Optoelectronics Packaging Research 2001 Peter Borgesen 1 Optoelectronics

Optoelectronics Packaging Research 2001 Peter Borgesen 1

Optoelectronics Packaging Research Long term: Contribute to transition from low volume, partially manual and robot based assembly of ‘pieces’ to fully automated manufacturing like microelectronics. Requires combination of: design for manufacturing & quality materials development & characterization systems & equipment development process development Obviously won’t (can’t) do all this, but to contribute we must understand: Research and education. 2

Research & Education Research: Establish activities in ‘ 1 st-3 rd‘ level packaging Understand current practices Problem solving with/for manufacturers Research topics of generic, long term relevance to automated manufacturing Education: Relevant research experience for our students Work with university on manufacturing relevant curriculum 4 hour tutorial on basics of optoelectronics packaging 3

1 st-3 rd Level Packaging Optical Sub Assembly (OSA): Rotator, polarizer, birefringent crystals Laser attach Component assembly: TEC attach OSA attach Fiber: Handling & Reliability Pigtailing & Connectorization Component attachment: Adhesive Selective soldering? 4

Quality & Yields Initial impressions: Interesting combinations of high precision (active alignment) and manual or semi-manual assembly common. Reproducibility: Only know quality of part tested? Yields? 5

Coaxial Laser Module Lens holder Welding ring Isolator Sleeve 5. 6 mm Aspherical lens Laser SM Fiber 22 mm Not exactly designed for manufacturing Simultaneous active alignment along 6 axes: Lens holder in X, Y (for lens) and (isolator) Fiber sleeve X, Y, Z 6

Generic 10 Gbps Laser Module monitor diode on ceramic lens filter laser on ceramic modulator on ceramic OSA Al. N TEC isolator SELFOC lenses Nor is this one ! 7

Laser Diode Packaging Not all packages are that complicated. In it’s simplest form a laser diode package has a laser shooting into fiber The rest is a matter of optimizing performance and (too rarely) ‘manufacturability’ 8

Edge Emitting Laser 1 -3 gold wire bonds to same electrode electrical contact to substrate 9

Edge Emitting Laser Attach wire bond laser solder lensed fiber Edge emitting laser attached to prototyping substrate with In solder, and lensed fiber with Ruby ring 10

Prototype Laser Diode Package Very similar looking product still offered commercially fiber sleeve laser submount TEC ceramic ‘optical bench’ Ruby ring UV tack adhesive butterfly package 11

Optimizing Performance monitor diode on ceramic lens filter laser on ceramic modulator on ceramic Free space coupling to selfoc and fiber OSA Al. N TEC isolator SELFOC lenses Parts all help optimize, but we still have choices to make: Package details (sealing), order of assembly & alignment. 12 Free space/waveguides/lensed fiber/SSC (alignm. budget)?

Coupling of Edge Emitting Laser Modes are not well matched: Different sizes/divergencies Optimized butt coupling to cleaved SM fiber offers only 9% efficiency 13

SM Laser - Fiber Coupling Mode Field Diameter (MFD): 1/e 2 width, typically 15% larger than core Mode field mismatch loss (perfect alignment): Effects of misalignment also depend on mode field diameters and match! 14

Coupling of Edge Emitting Laser Greatly improved by insertion of lens(es), but at the expense of reduced alignment tolerances. 15

Optical Aligment Consider transverse misalignment ( x, y) only. Then excess (butt-)coupling loss In a sense lens may be viewed as increasing MFDlaser, x/y at the fiber surface, reducing sensitivity to x/y there reducing MFDfiber at the laser surface, increasing sensitivity to x/y there Laser-lens alignment in x-y now less tolerant. 16

Optical Aligment Angular alignment tolerance clearly depends on transverse and longitudinal misalignment. If both the latter are zero, loss contribution varies with So, expansion of either mode increases sensitivity to angular misalignment at location in question 17

Lensed Fibers Kyocera Lensed fiber offers >90% coupling efficiency, but is very expensive and reduces alignment tolerance 18

Lenses Single aspheric lens may offer 55% coupling efficiency Combine with GRIN lens: >90% Expensive, 0. 2 m alignment GRIN: Parabolical variation in n(r), flat surfaces 19

Ball Lenses Ball lenses are clearly less effective than aspheric and, in particular, GRIN lenses. However, they are cheaper and easy to use 20

Optical Aligment Alignment of fiber and/or lens to within 0. 1 -0. 2 m often required for 85 -95% SM coupling. About 50% often achievable with 1 m. Fiber variations (diameter, core concentricity, cladding elliptricity) seem to prohibit passive alignment to better than 1 m. 21

Spot Size Conversion Blumenthal, UCSB Spot-size converter (in/at waveguide, amplifier, coupler, . . . ) may improve coupling (mode matching). Laser spot conversion raises lateral & longitudinal alignment tolerances (reduces angular tolerances). 22

Optical Aligment Coupling Efficiency (d. B) Expanding 0. 3 m laser spot to 1 -2 m before emission may raise tolerances to more than 1 m. 0 1. 4 m -2 w. spot-size converter -4 0. 75 m -6 Regular laser -8 -3 -2 -1 0 1 Position ( m) 2 3 Fish et al. , UCSB 23

Optimize/improve Manufacturability ‘Classical’ butterfly package design is not equally suitable to all coupling schemes: Fiber manipulation through hole in wall unnecessarily ackward 24

Generic 10 Gbps Laser Module Sensitivity to warpage/creep depends on alignment/coupling scheme (often not considered in design). Real products may use adhesives or Au. Sn solder & welding. Overall assembly issues very sensitive to choice. 25

Optical Bench Structure in Cooling Silicon or ceramic bench: Always CTE mismatches Almost always warpage FEM: ‘Typical’ structures may warp 0. 2 -0. 4 m/o. C Sensitive to adhesive properties, but always significant (Plastic) creep properties important. 26

Laser Diode Package Contents Typical components include laser, monitor, modulator, lens, isolator, pigtail, filter, detector, amplifier, cooler, driver chip Let’s consider optical isolators: ‘Optical diodes’ 27

Reflections ‘External cavity’ may create extra modes in SM laser. 28 -30 d. B (0. 1%) reflection may destabilize laser

Polarcor Rotator intensity Polarcor Polarization Sensitive Optical Isolator polarization Faraday rotator: Non-reciprocal rotation of polarized light 29

Polarization Sensitive Optical Isolator Magnet 30

Polarization Insensitive Isolator Also developing polarization insensitive isolator and circulator. Separate components, just keep adhesive from optical path. Current (manual) practice showed obvious manufacturing issues: Quality/yields? Automatability? 31

Polarization Insensitive Optical Isolator Primary light transmission 32

Polarization Insensitive Optical Isolator GRIN lens Primary light focussed to minimum spot by GRIN lens. Exit beam focussed back into same fiber from 16 m spread 33

Polarization Insensitive Optical Isolator Deflection of backward light: Divergent beams not focussed back into upstream fiber. 34

Polarization Insensitive Optical Isolator At a minimum optical path must be epoxy free 1 mm adhesive View of interface between rotator and wedge Wedge/rotator/wedge sandwiches each glued together along edges. Minute gaps left by surface morphologies: Capillary action. Uncontrollable 35

Wideband Polarization Insensitive Isolator Optical isolator assembly with epoxy: Very small parts, awkward locations, lots of active alignment. 36

Wideband Polarization Insensitive Isolator coated fiber epoxy ferrule cylinder Pigtail prepared by inserting stripped into epoxy in ferrule. Ferrule epoxied into steel cylinder. 37

Wideband Polarization Insensitive Isolator AR coating epoxy GRIN lens epoxied into other end of steel cylinder 38

Wideband Polarization Insensitive Isolator cylinder fiber ferrule GRIN 8 o GRIN lens inserted into magnet and gap to cylinder filled with epoxy. magnet 39

Wideband Polarization Insensitive Isolator magnet GRIN lens Another GRIN lens actively aligned with exit side 40

Wideband Polarization Insensitive Isolator Steel cylinder with GRIN lens and fiber ferrule (at other end) soldered to outside cylinder 41

Polarization Insensitive Optical Isolator GRIN lens Remember offset? But this manufacturer did not bother 212 m offsetting fiber opening at end of cylinder. End segment tilted, fiber bent to 1” radius. Uncontrolled! How well is stripped section protected from bending? (1/10 stripped fibers bent to 1” would last 11 days at 50%R. H. ) 42

Adhesives offer some obvious attractions for automation. However, dispensing is a bit of an art form. Also, there are numerous issues with properties. 43

Adhesive Projects Importance of deposition process control: FEM Dipping, pin transfer and dispense of small volumes: Fundamentals and applied. Shifts in placement and cure: Effects of deposition control & materials properties (just started) Gap & constraint dependent cure kinetics & properties: Realistic configurations vs. DSC, DMA & data sheets Creep & misalignment: FEM & experiments (to come) 44

Adhesive Deposition Process Control Effect of temperature change on optical component with asymmetric adhesive fillets: Rotate 1 o per o. K ! 45

Polarization Insensitive Optical Isolator Magnet opening w. isolator structure adhesive Small adhesive volumes in awkward locations 46

Adhesive Fixturing of TO-can Fiber Adhesive TO-can Small adhesive volumes in awkward locations, uniformity critical 47

Small Adhesive Volumes 0. 25 mm How small is small volume? This ball lens needs 1 g of adhesive in small dot: wet-out important? 48

Flip Chip VCSEL by Dipping adhesive Au Alternative small volume application 49

Flip Chip VCSEL by Dipping adhesive Au Alternative small volume application 50

Adhesive Deposition Dipping, dispensing, or pin transfer preferred depending on volume & location control needed. Automation requires minimization of scatter. All dominated by dynamic wetting. 51

The 3 Stages of Dipping 1 a) 1 b) 3 a) 3 b) 2) 1) Insertion 2) Hold 3) Withdrawal 52

Dynamic Liquid-Solid Contact Angle Steady-state wetting withdrawal insertion Speed 53

Dynamic Liquid-Solid Contact Similar picture applies to pin transfer deposition step. Dispensing of dot may be viewed as transfer with hollow pin, in some respects. Steady-state contact angle scales with and slow model pin transfer experiments could be rationalized on basis of steady-state: More pick-up with faster withdrawal, etc. 54

Dynamic Liquid-Solid Contact Realistic applications may not reach steady-state in each step: Much faster withdrawal (and no hold) shortened time in adhesive, and thus time to reduce contact angle, giving less pick-up. The time to approach steady-state to a certain extent should scale with Short travel distance (dip depth) and high speed may define other ‘simple regime’ for flip chip dipping (dependence on hold time? ). Otherwise, dynamics can be calculated numerically and calibrated experimentally: Minimize sensitivity to variations (scatter). 55

Pin Transfer Adhesive dots on glass surface 56

Pin Transfer Process far from optimized yet, but 1 g dots deposited with diameter=1% volume=2% Strong sensitivity to substrate chemistry and morphology: Account for latter in choice of transfer height 57

Cure and Properties of Optical Adhesives Initial studies considered four UV/blue light curable adhesives Supplier indicates ‘seconds’ of light or 1 hour at 100 o. C (or 2 at 80 o. C), but suggests testing for performance. Careful: Shadowing may affect curing (need post cure) and formation of protective skin against oxygen inhibition of post cure. There also other reasons for cure kinetics and properties to depend on configuration. Sometimes DSC data only qualitatively relevant 58

DSC of Optical Adhesives Effect of duration of thermal cure at 100 o. C from subsequent DSC scans. Cure-% is defined as relative peak area. 59

Thin, Constrained Adhesive Layers Fiber Adhesive TO-can Cure kinetics and resulting properties unlikely to be the same as in ‘bulk’. 60

Thin, Constrained Adhesive Layers Silicon V-grooves for optical fiber array: 0 -90 m adhesive 61

Adhesives in Realistic Configurations . 125 mm adhesive 1. 5 mm Exposure to light from both sides (max. depth 0. 75 mm) Avoid shadowing. Take shear strength as measure of cure. 62

% cure DSC of A 146 T After Light Cure Light/distance 2 (sec/inch 2) Apparently no further curing above 120 s 63

Shear Testing of A 146 T After Light Cure Strength (lbf) 30 20 10 0 0 2 x 50 2 x 100 2 x 150 Light/distance 2 (sec/inch 2) 2 x 200 However, there is clear increase in strength after more light 64

DSC of A 146 T After Light + Thermal Cure +1 hr @115 C +1 hr @110 C % cure +1 hr @100 C no thermal Also, more light enhances efficiency of subsequent thermal cure Light/distance 2 (sec/inch 2) 65

A 146 T Effect of Light What’s ‘fully cured’? According to DSC we need lots of light: 1. 5 hours @ 1” distance + 1 hour @ 115 o. C for ‘complete’ 0. 5 -1. 0 hours @ 1” + 1 hour @ 110 o. C for 99% cure Interpretation: >60 s needed on each side to create reproducible ‘skin’ against oxygen inhibition (see next). Still need to see how much is ‘enough’ cure. 66

Failure Probability A 146 T: Effect of Thermal Cure After 2 x 60 s Light 0. 99 0. 5 30 min 0. 2 45 min 0. 1 0. 05 Broad statistical distributions of strength 0. 02 1 2 3 4 5 67 Strength (MPa) 67

A 146 T: Effect of Thermal Cure After 2 x 60 s Light Failure Probability 0. 99 0. 5 45 min 60 min 0. 2 0. 1 0. 05 Broad statistical distributions of strength 0. 02 1 2 3 4 5 6 7 8 9 10 Strength (MPa) 68

A 146 T: 1 Hour Thermal Cure After Light Failure Probability 3 min on each side enough 0. 99 for reproducible ‘skin’? 0. 9 2 x 60 s 0. 5 0. 2 2 x 180 s 0. 1 0. 05 0. 02 1 2 3 4 5 6 7 8 9 10 Strength (MPa) 69

Adhesives in Long, Narrow Gaps . 01 -. 4 mm adhesive 1 -11 mm Effects of shadowing and dimensions (thickness & depth) 70

Effect of Depth (Bond Length) A 4061 T 1 hour thermal cure No light. Ultimate shear strength should be proportional to bond area (shear stress should be independent of bond length). Initial increase could be oxygen inhibition? 71

Effect of Thermal Cure Ambient (A 4061 T) Slower thermal cure in air (oxygen inhibition? ) Much slower cure in closed DSC pan: Effect of configuration? 72

Effect of Light on Thermal Cure (A 4061 T) Light builds protective ‘skin’: Most important at shallow depths. Still doesn’t explain ultimate stress dropping with bond length! 73

Effect of Thickness (A 4061 T) Strength Light exposure + 1 hour thermal (in vacuum) Gap Size (mil) Of greater concern: Abrupt dependence on thickness (without change in failure mode). 74

Optical Adhesives Cure-% measured with DSC not directly useful. We wouldn’t know which % is ‘enough’, even if it applied to configuration of concern. Cure kinetics and properties vary in complex fashion with configuration. Some of this may be ascribed to ‘shadowing’ of light and the formation of a protective ‘skin’ on the surface. However, systematic variations with bond length and thickness were not. 75

Fluxless Au-Sn Soldering ‘Standard’ in hermetic applications Void less laser attach Feedthrough Attach sub-mm pieces in tight spaces 76

Au. Sn Maintain alignment: creeps less than epoxies and soft solders no swelling or densification Hermeticity: outgasses less than epoxies (and even Sn/Pb) 77

Laser Attach with Au. Sn Cooling & temperature stabilization: better thermal conductivity than filled epoxies and alternative hard solders Voids easily affect laser temperature control. 78

Au. Sn Soldering Process usually must be fluxless. Current approaches involve N 2(5 -10%H 2) atmosphere scrubbing static pressure? pressure variation & vacuum 79

Au. Sn Soldering Process Current approaches are not attractive for true volume manufacturing. Scrubbing breaks up surface films but may supposedly disturb molten solution and reduce homogeneity. Soldering in N 2 (without H 2) possible, but may enhance risk of voiding: Sn goes to surface of liquid, but H 2 scavenges O 2 Also works in vacuum H 2 less important if interface not exposed in melt? 80

Au. Sn Soldering With Preforms Don’t entrap voids: Don’t touch/damage ‘bottom’ of preform before placement. Don’t let contact pad touch top of solid preform: Heat before place enhances throughput too Sn on liquid surface exposed to ambient! 81

Au. Sn Soldering From Multilayer Au/Sn/Au thin film structure may allow pad contact (protection of interface) before melt. Also allows for patterning of small contact areas (see isolator). No oxide layer before reflow. Design structure for stability ‘on shelf’ and bonding before ‘freezing’. Small thickness should help reflow hierarchy? 82

Au-Sn Phase Diagram Massalski, 1990 83

Au. Sn Soldering From Multilayer Polarcor (borosilicate glass) Adhesion layer Au Sn Au As deposited Au Adhesion layer Rotator (thin film garnet crystal) 84

Au. Sn Soldering From Multilayer Polarcor (borosilicate glass) Adhesion layer Au Au. Sn After aging at RT? Au Au Adhesion layer Rotator (thin film garnet crystal) Sn is still protected from ambient ! 85

Au. Sn Soldering From Multilayer Polarcor (borosilicate glass) Adhesion layer Au After reflow Au 71 Sn 29 Au Adhesion layer Rotator (thin film garnet crystal) Still enough Sn to mix with Au on rotator pad 86

Au. Sn Soldering From Multilayer Polarcor (borosilicate glass) Adhesion layer Ni Au Au Sn Sn Au. Au As deposited Au Ni Adhesion layer Rotator (thin film garnet crystal) 87

Soldering With Predeposited Au. Sn Alloy Shares some advantages with multilayers, but less protection against oxidation before reflow does not require as careful design of thicknesses Eutectic or Au-poor? At least one supplier suggests reflow in air!? Currently under investigation. 88

Au. Sn Soldering Process Mechanical properties depend on ? Effects of reflow parameters on creep TBD Au at-%Sn Revised, Ciulik & Notis 89

Optical Fiber Handling & Reliability Fibers are bent, squeezed, pulled, cleaved, stripped, cleaned, spliced, connectorized, polished Invariably mounted under thermal mismatch stress. 90

Optical Fiber Projects Quantify importance of damage in handling and packaging: Predict ‘life in service’ -effects of load, humidity, damage Quantify damage Identify non-damaging procedures 91

Optical Fiber Handling & Reliability Fibers may fail (apparently) instantaneously under sufficiently high load. You’ll notice that. Fibers may fail more gradually (minutes, hours, days, years) by subcritical growth of a surface defect. That’s scarier. The former is easy to test for. How do we test for the latter? 92

Optical Fiber Damage & Failure An optical fiber breaks when KI >KIc=0. 75 MPa*m 1/2 where at a surface defect of length a. The only significant sub-critical crack growth is caused by a chemical reaction with water. This is only significant if the effective activation energy is reduced by a tensile stress. If so, even an extremely low humidity level (hermetic package) may be sufficient. 93

Crack Growth Rate vs. KI Crack Velocity (m/s) 10 -1 moisture diffusion limited 10 -3 10 -5 critical crack growth stress & humidity limited 10 -7 0. 4 0. 5 0. 6 0. 7 0. 8 Stress Intensity Factor (MPa m) By the time we reach diffusion limited regime it’s essentially over! 94

Life & Damage Assessment Industry test data are most often interpreted based on a power law for sub-critical crack growth: This may overestimate service life by several orders of magnitude (years vs. days). We and others find best agreement with 95

Life & Damage Assessment: Humidity Based on dynamic strength measurements at moderate and high loading rates literature agrees on p 2 in but we find p to increase with decreasing loading rate: p 2 at 25 -1000 lbf/min p 4 at 0. 5 lbf/min p 4 under static load 96

Strength (lbf) Life & Damage Assessment: Humidity AI (%R. H. )2 AI (%R. H. )4 Humidity (%R. H. ) Dynamic tensile strength of pristine fiber at 1000 lbf/min 97

Strength (lbf) Life & Damage Assessment: Humidity AI (%R. H. )2 AI (%R. H. )4 Humidity (%R. H. ) Dynamic tensile strength of pristine fiber at 25 lbf/min 98

Strength (lbf) Life & Damage Assessment: Humidity AI (%R. H. )4 AI (%R. H. )2 Humidity (%R. H. ) Dynamic tensile strength of pristine fiber at 0. 5 lbf/min 99

Normalized Life (s) Life & Damage Assessment: Humidity AI (%R. H. )4 AI (%R. H. )2 %R. H. Normalized life of pristine fiber in static bending 100

Optical Fiber Damage & Failure Life is extremely sensitive to initial defect size: tf a 10 We are commonly concerned with sub-micron defects, and a 25% increase in size is significant! 101

Probability Initial Defects in ‘Pristine’ Fiber Initial Defect Size (nm) 102

Optical Fiber Testing We can predict life under given loading if we know initial defect size and (in case of bending) location. Static tests are by far the most sensitive to defect size, but they are often not very practical. For most (all? ) practical purposes defect sizes can be conveniently assessed through dynamic strength. Dynamic bending measurements are very common, but should be carefully interpreted in terms of statistics of defects. 103

Optical Fiber Bending Account for non-linear stress-strain relationship: ‘Effective gage length’ is much less than in tensile test, especially if considered in terms of maximum tensile stress: Slope of single mode Weibull distribution is the same, but mean is higher ! 104

‘Pristine’ Fibers in Static Tension & Bending 106 Bending Life (s) 104 102 Tension 100 10 -2 Maximum Tensile Stress (GPa) 105

Optical Fiber Damage For all practical purposes it is impossible to damage fiber significantly without penetrating coating penetrate coating without damaging fiber significantly Fiber is damaged by cleaving, stripping, wiping, splicing, . . . 106

Probability Initial Defects in Stripped Fiber Initial Defect Size (nm) 107

Initial Defects stripped only Probability spliced, Program 6 spliced, Program 1 Initial Defect Size (nm) 108

Optical Fiber Damage distribution is sensitive to stripping parameters (stripper, strip length, wiping, . . . ), but the largest defects may not be. Damage distribution is often non-uniform, making 2 -point bending tests more difficult/risky to interpret. Cleaving and/or splicing caused additional (more consistent) damage, but largest defects were similar. Damage distribution (strength of splice) was sensitive to combination of splicing recipe and fiber(s): SMF, EDF 109

‘Specs’ Taking reasonable levels of statistics into account we suggest (very conservatively, aside from optical considerations) limiting: Handling (Immediate Failure) Radius of curvature >70 mil for pristine fibers >3. 6” for splices (>>1”) ‘Permanent’ (35 years), non-hermetic Tension <1. 2 lbf, radius > 0. 5” for pristine fibers Tension <26 g (!), radius > 10” for splices Until further notice we suggest same ‘specs’ for cleaved+stripped fiber ends as for splices. Crack growth data are needed for very low humidity levels in hermetic packages. 110

So When Do We Care? As we gather more statistics crack growth formalism allows us to more quantitatively assess ‘what is important’: The weakest out of one set of 10 stripped fibers tested would last 11 days at 50%R. H. bent to radius of 1” (remember optical isolator? ). Different splicing recipes gave mean life of 15 vs. 30 years under 0. 6 lbf of tension, but the former also had lower Weibull slope. Extrapolations to first fail of 10, 000 under 60 g of tension (3. 8” bend) gave 25 minutes vs. 35 years. While undoubtedly unrealistic this illustrates concern. 111

Summary Problem solving for, and product development with, manufacturers help define generic research topics. Au. Sn soldering: Metallurgy, design of multilayers Experience with preforms Adhesives: Automated deposition Cure kinetics Effects of realistic configurations Fibers: Damage/defects Quantitative assessment of consequences Cleaving, stripping, wiping, splicing, fiber type, . . . 112