`
`45
`
`Fabrication of a Mechanical Antireflection
`Switch for Fiber-to-the-Home Systetns
`
`James A. Walker, Keith W. Goossen, and Susanne C. Arney
`
`Abstract-We present the methods used to fabricate a microme(cid:173)
`chanical silicon optical modulator for use in a fiber-to-the-home
`applications. We emphasize the efforts made to realize a practical,
`robust, manufacturable, and easily packaged device. In addition,
`recent speed, temperature stability, and reliability results are
`presented. Rise and fall times of 132 and 125 ns, respectively,
`have been observed in response to a square wave drive signal.
`The device has been temperature cycled from -50°C to 90°C
`and shown greater than 10-dB optical contrast ratio over this
`temperature range. Finally, the device has been cycled at 500kHz
`for a period of nearly two months (two-trillion cycles) without a
`noticeable loss in performance. [134]
`
`I. INTRODUCTION
`
`T HERE has b~en mu~h rece~t emphasis _on providing
`
`the world with an mformation super highway. It is
`believed by many that this will consist of a backbone of
`fiber optics that will extend throughout the network, all the
`way to the subscriber location or home. Recent work in the
`fiber-optics community has focused on providing a fiber-to(cid:173)
`the-home (FTIH) system, which can provide the bandwidth
`required to enable the high-performance services promised by
`proponents of the information superhighway. In some system
`configurations, an optical modulator is used at the subscriber
`interface to imprint outgoing information on an incoming
`signal and reroute it onto a fiber back to the central office
`[1], [2]. Some advantages of using an optical modulator for
`this purpose are the maintenance of wavelength coherence
`of the signal, reliability of the optical device, temperature
`insensitivity, and, most notably, cost. We present here an
`optical modulator based on microelectromechanical systems
`(MEMS) principles that incorporates a moving dielectric film
`to provide optical modulation. We call this device a mechanical
`antireflection switch (MARS) device [3]. Attributes of the
`MARS device include high speed, high contrast compared
`to other surface normal modulators, temperature insensitivity,
`polarization independence (due to its surface normal nature),
`and low cost. In [3], we presented a device that illustrated
`the modulator principles, but was fabricated without emphasis
`on reliability or yield. Here we present fabrication sequences
`that result in practical, robust devices and demonstrate high
`reliability.
`
`Manuscript received December 22, 1994; revised December 21, 1995.
`Subject Editor, K. Petersen.
`J. A. Walker and K. W. Goossen are with AT&T Bell Laboratories,
`Holmdel, NJ 07733 USA.
`S.C. Arney is with AT&T Bell Laboratories, Murray Hill, NJ USA.
`Publisher Item Identifier S 1057-7157 (96)02082-3.
`
`Optical Network Unit
`
`Central Office
`
`R
`
`-mmom
`
`Fig. 1. Representation of RITE-net architecture [2].
`
`A. Fiber-to-the-Home Systems
`The FTTH systems under consideration here are packet-·
`switched data networks. In this type of system, information is
`divided into discrete blocks of information of 48 bytes, called
`packets. These packets are sent serially through the network
`and received at their destination, where they are reconstructed
`into the complete information stream. A schematic represen(cid:173)
`tation of one such network, called RITE-Net [2], is shown in
`Fig. I.
`In RITE-Net, each packet traveling from the central office
`consists of a front half, which contains the downstream in(cid:173)
`formation going to the subscriber. The second half of the
`packet consists of a steady stream of light. Many channels
`of information are contained in the signal leaving the centrall
`office (CO) by means of wavelength division multiplexing. At
`the remote node, these multiple wavelengths are separated by
`means of a passive element known as a Dragone router [ 4].,
`and each wavelength then travels separately to its intended
`optical network unit (ONU). The ONU could be an individuall
`subscriber node (home) or a group of up to four. At the
`ONU, the packet travels through a fiber splitter, sending the
`downstream signal to a receiver and simultaneously sending
`the packet to an optical modulator. The modulator blanks oult
`the first half of the packet (although this is not required) and
`imprints the upstream data on the CW light in the second
`half of the packet. This data then travels back to the remote
`node where the Dragone router multiplexes it with the other
`wavelengths and they are sent back to the central office
`again. Since there are relatively few central offices compared
`to the ONU's, it is possible to use high-performance (i.e.,
`expensive) transmitters and receivers in the CO's. However,
`to provide the same level of component at every ONU woulcl
`be prohibitively expensive, so it is therefore necessary to
`provide an inexpensive alternative able to provide the most
`basic service. Bandwidth requirements for several levels of
`services are shown in Table I.
`
`1057-7157/96$05.00 © 1996 IEEE
`
`Cisco Systems, Inc.
`Exhibit 1030, Page 1
`
`
`
`46
`
`JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 5, NO. 1. MARCH 1996
`
`TABLE l
`BAND\VIDTH REQUIREMEKfS FOR VARIOUS LEVELS OF TELEPHONY SERVICES
`
`Service Bandwidth Requirements
`
`Standard telephone service (DSO)
`
`Video conference service (6 DSOs)
`
`64 kBitls
`
`384 kBitls
`
`Compressed video transmission
`
`1.5-4.5 MBitls
`
`Local area computer networks (LANs)
`
`Uncompressed video transmission
`
`lOMBitls
`
`lOOMBit/s
`
`A great many consumers will only be initially interested in
`maintaining the type of telephone service presently available,
`which requires only 64 kBits/s bandwidth. Although it is
`necessary to upgrade the network to provide higher-level
`services, every customer cannot be charged to pay for that
`upgrade if they are not utilizing the higher bandwidth. The
`subscriber who wishes higher bandwidth services will be
`willing to pay a surcharge to obtain higher-performance-level
`components. Volumes in the United States for the general
`service units are estimated to be roughly 100 million, given
`the present number of subscribers to the telephone network.
`A five-year deployment would therefore entail fabricating and
`installing roughly 100000 units per day over the five years.
`Again, this underscores the need for a cost effective and
`reliable solution. We are proposing as one possible solution
`the modulator described here.
`
`II. MECHANICAL ANTIREFLECTION
`SWITCH (MARS) DEVICE
`The MARS device is a surface-normal modulation device
`based on a moving antireflection film as described below.
`Examples of other devices based on vertically moving me(cid:173)
`chanical films can be found in literature, such as vertically
`moving gratings [5], moving metallic reflectors [6], and vari(cid:173)
`able air-gap Fabry-Perot structures [7]. For a fiber-to-the-home
`application it is necessary that the modulator have wide
`fabrication tolerance, high speed, wide optical spectrum, and
`low total packaged cost. Although the TI digital micromirror
`device can form a very good display element, it is unlikely to
`achieve the Mbt/s response times required here. By the very
`nature of their modulation mechanisms, the deformable
`valve and variable Fabry-Perot device suffer from a narrow
`optical spectrum and high fabrication complexity, which leads
`to high fabrication cost. As will be described here, the MARS
`device can satisfy all of the above operating requirements as
`well as promise very low fabrication cost.
`The operating principle of the MARS device is fairly simple.
`The device is based on optical interference effects between a
`dielectric film suspended above a substrate and the substrate.
`Optical antireflection films are widely used in the optics field.
`An antireflection film is formed by a dielectric film of a
`
`1300
`
`1400
`wavelength (nm)
`
`1500
`
`1600
`
`Fig. 2. Modeled optical spectrum for a suspended 1950-A-thick film of
`silicon nitride.
`
`thickness equal to 114 wavelength ( >.) of the incident light and
`with a refractive index equal to the square root of that of the
`substrate. If this film is instead suspended above the substrate
`by an air gap of >.I 4, it then becomes a nearly perfect reflector.
`there exists
`More generally, if the air gap is of thickness
`an antireflection state for m even, and for m odd there is a
`reflecting state. Therefore, the reflectivity state of a device can
`be changed from reflecting to nonreftecting by a change in air
`gap of )..j4. For the most basic MARS device, we create a
`silicon nitride membrane suspended above a silicon substrate
`by >.14 using surface micromachining techniques. The nitride
`film is then electrostatically attracted toward the substrate
`creating a >.I 4 antireflection condition upon contact. 'The
`1 case is the most fundamental and provides the largest
`m
`optical spectrum, as shown in Fig. 2. However, as mentioned
`earlier, any structure that has an initial air gap of thickness ~>
`where m is odd would have similar operating characteristics
`at a given wavelength. Due to the nature of the intended
`application, one of our goals is to strive for a wide optical
`spectrum with contrast ratio greater than 10: 1. A model of the
`optical spectrum for a suspended film of 1950-A-thick silicon
`nitride with a refractive index of 1.87 is shown in Fig. 2.
`One can see that for the >.I 4 case, contrast ratios in excess
`of 25 : 1 are possible for wavelengths ranging from below
`1300 nm to greater than 1600 nm. This broad spectral width
`makes this device useful at both typical operating wavelengths
`used in fiber-optics communications systems-1320 and 1560
`nm. For the case where rn is increased to three, the optical
`bandwidth is considerably narrower. The contrast ratio at a
`given center wavelength, however, would still be very high.
`The basic operation of the MARS device is illustrated in
`Fig. 3.
`Our modeling indicates that the deflection over the optical
`window is within 5% of the design value. Plane wave modeling
`shows that the contrast at the center wavelength remains
`greater than 180: l. Since our actual contrast is less than
`this (20 : 1 ), deflection does not appear to introduce significant
`
`Cisco Systems, Inc.
`Exhibit 1030, Page 2
`
`
`
`WALKER et al.: MECHANICAL ANTIREFLECTION SWITCH FOR FIBER-TO-THE-HOME SYSTEMS
`
`47
`
`wire bond
`
`signal
`
`1141.. membrane ~\t
`electrode "
`
`1300
`
`1400
`wavelength (nm)
`
`1500
`
`1600
`
`Fig. 4. Modeled optical spectrum for a suspended composite film of
`1000-A-thick polysilicon and 1950-A-thick silicon nitride.
`
`Fig. 3. MARS device operation.
`
`n or p doped silicon
`
`Fig. 5. Topographical and side views of a typical MARS device.
`
`degradation by itself. In fact, curvature of the optical window
`may actually help since we operate with the deflected state as
`the antireflection state, and so it will tend to deflect light away
`from the optical fiber and augment contrast.
`Fig. 4 shows the model of the optical spectrum for a device
`having a suspended film which is a composite of 1000 A
`of polysilicon and 1950 A of silicon nitride with an index
`of 1.87. As one can see, the performance is very similar in
`nature to that of the single film [3]. This is because the poly(cid:173)
`silicon layer is assumed to have the same refractive index as
`that of the silicon substrate, so that when the membrane is in
`contact with the substrate (or for any m even), the structure
`is optically identical to the nitride-only membrane. Since the
`poly-silicon layer is >.j 4 thick, form odd it actually augments
`the reflectivity, leading to enhanced modulation. As will be
`discussed below, the use of this composite film structure
`provides some flexibility in tailoring mechanical properties
`without affecting optical performance.
`
`Ill. FABRICATION
`The fabrication of the MARS device is complicated by
`the interdependent nature of the mechanical properties, op(cid:173)
`tical performance requirements, and electrical performance of
`the device. For a given silicon nitride film deposited using
`plasma-enhanced chemical vapor deposition (PECVD), the
`residual film stress cannot be tailored without affecting the
`film's refractive index. Also, changes in film stress affect
`
`the required drive voltage for actuation and the ability to
`obtain reasonable yield after the sacrificial layer etch. For the
`simplest film structure consisting of an aluminum sacrificial
`film and silicon nitride membrane, these constraints provide a
`significant challenge.
`The structure of a representative MARS device is shown in
`Fig. 5. The mechanically active area of the device is defined
`as the area released from the substrate and consists of the
`central plate supported by thin support beams. An opening in
`the electrode material on the central plate defines the optical
`window of the device. Obviously, variations of this structure
`are possible and many have been fabricated. For the purposes
`of this paper, however, we are confining our discussion to this
`basic structure.
`In the simplest fabrication method, we begin by depositing
`a 3>./4-thick film (1.16 ttm thick for an operating wavelength
`of 1560 nm) of E-gun evaporated aluminum on a doped silicon
`substrate. A film of PECVD silicon nitride is then deposited
`on the aluminum using 200 seem of 2% silane in nitrogen,
`900 seem of pure nitrogen, and 3 seem ammonia at 250°C,
`60 mW/cm2 and 900 mTorr. This results in a 1950-A-thick
`silicon nitride film with an index of 1.9 ± 0.05, which is
`under moderate tensile strain. This film is patterned using an
`SF6 :CChF2 reactive ion etch to form the devices. Electrodes
`are then added using lift-off of 55 A of titanium and 1000 A
`of gold. The structure is then released in a 60°C bath of
`commercially available aluminum etch and rinsed in flowing
`
`Cisco Systems, Inc.
`Exhibit 1030, Page 3
`
`
`
`48
`
`JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 5, NO. 1, MARCH 1996
`
`by a 1000-A-thick film of pure gold. The purpose of the
`chrome is to aid the adhesion of the gold to the silicon
`nitride. Upon removal from the aluminum etch used to release
`the membranes, the contacts appeared to be good. While the
`devices were being electrically actuated, however, bubbling
`under the gold was observed. The bubbling would continue
`until finally the electrode, and therefore the device, would
`faiL The same electrical signal was applied to devices that
`had not been released with no observation of this effect
`It was concluded that the aluminum etch was attacking the
`chrome adhesion film during release, leading to failure. We
`have subsequently switched to titanium for the adhesion film,
`and this has resulted in a robust, reliable electrode.
`Since the structure is conceptually so simple, there are a
`myriad of other material systems from which to choose. Most
`of our work, however, has been with the aluminum/nitride
`structure due to available technology. We are presently eval(cid:173)
`uating the use of other metals (e.g., copper), polymers (e.g.,
`polyimide), and highly HF-soluble glasses as the sacrificial
`film. Obviously, as with any micromachining process, etch
`selectivity is a concern. This is especially true here, since any
`change in the nitride membrane thickness would change the
`optical properties of the device.
`Another material system we have explored consists of a
`sacrificial layer of silicon dioxide and a composite structural
`layer of a 1 000-A-thick film of polysilicon and a 1950-A-thick
`film of silicon nitride. There are several advantages to this type
`of structure. 1) The mechanical properties of the membrane
`can be decoupled from its optical performance, since the
`stress in the polysilicon film can be tailored independently
`of any properties of the subsequent silicon nitride film. When
`using just a nitride film, the residual stress in the film and
`its refractive index are coupled. 2) A local oxidation method
`(LOCOS) can be used to form an oxide island under the
`desired mechanically active region only. 3) The polysilicon
`can be made arbitrarily thick outside of the optical window,
`making the structure less compliant and making the release
`procedure more reliable.
`To this point, stiction upon release of the polysilicon/nitride
`structure has been a problem. In part, this is due to our choice
`of Aj 4-thick oxide films for the sacrificial layers as well as the
`high compliance of the 1000-A-thick polysilicon film. We are
`presently exploring the use of a 3.A/4 gap as well as adding
`thicker structural polysilicon outside of the optical window.
`We are also exploring the use of supercritical C02 release
`procedures reported elsewhere [8].
`Because device cost is of such paramount importance for our
`intended application, we have kept our fabrication methods as
`simple as possible. Estimates of fabrication costs are in the
`range of a few pennies per device. We feel that packaging
`costs will be the dominant cost for any completed system.
`Fortunately, in this case the packaging costs should also be
`relatively low since the device is a surface normal device of
`very small size. In addition, the optical window can be made
`much larger than the core of a standard single-mode fiber
`resulting in alignment tolerances of tens of microns. Finally,
`the device is polarization independent All of these factors
`should result in very low-cost packaging.
`
`Fig. 6. Completed MARS device.
`
`DI water for five minutes. A sequential exchange of water
`with methanol is then used and the devices are left to air dry.
`An SEM photograph of a completed, released device is shown
`in Fig. 6.
`While the aluminum/PECVD nitride structure is convenient
`from the point of view of equipment capability and cost,
`this choice of material system does present some problems.
`Aluminum, while easily deposited and etched, is a less than
`ideal sacrificial film for this device. When depositing thick(cid:173)
`nesses approaching 1 JLm, the surface of aluminum can become
`quite rough and hillocks of greater than 1 JLm can form. We
`have explored the use of both thermally evaporated aluminum
`and E-gun evaporated aluminum. The surface structure of the
`E-gun aluminum was found to have a considerably smoother
`surface with an average surface roughness of ±60 A, as
`opposed to ±250 A for thermal aluminum. These roughness
`values are for as deposited films. During PECVD nitride
`deposition they become rougher due to the heating and induced
`stress associated with the PECVD process.
`In addition, because we are using PECVD silicon nitride,
`pinhole formation in the dielectric is a potential problem.
`When coupled with the surface roughness of the aluminum,
`pinhole density becomes an even greater issue. From Fig. 3,
`one can see that if the sacrificial film is a conductor, the electric
`field for a given voltage is much higher under the wire bond
`pad area of the device than in the moving area. Therefore, the
`existence of pinholes in the nitride or aluminum hillocks under
`the nitride can lead to shorting of the electrodes and failure
`of the device. In addition to changing the sacrificial film,
`we have tried two approaches to circumventing this problem.
`1) We have added a 1000-A-thick film of evaporated silicon
`monoxide under the wire bond pad area as an extra insulator.
`2) We are patterning the aluminum so that it only exists in
`the mechanically active area of the device. We have found
`both methods can be successful, however the aluminum island
`approach requires a conformal silicon nitride and complicates
`the mechanical properties of the device.
`In our first attempts to fabricate the device, we chose an
`electrode structure of a 50-A-thick layer of chrome followed
`
`Cisco Systems, Inc.
`Exhibit 1030, Page 4
`
`
`
`WALKER et al.: MECHANICAL ANTIREFLECTION SWITCH FOR FIBER-TO-THE-HOME SYSTEMS
`
`49
`
`0.9
`
`0.8
`
`0.7
`
`0.6
`
`:;.
`·::;: 0.5
`~ -
`0.4
`~
`0.3
`
`0.2
`
`40V
`
`20V
`
`/
`
`0
`
`15 ~
`"iii
`f!
`1:: 10
`0
`(.)
`
`5
`
`wavelength (nm)
`
`Fig. 7. Measured optical spectra for a I>.-thickness air gap device.
`
`IV. RESULTS
`All of the results discussed here were obtained using devices
`fabricated with the aluminum/PECVD nitride process. As
`mentioned above, variations of beam width, length, and central
`plate size have been explored. The central plate has been varied
`from 10 x 10 11m to 100 x 100 11m. The beam width has been
`varied from 2-21 11m, and beam lengths range from 5-75 11m.
`A model of the mechanical performance is being developed
`in order to optimize the device structure.
`We have fabricated samples for operating wavelengths rang(cid:173)
`ing from 1300-1560 nm, determined primarily by the thickness
`of the sacrificial film. The spectral bandwidth for an alu(cid:173)
`minum/nitride device with a zero-bias air gap of roughly 1.37
`11m is shown in Fig. 7. The spectra for bias conditions of 20
`and 40 V show a measured contrast ratio of roughly 15: 1 at
`1370 nm for this device. Using a minimum contrast ratio of
`10: 1, the spectral width of the device is approximately 40 nm,
`as would be expected based on the modeling shown in Fig. 3
`showing that as the air gap thickness is increased, the spectral
`width decreases.
`Since a device in service in a FTTH system would be
`subjected to temperatures ranging from summer in Arizona to
`winter in Alaska, it was necessary to examine the temperature
`variation of the operating parameters. Therefore, the ambient
`temperature of this device was varied from -50°C to 90°C in
`order to determine what effect this would have on the spectral
`response. DC measurements were done in vacuum in order to
`avoid the formation of frost on the device. As can be seen
`in Fig. 8, although there is a shift in central wavelength, the
`contrast ratio remains relatively high over the desired range of
`wavelengths. We believe the shift is due to thermal expansion
`of materials and changes in the air gap due to temperature
`variant stresses. It may be possible to alleviate some of these
`issues with modifications to the device structure, as well as by
`broadening the optical bandwidth of the device by reducing
`the air gap.
`We performed speed measurements on one of the m = 4
`devices (air gap = ,\) with a 30 x 30 11m central plate, 20-
`JLm-long x 5-JLm-wide support beams, and a drive voltage of
`
`1300
`
`1320
`
`1340
`
`1380
`1360
`wavelength (nm)
`
`1400
`
`1420
`
`1440
`
`Fig. 8. Modulator contrast ratio spectra at temperatures from -50°C to
`90°C.
`
`1 '1!2.00\l' 2 5.00V
`
`.-o.oos
`
`~;OO::l/
`
`I
`
`I I< I
`
`I
`
`•
`
`o • : I I ' 0
`
`I
`
`I
`
`J
`
`I
`
`I
`
`I I \ ' ' I
`
`I
`
`I
`
`I
`
`I
`
`I
`
`I
`
`I
`
`I
`
`I I \ I
`
`I
`
`I
`
`o
`
`J
`
`I
`
`I
`
`I
`
`I
`
`f2 RUN
`
`1 o
`
`o
`
`I
`
`o
`
`- - - ~- + - - - - - - - - - -t··~--!!··:·~~~\- -- --
`I
`\
`:\I
`-----------~-\~ __________________ j__
`:I
`I----.,: I
`::I
`::I
`::I
`·::I
`,: I
`
`1
`
`-
`
`Freq C2) =308. 2kHa
`
`RlseCD=132.5ns
`
`Fal!CD=125.0ns
`
`Fig. 9. Temporal response to a square wave drive signal.
`
`25 V. The electrode area extends down the support beams and
`covers the outer edge of the central plate, leaving an opticall
`window of 20 x 20 11m. Since the required drive voltage scales
`as the inverse of the square of air gap, the drive voltage can be
`significantly reduced by going to an air gap of m = 1 or using
`a tiered structure that has an electrode spacing of m = 1, but
`a different optical window spacing.
`Device response to a square wave drive pulse train is
`shown in Fig. 9, where rise and fall times for the device
`are 132 and 125 ns (as measured in a standard laboratory
`environment), respectively. For a data communications system,
`these translate to a bitrate greater than 2 Mbit/s. Devices from
`the same wafer with central plates of 40 x 40 11m were tested
`and found to have rise and fall times of roughly 350 ns. Similar
`devices were tested in both an ambient air environment and in
`vacuum, and it was found that air damping is a dominating
`factor for device speed. The trace for the 20 x 20 11m
`device shows some ringing due to an underdamped oscillator
`condition. A clean signal can theoretically be obtained through
`the use of simple signal-conditioning techniques [3].
`A 40 x 40 11m plate device was placed in a measurement
`setup and run at 500 kHz for 50 days or over 2 trillion cycles ..
`
`Cisco Systems, Inc.
`Exhibit 1030, Page 5
`
`
`
`50
`
`JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 5, NO. 1, MARCH 1996
`
`There was no sign of any degradation of signal over the entire
`time, nor was there any change in required drive voltage. The
`device ultimately failed due to electrode failure as will be
`discussed below, not due to mechanical failure as might be
`expected. The experiment was run in a standard laboratory,
`with no special environmental conditions. It is important to
`note that this experiment was performed on a device with an
`m = 4 air gap, and the membrane did not come into contact
`with the substrate during operation.
`
`2
`
`V. DISCUSSION
`The MARS device has been shown to be a viable choice
`for a communications system optical modulator. The advan(cid:173)
`tage of the use of a simple material system such as the
`aluminumJnitride structure lies in the ease of manufacture, and
`therefore cost, of the device. There are some aspects to its
`fabrication that require further investigation and discussion.
`As mentioned earlier, the use of a conductor (specifically
`aluminum) as a sacrificial layer causes several difficulties,
`especially when coupled with the use of PECVD silicon nitride
`as the structural material. By moving the conductive surface
`of the silicon up to the bottom of the silicon nitride, the
`electric field strength is made unnecessarily high in the region
`of the wire bond pad. The pressure on the membrane for a
`2
`given voltage goes as (x, _;e:x
`) 2 where c is the relative SiNx
`static dielectric constant, x 1 is the thickness of the SiNx, and
`x 2 is the thickness of the air gap. For an m = 4 device,
`this means the electrostatic force under the wire bond pad is
`2800 times higher than that in the active area of the device.
`Since PECVD nitride is prone to pinhole formation and the
`aluminum surface can be quite rough, this can result in shorting
`of contacts through pinholes. Also, PECVD silicon nitride
`is a poor electrical insulator in general, which can lead to
`current leakage. By adding a 1000-A-thick silicon monoxide
`pad between the silicon nitride and the electrode, we have
`overcome these problems. We are presently exploring other
`sacrificial films such as copper, silicon dioxide, and polyimide.
`Theoretically, the use of an oxide/polysilicon/nitride struc(cid:173)
`ture would alleviate all of the above problems. There are
`other problems associated with this structure that would need
`to be addressed, however. Primarily, the etch selectivity of
`hydrofluoric acid to silicon dioxide and silicon nitride must
`be taken into account. Any change in silicon nitride thickness
`not only changes the mechanical performance of the device,
`but the optical performance as well. In addition, release
`methods for this structure are much more complicated than
`with the aluminumJnitride structure due to the well-known
`stiction problem of polysilicon upon release. As discussed
`earlier, the use of very thick (2 p,m) support beams outside
`the optical window as well as the use of supercritical C02
`release techniques show promise.
`
`VI. CONCLUSION
`We have presented results for a micromechanical modulator
`that demonstrate its utility as a communications device for
`fiber-to-the-home applications. The MARS device has been
`shown to have high speed, greater than 2 Mbit/s data rates,
`
`temperature insensitivity with > 10 dB contrast from -50 to
`90°C, long-term reliability, and low cost. We estimate that
`the cost for a MARS device itself will be a few pennies, and
`therefore the cost of a packaged device will be dominated
`by the packaging costs themselves. Due to the polarization
`insensitivity and large optical window area relative to an
`optical fiber core, however, these should also remain relatively
`low.
`
`REFERENCES
`
`[1] T. Wood, E. C. Carr, B. L. Casper, R. A. Linke, C. A. Burrus, and
`K. L. Walker, "Bi-directional optical fiber transmission using a multiple
`quantum well (MQW) modulator/detector," Electron. Lett., val. 22,
`pp. 528-533, 1986.
`[2] N. J. Frigo, P. D. Magill, T. E. Darcie, P. P. Iannone, M. M. Downs,
`B. N. Desai, U. Koren, T. L. Koch, C. Dragone, and H. M. Presby,
`"RITE-net: A passive optical network architecture based on the remote
`interrogation of terminal equipment," in Optical Fiber Conf, San Jose,
`CA, 1994, post-deadline paper PD8, pp. 43-47.
`[3] K. W. Goossen, J. A. Walker, and S.C. Arney, "Silicon modulator based
`on mechanically-active anti-reflection layer with 1 MBit/sec capability
`for fiber-in-the-loop applications," IEEE Photon. Techno/. Lett., val. 6,
`no. 9, pp. 1119-1121, 1994.
`[4] C. Dragone, "AnN X N optical multiplexer using a planar arrangement
`of two star couplers," IEEE Photon. Techno!. Lett., val. 3, pp. 812-815,
`1991.
`[5] 0. Solgaard, F. S. A. Sandejas, and D. M. Bloom, "Deformable grating
`optical modulator," Optics Lett., val. 17, no. 9, pp. 688-690, 1992.
`[6] G. McDonald, M. Boysel, and J. Sampsell, "4 X 4 fiber-optic crossbar
`switch using the deformable mirror device," in OSA Tech. Dig. Spatial
`Light Modulators and Applicaitons, 1990, val. 14, pp. 80-83.
`[7] K. Aritani, P. J. French, P. M. Sarro, R. F. Wolffenbuttel, and S. Mid(cid:173)
`dlehoek, "Process and design considerations for surface rnicromachined
`beams for a tuneable interferometer array in silicon," in Proc. IEEE
`MEMS Workshop, Fort Lauderdale, FL, Feb. 7-10, 1993, pp. 230-235.
`[8] G. T. Mulhern, D. S. Soane, and R. T. Howe, "Supercritical carbon
`dioxide drying of microstructures," in 7th Int. Conf Solid-State Sensors
`and Actuators-Transducers '93, Yokohama, Japan, 1993, pp. 296-299.
`
`James A. Walker received the B.S. and M.S. de(cid:173)
`grees in electrical engineering from Rutgers Uni(cid:173)
`versity, New Brunswick, NJ, in 1984 and 1989,
`respectively.
`He joined AT&T Bell Laboratories in 1984 and
`is currently a Member of Technical Staff in the
`Advanced Photonics Research Department. He has
`been active in the field of rnicroelectromechanical
`systems (MEMS) since 1986. His research inter(cid:173)
`ests have included high-frequency LED's, thin-film
`shape memory alloy and, presently, the integration
`of ill-V photonic devices to VLSIIULSI silicon electronics by flip-chip
`bonding and hetero-epitaxy.
`
`the B.S. degree
`Keith W. Goossen received
`in electrical engineering from the University of
`California, Santa Barbara, in 1983, and the Ph.D.
`degree in electrical engineering from Princeton
`University, Princeton, NJ, in 1988.
`In 1988 he joined AT&T Bell Laboratories,
`Holmdel, NJ, where his work has been primarily
`on surface-normal optical modulators and their in(cid:173)
`tegration with electronics for optical interconnects.
`the first demonstration of high(cid:173)
`This
`includes
`speed, VLSI-density optical transceivers utilizing
`silicon CMOS- and GaAs-based modulators via flip-chip bonding. He has
`also investigated integration of quantum-well modulators on silicon via
`direct epitaxy and produced extensive research on quantum-well modulator
`technology.
`
`Cisco Systems, Inc.
`Exhibit 1030, Page 6
`
`
`
`WALKER et al.: MECHANICAL ANTIREFLECTION SWITCH FOR FIBER-TO-THE-HOME SYSTEMS
`
`51
`
`Susanne C. Arney received the B.S. degree in
`materials science and engineering from the Mass(cid:173)
`achusetts Institute of Technology, Cambridge, MA,
`in 1981. She received the M.S. and Ph.D. degrees
`in electrical engineering from Cornell University,
`Ithaca, NY, in 1988 and 1992, respectively. Her
`Ph.D. dissertation evaluated the nano-electrical and
`nano-electromechanical applications of the silicon(cid:173)
`on-insulator (SOl) technique.
`From 1981 to 1984, she worked at the Mo(cid:173)
`torola Bipolar Technology Center in Mesa, AZ,
`as a Process and Device Development Engineer. Her work involved the
`development of a trench isolation process. She is currently a Member of
`Technical Staff at AT&T Bell Laboratories in Murray Hill, NJ, with an active
`research program in silicon-based micromachining and microelectromechan(cid:173)
`ical systems (MEMS).
`
`Cisco Systems, Inc.
`Exhibit 1030, Page 7