90 % maximum) of the intensity profile is determined to be approximately 3.2 µm. This indicates that the fiber pigtail or receptacle attachment must be done within ±1.6 um of post attachment shift if the coupled power is to be maintained within 90% of the peak value. When the lens placement has lateral error from the ideal position along the optical axis, the fiber can be moved to the point where the skewed beam is focused and still reach relatively high coupling efficiency. Figure 8 is the plot of an XY profile with the chip offset with respect to the lens position while the pigtail fiber is realigned. As the plot indicates, the center-to-center tolerance of the lens and the chip is 30 µm to maintain 90% of the optimal coupling value. The lateral movement of the fiber was magnified by approximately 2.5 times the movement of the lens. That is, when the chip was moved by 30 µm, the realigned fiber position was about 75 µm off of the original position. The coupling efficiency is a relatively slow function of the Z distance and thus any concern regarding the Z variation is not serious unless the tolerance is excessively large. 2.4 Manufacturing Tolerance Requirements for Passive Alignment and Attachment The purpose of the tolerance analysis is to determine the acceptable manufacturing tolerances of the parts to achieve a certain optical throughput criteria. We will assume that the targeted maximum power loss due to the packaging process is no more than 10 % of the optimal coupling value. Consider the passive attachment of the lensed cap. As Figure 8 suggests, the chip and the lens positional tolerances must be within 30 µm to maintain at least 90 % of the optimal coupling. For the TO-46 design, the typical diametrical tolerances of the inner diameter of the cap and Normalized efficiency Lateral offset (um) Measurement Fitted data Figure 8. Peak coupling efficiency as a function of the lateral chip placement offset error the outer diameter of the header are 25 to 50 µm. Thus, just attaching the cap on the header already uses up the maximum allowable tolerance value. Therefore, redesign of the cap and the header will be required. Packaging on the TO-56 header may prove to be simpler than on the TO-46 can. Without the help of machine vision, the placement accuracy of the cap on the header is determined by the repeatability of the seam sealer parts loading, with the typical value being around 50 µm. Employing vision software may improve the placement accuracy significantly through pattern recognition of a fiducial with a high positional accuracy. The vision alignment of a surface-emitting laser may prove to be simpler than an edge-emitting laser, because fiducials can be easily fabricated on the wafer. The vision system may even detect the laser active area or the fiducials directly through the lens, achieving a very accurate alignment. The challenge with an edge-emitting laser is to identify the laser diode chip and the active area when the chip is placed vertically on the pedestal. Intensity (uw) Offset (um) Figure 7. A plot of line scans in the x and y direction of the SMF-28 fiber, while the chip and the lens are stationary, measured using a single mode DFB laser emitting at 1310 nm and a 1.27 mm diameter ball lens. Px Py Photonics 2004 Full Text 5

6 Die bonders capable of attachment accuracy on the order of 1 µm have become more readily available from several vendors, including Newport Corporation. The MRSI 5005, shown in Figure 9, is capable of attaching a chip within 5 um of attachment tolerance through the vision system. The allowable tolerances for each component to maintain 90 % coupling can be obtained by considering the sensitivity of the both. The result is summarized in Table 1. It is clear that the tolerances must become much tighter in this packaging approach and it is the reason why the technology is currently limited to multimode laser packages. Chip placement tolerance (um) Lens and receptacle center-to-center tolerance (um) Maintained coupling level (%) 0.0 ± ± 0.4 ± ± 0.8 ± ± 1.2 ± ± Figure 9. MRSI 5005 Die Bonder from Newport Corporation. Table 1. Allowable tolerances of the chip placement and of the lens center to the receptacle center when the lens and the fiber (receptacle) are tied together. CASE STUDY #3: FAST ACTIVE ALIGNMENT IN PHOTONICS DEVICE PACKAGING 3.1 Frame of reference for active alignment When talking about optical alignment, it is important to establish a frame of reference and terminology for what alignment really means. Figure 10 shows a typical plot of power intensity vs. posit ion across a laser beam, which in most telecom applications is a single -mode Gaussian distribution. Although aligned is defined as achieving the absolute peak intensity, for most optical alignment applications some noise and measurement errors can contribute to about 1% fluctuations in intensity, so aligned can be defined as anywhere within the FW(99%)M area, or within 1% of the peak intensity. For practical applications, where the optics are to be 100% FW95%M secured after alignment, the standard for FWHM aligned and attached is usually somewhere 50% within the FW(90-95%)M range, depending on the application. There are several different types of active 1/e2 alignment techniques: Intensity (% of Peak) Step-and-Read: This approach takes an intensity measurement reading after each Position (microns) step of the motion system that is positioning the optics (which is typically the fiber) 3. This is more time intensive, but also more robust. Figure 10. Intensity vs. Position profile for a laser beam Dithering: This method imposes small, continuous movements on the fiber (or other optical element) to generate a signal that gives information about the direction of the peak. For example, if the movement causes only decreases in power, you are at the peak. If you are not a the peak, then the algorithm follows increasing intensity Scanning: Scanning algorithms vary with the type of scanning pattern (raster, spiral, cross) and data taken during the scan. In some applications, the scan is used only to find a minimum threshold intensity level (commonly 0% Photonics 2004 Full Text 6

7 called first light ) so that other algorithms can take over for the optimization. In other applications, comprehensive data can be taken on time or position triggers to provide feedback to the optimization algorithms. Profile scans are useful both as alignment and metrology tools, since they typically record detailed, high-resolution data regarding intensity vs. position, and after scanning of the profile, the motion system can return to any commanded point Alignment Sequencing: A combination, or sequence of different algorithms is nearly always utilized to achieve peak alignment with a minimum cycle time. Variation of process parameters such as starting position, stage velocity, scan length, step size and/or data recording intervals can be utilized to balance process speed and performance. Before alignment begins, steps are taken to initially position the components in close proximity as to enable rapid detection of the optical signal. This is most commonly done by a combination of precision fixturing of the components and visual feedback under the guidance of a camera. The total alignment time itself is also typically broken down into two segments: First light (also known as blind search ) and fine alignment. First light is defined as achieving an initial coupling. Fine alignment constitutes the rest of the optimization process in all appropriate axes (x, y and z in this work). 3.2 Experimental Approach This work utilized different techniques to attack cycle time in the initial positioning and alignment regimes, and was tested on two different widely used optical trains (and correlating equipment set-ups) to analyze overall alignment performance. Data was taken on 30 samples to understand process variations Cycle Time Reduction Strategy: In a system without computer-vision aids, the initial positioning time can constitute as much as one minute, which can represent 25% or more of the total cycle time. For this study, computer-driven machine vision programming was implemented to reduce cycle time and improve process reliability. For fine alignment, we implemented specialized fast-scan algorithms (FastAlign process) that employ both embedded firmware and process control software. These scanning algorithms take real-time data on-thefly which is correlated to the motion system position, as read by a glass scale encoder on the moving carriage Optical Trains. The first optical train tested was direct coupling with a lensed fiber (Figure 11a), commonly utilized on 980 nm pump lasers. This is typically packaged into a 14-pin butterfly with the optical axis oriented along the horizontal plane. The second optical train was a (a) (b) Figure 11. (a) Direct-coupling of laser diode to lensed fiber and (b) lensed coupling of laser diode (in TO can) to angle-cleaved fiber. lensed coupling of a laser diode with the optical axis oriented vertically (Figure 11b). This is commonly packaged in a TO-style coaxial package, and used in many transmitter applications Equipment: The experiment was carried out on two different platforms. One is on Newport s Laser Welder LW4200, which is a two -beam system for horizontal device packaging (e.g. 14-pin butterfly type). The other is on Newport s LW4000 which is a three-beam laser welder for coaxial type of device packaging (Figure 12). Both systems employ 50 nm bi-directional repeatability motion control engines for active alignment and Newport Integra process control software for photonics packaging applications. 3.3 Results and Discussion Direct Coupling Figure 13 shows the results comparing several different initial positioning and alignment scenarios for the direct-coupling optical train (Figure 11a). There are three scenarios in the experiment. In Figure Beam LW4200 (left) and 3-Beam LW4000 (right) platforms used for alignment testing. Photonics 2004 Full Text 7

8 Case 1, the operator used a camera and manual control of the motorized positioning system to pre -position the fiber close to the laser diode, then used the existing built in alignment routines (raster scan, step-and-read, profile scan, etc.) to do the alignment, calling each part of the alignment sequence individually. This typically took about four minutes, or 240 seconds to complete and verify optimum alignment, and had a wide cycle time variation of as much as ±90 seconds, or nearly 38% of the total. This was primarily due to the inconsistency of starting position (with no computer assistance), and the resulting difficulty of getting first light and the variation of parameters and routines needed to achieve peak alignment once first light was attained. This represents a scenario where an automated alignment system without sequencing or computer vision capability is being used. In Case 2, an operator-assisted machine vision system was used for initial positioning. For fine alignment, a traditional raster scan first-light detection and hillclimb step-and-read optimization sequence was utilized, which included a built-in sequencer to advance the fiber in z (along the optical axis) and to adjust algorithm parameters as necessary to arrive at the focus. The average alignment time for Case 2 was slightly over 100 seconds (a >50% reduction), and cycle time variation was reduced to ±33 seconds, which while still relatively high from a percentage standpoint, was a significant improvement over the ±90 seconds of Case 1. Case 2 is more representative of many commercially available automated alignment systems where computer sequencing and some machine vision capabilities are available. In the last scenario, Case 3, fully automated machine vision was used for initial positioning. For this case, the computer automatically detected key fiducial marks on the fiber (edges and face) and laser facet (edges and front facet) and automatically positions them to a pre -determined distance without operator intervention. Positional repeatability for this technique was better than 10 µm. Laser diodes with prefabricated fiducial marks may allow even better positional tolerance. The alignment portion of the sequence includes the new FastAlign fast-scanning algorithms and spiral search routines along with a built-in beam propagation analysis to drastically accelerate the alignment process: Fast-scan algorithms: As mentioned previously, the fast-scan technique takes real-time data on-the-fly which is correlated to the actual motion system position, as read by a glass scale encoder on the moving carriage. By making a rapid sweep along a single axis, a complete map of position and intensity is derived and can return key values such a peak location, threshold values, FWHM, and the like to the process controller Spiral search: In order to detect first light, an expanding box spiral pattern built of a series of fast-scan can be effectively used. A spiral pattern is rapidly swept with the fast-scan routines returning a trigger when the appropriate threshold intensity value has been reached Data sampling optimization: Tests were run to analyze the sensitivity of sampling increment (distance between data points) to spiral search cycle time (Figure 14). In the fast alignment approach, the sampling rate is constant and determined by the electronics clock rate, so the sampling increment is directly proportional to the speed of the alignment stage. From the graph, we can see that once the Scan time (ms) sampling increment goes above 0.8 µm/data point, there is little reduction of scan time, but there is loss of resolution. Based on this data, it is appropriate to set the search speed with about 1µm/data in spiral search on this system. Time (s) Alignment Positioning Case 1 Case 2 Case 3 Figure 13. Comparison of cycle times for different initial positioning and alignment methods um/data Scan length (um) Figure 14. Scan time versus scan length at different data sampling increments (microns/data point) Photonics 2004 Full Text 8

9 Beam Propagation Analysis: For edge emitting lasers the cross section of the radiation beam is typically an elliptical shape. In laser diodes, when light exits from the device aperture, the long axis is usually along horizontal axis (we define as the x axis). Since the device has two different diverging angles, at one point the cross section of the propagation beam becomes a circle. This is termed the equal waist position (EWP). For a typical 980-pump laser diode with a cylindrically lensed fiber (radius approximately 10µ, see Figure 2a), we found out that the EWP is about 20µ from the front facet of the device (see Figure 15). During alignment, the EWP serves as a very good indicator as to how close the fiber tip is with respect to the device. For fast alignment, we can utilize these properties to predict where the EWP (and resultant optimum) will be and accelerate the search along the optical axis. In one case, we performed fast 2D scans at two points as the fiber approaches the laser. Then, based on the FWHM along both X and Y we can calculate where the EWP is and jump to that location Lensed-coupled optical trains: The fast alignment approach based on fastscans has bigger advantage for the lensed coupled devices because the focus is relatively far (mm) from the laser compared with the direct coupling device like pump lasers (µm). A typical example is the coaxial type of device shown in Figure 11b, commonly seen in VCSEL, Fabry-Perot and DFB lasers, as well as TOSA and ROSA assemblies. The test method used for this optical train was different from the lensed-fiber scenario. Instead of loading and unloading individual devices and using varying levels of machine vision to achieve initial position, we used only one device in the experiment and varied the initial position location. This enables us to quantitatively study the effects due to the variation of device loading as well as the mechanical or optical tolerances associated with LD position, lens position, and fiber location. The search for initial light is a very important step in coaxial device alignment. The proper search area depends on the variation of the optical axis location within the search plane. This is a function of (1) the accuracy of placing the die and the lens with respect to one another, which is primarily a function of both mechanical and assembly tolerances, (2) the optical tolerances of the device and lens, and (3) the mechanical tolerances of the components and tooling. In the experiment, 120 data points were generated by randomly positioning the fiber tip inside the abovementioned cylindrical region and then initiating the alignment. The alignment time measured from initial position (no light) to peak alignment for the data population was: Alignment Time = / seconds FWHM (um) Based on this data we can say that with 90% confident level we can finish the alignment under these conditions within an average of 8.58 seconds. The alignment time is closely related to both the alignment technique and the device characteristics. Specifically, the optical train design tolerances, mechanical tolerances and tooling fixtures play very important roles. For example, if you can only control your focus location to within a 600µ cylinder instead of 100um cube, the alignment time will be increased to about 30 seconds from 8 seconds, nearly a factor of Conclusions By introducing fully automated machine vision and a combination of embedded fast-scan algorithms and beam propagation analysis, cycle times were significantly reduced when compared with the traditional automated alignment scenario for a 980-pump laser diode (Case 2). The automated machine-vision positioning reduced the initial positionin g time from 20.8 to 6.2 seconds, a 70% savings, and the active alignment time was reduced from 85.1 seconds to 32.5 seconds, a 62% reduction. The combined time was reduced by 63.5%. For coaxial-type single-mode devices, 3-axis alignment cycle times of under 10 seconds were achieved when the initial starting position was controlled to within approximately 100µ of the focus. This work enables continued improvement in the cost-effective utilization of active alignment techniques where high coupling efficiency and optical performance are required Distance From Focus (um) Figure 15. Full width half maximum vs. distance from laser beam focus with a wedged fiber. X Y Photonics 2004 Full Text 9

10 4. SUMMARY Historically, active alignment techniques have been the state-of-the-art for single mode fiber alignment. The long cycle time required for active alignment has led to an increased interest in the use of passive alignment, to save on labor and machine time. However, passive alignment approaches for single mode packaging often just shift the cost from labor to high-precision manufacturing of piece parts and often results in lower yields, resulting in questionable net savings. High-performance packages, such as the DWDM package featured here, require active alignment due to the desire to maximize light power output. In these cases, increasing the speed of the active alignment is the path to cost reduction. Case Study #1 shows an excellent example of the need for active alignment, while Case Study #2 defines the tolerances required for passive alignment, and Case Study #3 provides data on the efficiencies that can be achieved with high-speed active alignment. Even in low-labor-rate markets, it still makes sense to automate critical manufacturing steps, such as single -mode fiber alignment. The techniques are mature and can effect significant cost reduction by decreasing cycle time and increasing yield. References: 1. Jay Jeong and Scott Trask, Design Requirements for Passive Alignment and Attachment of Single Mode Laser Coaxial Packaging, IEEE PhoPack Symposium, San Francisco, Aug Jingyan Guo and Randy Heyler, Fast Active Alignment in Photonics Device Packaging, Electronics Components and Technology Conference, Las Vegas, Nevada, June Kamran Mobarhan, Martin Hagenbuechle, and Randy Heyler, Fiber to Waveguide Alignment Algorithm, Newport Corporation Fiber Optics and Photonics Application Note #6, April Photonics 2004 Full Text 10