Advances in silicon-on-insulator optoelectronics

938 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 4, NO. 6, NOVEMBER/DECEMBER 1998

Advances in Silicon-on-Insulator OptoelectronicsB. Jalali, Senior Member, IEEE,S. Yegnanarayanan, T. Yoon, T. Yoshimoto, I. Rendina, and F. Coppinger

(Invited Paper)

Abstract—Recent developments in silicon based optoelectronicsrelevant to fiber optical communication are reviewed. Silicon-on-insulator photonic integrated circuits represent a powerfulplatform that is truly compatible with standard CMOS process-ing. Progress in epitaxial growth of silicon alloys has created thepotential for silicon based devices with tailored optical responsein the near infrared. The deep submicrometer CMOS processcan produce gigabits-per-second low-noise lightwave electronics.These trends combined with economical incentives will ensurethat silicon-based optoelectronics will be a player in future fiberoptical networks and systems.

Index Terms—CMOS, Gigabit Ethernet, integrated optics, op-toelectronic integrated circuit (OEIC), silicon-on-insulator (SOI),silicon-on-insulator photonic integrated circuit (SOIPIC).

I. INTRODUCTION

SILICON will play a critical role in the future optoelectron-ics industry. This trend is fueled by several developments;

namely: 1) the emergence of silicon-on-insulator (SOI) as aplatform for both photonic integrated circuits (PIC’s) as well asVLSI; 2) epitaxial growth of silicon based alloys with tailoredoptical properties; and 3) scaled CMOS technology offeringlow cost lightwave circuits with gigabits-per-second speed andlow-noise performance. This paper will describe critical issuesand recent results in these areas and their impact on fiber-opticsystems. Other applications including displays and imaging, aswell as MEMS technology are not covered in this paper. Anexcellent review of these topics has recently been provided bySoref [1].

II. SILICON-ON-INSULATOR PHOTONIC-INTEGRATED

CIRCUITS (SOIPIC)

SOI is a critical material for future electronic integratedcircuits. CMOS circuits fabricated on SOI benefit from reducedparasitics and absence of latch up [1]–[3] enabling high-speedoperation at low power. These represent key requirements inmany modern telecommunication and computation systems.The SOI structure also possesses unique optical propertiesowing to the large refractive index difference between silicon

Manuscript received July 27, 1998; revised September 14, 1998. This workwas supported by the Defense Advanced Research Projects Agency and bythe Office of Naval Research.

B. Jalali, S. Yegnanarayanan, T. Yoon, and F. Coppinger are with theDepartment of Electrical Engineering, University of California at Los Angeles,Los Angeles, CA 90095-1594 USA.

T. Yoshimoto is with the Department of Electrical Engineering, Hokkaido-Tokai University, Sapporo, Japan.

I. Rendina is with the National Council of Research of Italy, IRECE, 80124Naples, Italy.

Publisher Item Identifier S 1077-260X(98)09125-4.

and SiO . This has led to the investi-gation of the optical properties of SOI waveguides [4]–[6] andthe development of a number of photonic integrated circuits(PIC’s) [7]–[13]. Excellent optical properties as well as truecompatibility with silicon CMOS integrated circuit technologyis highly promising for future low-cost photonic integratedcircuits. While there are other methods for realizing a silicon-based waveguide, such as silicon-on-sapphire (SOS) and theglass waveguide technology, only SOI is truly compatiblewith VLSI processing. This section reviews the basics ofSOI photonic integrated circuits (SOIPIC) and presents recentresults on passive and active devices.

SOI substrates are fabricated using several different tech-nologies [2], [3]. Separation by implanted oxygen (SIMOX)technology uses implantation of oxygen at high doses (10cm ) followed by a high temperature anneal to form a buriedSiO layer in a silicon wafer. In bond-and-etchback SOI (BE-SOI) a silicon wafer is first oxidized using wet or dry oxidationtechniques followed by hydrophilic bonding to a bare siliconwafer, and a subsequent heat treatment. The first wafer is thenthinned and polished by mechanical and mechanical/chemicalprocesses to the desired thickness. A promising new SOItechnology is thesmart cutprocess [14]. The process startswith an oxidized silicon wafer followed by implantation ofhydrogen at doses in the range 10–10 cm followedby hydrophilic bonding to a bare wafer and subsequent heattreatment. During the heat treatment, the implanted wafer splitsinto two parts leaving a thin layer of SiOand Si bonded tothe second wafer. The splitting is a result of blistering, whichoccurs due to the formation of microcavities concentrated ata depth corresponding to the implantation range and spreadover a distance given by the implantation straggle [15]. Toobtain a smooth surface, the roughness associated with themicrocavities must be removed by polishing. The latter stepdoes create thickness nonuniformity although to a much lesserextent than the BESOI process. The main advantage of thesmart cut process for photonic integrated circuit applicationsis that the SiO layer can be as thick as in the BESOI, whereasthe silicon overlayer has better uniformity. Also, the siliconoverlayer can be thicker due to the larger implantation rangeof hydrogen compared to oxygen. Further, there is the potentialeconomic advantage afforded by the ability to reuse the secondwafer.

Silica glass waveguide based planar waveguides is a com-peting and more mature technology for lightwave circuits[16]–[19]. This technology uses index variation achieved bythe doping of SiO to realize a waveguide structure on either

1077–260X/98$10.00 1998 IEEE

JALALI et al.: ADVANCES IN SILICON-ON-INSULATOR OPTOELECTRONICS 939

Fig. 1. Comparison of glass and SOI waveguide technologies.

a silicon or glass substrate. Typical dopants are Ge, Ti, andP with the films deposited via chemical vapor deposition(CVD) or flame hydrolysis (FHD). Compared to SOI, thesilica waveguides are weakly confined structures. Typicalrefractive index step varies from 0.1% to 0.75% resultingin thick cladding layers. The overall waveguide thicknessis currently 50 m for the production process [20]. Thesilica waveguide technology has been applied successfullyto a wide range of applications and is a desirable platformfor photonic circuits. However as shown in Fig. 1, it is notfully compatible with electronic IC technology. There existsa large mismatch between the thickness of the waveguidelayers and the active electronic devices which have layerdimensions of a few micrometers or less. Further, the thickdielectric layers result in stress, which will be problematic inthe IC process. The weak optical confinement also preventsclose spacing of the waveguides; a property that is crucial forapplications in on-chip optical interconnects [21]. On the otherhand, because of the large refractive index difference in theSOI structure ( 58%), thin cladding layers can be used makingit compatible with the electronic IC technology. Single-modepropagation with low loss is a prerequisite for the operation ofPIC’s. Conventional wisdom suggests that the large refractiveindex step in SOI prevents single-mode propagation unlessthe waveguide has submicrometer transverse dimensions, inwhich case it will have extremely poor coupling efficiencyto optical fibers. However, as shown by Soref and Peter-mann, single-mode propagation is possible in SOI waveguideswith transverse dimensions that are large compared to theoptical wavelength in the material [4], [5]. This interestingphenomenon occurs in rib waveguides where the lateral slabregions can support guided modes. While the rib waveguidemay be multimode, the higher order modes “leak” into thesurrounding slab regions during propagation resulting in aneffective single-mode propagation in the rib region.

The loss mechanism limiting the ultimate bend radius inintegrated waveguide bends is radiation into the substrate. Theeffective index of the guided mode decreases in the waveguidebend region. Once the mode effective index becomes lowerthan the index of the lower cladding layer, substrate radiationloss occurs. Due to the large refractive index step between thesilicon guiding layer and the SiObottom cladding layer, SOIwaveguides can, in principle, achieve the sharpest bends inany current integrated waveguide technology. To attain a sharpbend and adequate fiber-to-waveguide coupling efficiency, anadiabatic taper (narrower in the bend) can be used [22].

(a)

(b)

Fig. 2. (a) Near-field image and computed mode profile for a SOI ribwaveguide. (b) Photonic integrated circuits in SOIPIC technology.

A. Passive Photonic Circuits

A number of SOI guided wave optical devices and circuitshave been demonstrated over the past few years as shown inFig. 2(b). The first directional couplers have exhibited excessinsertion losses of 1.9 dB with excellent uniformity [11].Large (5 9) star couplers with excess losses of 1.3 dB havealso been demonstrated [12]. Asymmetric Mach–Zehnder typewavelength filters with a channel spacing of 4 nm (free spectralrange (FSR) 8 nm) with suppression ratio of 18 dB have beenreported [13]. Low-loss multimode couplers with high fanoutand broad-band operation have been fabricated [23]. Opticalswitches based on the thermooptic effect [7] and free carrierinjection [9] have been reported. Wet chemical etching of theSiO has been used to realize movable SOI waveguides withpotential applications in switching and sensors [8]. Finally,asymmetrically coupled SOI and polymer waveguides havebeen used to demonstrate wavelength selective photodetection[10].

Currently, couplers and splitters represent the largest marketfor PIC’s. Most commercial couplers are based on the direc-tional coupler geometry whereas splitters use the Y-junction.A new and powerful integrated optics device is the multimodeinterference (MMI) coupler. The MMI coupler consists ofsingle-mode input and output waveguides separated by a slabregion. The slab region supports a large number of modes thatpropagate with different phase velocities leading to periodicself-imaging. This is shown in the beam propagation method(BPM) simulations of Fig. 3(a). The output waveguides areplaced at the positions of the intensity peaks. The length ofthe multimode section, scales as , where isthe number of maxima in the image (or the fanout). On the


(a)

(b)

Fig. 3. (a) 1�8 MMI splitter BPM simulation. (b) Near-field image at the output of 1� 8 MMI splitter.

other hand, the width of the multimode sectionscales asfor a given crosstalk, resulting in the length scaling as(orthe fanout). Due to the strong confinement in the SOI system,the output waveguides can be placed close together. Hence,extremely compact couplers can be realized in this system.Compared to a star coupler the MMI has several advantages.Due to the self-imaging, the input is distributed evenly amongthe output waveguides—in contrast the output distribution ina star coupler follows a Gaussian profile. In the MMI, thereis no diffraction loss such as that associated with the tails ofthe Gaussian intensity profile in the star coupler. The MMIuses Manhattan geometries whereas the star coupler requiresthe input and output waveguides to be radially distributed.This feature of MMI is extremely attractive for integrationwith electronic IC’s, as most standard computer-aided design(CAD) tools used for mask layout and fabrication in the siliconIC industry cannot handle continuously curved geometries.

Fig. 3(b) shows the near field image of the output waveg-uides for a SOI MMI coupler. The waveguide dimensionsare m, m and m. The dimensionsof the multimode section are m and m.Excellent agreement between the measured data and the BPMsimulations is observed. The MMI operation is independent ofthe wavelength in the 30-nm measured range. This propertyarises from the fact that the width of the multimode section,is large compared to the optical wavelength. This broad-bandoperation is critical for application in wavelength-division-multiplexed (WDM) networks.

Fig. 4(a) shows the photograph of a fabricated 4-in SOIwafer containing phased array waveguide gratings. The de-

vice consists of two back-to-back star couplers connectedby an array of waveguides of constant incremental pathlength difference [25]–[27]. The device disperses the differentspectral components of the signal to the different outputsand is an excellent wavelength multiplexer/demultiplexer forWDM applications. Fig. 4(b) shows the measured wavelengthresponse for a four-channel SOI phased array grating [28]. Thelight source is coupled into the center input channel througha polarizing fiber and collected at the four output channelsby a cleaved single-mode fiber (SMF). The measured FSR is7.6 nm and the channel spacing is 1.9 nm (237 GHz). Theadjacent channel crosstalk is22 dB and the on chip lossis 6 dB for all channels. We have also fabricated an eight-channel device with a channel spacing of 2 nm. The devicehas slightly higher loss otherwise its performance is similar tothe four-channel filter.

A very interesting feature of SOI waveguides is their weakpolarization dependence. Fig. 5(a) shows polarization sensitiv-ity measurements for the phased array grating [28]. A TE-TMshift of approximately 0.04 nm is observed. This to the bestof our knowledge is the lowest polarization shift observed inany waveguide technology without compensation techniques.The polarization sensitivity in integrated waveguides stemsprimarily from two sources: 1) intrinsic material birefringenceand the stress in the waveguiding layer and 2) the cross-sectional geometry of the waveguide. Unlike silica (glass)waveguides, SOI films do not have intrinsic stress. Therefore,we believe that the main source of polarization dependence inSOI waveguides is due to the asymmetry of the rib geometry.Therefore, the TE–TM dependence can be nearly eliminated


(a)

(b)

Fig. 4. (a) 4-in SOI wafer with AWG devices. (b) Measured spectralresponse for four-channel AWG. The channel spacing is 1.9 nm, the crosstalkis �23 dB and the chip loss is less than 6 dB.

with a deep waveguide etch as shown in Fig. 5(b). However,there is a limit to the rib height since a very deep etch canresult in multimode waveguiding and concomitant distortionof the spectrum.

B. Active Photonic Circuits

Photonic technology is an attractive solution to the problemassociated with distribution and processing of millimeter andmicrowave signals in modern phased-array antennas. Ad-vantages afforded by photonics include reduced weight andsize, immunity to electromagnetic interference and low RFtransmission loss [30]. Further, as the operating frequency ofthe radar increases, the incremental time steps needed to attainthe requisite beam direction resolution decreases. Guided-waveoptical delay lines provide the required precision and aremore compact than optical fibers. The waveguides are definedusing photolithography and their length can be controlledwith submicrometer precision. The accuracy of the waveguidelength is limited by processing variations or mask quantization[31]. Despite its great promise, all conventional schemes sufferfrom either the use of a lossy and expensive optical switchor an expensive fast tunable laser. The expensive, bulky andoften unreliable fast tunable lasers from the current wavelengthcontrolled time delay networks need to be eliminated whilemaintaining the ability to scale to systems with large antennacount.

(a)

(b)

Fig. 5. (a) TE–TM polarization sensitivity of four-channel AWG with a ribheight of 2�m and (b) with a rib height of 3�m.

In this section, we describe a novel SOIPIC—the self-routed electrically tunable time-delay device. The self-routingis obtained by developing an active phased array grating andby exploiting its intrinsic symmetries as shown in Fig. 6(a).Carriers injected in the phase-shifter cause a fixed wavelengthoptical signal to be routed to a different output port of thegrating. Because of this feature, the loss is independent of thenumber of delay channels so the device is readily scalableto a large number of channels. This active grating representsa generic tunable filter that is useful in other applicationsincluding WDM networks.

The electrooptic effect we use is the free-carrier plasmadispersion in silicon. Current injection in silicon creates aplasma of free-carriers which interacts with the optical field,causing the refractive index of silicon to be reduced. Thiscan be achieved, for example, in a vertical or lateral p-i-ngeometry as shown in Fig. 6(b). In order to create a uniformoptical phase-front tilt with a single control signal, the gratingwaveguides need to be integrated with incrementally longerphase-shift sections. An incremental phase-shift of 2isrequired between adjacent grating waveguide phase-shifters to


(a)

(b)

Fig. 6. (a) Self-routed true-time delay SOIPIC. (b) Vertical pin phase-shifterin SOIPIC.

achieve a tuning through the entire optical bandwidth (free-spectral range). We could also use a folded-over phase-shifterin a modulo scheme where we provide phase shift of atmost 2 . Any larger phase shift requirement is then computedmodulo . This results in drastic reduction in the powerconsumption in the phase-shifters and easy scalability to largergratings while imposing minor penalty on the optical crosstalk.

The fabrication sequence for the phase-shifter network hasas its starting point the 4-in BESOI wafer with passivefour-channel waveguide gratings already etched. Standardphotolithography processes are used to define oxide windowopenings that are etched through RIE for a pion-implant.Subsequently, the oxide mask is stripped and the sample isprepared for a n ion-implant. The sample is later annealedin a rapid-thermal anneal to activate the implanted electronicspecies. Subsequently, the surface is prepared for contactwindow opening and aluminum metallization is deposited in anelectron–beam evaporator. The metal is patterned by standardphotolithography and etched in an acid mixture.

Fig. 7 shows the wafer after the completion of phase-shiftnetwork fabrication. This array had pand n regions spaced5 m from the edge of the optical waveguide. The reversesaturation current of pin diodes was found to be between200–400 A for the linear phase-shift arrays. The breakdownvoltage was found to be between 20–40 V over the entirewafer.

The optical spectrum was obtained by exciting the devicewith a broad-band source [amplified spontaneous emissionfrom an erbium-doped fiber amplifier (EDFA)]. Since thegrating was designed to optimize the phase shifter power con-sumption, a very sparse number of 15 grating waveguides wereused to form the grating, resulting in a crosstalk of approx-imately 10 dB ( 20-dB electrical). Passive eight-channelgratings fabricated on the same wafer exhibit a crosstalk below

Fig. 7. Wafer after active SOIPIC fabrication. Inset shows a close up of thep-i-n electrode arrays.

TABLE IPEAK WAVELENGTH SHIFT WITH INJECTION CURRENT IN THE PHASE-SHIFTER

20 dB and a clean spectrum when using 60 waveguides inthe grating.

We also measured the spectral response with carrier injec-tion in the 5- m p-i-n phase shifter array. The shift in the peakwavelength as a function of the current into the linear phaseshifter array is shown in Table I. At low currents, the spectralresponse instantaneously recovers upon shutting off the currentdrive to the p-i-n diode. At higher current levels, the responseis slower to recover. This is also evident in the spectral shift,which in fact, reduces at currents above 100 mA due to thermaleffect competing with the free-carrier plasma effect. Free-carriers injected into the intrinsic region cause a reduction ofthe refractive index while the heat dissipated in the area causesthe refractive index to increase. We believe this is mainlydue to poor contact between metal and semiconductor andthe concomitant high turn-on voltage. Simulations show thatwith our next generation phase-shifter design, and improvedohmic contacts, the thermal problem will be eliminated.

C. Photodetector

The use of silicon-germanium heterostructures permits therealization of Si-based optoelectronic detectors within the1.2–1.6- m infrared wavelength window over which crys-talline Si is highly transparent [32]–[37]. Low-cost, monolithicoptoelectronic circuits will enhance the economic feasibility


of short-range optical interconnects based on parallel or mul-tiwavelength architectures. In addition they have applicationsin optical clock distribution in ULSI silicon chips. SOI waveg-uides are ideal for integration with active devices [32]. Whileefficient light generation in silicon remains an elusive goal,an efficient photodetector at 1300 nm is a real possibility.To attain absorption at 1300 nm, Ge can be used to extendthe absorption spectrum of silicon to longer wavelengths[32], [33]. To obtain adequate absorption, films with high Geconcentration are required. The 4% lattice mismatch betweenSi and Ge renders such films compressionally strained, whengrown on Si substrates. The critical layer thickness imposedby the strain energy limits the maximum thickness of theabsorption layer to 20 nm for Ge concentrations of morethan 40%. To attain a thicker absorption region, a superlatticeconsisting of alternating layers of GeSi and Si are used. Thecritical layer thickness for the superlattice is equal to that fora film whose Ge concentration is equivalent to the averagevalue of the superlattice. Although the maximum superlatticethickness depends on its design, it is typically below 1m fordetectors designed to operate at 1300 nm.

The waveguide geometry is effective in overcoming thelimitation in the absorption region thickness and the relativelylow absorption coefficient of group IV materials. In suchdetectors, the GeSi superlattice with a high average refractiveindex forms the waveguide core and Si buffer layers with lowerrefractive index constitute the top and bottom cladding layers.Top and bottom electrodes form a vertical p-i-n structure.The light propagates in the longitudinal direction until itis absorbed, ensuring a high internal efficiency. However,the external efficiency remains low due to the high modalmismatch between the optical fiber and the GeSi waveguidewith the strain-limited thin core region.

Integration of the waveguide detector with an SOI wave-guide is an attractive solution. The latter is designed tohave a large core with high numerical aperture. Using etchand regrowth, an integrated device based on a pin junctionphotodetector butt coupled to a passive SOI waveguide hasbeen demonstrated [36]. Integration of the active and passivewaveguides can also be attained using evanescent coupling.Evanescent coupling is preferred over butt coupling as iteliminates the need for the etch and regrowth process.

Fig. 8 shows the layer structure for such a photodetector.A 12 period undoped 6.6 nm Si Ge /48 nm Si is usedas the absorption region. The layers were grown by solidsource molecular beam epitaxy (MBE) [37] on top of an SOIwaveguide with a 2.5-m top silicon layer. The waveguidewas defined using reactive ion etching (RIE) of Si and SiGelayers. The detector area is 30500 m. Fig. 9 shows thewaveguide coupled photodetector structure. Light is coupledfrom a SMF into the cleaved facet of the SOI waveguide.In the photodetector region, the light transfers into the higherrefractive index SiGe absorption layer via evanescent coupling.

The device shows excellent current versus voltage (– )characteristics indicating the high quality of epitaxial layers.The reverse leakage current is 70 pA/mmat 10 V with asharp breakdown at the reverse bias of 20 V. Fig. 10 showsthe measured external responsivity of the device for 1300-nm

Fig. 8. Layer structure of the GeSi–SOI photodetector.

Fig. 9. The waveguide coupled photodetector.

Fig. 10. Measured external responsivity of the photodetector.

incident light. The external responsivity increases as the biasincreases in the small bias range due to the carrier trapping inthe potential barrier. The external responsivity is saturated at2 V. The origin of the saturation is that all the photogeneratedcarriers are swept out. The external responsivity at 20 Vreaches 0.2 A/W corresponding to the avalanche multiplicationfactor of . The responsivity can be further increased byseveral approaches. First, it is possible to enhance the respon-sivity using SiGe–Si superlattice layer with higher Ge content.Also, using a proper thickness of the superlattice layer and theSOI waveguide, the coupling efficiency between them can beoptimized [39]. To obtain absorption at 1550 nm, we haveinvestigated the use of carbon for strain compensation [34].

Another possible structure for efficient detection is the use ofGe layers grown directly on silicon. Typically, such structuresrequire a thick (and therefore time consuming) graded bufferlayer to be grown prior to the active layer growth. However, re-cent results have shown that unbuffered GeC –Si struc-tures can be grown [38]. GeSn–Si structures may also be po-tentially useful for high-efficiency infrared light detection [35].


(a)

(b)

Fig. 11. (a) Schematic of a single photonic bandgap cavity consisting of agrating with a defect “phase slip” period. (b) Schematic of cascaded photonicbandgap cavities, the grating contains several defects.

D. Periodic Structures

It is well known that placing the active device structureinside a resonant cavity enhances the performance of optoelec-tronic devices. In such structures, the presence of the cavitycauses a large enhancement of the resonant optical field atdiscrete wavelengths. In particular, this effect has been demon-strated in the realization of high-speed photodetectors withthin absorption layers [40]. Both wavelength selectivity andhigh-speed response make them ideal for WDM applications.

Usually, resonant cavity enhanced (RCE) photodetectorshave a vertical, i.e., normal-to-the-surface, cavity realized bymeans of epitaxial growth techniques. In particular, the activeabsorber section is made of a small bandgap semiconductorenclosed between parallel Bragg mirrors. These structuresare suitable for III–V compound semiconductor materialswhere a wide variety of lattice-matched materials with largerefractive index difference are available. Photodetectors withhigh reflectivity mirrors ( 95%) and wide bandwidth (50nm) have been demonstrated [41], [42].

A different class of resonant enhanced photodetectors canbe realized based on waveguide structures with a high-contrastBragg grating overlay. The photonic bandgap (PBG) effectcan be exploited by incorporating single or cascaded phase-slip regions in the grating (see Fig. 11). Bragg gratings witha phase-slip section have been developed for distributed-feedback (DFB) lasers [43], and their use as optical filtersin WDM systems has also been proposed [44]–[46]. Recently,periodic structures have been demonstrated in SOI waveguides[47], [48]. PBG based structures offer resonant cavities withhigh-quality factor and large free spectral range that aresuitable for integrated optics.

The reflection and transmission spectrum of single andcascaded PBG’s can be modeled using a variety of theoreticalapproaches [44], [45], [49]. The matrix formalism [50], [51]can predict not only the reflection and transmission spectra

Fig. 12. Simulated transmission of a PBG structure with three weaklycoupled�=2-defects.

but also the internal electric field distribution throughout PBGstructures [52]. We have used this technique to study singleand cascaded PBG’s for resonant cavity photodetectors.

Fig. 11 shows (a) single and (b) cascaded PBG cavities.The main feature of the structure in Fig. 11(a) is that ofexhibiting a narrow transmission peak centered at(Braggwavelength) in a stopband. This happens in analogy withthe effect caused by a defect in an electronic crystal, viz.,the appearance of an allowed energy state inside the crystalenergy bandgap [53]. For application to photodetectors, severalPBG cavities can be cascaded to obtain high efficiency. Thepossibility of cascading several PBG cavities would makepossible the realization of distributed photodetectors withvelocity-matched travelling-wave configuration. The latter hasbeen proposed and demonstrated in conventional nonperi-odic waveguide structures [54]–[56]. A distributed resonancephotodetector would exhibit extremely fast operation, highsaturation powers, wavelength selectivity, and high-quantumefficiency.

Recent studies on cascaded PBG’s report the presence ofcoupling causing a splitting of the resonant transmission peakinside the stopband [44]. For the simplest case of only twocascaded defects, when [see Fig. 11(b)], thecentral transmission peak at splits in two symmetricalpeaks. The separation is inversely proportional to . Thisresult can be generalized to the case ofcascaded cavitieswhere, under the stated condition, the central peak splits into

distinct peaks.Coupled cavities offer the exciting possibility of the

resonant-enhanced parallel detection of WDM optical signals.This promising feature can be reached in the weakly coupleddistributed-resonance structures in which each section isdesigned so that its resonant wavelength coincideswith one of the particular wavelengths to be detected. Fig. 12shows the optical intensity at different wavelengths along thePBG structure with three weakly coupled -defects. Thesections are tuned to Bragg wavelengths of 1550, 1552, and1554 nm. The transmission shows resonances at the threewavelengths. However, each cavity is resonant at a differentwavelength. This unconventional behavior occurs since theresonant peaks of each cavity falls outside the bandgap of


other cavities. Each wavelength is selectively enhanced in asingle defect. The device offers the possibility for paralleldetection of WDM signals.

III. FIBER-OPTIC CIRCUITS

Electronics plays an equally important role to photonics inan optoelectronic integrated circuit. These circuits representthe interface between the photonic device and the digitalelectronic world. Typical circuits include the preamp (tran-simpedance amplifier) and the quantizer (limiting amplifier)in the receiver and the laser driver in the transmitter. Becausethese circuits experience the full bandwidth of the optical link,they have traditionally been fabricated in exotic technologiessuch as GaAs or silicon bipolar. In addition to the bandwidth,noise is also critical in the optical receiver. As a result, thepreamp represents the most challenging electronic componentin an optical communication system.

CMOS technology continues to be scaled to smaller gatelengths. Currently, 0.25-m gate length CMOS is in produc-tion and 0.1 m process is in advanced stages of development.This trend is driven by the digital VLSI industry and is fueledby the need to increase the integration density. Fortuitously,the cutoff frequency, , of the transistor rapidly increaseswith scaling of the gate length. This dependence is forlong to moderate channel lengths and for short channeldevices. As a result, cutoff frequencies of 70 GHz havealready been reported for a scaled technologyLSI Logic ,allowing CMOS to challenge GaAs and bipolar technologiesfor gigabits-per-second lightwave applications. In addition tobandwidth, the cutoff frequency is also the central figure ofmerit affecting the noise performance of the preamp. At highbit rates, the minimum noise variance in a CMOS preamp(input referred noise) is proportional to

where is the bit rate and is the input capacitance ofthe transistor. The latter is also reduced with scaling of thegate length. These facts suggest that scaled CMOS shouldbe considered a viable technology for optoelectronic circuitsoperating at gigabits-per-second data rates. This conclusion issupported by recent results described below.

Fig. 13 shows the chip photo (a) and output eye diagram(b) at 1.0 and 1.25 Gb/s for a CMOS receiver consisting of atransimpedance preamplifier fabricated in 0.6-m CMOS, andan InGaAs pin photodetector with 1-pF capacitance [57]. Thecircuit had a sensitivity of 20 dBm and meets requirementsof the emerging Gigabit Ethernet network.

Fig. 14 shows the (a) chip photo and (b) output eye diagramsat 1.25 and 2.5 Gb/s for a CMOS laser driver fabricated inthe same process (0.6m) [57]. The modulated current is25 mA, into 25- load. The maximum modulation currentmeasured was 40 mA. The circuit also provides an adjustabledc bias current of 1–60 mA. This circuit also meets the GigabitEthernet specifications and is also a strong candidate for OC48WDM systems.

These results highlight the growing importance of siliconCMOS in high bit rate lightwave electronics. This is a critical

(a)

(b)

Fig. 13. (a) CMOS preamp chip photo. (b) Optical eye diagram at 1 Gb/sand 1.25 Gb/s. Horizontal: 200 ps/div. Vertical: 50 mV/div and 20 mV/div.Cp-i-n = 1 pF.

(a)

(b)

Fig. 14. (a) CMOS laser driver chip photo. (b) Output eye diagram at 1.25Gb/s and 2.5 Gb/s, Horz.: 200 ps/div and 100 ps/div, Vert.: 200 mV/div.

development as it inevitably results in lower cost and morereliable optoelectronic circuits which will in turn enhance theproliferation of optical networks and data links.

IV. SUMMARY

This paper has reviewed recent trends in silicon basedoptoelectronics for communication applications. SOI offers aplatform for high-performance guided-wave optical circuits. Incontrast to the silica-based waveguide technology, SOIPIC istruly compatible with the electronic VLSI processing. Optical


switching is possible by the thermooptical or free-carrierplasma effect and photodetection at 1.3m is achievableusing a combination of GeSi alloys and a waveguide structure.Combined with gigabits-per-second lightwave circuits, nowpossible with scaled CMOS, silicon based optoelectronics willcontinue to assume an increasingly important role in fiber-opticsystems.

REFERENCES

[1] R. Soref, “Applications of silicon-based optoelectronics,”MRS Bull.,pp. 20–24, Apr. 1998.

[2] J.-P. Colinge, Silicon-on-Insulator Technology: Materials to VLSI.Boston, MA: Kluwer Academic, 1991.

[3] L. Peters, “SOI takes over where silicon leaves off,”Semiconduct. Int.,vol. 16, pp. 48–51, 1993.

[4] R. A. Soref, J. Schmidtchen, and K. Petermann, “Large single-mode ribwaveguides in GeSi and Si-on-SiO2,” IEEE J. Quantum Electron., vol.27, pp. 1971–1974, 1991.

[5] J. Schmidtchen, A. Splett, B. Schuppert, K. Petermann, and G. Burbach,“Low-loss singlemode optical waveguides with large cross-section insilicon-on-insulator,”Electron. Lett., vol. 27, pp. 1486–1487, 1991.

[6] A. G. Rickman, G. T. Ree, and F. Namavar, “Silicon-on-insulator opticalrib waveguide loss and mode characteristics,”J. Lightwave Technol.,vol. 12, pp. 1771–1776, Oct. 1994.

[7] U. Fischer, T. Zinke, and K. Petermann, “Integrated optical waveguideswitches in SOI,” inProc. 1995 IEEE Int. SOI Conf., Tucson, AZ, Oct.1995, pp. 141–142.

[8] T. T. H. Eng, S. S. Y. Sin, S. C. Kan, and G. K. L. Wong, “Surface-micromachined movable SOI optical waveguides,”Sensors & ActuatorTech., vol. 1, pp. 348–350, 1995.

[9] C. Z. Zhao, G. Z. Li, E. K. Liu, Y. Gao, and others, “Silicon on insulatorMach-Zehnder waveguide interferometers operating at 1.3�m,” Appl.Phys. Lett., vol. 67, no. 17, pp. 2448–2449, Oct. 23, 1995.

[10] B. Pezeshki, F. Agahi, J. A. Kash, J. J. Welser, and others, “Wavelength-selective waveguide photodetectors in silicon-on-insulator,”Appl. Phys.Lett., vol. 68, no. 6, pp. 741–743, Feb. 5, 1996.

[11] P. D. Trinh, S. Yegnanarayanan, and B. Jalali, “Integrated opticaldirectional couplers in silicon-on-insulator,”Electron. Lett., vol. 31, p.2097, 1995.

[12] , “5� 9 integrated optical star coupler in silicon-on-insulatortechnology,”IEEE Photon. Technol. Lett., vol. 8, p. 794–796, 1996.

[13] P. D. Trinh, S. Yegnanarayanan, and B. Jalali, “Guided-wave opticalcircuits in silicon-on-insulator technology,” inTech. Dig. IntegratedPhotonics Research Conf., Boston, MA, 1996, vol. 6, pp. 273–277.

[14] M. Bruel, “A new silicon-on-insulator material technology,”Electron.Lett., vol. 31, no. 14, pp. 1201–1202, 1995.

[15] C. Maleville, B. Aspar, T. Poumeyrol, H. Moriceau, M. Bruel, A.J. Auberton-Herve, T. Barge, and F. Metral, “Physical phenomenainvolved in the smart-cut process,” inProc. Seventh Int. Symp. Silicon-on-Insulator Technology and Devices, Los Angeles, CA, May 1996, pp.34–46.

[16] Y. Shani, C. H. Henry, R. C. Kistler, and R. F. Kazarinov, “Integratedoptic adiabatic polarization splitter on silicon,”Appl. Phys. Lett., vol.56, no. 2, pp. 120–121, Jan. 1990.

[17] M. Kawachi, “Silica waveguides on silicon and their application tointegrated optic components,”Opt. Quantum Electron., vol. 22, pp.391–416, 1990.

[18] T. Tsuchiya, K. Ohno, and J. Sato, “A study of optical functionalintegrated circuit that uses silica-based waveguide technique,”IEICETrans. Commun., vol. E75-B, no. 9, pp. 871–879, 1992.

[19] C. H. Henry, “Silica planar waveguides,” inProc. 19th Austral. Conf.Optical Fiber Technology (ACOFT’94), Melbourne, VIC, Australia, Dec.1994, pp. 326–328.

[20] Product Catalog for Photonic Integration Research Inc. (PIRI), Colum-bus, OH.

[21] I. Hayashi, “Future OEICs: The basis for photoelectronic integratedsystems,” inIntegrated Optoelectronics, M. Dagenais, R. F. Leheny,and J. Crow, Eds. New York: Academic, 1995, pp. 645–676.

[22] L. H. Spiekman, Y. S. Oei, E. G. Metaal, F. H. Groen, P. Demeester,and M. K. Smit, “Ultrasmall waveguide bends: The corner mirrors ofthe future?,”IEE Proc. Optoelectron., 1994, vol. 142, pp. 61–65.

[23] P. D. Trinh, S. Yegnanarayanan, F. Coppinger, and B. Jalali, “Compactmulti-mode interference couplers in silicon-on-insulator technology,” inConf. Lasers and Electro-Optics, CLEO’97, Baltimore, MD, May 1997,p. 441, paper CThV4.

[24] U. Fischer, T. Zinke, J.-R. Kropp, F. Arndt, and K. Petermann, “0.1dB/cm waveguide losses in single-mode SOI rib waveguides,”IEEEPhoton. Technol. Lett., vol. 8, pp. 647–648, May 1996.

[25] H. Takahashi, S. Suzuki, K. Kato, and I. Nishi, “Arrayed-waveguidegrating for wavelength division multi/demultiplexer with nanometerresolution,”Electron Lett., vol. 26, no. 2, pp. 87–88, 1990.

[26] A. R. Vellekoop and M. K. Smit, “Four-channel integrated-optic wave-length demultiplexer with weak polarization dependence,”J. LightwaveTechnol., vol. 9, pp. 310–314, Mar. 1991.

[27] C. Dragone, “AnN�N optical multiplexer using a planar arrangementof two star couplers,”IEEE Photon. Technol. Lett., vol. 3, pp. 812–814,1991.

[28] P. D. Trinh, S. Yegnanarayanan, F. Coppinger, and B. Jalali, “Silicon-on-insulator (SOI) phased-array wavelength multi-demultiplexer withextremely low-polarization sensitivity,”IEEE Photon. Technol. Lett.,vol. 9, pp. 940–942, July 1997.

[29] R. A. Soref and B. R. Bennett, “Electrooptical effects in silicon,”IEEEJ. Quantum Electron., vol. QE-23, pp. 123–129, 1987.

[30] W. Ng, A. A. Walston, G. L. Tangonan, J. J. Lee, I. L. Newberg, and N.Bernstein, “The first demonstration of an optically steered microwavephased array antenna using true-time-delay,”J. Lightwave Technol., vol.9, pp. 1124–1131, 1991.

[31] S. Yegnanarayanan, P. D. Trinh, F. Coppinger, and B. Jalali, “Compactsilicon-based integrated optic time delays,”IEEE Photon. Technol. Lett.,vol. 9, pp. 634–635, 1997.

[32] B. Jalali, L. Naval, and A. F. J. Levi, “Si-based receivers for opticaldata links,”J. Lightwave Technol., vol. 12, pp. 930–935, 1994.

[33] L. M. Giovane, L. Liao, D. R. Lim, A. M. Agarwal, E. A. Fitzgerald,and L. C. Kimerling, “Si/0.5/Ge/0.5/relaxed buffer photodetectors andlow-loss polycrystalline silicon waveguides for integrated optical inter-connects at lambda=1:3 �m,” in Proc. SPIE—Int. Soc. Opt. Eng., SanJose, CA, 1997, vol. 3007, pp. 74–80.

[34] F. Y. Huang, K. Sakamoto, K. L. Wang, and B. Jalali, “Epitaxial SiGeCwaveguide photodetector grown on Si substrate with response in the1.3–1.55�m wavelength range,”IEEE Photon. Technol. Lett., vol. 9,pp. 229–231, Feb. 1997.

[35] G. He and H. A. Atwater, “Interband transitions in SnxGe(1�x) alloys,”Phys. Rev. Lett., vol. 79, pp. 1937–1940, Sept. 1997.

[36] A. V. P. Kesan, P. G. May, E. Bassous, and S. S. Iyer, “Integratedwaveguide-photodetector using Si/SiGe multiple quantum wells for longwavelenth applications,” inProc. IEDM, San Francisco, CA 1990, pp.637–640.

[37] T. Yoshimoto, S. Thomas, K. L. Wang, and B. Jalali, “Waveguide-fed SiGe avalanche pin photodetector grown on SOI substrate with0.2A/W external responsivity at 1.3�m,” in Tech. Dig. IntegratedPhotonic Research (IPR’98), Victoria, BC, Canada, Mar./Apr. 1998, pp.285–288.

[38] A. Sakai, T. Tatsumi, and K. Aoyama, “Growth of strain-relaxed Gefilms on Si(001) surfaces,”Appl. Phys. Lett., vol. 71, no. 24, pp.3510–3512, 1997.

[39] L. Naval, B. Jalali, L. Gomelski, and J. M. Liu, “Optimization of GeSiwaveguide detectors for 1.3�m operation,”J. Lightwave Technol., vol.14, no. 5, pp. 787–797, 1996.

[40] M. S. Unlu and S. Strite, “Resonant cavity enhanced photonic devices,”J. Appl. Phys., vol. 78, no. 2, pp. 607–639, 1995.

[41] A. G. Dentai, R. Kuchibhotla, J. C. Campbell, C. Tsai, and C. Lei,“High quantum efficiency, long wavelength InP/InGaAs microcavityphotodiode,”Electron. Lett., vol. 27, pp. 2125–2126, 1991.

[42] R. Kuchibhotla, A. Srinivasan, J. C. Campbell, C. Lei, D. G. Deppe,Y. S. He, and B. G. Streetman, “Low-voltage high-gain resonant-cavity avalanche photodiode,”IEEE Photon. Technol. Lett., vol. 3, pp.354–356, 1991.

[43] Guided-Wave Optoelectronics, 2nd Edition, ed., T. Tamir, Ed. Berlin,Germany: Springer-Verlag, 1990.

[44] H. A. Haus and Y. Lai, “Theory of cascaded quarter wave shifteddistributed feedback resonators,”IEEE J. Quantum Electron., vol. 28,pp. 205–213, Jan. 1992.

[45] R. Zengerle and O. Leminger, “Phase-shifted bragg-grating filters withimproved transmission characteristics,”J. Lightwave Technol., vol. 13,pp. 2354–2358, Dec. 1995.

[46] J. N. Damask, V. V. Wong, J. Ferrera, H. I. Smith, and H. A. Haus,“Highly coherent electron-beam-written quarter-wave-shifted distributedBragg resonators for channel dropping filters,” inTech. Dig. OFC’95,1995, pp. 225–226.

[47] J. S. Foresi, P. R. Villeneuve, J. Ferrera, E. R. Thoen, G. Steinmeyer, S.Fan, J. D. Joannopoulos, L. C. Kimerling, H. I. Smith, and E. P. Ippen,“Photonic-bandgap microcavities in optical waveguides,”Nature, vol.390, pp. 143–145, 1997.


[48] J. Tidmarsh and J. Drake, “Silicon-on-insulator waveguide Bragg grat-ings,” in Tech. Integrated Photonic Research (IPR’98), Victoria, BC,Canada, Mar. 30–Apr. 1, 1998, pp. 290–292.

[49] G. P. Agrawal and S. Radic, “Phase-shifted fiber Bragg gratings and theirapplications for wavelength demultiplexing,”IEEE Photon. Technol.Lett., vol. 6, pp. 995–997, Aug. 1994.

[50] A. Yariv and P. Yeh,Optical Waves in Crystals. New York: Wiley,1984.

[51] G. Cocorullo, F. D. Corte, I. Rendina, and A. Cutolo, “New possi-bilities for efficient silicon integrated electro-optical modulators,”Opt.Commun., vol. 86, pp. 228–235, 1991.

[52] M. A. Muriel and A. Carballar, “Internal field distributions in fiber Bragggratings,”IEEE Photon. Technol. Lett., vol. 9, pp. 955–957, July 1997.

[53] E. Yablonovitch and T. J. Gmitter, “Photonic band structure: The face-centered-cubic case,”Phys. Rev. Lett., vol. 63, no. 18, pp. 1950–1953,Oct. 1989.

[54] H. F. Taylor, O. Eknoyan, C. S. Park, K. N. Choi, and K. Chang,“Traveling wave photodetectors,” inProc. SPIE, 1990, vol. 1217, p.5963.

[55] V. M. Hietala, G. A. Vawter, T. M. Brennan, and B. E. Hammons, “Trav-eling wave photodetectors for higher-power, large-bandwidth applica-tions,” IEEE Trans. Microwave Theory Tech., vol. 43, pp. 2291–2298,Sept. 1995.

[56] L. Y. Lin, M. C. Wu, T. Itoh, T. A. Vang, R. E. Muller, D. L. Sivco,and A. Y. Cho, “Velocity-matched distributed photodetectors with highsaturation power and large bandwidth,”IEEE Photon. Technol. Lett.,vol. 8, pp. 1376–1378, Oct. 1996.

[57] T. Yoon and B. Jalali, “1.25 Gb/s CMOS differential transimpedanceamplifier for gigabit networks,” inProc. 23rd Eur. Solid State CircuitsConf., ESSCIRC’97, Southampton, U.K., Sept. 1997, pp. 140–143.

B. Jalali (S’86–M’89–SM’97), photograph and biography not available at thetime of publication.

S. Yegnanarayanan, photograph and biography not available at the time ofpublication.

T. Yoon, photograph and biography not available at the time of publication.

T. Yoshimoto, photograph and biography not available at the time ofpublication.

I. Rendina, photograph and biography not available at the time of publication.

F. Coppinger, photograph and biography not available at the time ofpublication.

Advances in silicon-on-insulator optoelectronics

Documents

Transcript of Advances in silicon-on-insulator optoelectronics