Chapter 6 Optical Backplanes, Board
`and Chip Interconnects
`
`Rainer Michalzik
`University of Ulm, Optoelectronics Dept., Ulm, Germany
`
`6.1.
`
`Introduction
`
`The present chapter is intended to give some insight into the rapidly ad-
`vancing field of research into optical interconnects within data processing
`systems. This topic is dealt with in a huge number of publications at con-
`ferences, in scientific journals and their special issues, as well as book
`chapters, good access to which is obtained through [1-14]. Given the ex-
`tremely dynamic environment, it becomes clear that we cannot attempt to
`provide a complete overview but rather have to restrict ourselves to sev-
`eral examples from current research, which the readers hopefully will find
`representative. We start off by looking into recent developments in the
`10 Gbit/s data rate regime at the intersystem level. These are high-speed
`850-nm vertical-cavity surface-emitting lasers, parallel optical links, and
`new-generation silica as well as plastic optical multimode fibers. Within a
`single box environment we follow down the usual hierarchy from optical
`backplanes to intraboard to inter- and perhaps even intrachip levels. For all
`these categories we attempt to list available technology options and give
`practical examples from ongoing project work.
`The overwhelming majority of digital data inside and between even peak
`performance computer systems is nowadays carried by electrical signals
`traveling on metallic lines. Besides the fact that high-speed electronic rather
`than photonic data transmission has historically been the first available
`
`216
`
`FIBER OPTIC DATA COMMUNICATION:
`TECHNOLOGICAL TRENDS AND ADVANCES
`$35.00
`
`Copyright (cid:14)9 2002 by Academic Press.
`All rights of reproduction in any form reserved.
`ISBN: 0-12-207892-6
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 1 of 54
`
`

`

`6. Optical Backplanes, Board and Chip Interconnects
`
`217
`
`technique, clearly there are a number of advantages that speak in its favor.
`In the computer environment the most important ones are ease of imple-
`mentation and handling, low cost, and high reliability. However, with the
`exponentially increasing performance of electronic processing systems,
`more and more drawbacks of the conventional approach become apparent.
`The pronounced waveguide dispersion characteristics of electrical lines
`correspond to a relatively small bandwidth-distance product so that at
`higher clock rates it is an increasingly difficult task to bridge the required
`distances even within a single box. Electromagnetic cross-talk is forcing
`circuit designers to increase the width of data buses rather than the bus
`frequency. Susceptibility to electromagnetic interference requires proper
`shielding and grounding, which can turn thin wires into bulky copper
`strands, thus limiting the overall interconnect density. Finally signaling
`rates in the hundreds of megahertz regime make impedance matching a ne-
`cessity, which is not easy to achieve in practice and only works over a very
`limited frequency range. Apart from the clock rate's driving force, novel
`distributed or parallel computing approaches fuel the need for high band-
`width linking of individual data processing subsystems. Bearing these chal-
`lenges in mind it is widely recognized that the consequences of an apparent
`electrical signaling bottleneck are experienced at shorter and shorter link
`lengths and that optical interconnects hybridly integrated with electronics
`have attractive solutions at hand or are at least potentially able to offer them.
`It has become customary to classify optical interconnects into distinct,
`albeit overlapping, categories. Some of those are illustrated in Fig. 6.1,
`ranging from the longest to the shortest transmission distance:
`
`~ Rack-to-rack, also called frame-to-frame;
`(cid:12)9 Board-to-board;
`~ Multi-chip module (MCM)-to-MCM or intraboard;
`(cid:12)9 Chip-to-chip on a single MCM;
`(cid:12)9 Intra- or on-chip.
`
`On the frame-to-frame level, it is a relatively easy task to replace space-
`consuming and performance-limited copper cables with lightweight fibers.
`Single-channel or space-parallel optical transceiver modules are already
`commercially available at reasonable cost. Board-to-board interconnection
`within a rack can be accomplished via edge connections to optical wave-
`guides placed on a hybrid electrical/optical backplane or via free-space trans-
`mission. Within a printed circuit board, routed fiber circuits or integrated
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 2 of 54
`
`

`

`218
`
`Rainer Michalzik
`
`Transmitter / Receiver
`
`Fiber-parallel optical interconnect
`
`3card Leve
`
`Electric~ / optical
`Intraboard
`backplane
`optical\ ~ 1
`c h a n n e ~ , ~ * ~ ~
`
`Optical
`waveguides
`
`Free-space connections
`MCM and Chip Levels
`Optical
`interconnectio
`board -------~/'/,~:~~~~'~~ Optical
`connections
`
`T r a n s m i t t ~ ~ ~ ~ / / ~
`Receiversrs ~~;~=~',~~'~~" Electrical
`Logic IC ~ , / ~ ~ ~ / v
`~
`circuit board
`
`Fig. 6.1 Major optical interconnect categories.
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 3 of 54
`
`

`

`6. Optical Backplanes, Board and Chip Interconnects
`
`219
`
`channel waveguides can be applied for inter-MCM communication. Data
`transfer between or even within chips finally might be achieved by optical
`fiber bundles or an optical overlay providing guided-wave or free-space
`data transport. In most of these scenarios the demand for high-throughput
`interconnection can be satisfied only by one- or two-dimensional high-
`density arrays of optoelectronic components, the surface-normal operation
`of which usually proves to be extremely beneficial.
`
`6.2. F r a m e - t o - F r a m e
`
`I n t e r c o n n e c t i o n s
`
`Owing to the benefits addressed before, it is an increasingly common prac-
`tice to take advantage of optical fibers to implement the data links between
`a high-performance computer box and its outside world, which might con-
`sist of other processing units, a storage farm, or a network server. In this
`section we thus review some recent achievements in graded-index fiber-
`based optical interconnects approaching the 10 Gbit/s data rate regime,
`which in the future are likely to be employed at the frame-to-frame level,
`replacing 1 Gbit/s or lower speed modules. In particular these advances
`rely on progress in the fabrication of high-speed vertical-cavity surface-
`emitting laser arrays and new-generation silica as well as polymer optical
`multimode fibers.
`
`6.2.1. ONE-DIMENSIONAL VCSEL A R R A Y
`D E V E L O P M E N T FOR NEXT-GENERATION
`PARALLEL OPTICAL DATA LINKS
`
`The vertical-cavity surface-emitting laser (VCSEL) [15-15b] is a fine spec-
`imen of a novel compound semiconductor device that has been success-
`fully commercialized in the last few years. Operation principles and laser
`technology are treated in some detail in Vol. 1 Chapters 2 and 16 of this
`handbook. Among the various VCSEL applications, optical datacom is
`the primary driving field. Especially Gigabit Ethernet (GbE) and related
`transceivers for graded-index (GI) multimode fiber (MMF) data trans-
`mission have become inexpensive mass products by relying on 850 nm
`short-wavelength VCSEL technology. Generally speaking, the most at-
`tractive features of datacom VCSELs include on-wafer testing capabil-
`ity, mounting technology familiar from the low-cost light-emitting diode
`market, circularly symmetric beam profiles for ease of light focusing and
`fiber coupling, high-speed modulation with low bias currents, driving volt-
`ages well compatible to silicon VLSI electronics, temperature insensitive
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 4 of 54
`
`

`

`220
`
`Rainer Michalzik
`
`operation characteristics, and obvious forming of one- or two-dimensional
`arrays. Transceiver modules are employed for a variety of tasks such as in-
`building backbone links, interconnection of computer clusters or of telecom
`gear in central offices. The aggregate data throughput of single-channel
`modules can easily be increased by using the space-division multiplexing
`technique described, e.g., in Vol. 1 Chapter 11, where optical signals are
`transmitted in parallel through a MMF ribbon cable. A useful overview
`with many references on early fiber ribbon data links has been compiled
`in [ 16]. Although significant cost breakthroughs have been achieved with
`high yield one-dimensional laser [ 17] and photodetector arrays as well as
`alignment tolerant packaging approaches [ 18], a GI MMF ribbon cable is
`still a relatively expensive component, especially if interchannel skew is to
`be minimized. Therefore parallel links are currently competitive only for
`several tens of meter transmission length. State-of-the-art modules operate
`at 2.5 Gbit/s channel data rate and thus achieve 30 Gbit/s throughput for
`a 12-channel system [19-21]. Obviously, intensive work toward modules
`with 10 Gbit/s individual channel data rate has commenced.
`Figure 6.2 shows a photograph, bit error rate (BER) characteristics,
`and an eye diagram of a 1 (cid:12)9 10 elements VCSEL array that is being de-
`veloped for these next-generation parallel optical transceivers. The oxide-
`confined VCSELs are arranged on a 250-#m pitch that is compatible to MT
`(Mechanical Transfer) -type multifiber connectors [22] and each unit cell
`of the array contains p- and n-contacts separated by 125 #m. The etched
`VCSEL mesas are planarized by polyimide to obtain a low parasitic capaci-
`tance, coplanar contact layout. With this design, modulation corner frequen-
`cies in excess of 12 GHz are achieved [23] which are well-suited for data
`transmission in the 10 Gbit/s regime. In the given experiment, about 3 # m
`active diameter lasers with an average threshold current Ith - - 340 #A, emit-
`ting in a single transverse and longitudinal mode at 850-nm wavelength,
`have been driven at identical 1.65 mA bias current and 0.65 Vpp modulation
`voltage [24], yielding a dynamic on-off ratio of 6 dB. Figure 6.2 reveals that
`the BER curves thus obtained for back-to-back operation almost coincide
`and that error rates of 1 0 - 9
`a r e reached with less than - 1 5 dBm opti-
`cal power incident onto a pin photodiode and transimpedance-amplifier-
`based fiber pigtailed receiver. Although 10 Gbit/s-compatible prototype
`VCSEL arrays are available today, the manufacturing of complete inter-
`connect modules still requires some challenges to be addressed. Among
`those are the realization of high-sensitivity MMF-compatible photore-
`ceiver arrays and the dense hybrid integration of optoelectronic chips with
`high-speed, probably silicon germanium (SiGe) based electronics [24a].
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 5 of 54
`
`

`

`6. Optical Backplanes, Board and Chip Interconnects
`
`221
`
`VCSEI
`
`p-bondpa,
`
`~tact
`
`mide
`
`l.dpad
`
`.......... (cid:12)9 ...................... I::: ........
`...... ! ............ (cid:12)9 ............... I ...............
`B-T-B operation
`of all 10 channels
`at 10 Gbit/s data rate
`
`i ................ (cid:12)9 ............... ! .................. (cid:12)9 ................ ! ...........
`[th = 340 l.tA,
`~i"
`* ~ Z - 850 nm
`.....
`
`-2
`10
`
`-3
`10
`
`10 "4
`
`O
`0 - 5
`e 1
`m
`
`10 -v
`-9
`10
`-11
`10
`
`I
`
`(cid:12)9
`
`I
`
`(cid:12)9
`
`|
`
`(cid:12)9
`
`I
`
`,
`
`I
`
`-16
`-20
`-24
`-28
`-32
`Received Optical Power (dBm)
`
`I
`
`-12
`
`F i g . 6.2 C l e a v e d u n i t cell o f a l i n e a r 1 (cid:12)9 10, 2 5 0 - # m s p a c i n g V C S E L a r r a y w i t h c o p l a n a r ,
`1 2 5 - # m p i t c h c o n t a c t a r r a n g e m e n t (top) a n d bit e r r o r rate c h a r a c t e r i s t i c s for b a c k - t o - b a c k
`( B - T - B ) o p e r a t i o n o f all 10 c h a n n e l s at 10 G b i t / s m o d u l a t i o n w i t h a r e p r e s e n t a t i v e e y e
`d i a g r a m (on a 2 0 0 ps h o r i z o n t a l a n d 9 0 0 m V v e r t i c a l s c a l e ) in the i n s e t ( b o t t o m ) .
`
`Figure 6.3 shows a typical example of a complete parallel optical link
`module. Generally, effective heat sinking is required to remove the ex-
`cess power generated by the driver and receiver electronics as well as the
`multiplexing and demultiplexing circuits.
`
`6.2.2. L O N G - D I S T A N C E 10 GBIT/S M M F
`DATA T R A N S M I S S I O N
`
`Even before the GbE standard had been finalized it had already become
`evident that backbones operating at 1 Gbit/s speed would only for a short
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 6 of 54
`
`

`

`222
`
`Rainer Michalzik
`
`Infineon Technologies' parallel optical link (PAROLI) module with and without
`Fig. 6.3
`an attached 12-fiber ribbon cable [19] (left, (cid:14)9 2000 IEEE) and bottom view of an open
`module [19a] (right).
`
`time be able to satisfy the ever-increasing bandwidth demand in local area
`network environments. Thus, in March 1999 the Institute of Electrical and
`Electronics Engineers, Inc. (IEEE) established a Higher Speed Study Group
`to explore available technology options, and the current 10-Gigabit Ethernet
`proposal is expected to be adopted as a standard termed IEEE 802.3ae [25]
`during 2002. With regard to a multimode fiber medium that is attractive
`for low apparatus cost in-building cabling, several options exist to reach
`the target data rate on the order of 10 Gbit/s. Coarse wavelength-division
`multiplexing (CWDM, sometimes also called wide-WDM), usually with
`more than 10 nm channel spacing, is an attractive option for increasing the
`data throughput while utilizing the often low-bandwidth installed MMF
`base. CWDM modules based on 780 to 860 nm wavelength VCSELs for
`4 (cid:12)9 2.5 Gbit/s data transmission over 100 m of 62.5/zm core diameter MMF
`[26] or for 8 (cid:12)9 155 Mbit/s operation [27] have been demonstrated. For the
`1300 nm long-wavelength regime, the use of edge-emitting DFB lasers
`has been proposed for 4 . 2 . 5 Gbit/s transmission over either multimode
`or single-mode fiber [28]. However, several demonstrations of 10 Gbit/s
`VCSEL transmitters and MMF receivers together with the progress of
`CMOS and SiGe electronics clearly show the feasibility of a straight for-
`ward serial high-speed solution without the added complexity of optical
`multi- and demultiplexing. Indeed, the short-wavelength CWDM approach
`is presently not considered for the IEEE 802.3ae physical layer [28a].
`Subcarrier multiplexing and multilevel coding [29] or adaptive electronic
`equalization [30] as alternative upgrading methods have not yet been suf-
`ficiently evaluated for practical use but might become competitive in the
`future. On the other hand, an adoption of the 850 nm serial solution into the
`10-GbE standard required the availability of a MMF with improved modal
`bandwidth.
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 7 of 54
`
`

`

`6. Optical Backplanes, Board and Chip Interconnects
`
`223
`
`The maximum transmission distances achievable with a MMF are lim-
`ited by intermodal dispersion resulting in pulse broadening and thus inter-
`symbol interference. As an example, for the 850-nm short-wavelength
`regime the GbE standard defines an operating range of 275 m for a 62.5-#m
`core diameter GI MMF with a bandwidth-distance product of 200 MHz (cid:12)9
`km. Due to lower numerical aperture and smaller core area and thus better
`manufacturing control over the refractive index profile, a 50-/~m diameter
`fiber featuring 500 MHz (cid:12)9 km normalized bandwidth will suffice up to
`distances of 550 m [31, 32]. Although transmission over longer lengths has
`been demonstrated by taking advantage of restricted fiber launch from a
`laser source [33], average quality MMFs are not able to support 10 Gbit/s
`signals over sufficiently long distances.
`Minimum cabling length requirements can be extracted from the results
`of a survey of 107 U.S. companies of diverse size, location, and industry
`conducted by the IEEE in July 1996 during the development of GbE [34].
`According to Fig. 6.4, more than 90% of the in-building fiber backbones
`are less than 300 m in length, which is thus defined as a target value. In early
`2000, Lucent Technologies was the first company to introduce a 50-#m core
`diameter MMF with an increased bandwidth-distance product of 2.2 GHz (cid:12)9
`km at 850-nm wavelength [34]. Besides the adoption of GbE and related
`standards, commercialization of VCSEL technology, and the realization
`of low-cost connector and other hardware for fiber local area networks
`(LANs), this development is regarded as a key opportunity for further
`
`8000
`
`7000
`
`6000
`
`5000
`o 4000
`
`E 3000
`=
`Z 2000
`
`1000
`
`_
`
`//
`1
`
`lfli
`lll l
`100- 200- 300- 400- >500
`0-
`-199
`-299
`-399
`-499
`-99
`Backbone link length interval (m)
`
`100
`
`80 ~
`0
`
`o , , , ~
`
`60 x~
`
`(cid:12)9 , . . . i
`
`40 ~
`
`. , , , o
`
`20 ~
`
`0
`
`Fig. 6.4 In-building fiber backbone distance distribution from a 7/1996 IEEE survey [34].
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 8 of 54
`
`

`

`224
`
`Rainer Michalzik
`
`penetration of optoelectronics into data networks [35]. Fiber optimization
`consists of a reproducible and improved realization of a close to parabolic
`refractive index grading that minimizes the group delay differences between
`the multiple guided modes over a certain target wavelength regime. Of
`special importance is the suppression of the center dip in the index profile,
`which is easily established during fiber preform preparation [36], leading
`to propagation delay particularly for the fundamental mode. Despite being
`designed for the short-wavelength region, the bandwidth requirements of
`1.3 #m wavelength signals as defined in the IEEE 802.3z GbE standard
`1000BASE-SX physical layer are still satisfied by the new fiber.
`Figure 6.5 displays the results of time-domain differential mode delay
`(DMD) measurements carried out on a conventional as well as a new gen-
`eration MMF [37]. In this technique, an optical pulse is launched from a
`single-mode fiber into the MMF under test at different radial offset posi-
`tions and the resulting pulse shapes at the MMF end are recorded. Due to the
`excitation of different mode groups with different group velocities, the con-
`ventional fiber shows significant pulse splitting and wandering of the peak
`
`'
`'
`Current generation MI~F
`/ ~ ~ 2 0 ~tm radial offset
`
`15 ~tm
`
`O11li
`
`I
`I
`I
`New generation MMF
`
`~
`I
`
`I
`
`0 lam
`I
`
`reference
`pulse
`
`l
`
`20 lam
`
`10 lum
`
`0 ~tm
`
`0
`
`0.1
`
`0.2
`
`0.4
`0.3
`Time (ns)
`
`0.5
`
`0.6
`
`0.7
`
`Fig. 6.5 850-nm wavelength differential mode delay characteristics of 500-m spools of a
`typical deployed MMF (top) and of a new-generation optimized MMF (bottom), both with
`50-/zm core diameter [37].
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 9 of 54
`
`

`

`6. Optical Backplanes, Board and Chip Interconnects
`
`225
`
`10-2
`
`,l,,,,a
`
`~D
`r
`
`o
`
`, ~ . , ~
`
`0-4
`1
`
`10-6
`
`10-8
`
`10-10
`
`LazrSPEED TM MMF,
`to 5 spools in series;
`830 nm SM VCSEL,
`10 Gb/s data rate
`
`(cid:12)9 B-T-B
`o 900 m MMF
`(cid:12)9 1.6 km MMF
`
`10-12
`. . .
`-25
`
`- 1 5 " " i 0 ....
`-20
`....
`-;
`Received Optical Power (dBm)
`
`Fig. 6.6 Bit error rate characteristics for 10 Gbit/s data transmission with an 830-nm
`wavelength transverse single-mode (SM) V C S E L over up to 1.6 km of a high-performance
`multimode fiber as well as for back-to-back (B-T-B) operation [38] ((cid:14)9 1999 IEEE).
`
`position with varying launch offset. In contrast, the bandwidth-optimized
`fiber features nearly ideal pulse propagation behavior over almost the en-
`tire core area with only slight broadening versus the directly detected ref-
`erence pulse. Thus, alignment-tolerant data transmission properties can be
`expected, which retains the cost benefits of a MMF-based system.
`Figure 6.6 shows bit error rate measurements for 10 Gbit/s data trans-
`mission with an 830-nm VCSEL source over this fiber medium [38]. Error-
`free operation over up to 1.6 km distance is achieved with only 3 dB power
`penalty. Five connectorized fiber spools of 500, 400, 300, 300, and 100 m
`length are used in the experiment. Further including a 2-km-long fiber
`piece, transmission over even 2.8 km was shown to be possible [39]. Al-
`though in these two demonstrations single-mode emission turned out to be
`an essential feature due to chromatic dispersion limitations, highly multi-
`moded VCSELs can be employed as well for transmission over distances
`that will be required for 10-GbE [40].
`With a further increase of bandwidth demand beyond the direct cur-
`rent modulation capabilities of laser transmitters, CWDM approaches as
`mentioned before will certainly gain importance for premises backbone
`links. With a channel spacing of several nm, the requirements for wave-
`length stabilization of optical sources as well as on the demultiplexing
`process can be considerably relaxed compared
`to dense W D M
`in
`telecommunications. For a first laboratory realization of a 40 Gbit/s M M F
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 10 of 54
`
`

`

`226
`
`Rainer Michalzik
`
`10-2
`
`1
`0-4
`
`10-6
`
`10-8
`
`~D
`,4,-,a
`
`0
`
`1 0 ' ~ t
`10-12 _
`-20
`
`VCSEL W D M over 310 m M M F cable,
`4 • 10 Gbit/s simultaneously
`\
`
`n 815 nm, B-T-B
`A 815 nm, M M F
`<> 822 nm, B-T-B
`(cid:12)9 822 nm, M M F
`o 828 nm, B-T-B
`(cid:12)9 828 nm, M M F
`o 835 nm, B-T-B
`(cid:12)9 835 nm. M M F
`
`-4
`-8
`-
`-12
`-16
`Received Optical Power (dBm)
`
`0
`
`Fig. 6.7 Bit error rates in a 4-channel C W D M experiment showing 4 (cid:12)9 10 Gbit/s data
`transmission over 310 m of a 5 0 - # m core diameter M M F cable [41 ].
`
`system [41 ], oxide-confined high-speed transverse single-mode VCSELs
`in the 3 to 4 # m active diameter range with emission wavelengths of 815,
`822, 828, and 835 nm have been selected. VCSEL beams are combined
`through polarizing beam splitters and a MMF 3-dB coupler and a grating
`is used for free-space demultiplexing. The transmission experiment is per-
`formed by driving all VCSELs simultaneously with decorrelated 10 Gbit/s
`pseudo-random bit sequences from a pattern generator, whereas only one
`of the channels is detected. Figure 6.7 shows the measured BERs before
`and after the insertion of a 310-m-long high-bandwidth MMF cable, where
`error-free operation is achieved for all channels. Power penalties between
`1 and 3 dB are mainly caused by polarization noise generated during beam
`multiplexing and a below 15 dB adjacent wavelength suppression in the
`present demultiplexing setup. Integration of devices as demonstrated at
`lower speeds in [27] or [28] is likely to yield very compact affordable mod-
`ules for 40 Gbit/s or higher data rates in the future.
`
`6.2.3. DATA T R A N S M I S S I O N OVER P L A S T I C
`OPTICAL F I B E R S
`
`Traditionally plastic (sometimes also denoted polymer) optical fibers
`(POFs) are mainly applied in lighting, display, image transmission, or
`simple optical interconnection of consumer electronics equipment [42].
`For several years, however, POFs have produced much interest for use in
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 11 of 54
`
`

`

`6. Optical Backpla/nes, Board and Chip Interconnects
`
`227
`
`high-speed data communications. POFs can be fabricated with larger core
`diameter than any silica glass fiber while being much more flexible and
`resistant to breaking. As a consequence, large alignment tolerances allow
`the use of injection-molded parts in connectors and transceivers, which re-
`duces the apparatus cost to a minimum. At the same time, higher bandwidth,
`lighter weight, and inherent immunity to electromagnetic interference can
`give plastic fibers an edge over conventional copper cable technology.
`Progress in POF fabrication, system applications, and corresponding pas-
`sive and active components is well documented in the Proceedings of the
`International Plastic Optical Fiber Conferences [43-45a]. Fiber manufac-
`turing is concentrated mainly in Japan with Mitsubishi Rayon Co., Ltd.,
`Asahi Kasei Co. (previously Asahi Chemical Industry Co., Ltd.), Toray
`Industries, Inc., and Asahi Glass Company as the main players.
`One of the most important benchmarks for an optical fiber is its spectral
`attenuation characteristics. Figure 6.8 shows such spectra for the well-
`known silica glass fiber and the very low-cost PMMA step-index (SI)
`POF [46]. For data communication wavelengths of around 900 nm, these
`fiber types have hugely different attenuation coefficients of about 1 dB/km
`and 104 dB/km, respectively. The various attenuation peaks in the PMMA
`
`100000
`
`10000
`
`~ ,
`
`~
`
`~
`
`=
`.o
`
`q"~
`
`1000
`
`E
`
`100
`
`l O -
`
`~
`
`]
`
`-
`
`A SI-POF
`
`PF GI-POF
`
`~.
`
`~
`
`I ~
`
`I
`
`i
`
`I
`
`t
`
`I
`
`I
`I
`- .... ~
`
`|
`
`s
`
`.
`
`Glass Optical Fiber
`
`0.1
`
`t
`t
`t
`t
`i
`t
`300 500 700 900 1100 1300 1500
`Wavelength (nm)
`
`Fig. 6.8 Currently achieved as well as potential attenuation characteristics of a perflu-
`orinated polymer (CYTOP) fiber compared with standard silica glass and polymethyl
`methacrylate (PMMA) material [46].
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 12 of 54
`
`

`

`228
`
`Rainer Michalzik
`
`spectrum are mainly caused by resonances of carbon-hydrogen bonds. In
`the more recent perfluorinated (PF) POF, basically hydrogen is replaced by
`heavier fluorine atoms, thus effectively suppressing most of the resonances
`and allowing achievement of intermediate (on a logarithmic dB-scale) at-
`tenuation coefficients of about 100 dB/km. The potential of the promis-
`ing perfluorinated material is exploited in graded-index fibers containing
`a parabolic-like refractive index profile in the core region. The so-called
`Lucina TM fiber with 120-/zm core diameter, fabricated from a glass-state
`PF polymer by Asahi Glass, offers an over-filled launch bandwidth in the
`200 MHz (cid:12)9 km range (see [47] for a discussion of maximum theoretical bit
`rates) and losses below 50 dB/km throughout the entire data communica-
`tions window from 700 to 1300 nm. Minimum losses down to 10 dB/km
`seem to be within reach [46].
`Table 6.1 provides an overview of recent system experiments performed
`with various kinds of POF and either Fabry-P6rot-type, DFB, or V C S E L
`sources in different wavelength regimes. Due to high attenuation of 1 to
`3 dB/m in the usual 780- to 860-nm short-wavelength datacom operation
`window (and even considerably higher losses for longer wavelengths),
`data transmission over PMMA-based fibers is feasible only for meter-long
`links within a cabinet [57] or on the circuit board level, as mentioned in
`
`Table 6.1 POF-Based Data Transmission Experiments
`Fiber Types are PMMA SI-POF (PMSI), PMMA GI-POF (PMGI),
`or Perfluorinated GI-POF (PFGI)
`
`Fiber
`type
`
`PMSI
`PMSI
`PMSI
`PMGI
`PFGI
`PFGI
`PFGI
`PFGI
`PFGI
`PFGI
`
`Core
`diameter
`
`1 mm
`1 mm
`120 #m
`500 #m
`170 #m
`130/zm
`130 #m
`130 #m
`155 #m
`130 #m
`
`Bit rate
`
`Distance
`
`Laser
`
`Wavelength Ref.
`
`0.5 Gbit/s
`1 Gbit/s
`2.5 Gbit/s
`2.5 Gbit/s
`2.5 Gbit/s
`2.5 Gbit/s
`2.5 Gbit/s
`9 Gbit/s
`7 Gbit/s
`11 Gbit/s
`
`100 m
`15 m
`2.5 m
`200 m
`300 m
`550 m
`550 m
`100 m
`80 m
`100 m
`
`FP
`VCSEL
`VCSEL
`FP
`FP
`VCSEL
`DFB
`VCSEL
`VCSEL
`FP
`
`650 nm
`780 nm
`835 nm
`645 nm
`645 nm
`840 nm
`1310 nm
`830 nm
`935 nm
`1300 nm
`
`[481
`[491
`[501
`[51]
`[52]
`[531
`[531
`[54]
`[551
`[56]
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 13 of 54
`
`

`

`6. Optical Backplanes, Board and Chip Interconnects
`
`229
`
`Sect. 6.5. Losses of about 0.2 dB/m allow for somewhat longer distances
`at around 650 nm, however, here the fabrication of array type laser sources
`with sufficient temperature stability imposes great difficulties. Although
`remarkable progress has been reported on red-emitting VCSELs in the
`last few years [58-60], high-speed operation at elevated temperatures still
`needs to be demonstrated. In addition, the PMMA attenuation spectrum in
`Fig. 6.8 exhibits a rather narrow dip at around 650 nm and unfortunately,
`laser performance degrades with almost every nanometer below 670 nm
`wavelength [61 ].
`Due to comparatively low attenuation over a several 100-nm-wide spec-
`tral range, the PF GI-POF appears to be a candidate for data transmission
`even up to 10 Gbit/s over 100 m distance. Graded-index POFs indeed gen-
`erally show higher bandwidth-distance products than would be expected
`from their refractive index profile. Both strong mode coupling between low
`order mode groups [62] and differential mode attenuation [63] are discussed
`as underlying mechanisms. Furthermore, the bandwidth can be enhanced
`through under-filled launch from a small numerical aperture source such
`as a VCSEL. Apart from the special modal dispersion properties it is also
`interesting to note that the material dispersion of perfluorinated material is
`lower than that of silica glass or PMMA [64]. However, as can be concluded
`from Table 6.1 there is no standardized fiber diameter as yet and long-term
`stability is still a much-debated issue. These factors all contribute to limited
`commercial fiber availability to date.
`With the aforementioned fiber properties, the application fields included
`in Fig. 6.9 are envisioned for POFs [64]. Step-index POFs with conventional
`or lower (0.2 to 0.3 rather than 0.4 to 0.5) numerical aperture (NA) could
`compete with twisted pair and coax electrical cables and be employed in
`much discussed [43-45a] digital home area networks (HANs) based on the
`IEEE 1394 (FireWire) standard. Also, data buses for entertainment, control,
`or communication in mobile systems such as cars, trains, and airplanes
`have moderate bandwidth and length requirements but do benefit from
`electromagnetic interference immunity and small size and weight of POFs
`[65, 66]. PMMA GI-POFs could be used for data rates exceeding 400 Mbit/s
`but required 650-nm transceiver components. Fiber cables for horizontal as
`well as vertical connections with below 300 m length in premises networks
`are potential applications of PF POFs. With further expected increases in
`bandwidth and reductions in loss, these fibers might even be used in campus
`LANs or penetrate into access networks for HANs, business districts, or
`residential areas. Currently however, the performance of PF POFs is by far
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 14 of 54
`
`

`

`230
`
`Rainer Michalzik
`
`I
`
`:: ~
`
`I
`
`'i""'"1['I'"""1' I':: I
`
`I
`
`I I
`
`lo G iiiiiiiiiii ~PMMA base~i]':i~::~i,::,:~ i ~: ]
`.~
`
`
`L~,~:lii',i5OdB!:kii~,i;iiii'~ii !i !! ~ili!i!iiiii!i! i (Pwrer~ g e~C~ p u~
`iii!!i!i i,i!iii iii
`
`~ IOOM
`IOM 10
`
`100
`Link length(m)
`Fig. 6.9 Bit rate vs length limits and future application areas of POFs [64].
`
`1000
`
`inferior to that of silica fibers, especially when considering the progress
`in high-bandwidth MMFs capable of 10 Gbit/s transport over more than
`300 m, as reported in the previous section. Also it has been shown [67]
`that non-standard, extended core diameter (148 or 185 #m) and thus more
`alignment tolerant GI silica MMFs allow data transmission over much
`larger distances (2.5 Gbit/s over 4 km at 1300 nm wavelength) than PF
`POFs owing to significantly lower losses.
`In any case, whereas it is relatively easy to launch the light from a laser
`source into a POE the receiving end remains the most demanding part of
`a high-speed Gbit/s plastic-based data link. First of all, the diameter of the
`photodiode has to be chosen large enough in order not to waste signal power
`and to minimize modal noise penalties, which incurs bandwidth limitations
`through the increased diode capacitance. Mostly, additional optics have to
`be used. On the other hand, the higher fiber loss compared to silica decreases
`the available link power margin so that high-sensitivity receivers have to
`be employed, especially if the links are intended to be operated within the
`eye safety margin of the infrared short-wavelength regime.
`
`6.3. Optical Backplanes
`
`The first penetration step of optical data links into single-box, high-
`performance data processing systems most likely will be the introduc-
`tion of optical backplanes to interconnect the usual multitude of printed
`
`Ex. 1023
`CISCO SYSTEMS, INC. / Page 15 of 54
`
`

`

`6. Optical Backplanes, Board and Chip Interconnects
`
`231
`
`circuit boards (PCBs). For clock frequencies exceeding values of roughly
`100 MHz, parallel electrical lines already suffer from well-known draw-
`backs such as RC time constant limitations, ohmic losses, cross-talk, re-
`flections due to impedance mismatch, or ground bounce. The usual design
`strategy consists of increasing the data bus width rather than the data rate.
`Operating a bus at higher frequencies is usually not possible, requiring a
`complete redesign. Apart from pure intra-backplane transmission, the inter-
`f