Keynote Paper
`
`Reliability, Packaging, Testing, and Characterization of MEMS/MOEMS VII, edited by Allyson L. Hartzell,
`Rajeshuni Ramesham, Proc. of SPIE Vol. 6884, 68840B, (2008) · 0277-786X/08/$18 · doi: 10.1117/12.773640
`
`Proc. of SPIE Vol. 6884 68840B-1
`
`A Perspective on the Reliability of MEMS-Based Components for
`Telecommunications
`
`John C. McNulty∗
`DfR Solutions LLC, 5110 Roanoke Place, College Park, MD, USA
`
`
`
`ABSTRACT
`
`Despite the initial skepticism of OEM companies regarding reliability, MEMS-based devices are increasingly common
`in optical networking. This presentation will discuss the use and reliability of MEMS in a variety of network
`applications, from tunable lasers and filters to variable optical attenuators and dynamic channel equalizers. The failure
`mechanisms of these devices will be addressed in terms of reliability physics, packaging methodologies, and process
`controls. Typical OEM requirements will also be presented, including testing beyond of the scope of Telcordia
`qualification standards. The key conclusion is that, with sufficiently robust design and manufacturing controls, MEMS-
`based devices can meet or exceed the demanding reliability requirements for telecommunications components.
`
`Keywords: MEMS, tunable laser, tunable filter, VOA, DCE, reliability, optical, Telcordia
`
`
`1.0
`
`INTRODUCTION
`
`
`MEMS-based devices have been used for decades in applications like accelerometers and other sensors, demonstrating
`high reliability in demanding environments. More recently, Texas Instrument’s digital micro-mirror devices (DMD)
`have enabled the successful deployment of over 1.5 million light projection systems, with estimated mean lifetimes of
`over 50 years [1]. In the telecommunications industry, the use of MEMS-based devices has proliferated over the past ten
`years, with applications in optical switching [3,7], tunable lasers and filters [2-5], optical cross-connects, attenuators,
`add/drop multiplexers, and DWDM [6].
`
`There is a growing body of literature on the fundamental materials properties and size effects of MEMS structures [8-
`11], as well as their performance in extended cycling [12-13] and extreme shock and vibration environments [14-16].
`However, given the wide variety of MEMS device structures and their relevant failure modes, there is a general lack of
`MEMS-specific qualification test requirements. In the absence of standardized testing, more extensive knowledge of
`failure modes, and the associated reliability of MEMS-based devices, many OEMs have developed their own
`requirements for device manufacturers.
`
`
`2.0 MEMS STRUCTURES AND PROCESSING
`
`
`The author has worked with two main types of MEMS structures: comb-driven actuators and diffractive mirrors. Both
`structures are composed of single crystal silicon and produced using deep reactive ion etching (DRIE), with typical out-
`of-plane thicknesses less than 100 microns and minimum in-plane thicknesses of comb and support elements of 5
`microns. Scanning electron micrographs of each structure are shown in Figures 1 and 2.
`
`
`
`
`∗ jmcnulty@dfrsolutions.com; 415-806-7704
`
`Capella 2013
`Fujitsu v. Capella
`IPR2015-00727
`
`1
`
`

`
`I
`
`/T
`
`—-
`
`—
`-
`
`-
`
`—
`
`7:
`
`A
`
`- .—
`--Th -
`,c.
`-
`____
`
`-
`
`Proc. of SPIE Vol. 6884 68840B-2
`
`The actuator structures are used to move adhesively-attached mirrors that are also produced by DRIE etching. The first
`structure (Fig. 1(a)) is used in a Fabry-Perot tunable laser with a Littman-Metcalf external cavity, where the cavity is
`defined by the distance between the laser diode facet, grating, and MEMS mirror. The second structure (Fig. 1(b)) is
`used to modify the optical path length in a tunable filter. Both devices have closed loop control (the first with feedback
`from a wavelength locker subassembly, and the second with position detection off of the back surface of the mirror with
`a LED/split detector subassembly), and each enables tuning over 100+ channels of the C and/or L band, with 25 to 50
`GHz channel spacing. The MEMS structures of both devices are co-packaged with other optical subcomponents and
`then hermetically sealed in conventional butterfly packages.
`
`
`
`
`(a)
`
`
`
`(b)
`
`
`
`Figure 1: MEMS actuators for (a) tunable lasers and (b) tunable filters/receivers.
`
`
`The diffractive mirror structures are used in variable optical attenuators (VOAs, Fig. 2(a)) and dynamic channel
`equalizers (DCEs, Fig. 2(b)). In both cases, the optical beam path is perpendicular to the plane of the MEMS structures,
`and the mirrors tilt out of plane to affect attenuation of the reflected optical power. The VOA attenuates wavelengths
`more or less equally, whereas the DCE selectively attenuates up to 100 independent wavelengths spaced over the C
`and/or L band, similar to the tunable laser and filter. The MEMS components of each device are packaged individually
`into hermetic subassemblies (TO cans and 96-pin packages, respectively), and then integrated into larger structures.
`
`
`(a)
`
`
`
`
`(b)
`
`
`
`Figure 2: Diffractive MEMS structures for (a) VOAs and (b) DCEs (showing both the mirror pivots and combs).
`
`
`
`
`
`2
`
`

`
`Proc. of SPIE Vol. 6884 68840B-3
`
`3.0 MEMS FAILURE MECHANISMS
`
`
`There are a variety of failure mechanisms that can affect the reliability of MEMS structures: flaw sensitivity (arising
`from low fracture toughness) and the associated size dependence of strength, fatigue, stiction, wear, and stress corrosion
`cracking (due to water vapor). Depending on the MEMS design, stiction and wear may be eliminated – all of the
`structures discussed in Section 2 are non-contacting except in extreme shock and vibration events. Since elevated water
`vapor concentrations tend to exacerbate all failure mechanisms (stiction is usually inhibited by adsorbed water), the vast
`majority of MEMS devices in telecom applications are hermetically packaged. Fatigue can be effectively eliminated by
`maintaining a sufficient margin of safety between the design stresses and the stress associated with the fatigue crack
`growth threshold.
`
`The fracture toughness of DRIE structures is roughly equivalent to that of bulk single crystal silicon. Literature values
`are below 1.0 MPa√m independent of the cleavage plane [11], compared to values as high as ~4 MPa√m for
`polycrystalline silicon. The higher values for polysilicon are due to grain boundary toughening and other crack
`deflection mechanisms; such mechanisms are completely absent in single crystal silicon. The generic relationship
`between fracture toughness, Kc, strength, σc, and flaw size, a, is given by:
`
`Kc = Y·σc·√πa
`
`
`where Y is a geometric factor that varies with loading conditions and sample size. Since fracture toughness is generally
`constant (in the absence of environmental factors), the strength is controlled by surface defects (either etching defects or
`intentional stress concentrators), as compared to both surface and bulk defects for polycrystalline silicon.
`
`Size dependence of strength is a direct consequence of flaw sensitivity, since the likelihood on encountering a critical
`flaw increases as the surface area (or volume, for materials with bulk defects) increases. The failure probability, Ps, as a
`function of stress, σ, and surface area, A, is generally described using a two-parameter Weibull distribution of the form:
`
`
`
`
`
`(1)
`
`(2)
`
`Ps(A) = exp(-(A/Ao)·(σ/σo))m
`
`
`where m and σo are the Weibull shape parameter and reference strength, respectively, and Ao is the sample surface area
`on which both are experimentally determined. Both the strength and Weibull modulus vary significantly based on the
`surface roughness and defect density, but generally values of 1.2 to 8 GPa and 2.7 to 12, respectively, can be obtained
`with secondary processes to improve sidewall quality [9,10].
`
`The impact of MEMS element size and geometry on the resulting strength of simple unnotched and notched cantilever
`structures is shown in Figure 3, using data adapted from Minoshima et al [9]. The increasing spread in strength values
`as cantilever dimensions decrease is likely an impact of both slight variations in the nominal dimensions due to etching
`variations as well as the diminishing difference between surface flaw size and sample dimensions (a ratio of ~0.002 for
`the largest cantilever and ~0.1 for the smallest). The notches used in that study simulate single surface defects, and
`provide a vivid illustration of the effect of surface quality on strength – see Figure 5 for images of typical sidewall
`quality.
`
`Water vapor- and particulate-induced shorting (or blocking) of comb and diffractive MEMS structures are not intrinsic
`mechanisms, but nonetheless are likely the most commonly observed types of failures. Water vapor-induced failures can
`be eliminated using conventional hermetic packaging techniques, and hence such failures are strong indicators of poor
`process control and/or package design. Reduction of particulates in packaged devices is more difficult for a number of
`reasons: first, the silicon MEMS device itself is a large source of particulates, either as slivers from roughly-etched
`
`3
`
`

`
`Proc. of SPIE Vol. 6884 68840B-4
`
`sidewalls that break loose during contact or as chips broken off from device edges and surfaces during handling; second,
`particulates from other ceramic, metallic, or glass subcomponents (arising from handling) are common even in well-
`controlled cleanroom environments. As discussed in Sections 4 and 5, more extensive screening tests are required to
`reduce the impact of particulates.
`
`
`012345678
`
`Fracture Strength (GPa)
`
`200
`
`250
`
`0
`
`0.1
`
`0.2
`
`
`
`150
`100
`Cantilever Width, w (um)
`(a)
`
`0
`
`50
`
`012345678
`
`Fracture Strength (GPa)
`
`0.3
`Notch Length, a (um)
`(b)
`
`0.4
`
`0.5
`
`0.6
`
`
`
`
`Figure 3: Measured fracture strength of etched single crystal Si as a function of (a) cantilever width and (b) notch depth
`(adapted from Minoshima et al. [9]).
`
`4.0 MEMS DESIGN, SCREENING, AND QUALIFICATION ISSUES
`
`
`
`
`Much like any active opto-electronic component, substantial design, process control, and qualification testing is required
`to produce a reliable MEMS-based component. One of the goals of this paper is to highlight key test techniques during
`the development process, and thereby provide guidance for future MEMS-based telecom devices. Descriptions of
`various analyses, measurements, and tests are detailed in Table 1 below.
`
`Mechanical design and finite element analysis (FEA) software packages are indispensable tools for MEMS development.
`The most recent packages explicitly address surface roughness effects to identify the locations of highest stress
`concentration and subsequent fracture, which, as discussed in Section 3, are critical to evaluating MEMS processing
`parameters and their impact on flaw size distributions & subsequent fracture locations. On-wafer test structures and
`SEM or interferometric measurements of surface quality can then be used to provide accurate process capability
`assessments of material quality both within a given wafer and across multiple wafer lots. Relatively simple FEA
`analyses of a typical on-wafer test structure are shown in Figure 4. Representative SEM micrographs of wafers
`produced with two different processes are presented in Figure 5, overlayed on a graph showing the cumulative failure
`distributions (fracture load, as applied to the center of the structure in Figure 4(a)) for multiple test structures on each
`wafer. In this example, the fracture load in grams is roughly equivalent to the fracture stress in GPa. The values are
`consistent with those reported in other studies [9, 10].
`
`
`4
`
`

`
`•HHI
`
`Proc. of SPIE Vol. 6884 68840B-5
`
`Table 1: Summary of analysis and testing performed during development of MEMS devices for telecom applications.
`
`Development Step
`
`Design
`
`Process Development &
`Control
`
`Evaluation
`
`Screening
`
`Qualification
`(per relevant Telcordia
`standard)
`
`Description and Objectives
`• Mechanical/finite element analysis: dimensional parametric studies to assess resonant
`frequencies, peak stresses, and critical feature/flaw sizes
`• Wafer processing: SEM/interferometry of etch quality on actual and test structures; adhesion and
`strength measurements on test structures
`• Actuation characterization: curve tracer measurements of displacement as a function of I/V
`• Swept-frequency vibration (operational and non-operational): confirm resonant frequencies;
`assess optical performance variation; iterative development of closed loop controls (if necessary)
`• MEMS on submount: stepped shock testing to failure (500-5000 g+ acceleration) to identify
`functional limits
`• MEMS on submount: “proof” testing at 750 g+ acceleration to identify low level defects
`• Device: “proof” testing at 500 g+ acceleration to identify low level defects
`• Random locking (fixed or variable frequency): assess stability of actuator movement, particularly
`particulate-induced blockage or shorting
`• Operational vibration (5 g, 10-100 Hz; 2 g, 100-500 Hz): GR-468
`• Non-operational vibration (20 g, 20-2000 Hz): GR-468,-1073, and -1221
`• Operational shock (10g, 0.3 msec half sine; customer requirements up to 50 g, 0.1 msec half sine):
`GR-468
`• Non-operational shock (500 g, 1 msec half sine with TEC; 1500 g, 1 msec half sine with no TEC):
`GR-468;-1073, and -1221
`• Endurance locking/actuation (10 k-10 M cycles): GR-1073
`
`(a)
`
`
`
`(b)
`
`
`
`Figure 4: Typical FEA analysis of an actuator test structure, indicating the location of maximum stress. Note that
`these calculations do not explicitly account for surface roughness and defects, but that such can easily be
`implemented in FEA as a design/verification tool.
`
`
`
`
`
`
`
`
`
`5
`
`

`
`—Open Loop F r.quency E To, 4 5
`profile
`
`4 S2
`
`.
`
`—Closed Loop Frequency Error -
`—AcceIeration Profile
`
`g
`5z
`
`4
`
`2b -
`
`02
`
`C
`a)
`
`a)
`
`5-
`
`- 25z
`
`iáu
`k
`Vibration Frequency (Hz)
`
`10k
`
`Proc. of SPIE Vol. 6884 68840B-6
`
`1
`
`2
`
`4
`3
`Fracture Load (grams)
`
`5
`
`6
`
`7
`
`
`
`1.0
`
`0.8
`
`0.6
`
`0.4
`
`0.2
`
`Cumulative Probability
`
`0.0
`
`0
`
`Figure 5: Fracture loads of on-wafer test structures as a function of sidewall quality. The test structure geometry is
`shown in Figure 4.
`
`
`
`
`An example of swept frequency vibration testing on the tunable laser device is shown in Figure 6. In this particular
`example, the acceleration profile follows that of typical qualification testing, where laser stability is measured through
`variations in the output wavelength (arising from MEMS vibration and cavity length changes); in other designs, output
`power may be a more sensitive metric [5]. In open loop operation (top graph), significant resonances are observed near
`200 Hz (within typical specified vibration ranges for Telcordia) and slightly below 3 kHz (which can be induced during
`operational shock). The bottom graph demonstrates how closed loop control can dampen these resonances and
`significantly stabilize tunable laser output.
`
`
`Figure 6: MEMS performance of tunable laser during vibration under open-loop and closed-loop control.
`
`
`
`
`
`
`
`6
`
`

`
`I
`
`IL. l
`I IL'?
`
`Jl&..
`•11c.
`I II / '
`I )W JLr
`
`I
`
`I
`
`I
`
`I
`
`II
`II
`
`V
`
`I
`
`I
`
`I
`
`I
`
`I
`
`I
`
`Proc. of SPIE Vol. 6884 68840B-7
`
`Given that wafer test structures and SEM measurements provide only a limited snapshot of overall wafer quality and that
`subsequent MEMS damage can occur due to handling, evaluation and screening tests need to be performed at the MEMS
`and package-level to ensure reliable performance. Stepped-acceleration shock tests on multiple devices from a given
`wafer help to identify damage thresholds for the actual device structure. Proof testing at levels above expected field
`conditions (500 to 1,500 g, depending on whether or not thermo-electric coolers (TECs) are used in the device) provides
`further assurance that there are no ‘quality escapes’ due to low-level defect concentrations on the wafer. The sampling
`levels of both types of tests can be reduced if no failures are observed and/or significant correlation is established with
`the test structure results and SEM observations. The (not insignificant) limitations of shock testing are relative
`uncertainty of actual acceleration values at the MEMS level as well as the difficulty of testing enough samples per test
`run to not sacrifice production throughput.
`
` To date, there are no explicit Telcordia requirements for MEMS-based devices. However, due to increasingly prevalent
`special testing required by OEMs, implicit requirements were added into the latest revision of GR-468-CORE (Issue 2)
`to address operational shock and vibration. The conditions were based on shock events observed on operational line
`cards during adjacent line card insertions (to account for worst-case conditions during installation) as well as vibrations
`associated with usage and earth movements (adjusted to account for amplification of accelerations due to rack
`compliance). An example of actual measurements taken during line card insertion is shown in Figure 7(a), with the
`associated customer-specified operational shock condition requirements shown in Figure 7(b). The Telcordia conditions
`for operational shock represent the lowest common denominator of special OEM requirements (10 g acceleration with a
`0.3 msec pulse), whereas the most aggressive OEM-specified conditions range up to 50 g acceleration with a 0.1 msec
`pulse. The OEM-specified conditions for operational vibration are also typically higher than the Telcordia requirement
`due to substantial variations from manufacturer to manufacturer in rack compliance.
`
`
`(a)
`
`(b)
`
`
`
`
`
`
`
`Figure 7: Measured acceleration spectra during (a) line card insertion and (b) customer-specified 35g operational
`shock test. The vertical axis is acceleration in g’s, and the horizontal axis is time (1 msec/division).
`
`
`Some OEMs specify ‘endurance switching’ testing for MEMS devices, similar to the requirements of GR-1073-CORE
`for optical switches. Section 5 of that document indicates required durations of 10,000 cycles, with conditional
`requirements up to 1,000,000 cycles. However, OEMs may specify variations on that test for multi-channel devices,
`involving sequential switching (up and down the channel range) and/or random switching with possible variable
`frequency. For MEMS-based devices involving physical contact of the MEMS surfaces, such testing should be regarded
`
`7
`
`

`
`Proc. of SPIE Vol. 6884 68840B-8
`
`as a requirement for device suppliers due to concerns regarding stiction and/or generation of wear debris (with
`subsequent erosion and fracture of the structure). For non-contacting MEMS devices, the requirements should be based
`on an analysis of the peak stresses experienced during operation and anticipated shock/vibration events as well as the
`relative humidity in the device package to address the risk of fatigue. For hermetically sealed devices, the fatigue
`threshold for single crystal silicon is generally greater than 90% of the pristine fracture strength, and studies have
`demonstrated the lack of stress corrosion cracking-assisted fatigue in such environments [13]. Hence, fatigue fracture
`can be explicitly addressed through design of the MEMS device, and suppliers may thereby use theoretical arguments in
`lieu of actual testing. However, some level of endurance testing is still recommended to satisfy OEM concerns.
`
`
`5.0
`
`QUALIFICATION AND LONG-TERM RELIABILITY RESULTS
`
`
`Many authors have reported generic reliability data for MEMS-based telecom devices, in the form of either pass/fail
`numbers in individual tests (invariably only reporting results when all units passed) or extrapolated reliability estimates
`using limited test data. One important aspect of this paper is the opportunity to report Telcordia & special test data as
`well as actual field data on deployed devices. While the following results pertain only to the tunable laser using the
`MEMS structure shown in Figure 1(a), some general comments will be made regarding the other devices mentioned in
`Section 3.
`
` A
`
` summary of Telcordia, field deployment, and special test results is presented in Tables 2 and 3. For the first
`generation design of the tunable laser device, failures occurred during a variety of the tests required by GR-468-CORE.
`However, none of these failures occurred due to MEMS, but rather commonly encountered device packaging issues.
`
`Field deployment of a larger population of lasers revealed additional failure modes, only two of which related to MEMS.
`The first was due to water vapor condensation on, and subsequent shorting of, the MEMs actuator. During failure
`analysis, the internal water vapor level was measured at >80,000 ppm, which is sufficient to cause condensation on the
`actively cooled MEMS structure at case temperatures above ambient; upon cooling, the operation of both failed MEMS
`devices recovered due to evaporation of the condensed water . The true root cause of failure was solder joint fracture
`around the package window, allowing ingress of water vapor above of the nominal level of ~3,000 ppm. The second
`observed failure mode was particulate-induced shorting of the actuator. The source of the particulate was not
`definitively identified, but could have been handling-induced chipping of the MEMS and other subcomponents, or from
`outside of the package prior to sealing.
`
`Substantial process improvements were made to the second generation design, with additional inspection steps to reduce
`initial particulate concentrations. Furthermore, shock and vibration screening tests were added to identify weak
`structures as well as stimulate particulate movement within the package, thereby inducing potential failures before final
`module testing. As a result of all of these improvements, the tunable laser passed all Telcordia GR-468 tests during
`initial qualification and subsequent requalification.
`
`Field deployment of a much larger population of second generation lasers revealed some old and new failure modes.
`Additional devices failed due to particulate-induced shorting (though at roughly half the failure rate of the first
`generation), and one device failed due to fracture of a suspension in the MEMS structure. The fractured actuator may
`have had initial defects that were not effectively identified (or perhaps even induced) by shock and vibration screening.
`Given the uncertainty of actual field shock conditions, previous evidence of rough handling of other lasers, and the
`demonstrated endurance performance of the MEMS structure (Table 3), handling-induced damage in the field was
`nonetheless believed to be the root cause. Most notably, the greatest number of field failures was attributed to
`subcomponents on the tunable laser driver board, including capacitors, FETs, and ICs.
`
`
`8
`
`

`
`Proc. of SPIE Vol. 6884 68840B-9
`
`Table 2: Summary of Telcordia and field deployment failures in two generations of a tunable laser product, with the
`associated root causes.
`
`Device Design
`
`Condition
`
`Units Tested
`
`Telcordia GR-468
`
`129
`
`Generation 1
`
`Field Deployment
`
`368
`
`Telcordia GR-468
`
`251
`
`Generation 2
`
`Field Deployment
`
`5,300+
`
`Failures (in test)
`4 (accelerate aging)
`1 (high temp storage)
`1 (damp heat storage)
`1 (thermal cycling)
`3 (mechanical shock)
`2 (non-op vibration)
`5
`5
`5
`3
`1
`1
`0
`14
`7
`1
`
`MEMS Failures
`0
`0
`0
`0
`0
`0
`0
`0
`0
`2
`0
`1
`0
`0
`7
`1
`
`Root Cause
`Diode drift
`Alignment shift (adhesive)
`Alignment shift (adhesive)
`Adhesive debonding
`Adhesive debonding
`Wirebond failure
`Pigtail weld fracture
`Alignment shift (adhesive)
`Adhesive debonding
`Package leak (shorting)
`Wirebond failure
`Particulate shorting
`
`PCBA component failure
`Particulate shorting
`Suspension fracture
`
`Table 3: Constant frequency fatigue qualification testing on MEMS actuators in a tunable laser product. N and Nf
`are the total sample size and number of failures, respectively. The reliability estimate (FIT) is in failures per
`billion cycles, rather than hours.
`
`
`
`
`
`
`
`Test
`
`N
`
`Nf
`
`Cumulative Cycles
`
`FIT @ 90%
`Confidence
`
`Endurance cycling at
`200 Hz and 25ºC
`(ambient humidity, ~50%)
`
`20
`
`0
`
`95,000,000,000
`
`< 1
`
`
`The subcomponent failure rates are summarized in Table 4, based on estimated field deployment hours. It is difficult to
`strictly categorize each as “infant mortality” (decreasing rate with time) or “random” (constant failure rate): the
`particulate-induced failures would traditionally be labeled the former, since particulates are invariably pre-existing
`defects, whereas the PCBA component and MEMS fracture failures could be induced by a combination of pre-existing
`defects (for example, voids in capacitors or larger etch defects in MEMS) and aggressive field conditions (“acts of god &
`war” that typically define random failure). Nonetheless, the modes are treated as equal for comparison. Some key
`conclusions are:
`• The failure rates for the PCBA components are likely overestimates due to relatively limited field hours, but
`generally consistent with test data from the component manufacturers and/or generic estimates from Telcordia
`SR-332.
`• Particulate-induced failure is the most prevalent MEMS-based failure mode, but the failure rate is still
`comparable to that of the much more heavily deployed PCBA components.
`• The failure rate due to comb fracture is likely an anomaly due to unknown field conditions, but again
`comparable with the PCBA components.
`• The overall failure rate of the MEMS-based tunable laser module (~600 FITs at 90% confidence) is similar to
`that of many more mature opto-electronic components, while possessing much greater complexity and utility.
`
`
`
`9
`
`

`
`Proc. of SPIE Vol. 6884 68840B-10
`
`Table 4: Summary of MEMS and PCBA component failure rates in a Telcordia-qualified tunable laser product.
`FIT is defined as failures per billion hours.
`
`
`
`Devices Shipped
`Cumulative Field Hours
`
`Failure Type
`
`5,300+
`49,000,000+
`
`Description
`
`Occurrence
`
`Packaging-Induced
`
`MEMS: shorting or blockage due to particulates
`
`Subcomponent
`
`PCBA components (capacitors, diodes, FETs,
`potentiometers, oscillators, and ICs)
`
`MEMS: comb fracture
`
`7
`
`14
`
`1
`
`FIT @ 90%
`confidence, 25ºC
`240
`79-215
`(per component);
`410 (total)
`79
`
`
`The field reliability of MEMS devices in the other products (tunable filters, VOAs, and DCEs) is generally comparable
`to that of the tunable laser. Design and packaging similarity of the tunable filter (hundreds shipped) with the tunable
`laser leads to the same failure modes and overall occurrence rates. The MEMS-based VOAs are much more heavily
`deployed (>100,000 units shipped), and the only observed failure mode for the diffractive MEMS structure is water
`vapor-induced shorting due to package leaks in the TO can and subsequent water vapor ingress. Improvements to the
`sealing process have substantially reduced the occurrence of such failures. Failures in the DCE (hundreds to thousands
`shipped) from MEMS fractures have been directly attributable to excessive handling damage during transport,
`installation, and/or use.
`
`
`6.0
`CONCLUSIONS
`The use of MEMS-based components is proliferating in the telecommunications industry. MEMS structures exhibit a
`variety of failure mechanisms, but many can be eliminated through design and existing packaging methods. The
`occurrence of the remaining mechanisms can be reduced or eliminated with sufficient evaluation and screening tests.
`Specific and anecdotal results for a number of MEMS-based opto-electronic components demonstrate that such devices
`can meet or exceed the demanding reliability requirements of telecom OEMs.
`
`Acknowledgments
`
`
`
`
`The author would like give special thanks to Hal Jerman and Joe Drake of Coherent, Inc., for valuable discussions, data,
`and analysis, as well as Kevin Yasamura of FormFactor and Jocelyn Nee of NeoPhotonics for discussion of MEMS
`designs and testing.
`
`
`References
`
`
`1. M.R. Douglas, “DMD Reliability: a MEMS Success Story,” Proceedings of the SPIE, 4980 (2003).
`2. H. Jerman and J.D. Grade, “A Mechanically-Balanced DRIE Rotary Actuator for a High-Power Tunable Laser,”
`Technical Digest 2002 Solid State Sensor and Actuator Workshop, Hilton Head, SC, 7-10 (2003).
`J.D. Grade, H. Jerman, and T.W. Kenny, “A large-deflection electrostatic actuator for optical switching
`applications,” Technical Digest 2000 Solid State Sensor and Actuator Workshop, Hilton Head, SC, 97-100 (2000).
`4. J.D. Berger and D. Anthon, “Tunable MEMS Devices for Optical Networks,” Optics and Photonics News, March,
`43-49 (2003).
`
`3.
`
`10
`
`

`
`Proc. of SPIE Vol. 6884 68840B-11
`
`5. B. Pezeshki, E. Vail, J. Kubicky, G. Yoffe, S. Zou, J. Heanue, P. Epp, S. Rishton, D. Ton, B. Faraji, M. Emanuel, X.
`Hong, M. Sherback, V. Agrawal, C. Chipman, and T. Razazan, “20-mW Widely Tunable Laser Module Using DFB
`Array and MEMS Selection,” IEEE Photonics Technology Letters, 14 (10), 1457-1459 (2002).
`6. C.-P. Chang, “MEMS for Telecommunications: Devices and Reliability,” IEEE Custom Integrated Circuits
`Conference, 199-206 (2003).
`7. P. De Dobbelaere, K. Falta, L. Fan, S. Gloeckner, and S. Patra, “Digital MEMS for Optical Switching,” IEEE
`Communications Magazine, 88-95 (2002).
`8. C.J. Wilson and P.A. Beck, “Fracture Testing of Bulk Silicon Micro-cantilever Beams Subjected to a Side Load,”
`Journal of Microelectromechanical Systems, 5 (3), 142-150 (1996).
`9. K. Minoshima, S. Inoue, T. Terada, and K. Komai, “Influence of Specimen Size and Sub-Micron Notch on the
`Fracture Behavior of Single Crystal Silicon Microelements and Nanoscopic AFM Damage Evaluation,” Materials
`Research Society Symposium Proceedings, 546, 15-20 (1999).
`10. K.-S. Chen, A. Ayon, and S.M. Spearing, “Controlling and Testing the Fracture Strength of Silicon on the
`Mesoscale,” Journal of the American Ceramic Society, 83 (6), 1476-84 (2000).
`11. P. Chen and M.H. Liepold, “Fracture Toughness of Silicon,” American Ceramic Society Bulletin, 59, 469-72 (1980).
`12. O.N. Pierron and C.L. Muhlstein, “The Critical Role of Environment in Fatigue Damage Accumulation in Deep-
`Reactive Ion-Etched Single-Crystal Silicon Structural Films,” Journal of Microelectromechanical Systems, 15 (1),
`111-119 (2006).
`13. W.W. Van Arsdell and S.B. Brown, “Subcritical Crack Growth in Silicon MEMS,” Journal of
`Microelectromechanical Systems, 8 (3), 319-327 (1999).
`14. V.T. Srikar and S.D. Senturia, “The Reliability of MEMS in Shock Environments,” Journal of
`Microelectromechanical Systems, 11 (3), 206-214 (2002).
`15. D.M. Tanner, J.M. Walraven, K.S. Helgesen, L.W. Irwin, F. Brown, N.F. Smith, and N. Masters, “MEMS
`Reliability in Shock Environments,” 38th Annual International Reliability Physics Symposium, San Jose, CA, 129-
`138 (2000).
`16. D.M. Tanner, J.M. Walraven, K.S. Helgesen, L.W. Irwin, D.L. Gregory, J.R. Stake, and N.F. Smith, “MEMS
`Reliability in a Vibration Environment,” 38th Annual International Reliability Physics Symposium, San Jose, CA,
`139-145 (2000).
`
`
`
`11