218

Optoelectronics

Scanning-electron micrograph of the patterned second layer of a 3-D photonic-bandgap structure, showing amorphous Si ontop of the first-layer structure, with groves filled with SiO2. The second layer structure is shifted by half of the periods inboth x and y directions. An intentional overetch into the first layer is also observable.Courtesy: M. Qi (J. Joannopoulos and H. I. Smith).

219

Optoelectronics

• Fabrication and Characterization of Bipolar Quantum Cascade Lasers

• Development of Semiconductor Saturable Absorber Mirrors for Mode-locked Fiber Lasers

• Development of High Speed DFB and DBR Semiconductor Lasers

• Overgrowth of Rectangular-Patterned (In,Ga)(As,P) Surfaces

• Low Temperature Growth of Aluminum-Free InGaP/GaAs/InGaAs LED and Laser DiodeHeterostructures by Solid Source MBE using a Special GaP Cell

• Monolithic Integration of Surface-Emitting Laser Diodes on GaAs VLSI Electronics

• Design and Analysis of VCSEL-Based Resonant Cavity Enhanced Photodetectors

• MSM Photodetectors compatible with the Vitesse HGaAs-4 VLSI Process

• Monolithic Integration of 1550 nm Photodetectors on GaAs Transimpedance Amplifier Chips

• GaN /AlGaN UV Photodetectors

• High Performance (1Gbps/channel) Laser Drivers, Photoreceivers, andPhotoreceiver Arrays

• Development of Semiconductor Optical Amplifiers for All-Optical Signal Processing

• Epitaxy on Electronics Integration Technology

• Aligned Pillar Bonding:Full-wafer Selective-area Bonding of Optoelectronic Devices on GaAs Integrated Circuits

• Si-on-GaAs: Monolithic Heterogeneous Integration of Si CMOS with GaAs OptoelectronicDevices using EoE, SOI, and MEMS Techniques

• The OPTOCHIP Project

• Normal-Incidence Quantum Well Infrared Photodetectors (QWIPs) forMonolithic Integration

• The Theory of Microscale Thermophotovoltaic Energy Conversion

continued

220

• Fabrication and Characterization of Microscale Thermophotovoltaic Cells

• Photonic Bandgap Structures

• A Two-Dimensional Photonic-Band-Gap Super-luminescent Light-Emitting Diode

• Silicon Photonic Band Gap, Microcavity and Waveguide Structures

• Fabrication of 3-D Photonic Bandgap Structures

• Fabrication of an Integrated-Optical Grating-Based Matched Filter forFiber-Optic Communications

• Fabrication and Design of an Integrated Channel-Dropping Filter in InP

continued

221

Fabrication and Characterization of Bipolar Quantum Cascade Lasers

PersonnelG. S. Petrich and S. G. Patterson (L. A. Kolodziejski and R. J. Ram)

SponsorshipONR/MURI, MIT Lincoln Laboratories

The formation of series-coupled diode lasers allows thecreation of high power optical sources and two-dimen-sional laser arrays. These series-coupled lasers canconsist of discrete individual lasers interconnectedtogether during the post-growth fabrication stage orindividual lasers integrated together during the epitaxialgrowth process. However, interconnecting togetherdiscrete lasers suffers from the effects of parasitic resis-tances and capacitances. In bipolar quantum cascadelasers, a low resistance Esaki (tunnel) junction is createdbetween the separate quantum well lasers during theepitaxial growth process to minimize these parasiticeffects.

The bipolar quantum cascade laser was grown by gassource molecular beam epitaxy and consists of twoidentical epitaxially integrated lasers. Each laser iscomposed of 0.75 µm Si- and Be-doped InGaP claddinglayers (nominally lattice-matched to a GaAs:Si substrate)with a single In0.2Ga0.8As quantum well residing withinthe center of a 0.22 µm GaAs confinement layer. AnEsaki junction, consisting of heavily Be- and Si-dopedGaAs layers, each layer being 25 nm thick, couples thetwo lasers epitaxially. The fully processed lasers are300 nm SiO2 oxide-stripe, gain-guided, Fabry-Perotdevices which were mounted to a heat sink with Insolder.

A 5 µm wide, 450 µm long bipolar quantum cascade laserhas been optically and electrically characterized at roomtemperature while operating in a continuous wavemode. From the voltage versus current measurements,the incremental resistance of the cascade laser is 6 ohmsand the measured voltage drop across the device isdouble that of a single laser junction. The optical spec-trum of a bipolar cascade laser that was biased justbeyond the threshold of the second junction, exhibits twopeaks at 991nm and 993nm, which correspond to theemission wavelengths of the two lasers. The laser’soptical output power versus current clearly reveals twodistinct thresholds at 1.95 kA/cm2 and 2.18 kA/cm2. An

abrupt change in the output power was observed and islikely due to the p-n-p-n structure exhibiting a thyristor-type effect in which the location of the voltage dropwithin the device changed. The relatively large thresh-old currents are primarily due to current spreading,resulting in a wider injection region. Similar, but widerdevices without heat sinking, have typically demon-strated threshold current densities of approximately450 A/cm2. The different threshold current densities ofthe two epitaxially-integrated laser junctions are be-lieved to be indicative of the greater current spreadingin the lower laser junction. Upon reaching the secondlaser’s threshold, the slope efficiency abruptly switchesfrom 0.313 W/A to 0.622 W/A per facet, correspondingto a differential quantum efficiency of 99.3%. Thesehigh efficiencies are characteristic of cascade devices. Inaddition, the clear doubling of the slope efficiency as thesecond laser diode junction exceeds threshold, providesstrong evidence of the successful implementation of abipolar cascade laser.

Further work will include optimizing of the device bythoroughly characterizing the existing devices andmaterial properties, increasing the number of activeregions in the stack, including oxidation layers toalleviate current spreading, and investigating themodulation properties of the device.

222

Development of Semiconductor Saturable Absorber Mirrors forMode-locked Fiber Lasers

PersonnelL. A. Kolodziejski, E. P. Ippen, F. X. Kärtner, D. J. Jones,* M. Joschko, P. Langlois, E. M. Koontz, E. R. Thoen, andT. Schibli * University of Colorado and National Institute of Science and Technology, Boulder, CO

SponsorshipMIT Lincoln Laboratory, MURI/AFOSR

Passive mode-locking is currently the most robustmethod of generating ultrashort (femtosecond) pulses inlasers. Due to the widespread use of semiconductorsaturable absorber mirrors, it is important to characterizethe devices so as to fully understand their principles ofoperation.

The saturable absorber mirrors fabricated thus far consistof gas source molecular beam epitaxially-deposited InP-based absorbers on GaAs-based Distributed BraggReflectors (DBRs). The DBR contains 22 pairs of quarter-wave AlAs/GaAs layers, producing a reflectivity greaterthan 99% centered at λ~1.55 µm. The InP-based absorberconsists of four InGaAs quantum wells (λ~1.53 µm)centered within an InP half-wave layer. The absorptioncharacteristics of the structure are tailored to specificlaser cavities via the deposition of dielectric coatings.

Saturation fluence measurements (reflectivity change vs.incident energy density) of the aforementioned saturableabsorber mirrors reveal the existence of Two-PhotonAbsorption (TPA). At low energy densities, the reflectiv-ity change is due to absorption bleaching. As the energydensity is increased, a large drop in reflectivity is ob-served, suggesting the presence of TPA. Incorporation ofTPA in the commonly used fast absorber model gener-ates an accurate fit to the experimental data, confirmingthe significance of TPA for these absorber mirror struc-tures.

Since the magnitude of TPA is linearly dependent on theintensity incident on the absorber mirror, for a givenstructure, the TPA threshold will be determined by itsabsorption characteristics. Alteration of the absorptionbehavior by dielectric coatings is observable at bothenergy density extremes. Distributed Bragg reflectorswithout an absorber that were also measured over thesame energy density range, exhibited no change inreflectivity and hence, verifying the presence of TPAwithin the half-wave InP absorber region.

As revealed by the saturation fluence measurements,TPA effectively limits the range of energy densities inwhich the absorber mirror will behave as a bleachedabsorber independent of other effects. Thus, for a givensemiconductor saturable absorber mirror structure, thereexists a limited energy density range over which it willprovide a mode-locking mechanism for fiber lasers.

223

Development of High Speed DFB and DBR Semiconductor Lasers

PersonnelF. Rana, M. H. Lim and E. Marley (R. Ram, H. I. Smith and L. Kolodziejski)

SponsorshipMIT Lincoln Laboratory

High-speed semiconductor DFB and DBR lasers arecrucial for high-speed optical communication links.These lasers can be directly modulated at frequenciesreaching 20 to 30 GHz. They have important applicationsin optical links based upon WDM (Wavelength DivisionMultiplexing) technology. Direct laser modulationschemes are much simpler to implement and integratethan modulation schemes based upon external modula-tors. However, modulation bandwidth of externalmodulators can easily go beyond 60 GHz. Thus, it istechnologically important to have DFB/DBR laserswhose modulation bandwidths compete with those ofexternal modulators. The goal of this project is to de-velop DFB and DBR lasers capable of being modulatedat high speeds with low distortion and chirp.

High performance DFB and DBR lasers demand thatcareful attention be paid to the design of the gratings,which provides the optical feedback. Spatial-holeburning, side-mode suppression, radiation loss, laserlinewidth, spontaneous emission in non-lasing modes,lasing wavelength selection and tunability, laser-relax-ation oscillation frequency etc. are all features that are

very sensitive to the grating design. Improved gratingdesign can significantly enhance laser performance,especially at higher modulation frequencies. In the lastfew years various techniques have been developed inMIT NanoStructures Laboratory (NSL) that allowfabrication of gratings with spatially varying characteris-tics and with long-range spatial-phase coherence.Chirped optical gratings with spatially varying couplingparameter can be made using a combination of Interfero-metric lithography, spatially phase-locked electron-beamlithography and X-ray lithography. This provides us aunique opportunity for exploring a wide variety ofgrating designs for semiconductor DFB and DBR lasers.We plan to explore laser devices suited for high speed aswell as for low noise operation.

We have developed techniques for fabricating high-speed, polyimide-planarized ridge waveguide laserstructures that have low capacitance, and are thereforeideally suited for high frequency operation. Figure 1shows the cross section of a polyimide-planarized InPDFB laser. The active region consists of strain compen-

Fig. 1: Schematic of a polyimide-planarized InP DFB ridge-waveguide laser.

Fig. 2: Scanning electron micrograph of apolyimide-planarized InP ridge-waveguide laser.

continued

224

sated InGaAsP multiple quantum wells. The grating andthe ridge are dry etched in RIE using a mixture ofhydrogen and methane. Planarization is achieved byspinning multiple coatings of polyimide followed by ahigh temperature cure. Cured polyimide is dry etched inRIE using a mixture of oxygen and carbon tetra-fluorideuntil the top of the ridge gets exposed. Ohmic contact tothe ridge is made by lift-off on top of the polyimidelayer. The thick layer of polyimide significantly reducesthe capacitance between the top metal contact and thesubstrate. A large value of this capacitance can short outthe active region at high frequencies. Figure 2 shows aScanning-Electron Micrograph (SEM) of a laser structurefabricated using this process.

We are also developing techniques to fabricate co-planarstripline structures for high-speed DFB/DBR lasers. Co-planar striplines offer improved microwave performancecompared to microstrip structures. Figure 3 shows apolyimide based co-planar stripline laser structure.Fabrication of this structure requires etching polyimidesuch that the sidewalls do not become too steep so thatmetal interconnects can be continuous over them. Wehave successfully developed etching techniques forpolyimide that allows us to control the sidewall angle.Figure 4 shows SEM of a metal interconnect runningover the sidewall of a polyimide layer.

Fig. 3: Co-planar stripline laser structure.

Fig. 4: Scanning electron micrograph of a lowcapacitance polyimide based metal interconnect.

continued

225

Overgrowth of Rectangular-Patterned (In,Ga)(As,P) Surfaces

PersonnelG. S. Petrich, E. M. Koontz, G. D. U’Ren*, M. H. Lim, T. E. Murphy, and M. J. Khan (L. A. Kolodziejski, H. I. Smith,H. A. Haus and M. S. Goorsky, Department of Materials Science and Engineering, UCLA, Los Angeles, CA)

The use of Wavelength Division Multiplexing (WDM) toexpand the capacity of optical fiber communicationsystems is further enhanced by the minimization of thechannel separation. This, in turn, leads to the need ofnarrow-band optical filters capable of selecting a singlewavelength channel, while simultaneously leaving theremaining channels undisturbed. Waveguide-coupledBragg-resonant filters provide a means of achievingnarrow bandwidth filtering as required by high-densityWDM.

In order to fabricate the Bragg-resonant filters, ~3.0 µmof InP must be epitaxially deposited on InGaAsP rectan-gular-patterned gratings without altering the gratingprofile. A low temperature atomic hydrogen-assistedoxide removal technique provides the means of minimiz-ing the time at which the grating is exposed to thetemperatures that are required for gas source molecularbeam epitaxial growth, thereby minimizing mass trans-port effects. Using this technique, a grating similar tothat required for the optical filter was fabricated, over-grown and analyzed to assess the grating fidelity.

The structure was microstructurally analyzed via recip-rocal space maps that were generated by triple axis X-raydiffractometry. For an In0.77Ga0.23As0.43P0.57 rectangu-lar-patterned grating that was fabricated via x-raylithography and reactive ion etching, and overgrownwith ~1 mm of InP, satellite reflections, which are due tothe grating, are apparent in the reciprocal space mapsnear both the InGaAsP and InP peaks. The presence of±2nd order satellite reflections indicates that the dutycycle was not 50%; the duty cycle was measured to beapproximately 42% using scanning electron microscopy.As simulations of the actual grating showed, the inten-sity of the ±2rd order reflections should be similar to theintensity of the ±3nd order reflections. For the over-grown structure, the ±2nd order reflections are moreintense than the ±3rd order reflections and the additionalintensity is attributed to the presence of strain in both theInGaAsP grating teeth and the InP grating trenches.

In addition to the ±2 order satellite features, the asym-metric reciprocal space map exhibited additional inten-sity in the -1st order satellite reflection about the InPBragg peak, which corresponds to a shift in the diffrac-tion envelope to negative ∆qx values. Similarly, the +1st

order satellite reflection about the InGaAsP Bragg peakshowed an intensity enhancement, corresponding to ashift of the envelope to positive ∆qx values. The pres-ence of an asymmetric intensity distribution suggeststhat strain is present and is opposite in sense and ofsimilar magnitude. A loss of in-plane symmetry occursdue to the one-dimensional surface grating, and hence,the unit cell of the material residing within the gratinghas assumed an orthorhombic Bravais lattice. Theminimal orthorhombic strain is expected to influence theindex of refraction, but by an amount that will notimpact the optical performance of the Bragg-resonantfilter. Thus, the successful preservation of rectangular-patterned surfaces following epitaxial growth suggeststhat devices incorporating abrupt refractive indexvariations will operate as designed.

SponsorshipMIT Lincoln Laboratory, MURI/AFOSR

226

Low Temperature Growth of Aluminum-Free InGaP/GaAs/InGaAsLED and Laser Diode Heterostructures by Solid Source MBE using aSpecial GaP Cell

PersonnelP. A. Postigo (C. G. Fonstad, Jr., in collaboration with D. Braddock, EScience, Inc.)

SponsorshipDARPA, NSF

The use of InGaP instead of AlGaAs for the fabrication oflight emitting heterostructures as Light Emitting Diodes(LEDs) and Laser Diodes (LDs) presents importantadvantages, such as the reduction of deep donor levelsand lower InGaP/GaAs interface recombination velocity.In addition, InGaP is more suitable for the reduced-temperature Molecular Beam Epitaxy (MBE) required forthe Epitaxy-on-Electronics (EoE) integration technology(475°C), since it is aluminum-free.

The use of phosphorous in MBE has traditionally beendone through the introduction of phosphine (PH3) as agaseous source. However, the use of a solid source isalso very attractive since it is easier to implement and tomaintain in the MBE system, but it must be a solidsource which gives a very high ratio of dimers to tetram-ers, i.e., P2 to P4 . The dimers have a higher stickingcoefficient and are much better for MBE growth. Twomethods have been used to produce P2 from solidsources. One is based in a two-zone-cracker cell wherethe P4 is cracked in P2 by a very high temperature section(>1000°C), and where the source is solid red-phospho-rous. The other method is based on the sublimation ofphosphorous from phosphides as GaP, which producesthe best P2 to P4 rate (around 170 compared to the 3.5 to6 of the thermal cracker). The GaP decomposition cellhas the same design as a common group-III effusion cellthat operates at high temperatures (the typical tempera-ture used for phosphide growth is around 1000°C) and itcan be operated as a regular group-III cell. However,this type of cell can produce some residual amount ofGa, that can be reduced through an special design. Thisdesign implements a dome-shaped and disk-shapedPyrolytic Boron Nitride (PBN) cap on top of the normalcrucible, that acts as a trap for the Ga atoms. In collabo-ration with Dr. David Braddock, from EScience Inc., wehave successfully demonstrated a high-capacity (100 gr.)GaP decomposition source that has produced highquality epitaxial InP and InGaP. The epilayers have beenanalyzed by Double-Crystal X-Ray diffraction (DCXR),

Photoluminescence Spectroscopy (PL), and SecondaryIon Mass Spectroscopy (SIMS). The high purity of the P2beam obtained through this method and the goodbehavior of the cell have been used to produce InP/InGaAs photodetectors and InGaP/GaAs/InGaAs LEDsheterostructures at low growth temperature (475°C).Further work will focus on using this P2 source forreduced-temperature growth of InGaP/GaAs/InGaAslaser diodes for integration on GaAs VLSI chips usingthe Epitaxy-on-Electronics process.

227

Monolithic Integration of Surface-Emitting Laser Diodes onGaAs VLSI Electronics

facets and on fabricating IPSELs on the OEIC chip MIT-OEIC-6.

Vertical-cavity surface-emitting lasers (VCSELs) areparticularly suitable as light sources for optoelectronicsintegration technologies, such as our Epitaxy-on-Elec-tronics (EoE) process. The compact, vertical geometryand the completely epitaxial growth of VCSELs makemass fabrication and testing convenient and economical.The small size of the active region results in low thresh-old currents. Finally, the surface-emitting propertyresults in an excellent beam profile of the emission,which simplifies coupling to optical fibers. VCSELs areemerging as ideal light emitters for high-density free-space interconnection, but they are very difficult to growand present many challenges, especially for EoE integra-tion.

We have done extensive modeling of a variety of p-typemirror designs (it is the p-, rather than the n-, mirror thatis the more demanding), and are presently characterizingVCSEL mirrors and active regions grown at the reducedtemperatures compatible with EoE integration. Ourultimate objective is to integrate VCSELs on OPTOCHIP,as a second generation improvement on the originalLEDs.

PersonnelA. Postigo, H. Choy and J. Ahadian (C. G. Fonstad, Jr., in collaboration with S. Choy andW. Goodhue of MIT Lincoln Laboratory and U. Mass-Lowell)

SponsorshipNSF, Lincoln Laboratory

We are studying two surface-emitting laser diodedesigns for application in the Epitaxy-on-Electronicstechnology. The first is an In-Plane, Surface-EmittingLaser (IPSEL) design developed at MIT Lincoln Labora-tory, and the second is the more common Vertical-Cavity, Surface Emitting Laser (VCSEL).

An important challenge we face with using aluminum-free heterostructures to fabricate IPSELs is the dryetching of the vertical end-mirror facets and of theangled deflector structures, because of the very differentchemical make-up of the layers. In particular, the widerbandgap InGaAsP layers contain significant amounts ofIn and P, and relatively little or no As, whereas thenarrow-gap GaAs and InGaAs layers contain roughly50% As, no P, and relatively little or no In. Conventionalchlorine-based and methane-based dry etch techniquesdo not work well with the aluminum-free heterostruc-tures. We find, for example, that ion-beam assistedchlorine etching of InGaP is very slow at room tempera-ture; at elevated temperatures where the InGaP etchessatisfactorily, GaAs layers are etched without the needfor the ion beam and severe lateral etching occurs, i.e.,the etch is not directional and not anisotropic. While wecan make use of this feature in the EoE process forremoving polycrystalline deposits, it is not useful forfacet etching.

The solution to this problem lies in changing the etchantfrom chlorine to bromine because the vapor pressures ofthe relevant bromides are much more similar than arethose of the corresponding chlorides. Consequently, it ispossible to find etch conditions for which the etch ratesof InGaP and GaAs are sufficiently similar that verticalmirror facets can be successfully etched. We have usedthese results to produce the first etched-facet aluminum-free laser diodes. The threshold current densities ofbroad-area etched-facet laser diodes are a factor of twohigher than adjacent cleaved-facet lasers. We are cur-rently working on improving the quality of the etched

228

Design and Analysis of VCSEL-Based Resonant Cavity Enhanced Photodetectors

PersonnelT. Knoedl and K. H. Choy (C. G. Fonstad, Jr.)

SponsorshipUniversity of Ulm, NSF

It is extremely desirable for monolithic optoelectronicintegration to have emitters and detectors operating atthe same wavelength, which can be fabricated from thesame basic heterostructure. To this end we have mod-eled and studied (theoretically at this stage) the design ofResonant Cavity Enhanced Photodetectors (RCEPs)using heterostructure designed primarily for VerticalCavity Surface Emitting Lasers (VCSELs).

A transmission matrix model was developed to computethe optical electric field and power in a complex, multi-layered heterostructure. The input to the program is thecomposition profile of the structure, and the programcalculates the spectral response of the detector after firstcalculating the appropriate refractive indices, absorptioncoefficients, etc. The program can accommodate gradedinterfaces and doping profiles, as well as variations intemperature and in incident angle.

Simulation indicates that the most important parameterfor the device designer is the top mirror reflectivity. Asexpected, there is a direct competition between the peakquantum efficiency of an RCEP and its spectral band-width. It is clear from the analysis that the narrowspectral bandwidth inherent in achieving a useful peakquantum efficiency from a single-resonant-cavity RCEfabricated from a modified VCSEL structure, and thesignificant shift of the resonance to longer wavelengthseen with increasing temperature, that the usefulness ofsimple RCEP/VCSEL combinations is very limited.Consequently, we have also studied VCSEL-basedRCEPs with broadened response spectra created bydepositing additional, multiple-resonance mirror stackson the top surface, after first modifying the originalVCSEL top mirror stack. Initial indications from thesestudies, which are still in progress, indicate that thisapproach is very attractive.

A simple model allowing one to make a first orderapproximation to estimate the high speed behavior ofRCEPs was also developed. The indication is that RCEPscan be twice as fast as PIN detectors with comparablequantum efficiencies.

This research was initiated by Thomas Knoedl during asix-month stay in our laboratories at MIT and waspresented, in part, as his minor thesis (Studienarbeit) tothe University of Ulm in January 1998. Mr. Knoedlreturned to MIT in the Summer of 1998 and continuedhis work with an experimental investigation of resonantcavity enhanced photodetectors fabricated from VCSELheterostructures.

229

MSM Photodetectors compatible withthe Vitesse HGaAs-4 VLSI Process

PersonnelJ. Ahadian (C. G. Fonstad. Jr.. in collaboration withJ. M. Mikkelson of Vitesse Semiconductor)

SponsorshipVitesse Semiconductor

A clear choice for a photodetector to be used in a GaAsMESFET integrated circuit technology would be theMetal-Semiconductor-Metal (MSM) photodetector sinceit uses an interdigitated array of Schottky barriers on thesurface of GaAs which structurally is very similar to theMESFET gate metal pattern, and it would appear thatMSM detectors could be readily fabricated as part of astandard MESFET process. However, because a thegates in a modern GaAs IC process are self-aligned to thesource and the drain by ion implantation, this basicprocess must be modified so that this same implant doesnot occur over the MSM detectors, thereby degradingtheir performance significantly. Furthermore, becausethe ideal operating conditions for an MSM detectorinvolve fully depleting the regions under the fingerpattern, and because MSM performance can suffer if thephoto-injected carriers are not rapidly swept out of theactive device area, additional precautions must be taken.

We have worked with Jim Mikkelson, the Chief Techni-cal Officer of Vitesse Semiconductor, to make minimallyintrusive modifications to Vitesse’s new HGaAs-4 ICprocess which will allow Vitesse to produce GaAs ICswith integrated high performance MSM photodetectors.Cells containing these detectors have been included onthe MIT-OEIC-7 chip scheduled to be received from theMOSIS service in February 1999; testing and character-ization of the new MSM detectors will be undertaken inearly 1999. The HGaAs-4 process includes deep n-typeisolation implants to define separated p-type wells andheavy p-type implants and ohmic contacts to individualwells. These features, in conjunction with the processmodifications, are expected to yield high sensitivity, highspeed MSM detectors.

Monolithic Integration of 1550 nmPhotodetectors on GaAsTransimpedance Amplifier Chips

PersonnelH. Wang (C. G. Fonstad, Jr., in collaboration M.Kuznetzov of MIT Lincoln Laboratory and R. Demingand J. M. Mikkelson of Vitesse Semiconductor)

SponsorshipMIT Lincoln Laboratory

High data rate optical communication systems requireincreasingly complex integration of high performanceelectronic circuits with sophisticated optoelectronicdevices. In the short run these needs can be met byhybrid assemblies. However the cost, performancecompromises, and reliability concerns associated withhybrid integration, and the increasing need for special-ized subcircuits which are not commercially available,make development of the monolithic integrated circuittechnology extremely desirable.

We are using several techniques to monolithicallyintegrate 1550 nm photodetectors with Gallium Arsenide(GaAs) TransImpedance Amplifiers (TIAs) to formmonolithic OptoElectronic Integrated Circuits (OEICs)for fiber-based systems. Both Epitaxy-on-Electronics(EoE), described elsewhere in this report, and AlignedPillar Bonding (APB), also described elsewhere in thisreport, are being considered. In the EoE process, opticaldevices are epitaxially grown on fully processed GaAsintegrated circuits. For this application, high-speedphotodetectors based on the lattice-mismatched InGaAs/GaAs material system are being developed and evalu-ated. For the wafer bonding process, fully lattice-matched photodetector heterostructures, grown underoptimal conditions on InP, will be bonded onto the sameGaAs circuits. After EoE epitaxy or wafer bonding, thedevice heterostructures will be processed and monolithi-cally integrated with the pre-existing electronics, yieldinghigh speed, compact, reliable monolithic OEICs.

continued

230

The GaAs transimpedance amplifier test chip (MIT-OEIC-5/LL-MORX1), which incorporates modifiedversions of a commercial Vitesse transimpedance ampli-fier, has been obtained through the MOSIS service.Included on this chip are polarization diversity hetero-dyne photoreceivers, dual-balanced photoreceivers, andother functional cells; the chip is suitable for both EoEepitaxy and wafer bonding. Measurements of the highfrequency performance of the amplifiers on these chipsindicate that they will operate well to in excess of 600MHz.

The performance of 1550 nm In0.52Al0.48As/In0.53Ga0.47As pin photodiodes grown on GaAs sub-strates using linearly graded buffers has been comparedwith that of detectors incorporating Graded Short-PeriodSuperlattice (GSSL) buffers. Detectors on linearly-graded buffer were found to be superior to those onGSSL buffer. However, the best dark current levels seen(5µA for 50µm square detectors at 2V bias) must bereduced further to be acceptable for integration. Severalways of doing this have been investigated, but theirsuccess has been limited in the sense that device-to-device variations across a chip, as well as from chip tochip, remain large. Since the goal is to take advantage ofthe close matching promised by monolithic integration,this large variation is a serious concern. Initial indica-tions are that this variation arises from the need to dolattice-mismatched growth of long-wavelength detectorsbest grown on InP, on GaAs substrates instead, and maybe an intrinsic limitation. If this is true, the obvioussolution is aligned pillar bonding (APB).

continued GaN/AlGaN UV Photodetectors

PersonnelJ. Jacobs and R. Dutt (A. I. Akinwande)

Recent developments in III-nitride technology have ledto the fabrication of GaN and AlGaN UV photodetectorswith practical applications including missile detectionsystems and flame sensors. The AlGaN alloys are themost promising materials because of their directbandgap which is tunable between 3.4 eV (364 nm) forGan and 6.2 eV (200 nm) for AlN.

Several problems still exist which inhibit the develop-ment of these devices. Epitaxial layers of the materialsgrown by MOCVD still have high defect densitiesleading to high leakage currents. Other issues includepoor ohmic contacts and low carrier concentrations ofp-layers.

The objective of this project is to design and fabricateGaN solar blind UV detectors based on GaN/AlGaNnin/npn structures. The devices under consideration areback-illuminated (a) GaN/AlGaN heterojunction photo-diodes with high quantum efficiencies were demon-strated and (b) GaN/AlGaN heterojunction phototransis-tors. The structure of the diode is as shown in Figure 5below. The use of GaN/AlGaN heterojunction allowsphotons to be absorbed within the p-n junction. Thisimproves efficiency primarily due to the elimination ofcarrier losses by surface recombination and diffusionprocesses.

SponsorshipDARPA, Lockheed-Martin, Honeywell

Ni/Au p-Contact

pGaNiGaN Ti/Al n-Contact

AlGaNAlN

Sapphire

Fig. 5: Sxtructure of the diode.

231

High Performance (1Gbps/channel) Laser Drivers, Photoreceivers, andPhotoreceiver Arrays

PersonnelJ. Ahadian (C. G. Fonstad, Jr.)

SponsorshipOptoelectronics Industry Development Association (OIDA)

As important to producing monolithic OEICs operatingat 1 Gbps/channel, or more, as developing integrationtechnologies and high performance optoelectronicdevices, is designing suitable interface circuitry, i.e., laserand LED drivers and optical receiver circuits. We havedone extensive work in this area and most recently havedesigned several laser driver and optical receivers in thenew Vitesse Semiconductor HGaAs-4 process. Thesedesigns are included in three chips submitted to theMOSIS HGaAs-4 run in Fall 1998; the chips from this runare scheduled to be received in early February 1999. Thethree chips are a multi-design chip, MIT-OEIC-7, whichcontains several driver and receiver designs, and two4 x 4 receiver array chips, MIT-OEIC-8 and MIT-OEIC-9,that were designed for the Optoelectronics IndustryDevelopment Association (OIDA). These array chips aredesigned to operate at 1 Gbps/channel and each 2.3 mmsquare chip will be housed in a 34-pin quad-flat-pack.They will be made available to optical interconnectsystems researchers by OIDA through the Joint Opto-electronics Program (JOP).

The goal of any digital optical receiver is to generate alogic output on the basis of an optical input. To this end,the photocurrent generated by an illuminated MSMphotodetector must be compared to a reference level todetermine if the optical signal is high or low. In thiswork two basic types of receivers are being considered.One is an amplifier/comparator which continuallyproduces a valid logic output reflecting its input; theother is a clocked receiver which latches its input valueat a specified instant.

The array chips both use continuous-time receivers. OnMIT-OEIC-8 the design consists of three linear amplifierstages followed by a Direct-Coupled FET Logic (DCFL)inverter. The third stage is a differential amplifier whilethe first two stages are single-ended. To reject supplynoise, however, matched pairs of amplifiers are used inthe first two stages. This scheme is used instead of three

differential stages in order to eliminate level-shiftingbetween stages which would greatly increase powerconsumption and reduce bandwidth. This design issimulated to operate above 1 Gbps, consume 4.5 mW,and respond to input signals as low as 50 µW. It occu-pies a space 80 µm by 230 µm. The second array chip,MIT-OEIC-9 uses a fully differential receiver design anda novel line-driver design optimally suited for arrayapplications to maintain a high bandwidth. This designalso operates above 1 Gbps, is more sensitive (respond-ing to 25 µW input) than the other design and should bemore robust and less sensitive to noise, but it consumesconsiderably more power (23 mW).

The test chip, MIT-OEIC-7, contains several variations onthe circuits described above, as well as a clocked receiverwhich is much more compact (40 µm by 120 µm), and isexpected to consume much less power (1.8 mW), thaneither of the continuous-time designs. We are interestedin getting experience with receivers of this type andcomparing their performance with that of continuous-time receivers.

232

Development of Semiconductor Optical Amplifiers forAll-Optical Signal Processing

PersonnelL. A. Kolodziejski, K. L. Hall*, M. Kuznetsov*, J. Donnelly*, G. S. Petrich, E. M. Koontz andS. Saini* MIT Lincoln Laboratory

SponsorshipMIT Lincoln Laboratory

Ultrafast all-optical time-division-multiplexed communi-cation networks with data rates of 100 Gbits/s arecurrently being developed. At these speeds, significantdata processing can not be accomplished. Thus, networkelements incorporating all-optical switches that arecapable of operating at 100 Gbit/s and higher, areneeded. The use of a Semiconductor Optical Amplifier(SOA) as the nonlinear medium in all-optical switches, asopposed to a fiber amplifier, is attractive; semiconduc-tors have a greater intensity-dependent refractive indexmodulation and more amplification within a givenlength. The current work, in collaboration with MITLincoln Laboratories, involves the development of a SOAoptimized for an ultrafast nonlinear interferometric all-optical switch operating at 100 Gbit/s.

In preparation for processing the SOA, InP-based passivewaveguides were fabricated. The waveguide structureconsisted of an organo-metallic vapor phase epitaxy-grown InGaAsP/InP heterostructure (comprised ofInGaAsP guiding and cladding layers, surrounded byInP). The strip-loaded geometry was defined by stan-dard photolithographic techniques. The ridges weredefined via Reactive Ion Etching (RIE) and wet chemicaletching. Although RIE defines the ridges with straightsidewalls, the resulting trench basal plane can have aroughness of ∼ 0.1 ßµm. The role of the wet etch is toremove the remaining material within the trenches, andto smoothly terminate on the InGaAsP etch stop layer.This leads to a more uniform refractive index contrastresulting in better lateral waveguide confinement.Following the definition of the ridges, the waveguideswere covered with a protective oxide layer, and the InPsubstrate was thinned. Finally, the devices were cleavedto various lengths.

The passive waveguides were characterized by lossmeasurements. For these measurements, the outputfrom a distributed feedback laser was coupled into thewaveguide layer, and a Vidicon camera recorded thetransmitted intensity. Constructive or destructivereflections within the waveguide Fabry-Perot cavity, as afunction of wavelength, produces an oscillatory outputintensity. The facet reflectivity and the active layerabsorption loss were both determined from the outputintensity peak-to-valley ratio. A loss of a≈0.6 cm-1 and areflectivity of R≈0.3 were measured for the processeddevices. These measurements are comparable to thosewaveguides that were previously processed from thissame heterostructure. The InGaAsP loss calculationprovides a lower limit for the loss, which is determinedby the intrinsic material quality.

233

Epitaxy on Electronics Integration Technology

As in standard silicon technologies, the gallium arsenideVLSI process uses aluminum-based electrical intercon-nects. We have shown that these interconnects degradewhen exposed to temperatures in excess of 475˚C.Conventional MBE practice uses a 580˚C temperatureexcursion to desorb the native oxide on the GaAs surfaceprior to growth, and even this brief high temperatureexposure (which was used in previous EoE work) resultsin appreciable damage to the interconnect lines. Inter-connect degradation is now effectively eliminated byusing cracked hydrogen to remove the native oxide at aslow as 350˚C.

The epitaxy must also be carried out below 475˚C. Thisis not compatible with the growth of high qualityAlGaAs suitable for emitters (although it can be used inpassive applications) due to aluminum’s high affinity foroxygen. To circumvent this difficulty, current EoEefforts use the aluminum-free InGaAsP materials system,which is routinely grown a reduced temperatures.

Process innovations in the areas of DGW preparation,low temperature GaAs native oxide removal, and gas-source MBE growth of EoE compatible optoelectronicdevices have removed limitations present in previousEoE demonstrations. Ring oscillator measurementsmade before and after EoE processing have verified sub-100 picosecond gate delays, consistent with sub-nanosec-ond, multi-gigahertz electronics operation. The presentEoE technology is now being applied to a variety ofapplications benefiting from the integration of highperformance heterostructure devices with VLSI-com-plexity electronics.

PersonnelJ. Ahadian and P. A. Postigo (C. G. Fonstad, Jr., in collaboration with J. M. Mikkelson of Vitesse Semiconductor andS. Choy and W. Goodhue of MIT Lincoln Laboratory and U. Mass-Lowell)

SponsorshipDARPA, Lincoln Laboratory, ONR

The development of optical interconnects has beenhampered by the lack of a viable source of complexOptoelEctronic Integrated Circuits (OEICs), circuitswhich will ultimately need to contain thousands ofoptoelectronic devices tightly integrated with VLSI-complexity electronics. Hybriding, wafer-bonding, andepitaxial lift-off have made progress in addressing thisneed, however issues of density performance, reliability,and yield suggest that monolithic integration is the bestanswer, as it has been in conventional microelectronicsmanufacturing. To answer this need we have developeda process, termed Epitaxy-on-Electronics (EoE), formonolithically integrating optoelectronic devices oncommercially processed gallium arsenide ICs.

The EoE process begins with custom-designed GaAsVLSI circuits. The electronics technology (the VitesseSemiconductor HGaAs-3 and HGaAs-4 processes)provides enhancement- and depletion-mode MESFETsand four layers of aluminum-based electrical intercon-nect, as well as OPtical Field-Effect Transistor (OPFET)and Metal-Semiconductor-Metal (MSM) photodetectors.Molecular Beam Epitaxy (MBE) is used to grow deviceheterostructures on regions of the GaAs substrate whichare exposed by cutting through the interconnect dielec-tric stack. Established fabrication techniques completethe integration procedure. The unrestricted placement ofthe optoelectronic devices occurs as part of the routinelayout of the integrated circuit; the interconnect dielectricstack in the regions designated for these devices ispartially etched at the GaAs foundry forming DielectricGrowth Windows (DGWs). The etch is completed,exposing the underlying GaAs substrate, upon receipt ofthe ICs from the manufacturer. The source/drainimplant is used as the bottom n-contact of the optoelec-tronic device. Epitaxial material is then grown in theDGWs, while polycrystalline material is deposited on theoverglass. Standard processing techniques are then usedto remove the polycrystalline material, to fabricate theoptoelectronic devices, and to interconnect the top-sideelectrical contacts of the devices to the electronics.

234

Aligned Pillar Bonding:Full-wafer Selective-area Bonding of Optoelectronic Devices onGaAs Integrated Circuits

PersonnelG. Lullo, H. Wang and D. Crankshaw (C. G. Fonstad, Jr.)

SponsorshipNSF

A new integration technique is under developmentwhich uses aligned, selective-area bonding to integrateIII-V heterostructure devices, such as laser diodes andp-i-n detectors, with commercially-processed electronicintegrated circuits to create OptoElectronic IntegratedCircuits (OEICs) with VLSI levels of density and com-plexity. This technique has been named Aligned PillarBonding (APB). The APB technique is similar in functionto the Epitaxy-on-Electronics (EoE) technique in whichIII-V device heterostructures are first grown epitaxiallyin dielectric growth windows exposing the substratesurface in selected areas on fully-processed GaAsintegrated circuit wafers. The heterostructures are thenprocessed into optoelectronic devices interconnectedwith the pre-existing electronics to create monolithicOEICs. Like EoE, APB combines established commercialelectronics processes with standard optoelectronic devicefabrication sequences, to minimize the developmenteffort necessary to achieve complex optoelectronicintegration.

One of the limitations of the EoE process is that theepitaxy must be done at under 475˚C; another limitationis that the epitaxy must be done on a GaAs substrate.While the EoE process successfully addresses theselimitations, it would be very desirable in certain applica-tions to have a technique available which does not havethese two restrictions: APB fills this need.

A key element in the APB process is the use of bondingrather than direct selective area epitaxy. In the APBtechnique, the heterostructure for an optoelectronicdevice, such as a laser diode, is first grown underoptimal conditions on the optimal substrate. The hetero-structure is then patterned into pillars, which are in turnwafer fusion bonded into dielectric windows on asuitably prepared integrated circuit wafer (i.e., into thesame windows that would be used for epitaxy in the EoEprocess). Doing this requires something more complexthan encountered in previous III-V bonding experiments,

namely alignment. The wafer on which the deviceheterostructure is grown and the pillars are formed,called the “source” wafer, must be aligned with thedielectric windows on the integrated circuit wafer ontowhich the devices are to be bonded. This wafer is calledthe “target” wafer.

In the past year we have developed a technique forachieving this alignment, and have demonstrated for thefirst time the successful aligned fusion of III-V pillars indielectric windows on a GaAs IC chip. We have alsodemonstrated the subsequent removal of the substrate ofthe source wafer to complete the transfer of the pillars tothe target wafer. This initial work involved directsemiconductor-to-semiconductor bonding and requiredsubstantial pressure to achieve contact and bonding.Subsequent work has investigated palladium-to-GaAsbonding and, more recently, layered gold alloy-to-GaAsbonding. Both have been successfully demonstrated,and work is continuing to refine the process. At thesame time, contacts with groups growing Vertical-CavitySurface-Emitting Laser (VCSEL) layers are being pur-sued with the goal of establishing a collaboration withsuch a group to use the APB technique to integrateVCSELs on, for example, OPTOCHIP die.

235

Si-on-GaAs:Monolithic Heterogeneous Integration of Si CMOS with GaAsOptoelectronic Devices using EoE, SOI, and MEMS Techniques

PersonnelJ. London and J. Ahadian (C. G. Fonstad, Jr.)

SponsorshipDARPA, General Motors Fellowship

Researchers have long recognized the desirability ofmonolithically integrating III-V semiconductor optoelec-tronic devices, such as laser diodes and p-i-n detectors,with complex high density, high performance siliconelectronic integrated circuits, ideally CMOS, to economi-cally produce robust optoelectronic integrated circuitsfor a wide variety of applications. To this end, work wasbegun many years ago growing GaAs epitaxially onsilicon, and this GaAs-on-Si research continues to thisday, but with limited success. While most people focuson the lattice constant mismatch between Si and GaAs(and other III-V’s) as the main difficulty with this ap-proach, the more important issue is the large differencein Thermal Expansion Coefficient (TEC) between thesematerials. Very large stresses are induced in epitaxial III-V layers on silicon by even relatively small temperaturechanges because of the TEC difference; and such stressesare fatal to many optoelectronic devices.

We feel that the key to III-V/silicon monolithic optoelec-tronic integration lies in recognizing that optoelectronicdevices are intrinsically very thick devices (typically atleast a micron thick) which are very sensitive to stress,and silicon MOS transistors need only be a few tens ofnanometers thick and are much less stress sensitive.Silicon-On-Insulator, SOI, transistors, for example, areroutinely formed in Si films less than 10 nm thick. It isalso well known from Silicon-On-Sapphire, SOS, process-ing that such thin silicon can withstand very highstresses induced by a large difference in TEC, in this casebetween Si and sapphire (which has a TEC similar to thatof GaAs.)

Clearly the solution to monolithically integrating SiCMOS and GaAs-based optoelectronics is (1) to takeadvantage of the fact that one is not restricted to usingsilicon as the substrate for Si CMOS, and (2) to retain aGaAs substrate so that the intrinsically thick, inherentlystrain- and defect-sensitive optoelectronic devices seetheir optimum substrate, i.e., GaAs. In particular one can

imagine using wafer bonding and SOI techniques toproduce Si CMOS electronics on gallium arsenidesubstrates without sacrificing any of the performance ofthe CMOS while at the same time gaining access to state-of-the-art performance optoelectronic devices, e.g. laserdiodes and photodetectors. This is the approach we arepursuing in developing a technology we term Silicon-on-Gallium Arsenide (SonG). The SonG process com-bines silicon and gallium arsenide substrates by waferbonding. It builds expertise at MIT in the areas of waferbonding, Silicon On Insulator (SOI) MOS IC technology,and Epitaxy on Electronics (EoE) optoelectronic integra-tion technology.

In the past year we have demonstrated, for the first time,100 nm thick silicon single crystal layers bonded byintervening oxide layers on 4" diameter GaAs wafers.These Si-on-GaAs (SonG) wafers have been shown to beable to withstand temperature cycles to in excess of700˚C, and thus to be suitable for CMOS fabricationusing a reduced-temperature SOI process and forEpitaxy-on-Electronics (EoE) integration of optoelec-tronic devices using molecular beam epitaxy. The sameprocess used to create Si-on-GaAs wafers can also beused to transfer fully-processed SOI circuits to GaAswafers (to be followed by EoE optoelectronic integra-tion).

Work is now in progress to (1) demonstrate the growthof III-V device heterostructures on SonG substrates, and(2) planarize and bond processed SOI CMOS wafers toGaAs wafers using the SonG process. Success in theseareas will open the way subsequently for the full inte-gration of III-V light emitters (LEDs and laser diodes)with silicon CMOS electronics.

236

The OPTOCHIP Project

PersonnelJ. Ahadian (C. G. Fonstad, Jr., in collaboration with J. M. Mikkelson, Vitesse Semiconductor)

SponsorshipDARPA/ONR

The OPTOCHIP Project is a research foundry offeringintended to provide prototype OEICs to selected univer-sity groups doing research on optical interconnectsystems. The first generation OPTOCHIPs use InGaP/GaAs Light Emitting Diodes (LEDs) monolithicallyintegrated using the Epitaxy-on-Electronics (EoE)technology on commercially fabricated GaAs integratedcircuit chips containing Optical Field Effect Transistor(OPFET) and metal-semiconductor-metal (m-s-m)photodetectors, and enhancement- and depletion-modeMetal-Semiconductor Field Effect Transistors (MES-FETs). A solicitation for participation was made in late1995 and in early 1996 nine groups from eight universi-ties were selected to participate; the universities repre-sented are California Institute of Technology, ColoradoState University, George Mason University, McGillUniversity, Texas Christian University, University ofSouthern California, and University of Washington.These groups began work in February 1996 on thedesigns for 2 mm by 2 mm OEIC chips which werecombined into a larger die and submitted to the MOSISservice in May 1996. The chips were fabricated byVitesse Semiconductor Corporation, Camarillo, CA, inthe summer and fall of 1996, and the EoE integrationprocess was initiated early in 1997. Fabrication of theintegrated LEDs was completed in May 1997, and thecompleted OPTOCHIP die were sawn into individual2 mm by 2 mm chips and returned to the designers fordeployment in their optical interconnect architectures.We have continued working throughout 1998 with thevarious groups which did OPTOCHIP designs as theyhave done measurements on their chips and employedthem in systems demonstrations.

Electrical tests on the completed OPTOCHIP die showthat there was no degradation in the electrical perfor-mance of the circuitry due to the EoE process. Theperformance and yield of the LEDs were poorer than hadbeen achieved on previous test runs, due it appears toproblems in the growth system, but several functional

die were provided to each participating group. Theamount of work done with the completed chips variesbetween the various groups, but several groups haveperformed extensive testing of their OEIC chips andhave successfully employed them in systems-levelsituations. In one case, communication between a pair ofOPTOCHIPs has been demonstrated. Work is continu-ing on the testing of the chips, and our group at MIT isworking to be able to supply the participants withanother set of die processed using the new GaP phospho-rous source recently installed on our MBE.

The completed, overall 7mm by 7mm square OPTOCHIPdie each contain over 200 LEDs integrated with numer-ous different circuits and subsystems containing thou-sands of transistors. As such they represent some of themost complex LED-based monolithic OEICs ever fabri-cated. The OPTOCHIP project was also unique in thatthe OPTOCHIP die incorporate designs from a diverseselection of groups and in that minimal constraints wereplaced on the circuit designs. Our intention is that therewill be future OPTOCHIP offerings, and that the pro-cessing of subsequent OPTOCHIPs will be done on asemi-professional basis using the facilities of the Tech-nology Research Laboratory (TRL) of the MicrosystemsTechnology Laboratory (MTL) at MIT. We are alsoanxious to make surface-emitting lasers (SELs) availableto OPTOCHIP users in the near future.

237

Normal-Incidence Quantum Well Infrared Photodetectors (QWIPs) forMonolithic Integration

PersonnelJ. Pan and C. G. Fonstad, Jr.

SponsorshipNSF, ONR, Lockheed-Martin

Band gap engineering allows us to design the peakresponsivity of a Quantum Well Intersubband Photode-tector (QWIP) to be at anywhere in the infrared regionbeyond about 2 µm. This wavelength region is useful forspectroscopy and identification of unknown gases, aswell as for imaging in the Earth’s atmosphere in thetransparent spectral regions of 3 to 5 µm and 8 to 12 µm.Modern epitaxy techniques can achieve high uniformityof semiconductor parameters across entire III-V (GaAsand InP) wafers, which allows for the realization of largeFocal Plane Arrays (FPAs) of QWIPs with low spatial(fixed) pattern noise. Furthermore, the growth of QWIPson GaAs substrates is compatible with the integration ofQWIPs with standard GaAs detector circuits using one ofour monolithic optoelectronic integration technologies.

QWIPs which respond to normal incident radiationwithout the need for an optical grating, or other surfacepatterning, are of special interest because they can befabricated with fewer process steps and increasedexpected yield. An important contribution of our workhas been the demonstration of the first n-type QWIP(n-QWIP) which showed a significant detectivity withoutthe use of an optical grating. The detectivity of thesedevices was large enough that focal plane array perfor-mance would be limited by the uniformity of the pro-cessing rather than by the single pixel detectivity.

Another important result of our work in the past yearhas been the development of numerically accuratephysical models yielding simple analytical expressionsfor the QWIP leakage current and photocurrent, as wellas for the number of wells, and the distance, over whichcarriers are depleted from quantum wells whenever thephotocurrent is larger than the leakage current. Thisdepletion capacitance is expected to be important at highfrequencies, in situations where the photocurrent islarger than the leakage current, and in QWIPs designedwith a small number of quantum wells (as when thequantum efficiency is large or an optical cavity is used).

Studies of the microscopic physics of quantum wellswere undertaken to elucidate the physical origin of theintersubband absorption of normally incident radiation.A key result of this work was the derivation of selectionrules for intersubband absorption of normally incidentradiation by hole subbands in a p-doped QWIP(p-QWIP). It was found that the absorption of normallyincident radiation by holes in a p-QWIP is largest forheavy hole to light hole transitions. The intersubbandabsorption of normally incident radiation by electrons inan n-QWIP is found to be much smaller than that in a p-QWIP. It is also found that uniaxial strain does not havea large effect on the strength or the selection rules ofintersubband absorption because the Hamiltoniandescribing uniaxial strain has the same tetragonalsymmetry as that carriers confined to quantum wellsalong the growth direction.

Finally, common QWIP designs used in industry havebeen evaluated. The commonly used n-QWIP designwith superlattice confinement barriers intended toreduce thermionic leakage by pushing the three-dimen-sional continuum energy further up in energy whilehaving the photocurrent flow in the superlatticeminiband, was studied in detail. It was found that thisQWIP design will in fact have a tunneling leakagecurrent which is, at best, commensurate with (rather thanmuch better than) a conventional QWIP made with abarrier material having a band edge matching that of theaverage band edge of the superlattice.

The research on QWIPs described here will be continuedby Prof. Janet Pan at Yale University.

238

The Theory of Microscale Thermophotovoltaic Energy Conversion

PersonnelJ. Pan and K. H. Choy (C. G. Fonstad, Jr.)

SponsorshipUnfunded

The theoretical treatments of radiative heat transferbetween bodies in close proximity in the literature tendto be highly numerical and, consequently, relatively non-physical and non-intuitive; they also tend to be wrong,or at least to contain errors which are obscured by thenumerical calculations. The same problem has extendedto the much more limited literature on microscalethermophotovoltaic conversion. The proximity effect is,however, relatively easy to understand if one thinks interms of evanescent field coupling between two surfaces,and recalls a simple experiment involving two prismsand a laser which is often performed as a demonstrationin undergraduate electromagnetics courses.

When the two prisms are positioned with their long facesparallel, but separated, a laser beam incident on theinput face of the one prism will experience total internalreflection at the long face and exit the output face of thesame prism; this is normal prism action. However, if theseparation of the two prisms is decreased to a fraction ofa laser light wavelength, some of the laser beam willtransfer evanescently through the long face and into thesecond prism. As the separation decreases, the amountof transfer increases toward 100%.

To relate this experiment to microscale radiative transferone need simply realize that only a fraction of theradiation within a black body escapes from it and is“seen” by a distance cold surface; the fraction is thatwhich impinges on the radiating surface at less than thecritical angle for total internal reflection. If we canincrease the amount of the radiation within the body thatcan “escape”, we can extract more energy radiativelyfrom that black body. This is exactly what is done bybringing the cold surface close enough to the radiatingsurface that there is evanescent coupling across the gap.

We have recently developed an analytical model whichquantifies this picture and allows us to predict theproximity effect enhancement of the output of TPV cellsas a function of the separation between the cell and theblack body. The largest enhancement occurs when theindices of refraction of the blackbody and the TPV cellare the same, and we see that there is a maximumproximity factor, given by this index of refractionsquared. At a given separation the enhancement isgreatest for the longer wavelength cells and the celloutput is also greater for the smaller energy gap cells.However, if one does not want to have to cool the cell,one must realize that a 0.35 eV cell is about the narrow-est gap cell which can be used at room temperature, i.e.,without significant cooling. This appears to be the bestchoice of cell for the first experimental demonstration ofthis effect.

The enhancement is actually larger for the lower black-body temperatures, but the overall cell output increasesquickly as the blackbody temperature is increased, so itis desirable to operate with the highest blackbodytemperature compatible with a given application. Forthe initial experiments it is felt that it will be relativelystraightforward to operate with a blackbody as high as500˚C, but that going higher will unnecessarily compli-cate the initial experiments. It is also worthwhile notingthat significant levels of power can be generated evenfrom 200˚C blackbodies by taking advantage of theproximity effect. This means that it will be possible touse MTPV to recover waste heat energy, converting it toelectricity, in situations which are now impractical forTPV energy conversion.

239

Fabrication and Characterization of Microscale Thermophotovoltaic Cells

PersonnelK. Waithe and M. Masaki (C. G. Fonstad, Jr., in collaboration with R. DiMatteo of C. S. Draper Laboratory)

SponsorshipC. S. Draper Laboratory

We have begun a program of research to provide the firstexperimental demonstration that a dramatic increase inThermoPhotoVoltaic (TPV) energy conversion (i.e., a 5 to10x increase) is obtained by positioning the active diodesurface in extreme close proximity to the radiator (i.e., onthe order of a tenth of the wavelength of the radiation, orless). The demonstration of this proximity effect is thefirst step in the creation of a new class of MicroscaleThermoPhotoVoltaic (MTPV) devices which promise tomake the extraction of electrical energy from a widevariety of heat sources practical and to provide a newclass of compact, portable sources of electricity. More-over, MTPVs will be able to utilize thermal energy nowdiscarded as waste heat, and will enable significantincreases in the overall efficiency of many complexsystems.

The realization that the conventional blackbody radiation“law” is not valid within a wavelength of a source is notnew; that his model was limited to large separations wasappropriately pointed out by Max Planck in his originalderivation. The fact that the rate of radiative energytransfer is significantly higher than predicted by Planck’slaw when the source and sink are in such close proxim-ity, however, is a much more recent discovery, and is aphenomenon that has been studied by only a very fewindividuals interested in radiative heat transfer. Andwhile increased radiative heat transfer between closelyspaced bodies has been demonstrated experimentally,this effect remains largely unknown. In 1996, ourcollaborator, Robert DiMatteo recognized that this sameenhanced energy transfer can be exploited to advantagefor thermal-to-electrical energy conversion. However todate, there has been no experimental verification orexploration of this microscale thermophotovoltaic effect;our effort that will provide this confirmation and lay theexperimental foundation for further development ofMTPV systems.

We intend, first, to design and fabricate photovoltaiccells specifically for MTPV applications. These cells willbe narrow band gap diodes with highly planar frontsurfaces. These cells will then be placed in an apparatuswhich allows a flat, hot surface (the “source”) to bebrought from greater than 1 µm away to within 0.1 µmof the MTPV cell surface. A plot of the short circuitcurrent of the MTPV cell verses the cell-to-sourcespacing should show a many-fold increase in the currentas the spacing approaches zero. This result wouldunambiguously confirm the validity of the MTPVconcept, and mark the first milestone on the road topractical, widespread MTPV generation of electricity. Afull array of experiments are planned, in addition to thisbasic short circuit current measurement, to fully charac-terize and quantify MTPV cell operation.

240

Photonic Bandgap Structures

PersonnelG. S. Petrich, S. Fan, P. R. Villeneuve, G. Steinmeyer, K.-Y. Lim, and D. J. Ripin(L. A. Kolodziejski, J. D. Joannopoulos, E. P. Ippen, H. I. Smith)

SponsorshipNSF/MRSEC, ARO

This project represents the combined efforts of theresearch groups led by Professors John D. Joannopoulos,Leslie A. Kolodziejski, Erich P. Ippen, and Henry I.Smith. Prof. Joannopoulos’ research group designs thestructures and theoretically calculates the optical proper-ties. Prof. Kolodziejski’s group fabricates the variousdevices with embedded one- and two-dimensionalphotonic bandgap crystals using III-V compound semi-conductor technologies. Prof. Smith’s group providesthe expertise in nanoscale fabrication. Finally, Prof.Ippen’s research group optically characterizes thedevices. The complexity of the design, fabrication andcharacterization of these structures necessitates a stronginteraction between the various research groups. Inaddition, the Microsystems Technology Laboratory(MTL), the NanoStructures Laboratory (NSL) and theBuilding 13 Microelectronics Fabrication Laboratory arebeing used to fabricate the PBG structures.

A photonic crystal is a periodic dielectric structure thatprohibits the propagation of photons within a certainrange of frequencies in all directions. This forbiddenband of frequencies translates into a Photonic BandGap(PBG). A defect state can be introduced in the photonicbandgap when the selective removal or addition ofdielectric material breaks the dielectric periodicity of aphotonic crystal. This defect results in the spatiallocalization of the defect mode into a volume of approxi-mately one cubic wavelength, yielding a low modalvolume, high quality factor resonant microcavity.

• One-Dimensional Photonic Bandgap Devices:Monorail and Air-Bridge Microcavities

The one-dimensional photonic bandgap devices that arebeing investigated are referred to as monorail and air-bridge microcavity devices. In the monorail microcavity,the photonic crystal consists of a GaAs waveguide withan array of holes etched through the waveguide. Theholes are spaced periodically and by introducing addi-

tional separation between the holes in the middle of thearray, a defect is created. The photonic crystal resides ona layer of aluminum oxide (AlxOy) which has a muchlower refractive index than GaAs. In the air-bridgemicrocavity, a photonic crystal that is similar to thephotonic crystal in the monorail microcavity device, issuspended in air, resulting in a higher index contrastbetween the photonic crystal and its surroundings. Thedevices are designed to resonate at a wavelength of1550 nm. Both the monorail and air-bridge microcavitieshave waveguides coupled into and out of the devices tofacilitate the optical characterization of these devices.

The fabrication process for the devices utilizes a hetero-structure of GaAs/Al0.9Ga0.1As that was grown by gassource molecular beam epitaxy. Subsequently, a series ofplasma-enhanced chemical vapor deposition, electron-beam lithography, reactive ion etching, photolithographyand wet etch processing steps are employed to fabricatethe final devices. The samples are thinned and thencleaved to create the mirror-smooth facets at thewaveguide ends for optical coupling.

Using a continuous-wave NaCl:OH- color center laserwith an emission wavelength range of approximately1500 nm to 1670 nm as a source, the optical transmissionspectra from both the monorail and air-bridgemicrocavity devices were measured. The transmissionspectra of the monorail structures exhibited resonancesfrom 1522 nm to 1566 nm with Q ranging from 117 to 136depending on the cavity size. An increase in the defectsize resulted in a shift of the resonance peak to longerwavelengths. The transmission spectra of the air-bridgestructures exhibited resonances from 1521 nm to 1633 nmwith Q ranging from 230 to 360 depending on the cavitysize. As expected, due to the stronger modal confine-ment in the air-bridge microcavity, the cavity qualityfactors were larger for the air-bridge structures ascompared to the monorail structures.

continued

241

• Two-Dimensional Photonic BandgapLight-Emitting Diode

A two-dimensional photonic bandgap crystal inhibits thepropagation of light within a range of frequencies in anydirection within the plane of the crystal. In the devicecurrently being investigated, the two-dimensionalphotonic bandgap effect is being used to enhance theextraction efficiency of a Light Emitting Diode (LED) bycreating a two dimensional PBG crystal within the topcladding layer of an InGaP/InGaAs quantum wellstructure. The photonic crystal is designed such that theemission wavelength of the quantum well lies inside thephotonic bandgap and hence, does not couple to theguided modes within the active region. Coupling to theguided modes is a considerable source of loss in conven-tional light-emitting diodes. In the structure beingfabricated, this problem is expected to be greatly reducedand the amount of light radiated from the device isenhanced.

The 2-D PBG LED, which is designed to emit at a wave-length of 980 nm, consists of an InGaAs quantum wellthat is clad by InGaP, an AlxOy “spacer-layer” and anAlxOy/GaAs distributed Bragg reflector (DBR). Thevarious layers of the LED were first deposited on a GaAssubstrate using gas source molecular beam epitaxy. Atthis stage, the “spacer-layer” is a layer of Al0.98Ga0.02Asand the DBR consists of alternating layers of AlAs andGaAs. Using plasma enhanced chemical vapor deposi-tion, direct-write electron beam lithography and reactiveion etching, a 2-D PBG crystal is created within the upperInGaP cladding layer. After the device mesa is formedusing photolithography and reactive ion etching, the Alcontaining layers are oxidized at 435°C using steamcarried by N2. To further improve the extraction effi-ciency of the LED, the active quantum well region lies ontop of a AlxOy/GaAs DBR that is designed to reflect the980 nm light that was initially emitted towards thesubstrate. The high dielectric contrast between the GaAs

and AlxOy layers causes the DBR to have a large high-reflectivity bandwidth. By separating the active regionfrom the DBR with a low index AlxOy “spacer-layer,” alarger portion of the emitted light will be normallyincidence on the DBR, thereby increasing the amount oflight being extracted from the device.

Currently, the devices are being optically characterizedusing cathodoluminescence and scanning photolumines-cence. The intensity of the λ=980 nm light emitted froma device containing the 2-D PBG crystal is compared tothe intensity of the light emitted from a device withoutthe PBG crystal. Simulations have shown that thestructure with the PBG crystal should demonstrate atleast a five-fold increase in extraction efficiency ascompared to the same device without the presence of thephotonic crystal.

continued

242

A Two-Dimensional Photonic-Band-Gap Super-luminescentLight-Emitting Diode

PersonnelAlexi A. Erchak, S. G. Johnson, K-Y Lim, M. Mondol, D. J Rippen (Dr. S. Fan, Erich P. Ippen, John D. Joanopoulos,Leslie A. Kolodziejski, Dr. Gale Petrich, and Henry I. Smith

SponsorshipNSF, ARO

A Photonic Band Gap (PBG) is the optical analog of anelectronic band gap in a semiconductor. A periodicvariation in the dielectric constant forbids certain photonenergies within the semiconductor. Specifically, a twodimensional PBG inhibits the propagation of light withina certain range of frequencies in any direction in a plane.In this work, a two dimensional PBG is fabricated in thetop cladding layer of an InGaP/InGaAs quantum wellstructure emitting at λ = 980 nm. The photonic crystal isdesigned such that the emission wavelength lies insidethe photonic band gap and hence does not couple toguided modes within the semiconductor. Coupling tothe guided modes is a major source of loss in conven-tional light-emitting diodes. In the structure beingfabricated, this problem is greatly reduced and theamount of light radiated from the device is enhanced.

The 2-D PBG LED consists of an InGaP/InGaAs activeregion, an AlxOy spacer layer and an AlxOy/GaAsDistributed Bragg Reflector (DBR). Figure 6 shows aschematic of the structure along with a completedstructure. The 2-D photonic crystal is created by ahexagonal lattice of holes within the upper InGaPcladding layer with a hole-to-hole spacing of 315 nm, andhole diameter of 220 nm. These dimensions center thephotonic band gap around the 980 nm emission wave-length of an InGaP/InGaAs quantum well structure. Theholes do not penetrate the InGaAs quantum well region,however, to minimize surface carrier recombination.Moreover, the active quantum well region lies on top of aDBR designed to reflect the 980 nm light. The DBRconsists of alternating layers of GaAs and AlxOy toachieve high dielectric contrast between the layers whichleads to a DBR with a wide stop band. The active regionis separated from the DBR by an AlxOy buffer layer.Separation by a low index layer forces a larger portion ofthe emitted light to have normal incidence on the DBRwhich increases the amount of reflected light.

The fabrication of the 2D PBG LEDs utilizes gas-sourcemolecular beam epitaxy, direct-write electron-beamlithography, Reactive-Ion Etching (RIE) and oxidationprocesses. The device structure is grown using gas-source molecular beam epitaxy. The separation layer isinitially grown as Al.98Ga.02As and the DBR consists ofAlAs and GaAs layers. A SiO2 layer is deposited on thegrown structure using plasma enhanced chemical vapordeposition. The holes are defined in PMMA by direct-write electron-beam lithography. The electron beamwrites a square pattern in the PMMA to represent eachhole. The beam size, however, is larger than the step sizefor translating the electron beam. This leads to thedesired circular pattern following development.

The PMMA is used as a mask in transferring the hexago-nal pattern to the SiO2 layer using RIE. This is accom-plished by RIE with a CHF3 plasma using 15 secondsteps in between 1 minute cool-down steps, duringwhich the electrode is back-cooled with He gas flow.The purpose of the cool-down step is to prevent flowingof the PMMA mask. The SiO2 mask is subsequentlyused in the RIE of the holes into the upper InGaP clad-ding layer using RIE with a CH4/H2 plasma in a 1:4 gasflow ratio. The mesas are next defined using photoli-thography followed by RIE. The final step in the devicefabrication is the wet thermal oxidation of theAl.98Ga.02As separation layer and the AlAs DBR layers.Figure 7 is a scanning electron micrograph of the hexago-nal array of holes within the upper InGaP cladding layer.

continued

243

Currently, the devices are being tested using cathodolu-minescence. A high energy beam of electrons excitescarriers within the active region which subsequentlyrecombine in the quantum well. The intensity of the λ = 980 nm light emitted from the quantum well ismeasured using a Ge detector. The intensity of light

Fig. 6: a) 2-D PBG light emitting diode structure. The InGaP/InGaAs quantum well structure emits at λ = 980 nm. The activeregion is separated from the DBR by a low refractive index AlxOyseparation layer. This separation increases the amount of emitted

light reflected from the DBR. A hexagonal lattice of holes in the toplayer of the device eliminates coupling of emitted light to guidedmodes. b) SEM micrograph of 2-D PBG light emitting diode.

Fig. 7: a) Hexagonal array of holes formsa photonic crystal in the top InGaP layer

of the LED. The holes are designed to forbid theguiding of λ = 980 nm light and hence increasethe amount of light radiating from the device.

b) Cross-sectional view of 2-D PBG. The holes donot penetrate the InGaAs quantum well to avoid

surface recombination.

continued

extracted from a device containing the 2-D PBG iscompared to that of a device without the PBG. Simula-tions have shown that this structure should demonstrateat least a five-fold increase in extraction efficiency ascompared to the same device without the presence of thephotonic crystal.

244

Silicon Photonic Band Gap, Microcavity and Waveguide Structures

PersonnelT. Chen, K. Chen, D. Lim and K. Wada (L.C. Kimerling)

SponsorshipNSF/MRSEC, and NTT

In this project, we have been developing a key elementfor Si microphotonics, i.e. microcavities functioning inthe range of λ = 1.54 µm. This wavelength is compatiblewith fiber optic communication systems. Thesestructures are key devices for large scale WDM networkson Si LSI chips. In order to realize the microcavities, wehave employed two approaches: photonic bandgap andmicroring resonators. The PBG is the optical analog of asemiconductor. An electromagnetic wave travellingthrough a PBG structure encounters a large, periodicchange in the dielectric constant. This periodicity causesbands of frequencies to be disallowed for propagatingthrough the material. We have employed two methodsto fabricate photonic crystals: chemical vapor depositionand sol-gel method. First, we have fabricated thisstructure using chemical vapor deposition (CVD)processes. The periodic dielectric constant consists ofalternating films of Si and SiO2, which are most widelyused in Si-CMOS LSI processes. Er has been implantedinto a polysilicon microcavity layer. We have developeda polysilicon:Er process that yields emission of light atcomparable intensities to single crystals.

Based on coupling of photons with the µ-cavity, the Erspontaneous emission would be enhanced by the orderof quality factor Q of the cavities, which is around 100 to1000 in our case. We are currently measuring theemission enhancement by photoluminescence.

The periodic dielectrics are now being fabricated bywafer bonding and delamination. Single crystalline Silayers and thermally grown SiO2 layers were bonded infour layers to form a one-dimensional PBG. Preliminarydata show high quality PBG in excellent agreement withtheory prediction.

The sol-gel technique is an inexpensive method forproducing a wide variety of thin oxide-based filmsamenable to large-scale manufacturing. In brief, themethod involves the controlled hydrolysis and polymer-ization of an organometallic precursor followed bydensification at elevated temperatures. For thin filmapplications, a variety of techniques are available fordepositing the film. In our work, all films have beenspin-coated. We choose TiO2 (n=2.8) and SiO2 (n=1.45)for their relative ease of preparation and large refractiveindex contrast (Dn = 2.8/1.45 ª 2.0). Other materialssystems to consider include various ferroelectrics thathave high n. We have built an interference filter usingTiO2/SiO2 with a PBG centered around λ=1.5 µm andreflectivity greater than 85%. We are currently fabricat-ing the very first microcavity device based on the sol-geltechnique.

Three-dimensional PBGs operating in the near-IR andoptical regimes can be used for enhanced light emissionand waveguiding. To obtain a full, omnidirectionalbandgap, 3D PBGs must exhibit certain configurations ofrefractive index periodicity. Self-assembly of monodis-perse colloidal particles is an elegant method for produc-ing an FCC arrangement, which can lead to a fullbandgap when the index contrast between the particlesand matrix is Dn>3. With the aid of polyelectrolytesurfaces, it may be possible to pattern specific designs ona substrate on which to conduct the self-assembly. Anunderstanding of the self-assembly process is neededand techniques to obtain highly ordered particle arraysare being investigated.

continued

245

Novel microring resonator structures have also beendesigned, fabricated, and implanted with Er. Reso-nances with Qs of 250 were observed at λ = 1.55 µm withrings of 3 to 5 µm in radius) Large diameter rings (100mm radius) have recently been fabricated and implantedwith erbium. The large diameter rings are designed toresonate with the Er transition line at 1.54 µm. We arecurrently measuring the luminescence enhancementfrom a ring resonator.

1000 1500 2000 2500 3000

0.0

0.2

0.4

0.6

0.8

1.0

calculated

annealed 600 oC, 24 hrs

refl

ecta

nce R

wavelength (nm)

1500 1520 1540 1560 1580 16000.6

0.7

0.8

0.9

1.0

refl

ecta

nce R

wavelength (nm)

Er (1537 nm)

4? pairs

2 pairs

Silicon substrate

SiO2 (260 nm)

Si (110 nm)

ACTIVESi:Er (220 nm)

SiO2 (260nm)

Si (110 nm)

Fig. 8: (left) Si/SiO2 microcavity with Si:Er active layer. Thecavity has an active layer that is tuned for 1537 nm. (right)Reflectance spectrum of microcavity with resonance at 1537

nm. The cavity quality factor is ~300.

(fabricated at Polaroid-Norwood).

continued

Our ring resonators have demonstrated cavityresonances with Q’s around 300. They are primecandidate device structures for luminescenceenhancement of Er-doped silicon. Rings with 3 to 5 µmradii exhibit a Free Spectral Range (FSR) of about 20 µmbetween resonance peaks. Because the linewidth of theSi:Er main transition is less than 10 µm, a ring resonatorwith a smaller FSR is required. Ring resonators with aradius of 100 µm were computed to manifest a FSR ofless than 10 µm and were designed to have Q’s around10,000. These rings have been fabricated and doped withEr. Their spectral properties are currently underinvestigation.

Fig. 9: SEM micrograph of SiO2 spheres (diameter 0.7µm)deposited on bilayers of LPEI/PAA polyelectrolyte stripes.

The stripes are patterned via micrcontact printing, andhave a width of 4µm.

246

Fabrication of 3-D Photonic Bandgap Structures

PersonnelM. Qi (J. Joannopoulos and H. I. Smith)

SponsorshipNSF

Three-dimensional Photonic Bandgap Structures (PBG)are true optical analogs of semiconductors in a sense thatthey present omni-directional bandgaps for photons.Figure 10 illustrates the 3D structure we are attemptingto fabricate using planar fabrication techniques.Modeling indicates that it has a large, complete bandgap,and allows defects to be introduced in a controlledmanner. Figure 11 illustrates the completion of the firsttwo layers of the structure. The amorphous silicon layersare sputter deposited at room temperature, and the

Fig. 10: Depiction of a 3D photonic-bandgap (PBG) crystal to befabricated at a 790 nm period. The dark gray and light grayregions correspond to Si, with a high index (3.4), and SiO2,with low index (1.4), respectively. The SiO2 can be removedafter the structure is formed, and that will increase the idealbandgap from 14% to 21 % of the midgap frequency, which

c orresponds to a wavelength of 1.53 µm. The design enablesplanar fabrication techniques to be applied.

Fig. 11: Scanning-electron micrograph of the patternedsecond layer of a 3-D photonic-bandgap structure, showingamorphous Si on top of the first-layer structure, with groves

filled with SiO2. The second layer structure is shifted byhalf of the periods in both x and y directions.

An intentional overetch into the first layer is also observable.

patterns are aligned and defined using our VS2AScanning-Electron-Beam Lithography (SEBL). Themisalignment is shown to be within 45 nm, which issufficient to support the upper layer structures. ThenSiO2 is conformally deposited via high-density plasma-enhanced CVD at 80oC, which also providesplanarization. The etch back of the SiO2 to the level of theSi is done by a combination of controlled dry etchingwith CHF3/O2 and wet etching with diluted HF. Thisapproach should be extendable to five to seven layers,which are required to achieve the true 3-D bandgap.

247

Fabrication of an Integrated-Optical Grating-Based Matched Filter for Fiber-Optic Communications

PersonnelJ. Ferrera, M. H. Lim and T. E. Murphy, (J. N. Damask and H. I. Smith)

SponsorshipAFOSR

For future all-optical communications systems, filters areneeded for a wide variety of network functions includ-ing dispersion-compensation, wavelength multiplexing,gain flattening, and noise suppression. This projectseeks to develop the technology for building such filters,using integrated Bragg gratings. Integrated gratingsprovide a convenient way to perform filtering opera-tions, in a package that can be integrated on a chip, alongwith other electronic and optical components of thecommunications system.

Figure 12 depicts the structure of a novel integrated-optical filter we are developing in the NanostructuresLab. The filtering action is performed by the shallowcorrugation etched into the top surface of the Ge-dopedsilica waveguide. As depicted, the period of the gratingis ~530 nm, which places the spectral response in the1550-nm communications band. In order to separate thereflected filtered signal from the incident input signal, aMach-Zehnder interferometer is used. In this configura-tion, light is launched into the top waveguide and acodirectional coupler splits the signal between the upperand lower arms of the interferometer. The reflectedsignals are recombined in the coupling region andemerge in the lower port of the device.

The spectral response of this device can be tailored byadjusting the geometry of the Bragg grating. For ex-ample, to achieve a notch-type filter with high out-of-band rejection, the grating amplitude can be apodizedover the length of the device. To construct a gratingfilter for dispersion-compensation, the grating periodcould be gradually chirped over its length. In thisproject, our goal is to build a grating filter that has aspectral response that is matched to a communicationssignal of interest. The integrated Bragg-grating is an

ideal filter for such an application, because if the lengthand shape of the grating are properly selected, thereflection spectral response of the grating can be made tohave a characteristic sinc shape which is matched to thatof a binary communications signal.

Other laboratories are developing similar devices usingfiber-Bragg-gratings. In such devices, the grating isformed via a UV-induced index change in the core of anoptical fiber. Some researchers are attempting to con-struct integrated filters by applying the same UV-induced technique to a planar waveguide structure. Ourwork is unique in that the grating is formed by physi-cally etching a corrugation onto the integratedwaveguide structure. This approach allows us to: (1)tailor the shape of the grating on an almost tooth-by-tooth basis, (2) take advantage of existing nanoalignmenttechniques to achieve accurate alignment of the gratingto the waveguides without any kind of in-situ monitor-ing, and (3) circumvent the inherent limitations ofphotorefractive gratings. For these reasons, we believeour approach is not only more flexible, but also moresuitable for mass-production of highly integrated optics.

As described in our related article concerning the InPchannel-dropping filter, we have developed a flexibleand robust method of constructing integrated Bragg-grating-based devices which solves some of the criticalproblems of alignment, period selection and gratingfidelity. This approach is also applied to these silicawaveguide devices.

Our fabrication sequence uses photolithography todefine the waveguide features of the device, and acombination of interferometric lithography, e-beamlithography, and x-ray lithography to print the fine-period grating structures.

continued

248

At present, we have fabricated and measured a set ofpreliminary waveguide-coupler devices. These initialdevices, which do not have a corrugated grating on thetop surface, allow us to measure the propagation loss,fiber-coupling loss, polarization dependence andwaveguide-to-waveguide directional coupling. For

Fig. 12: Diagram of an integrated optical matched-filter. The waveguide consists of a germanium-dopedSiO2 core, 6.6 µm wide and tall, surrounded by SiO2 cladding. The 10 mm-long Bragg grating is formed by

etching a shallow, 535 nm-period grating onto the top of the waveguide before the upper cladding layer is deposited.The waveguide interferometer is designed to redirect the reflected, filtered signal to a separate output port.

these coupler devices, we measured a total insertion losslower than 1 dB, which includes propagation loss, fiber-coupling loss, and bending loss. Having characterizedthe waveguide structure, we have now turned ourattention to the Bragg grating. Our current efforts focuson investigating the overgrowth properties of silicacladding over fine-period grating structures.

continued

249

Fabrication and Design of an Integrated Channel-Dropping Filter in InP

PersonnelJ. Ferrera, T. Hastings, M. Jalal Khan, E. M. Koontz, M. H. Lim and E. Murphy(J. N. Damask, L. A. Kolodziejski, and H. I. Smith)

SponsorshipAFOSR

Wavelength-Division Multiplexing (WDM) is a strategyfor utilizing the enormous bandwidth capacity of opticalcommunications systems by multiplexing many datasignals, each placed at a different wavelength. In orderto realize the full potential of WDM, narrow-band filtersare needed to differentiate between the many wave-length channels of interest.

The Channel-Dropping Filter (CDF) is a critical compo-nent for such WDM systems. The function of the CDF isto drop (or add) one wavelength channel from a multi-channel bus. The key advantage of the CDF approachover many currently employed technologies is that onewavelength channel may be accessed without disturbingany of the remaining channels, and without ever con-verting the data into an electronic signal. Thus, it offers aflexible and extendable means of all-optical routing.

Although it is widely recognized that WDM will play anincreasingly important role in future communicationssystems, the fabrication of integrated optical componentsneeded for WDM presents some unique challenges thathave not been fully appreciated or adequately addressed.In this project, we demonstrate a novel process forfabricating integrated grating-based optical devices,which addresses some of these unique challenges.

Figure 13 depicts the Channel-Dropping Filter (CDF)which we are in the process of building in the Nano-structures Lab. The filtering action takes place in thequarter-wave-shifted Bragg gratings located above andbelow the bus waveguide. These gratings act as narrow-band resonators, which are only excited by one resonantwavelength channel.

Fig. 13: Diagram of the channel-dropping filter. As depicted in the upper portion, a multi-wavelength input signal islaunched along the bus waveguide. One resonant wavelength-channel is dropped in the upper port of the device. As shown in

the lower portion of the Figure, the waveguide is composed of InGaAsP, surrounded by an InP cladding.Note: the final top layer of InP is not depicted.

continued

250

Some of the fabrication challenges presented by thisdevice are: (1) a fine-pitch (244.4 nm) grating must beetched into the top of the much taller (1.2 µm)waveguide, and its period must be controlled to within0.1 nm. (2) The k-vector of the grating must be alignedwith the axis of the waveguide to within 1.7 milliradians;any misalignment greater than this will cause the modeto lose guidance and scatter excessively. (3) Abruptquarter-wave phase shifts must be placed at specifiedlocations in the grating. (4) The InP must be grown overthe grating without altering the square-wave shape ofthe underlying quaternary.

To meet these requirements, we have developed a “Dual-Hardmask Process” (DHP), depicted in Figure 14. ThisDHP enables us to avoid the difficult problem of litho-graphically patterning 0.1 micron features over extremetopography. We first lithographically define the gratingin a hardmask, using x-ray lithography. We then deposita second hardmask directly on top of the grating, anddefine the waveguide pattern in it using contact photoli-thography.

Once the two hardmasks are patterned, the waveguidesand gratings are formed by a sequence of dry etchingsteps. The only constraint on the choice of hardmaskmaterials is that the one on top can be removed withoutaffecting the one below. We use Ni for the waveguide-level mask and Ti for the underlying grating-level mask.Figure 14 includes a micrograph of a structure fabricatedusing this process.

The grating-level x-ray mask is written using a techniquewhich we call Spatially-Phase-Locked E-Beam Lithogra-phy (SPLEBL). In SPLEBL, interferometric lithographyis first used to generate a coherent grating pattern on thex-ray mask. Then e-beam lithography is used to writethe arbitrary patterns required by the device. During thee-beam writing, the interferometrically-generatedgrating pattern on the mask is sampled in order todetermine the absolute beam position. This techniqueensures that the e-beam writes a coherent grating, free ofstitching errors.

The waveguide mask is written at a commercial masksupplier using a MEBES tool. Both the x-ray mask andthe optical photomask have complementary alignmentmarks to enable angular alignment of the gratingk-vector with the waveguide axis.

We believe that the fabrication techniques developed forthis device will prove useful for constructing a variety ofrelated, active and passive grating-based devices.

Fig. 14: The Dual-Hardmask Process is used to pattern thefine-period gratings atop the InGaAsP waveguides. (a) First thegrating-level hardmask is patterned over the substrate, then thewaveguide-level mask is patterned over the grating-level mask.(b) The excess grating mask is stripped. (c) The waveguide featuresare dry-etched. (d) After removal of the waveguide mask, the gratingfeatures are dry-etched.

continued