Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
APPLIED PHYSlCS REVIEWS Resonant cavity enhanced photonic devices M. Selim lhiiia) Boston Universiv, Department of Electrical, Computer and SystemsEngineering, and Centerfor Photonics Research,Boston, Massachusetts02215 Samuel Strite IBM ResearchDivision, Zurich ResearchLaboratory, 8803 Riischlikon, Switzerland (Received 1 August 1994; accepted for publication 1 March 1995) We review the family of optoelectronic devices whose performance is enhanced by placing the active device structure inside a Fabry-Perot resonant microcavity. Such resonant cavity enhanced (RCE) devices benefit from the wavelength selectivity and the large increase of the resonant optical field introduced by the cavity. The increased optical field allows RCE photodetector structures to be thinner and therefore faster, while simultaneously increasing the quantum efficiency at the resonant wavelengths. Off-resonance wavelengths are rejected by the cavity making RCE photodetectors promising for low crosstalk wavelength division multiplexing (WDM) applications. RCE optical modulators require fewer quantum wells so are capable of reduced voltage operation. The spontaneous emission spectrum of RCE light emitting diodes (LED) is drastically altered, improving the spectral purity and directivity. RCE devices are also highly suitable for integrated detectors and emitters with applications as in optical logic and in communication networks. This review attempts an encyclopedic overview of RCE photonic devices and systems. Considerable attention is devoted to the theoretical formulation and calculation of important RCE device parameters. Materials criteria are outlined and the suitability of common heteroepitaxial systems for RCE devices is examined. Arguments for the improved bandwidth in RCE detectors are presented intuitively, and results from advanced numerical simulations confirming the simple model are provided. An overview of experimental results on discrete RCE photodiodes, phototransistors, modulators, and LEDs is given. Work aimed at integrated RCE devices, optical logic and WDM systems is also covered. We conclude by speculating what remains to be accomplished to implement a practical RCE WDM system. 8 1995 American Institute of Physics. TABLE OF CONTENTS I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II. Theoretical formulation of RCE parameters. . . . . . A. Formulation of the quantum efficiency for RCE photodetectors. . . . . . . . . . . . . . . . . . . . . . . B. Standing wave effect. . . . . . . . . . . . . . . . . . . . . . C. Exact numerical calculation of RCE quantum efficiency................................ D. Resonant cavity enhancement of quantum efficiency................................ E. Near unity quantum efficiency photodetection. . . . . . . . . . . . . . . . . . . . . . . . . . . . F. Wavelength selectivity of RCE photodetection. . . . . . . . . . . . . . . . . . . . . . . . . . . . G. Formulation of crosstalk. . . . . . . . . . . . . . . . . . . H. Angle dependence of quantum efficiency.. . . . . I. Formulation of inhibited spontaneous emission................................ III. Design criteria for RCE photodetectors. . . . . . . . . A. Material requirements for RCE photodetection. . . . . . . . . . . . . . . . . . . . . . . . . . . . “Electronic mail: [email protected] 608 608 608 610 611 612 613 614 614 615 616 616 616 1 B. Material system combinations for RCE photodetection. ........................... 1. AlGaAs/GaAs/InGaAs. ................. 2. InP/InGaAs/InAlAs. ................... 3. AlAsfGaAslGe. ....................... ; ..... 4. SilSiGe ........................ 5. Si/AlP/GaP. .......................... . High speed properties of RCE photodiodes. ..... A. High speed capabilities of conventional photodetectors. ........................... B. High speed performance of RCE photodetectors. ........................... C. Impulse response of RCE and conventional photodiodes .............................. D. Simulated high speed response of RCE photodetectors. ........................... V. Experimental results on discrete RCE devices ..... A. Schottky photodiodes. ..................... B. Avalanche photodiodes. .................... C. P-I-N photodiodes. ....................... D. Metal-semiconductor-metal photodetectors ..... E. Heterojunction phototransistors. ............. F. Bipolar inversion channel field effect .-......... phototransistors. ................ 617 617 618 618 618 618 619 619 620 621 621 624 624 624 625 625 626 627 607 0021-8979/95/78(2)/607/33/$6.00 8 1995 American Institute of Physics J. Appl. Phys. 78 (2), 15 July 1995 1 a Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp G. Wavelength demultiplexing with RCE photodiodes and phototransistors. ............ H. Optical modulators. ....................... I. Light emitting diodes. ..................... J. Optical amplifiers. ........................ VI. RCE optical logic and systems. ............... k Integration of detectors with LEDs. .......... B. Vertical cavity lasers integrated with photodetectors. ........................... C. Wavelength selective optical logic. .......... VII. Towards a practical RCE WDM system. ....... VIII. Conclusions. ............................. 627 628 629 630 631 631 632 633 634 637 I. INTRODUCTION Over the past five years a new family of optoelectronic devices has emerged whose performance is enhanced by placing the active device structure inside a Fabry-Perot resonant microcavity. In such structures, the device functions largely as before, but is subject to the effects of the cavity, mainly wavelength selectivity and a large enhancement of the resonant optical field. The increased optical field allows photodetectors to be made thinner and therefore faster, while simultaneously increasing the quantum efficiency at the resonant wavelengths. Since off-resonance wavelengths are rejected by the cavity, the photodetectors have both wavelength selectivity and high speed response making them ideal for wavelength division multiplexing (WDM) applications. Optical modulators situated in a resonant cavity require fewer quantum wells to absorb the same fraction of the incident light, and can therefore operate at lower voltages. In the case of emitters, the cavity modifies the spontaneous emission of light emitting diodes (LED) improving their spectral purity and directivity. For these reasons and others, devices of this genre are referred to as Resonant Cavity Enhanced (RCE). While the fundamental physics of RCE has been known for nearly 100 years,“’ and the first observation of RCE in semiconductor devices occurred nearly twenty years ago,3 recent developments have stimulated research activity in this area. Optical communication networks demand higher bandwidth photo& devices and devices capable of WDM.4 Molecular beam epitaxy (MBE) provides researchers with a growth technique capable of producing epitaxial Fabry-Perot microcavities within exacting specifications enabling the realization of RCE devices.5 This review attempts to provide a broad overview of RCE devices and systems, including theoretical, simulation and experimental results. All classes of RCE optoelectronic devices studied to date are covered. However, resonant devices for which cavity resonance is necessary for device operation, such as vertical cavity surface emitting lasers (VCSEL), are not included. In Section II, we assemble a largely analytical formulation of the RCE effect which is extended to numerical modeling when necessary. Section III lays out fundamental materials criteria for RCE devices and examines the suitability of common heteroepitaxial systems. In Section IV, a simple model compares the bandwidth of RCE and conventional photodetectors and corroborating results from simulations of high speed RCE photodiode response are overviewed. Section V is devoted to 608 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 the body of experimental work on discrete RCE photonic devices, including photodetectors, phototransistors, modulators, and LEDs. Section VI covers integrated RCE devices and their implementations for optical logic and WDM. In Section VII, we speculate over what remains to be accomplished towards implementing a practical RCE WDM system. II. THEORETICAL PARAMETERS FORMULATION OF RCE Important for the design and understanding of RCE based devices is the development of an analytical mathematical model describing the behavior of an active absorption (or gain) region inside a Fabry-Perot resonator. This challenge has been taken up by a number of workers3,6-‘3 whose efforts contribute to this section. Below, we present a formulation which allows the calculation of the quantum efficiency, finesse, and free spectral range of an .arbitrary RCE detector structure. Special consideration is given to solutions which yield near unity quantum efficiencies. We also consider the dependence of RCE detector properties on the placement of the active layer within the cavity and the angle of incidence of the detected radiation. The derivation for the lossy cavity (detection) can also be applied to a gain cavity without lasing (emitters) by replacing the absorption coefficient with the gain coefficient. Finally, a formulation for spontaneous emission within an optical cavity suitable for LEDs is presented. A. Formulation photodetectors of the quantum efficiency for RCE One of the desired features for photodetectors is high quantum efficiency. The quantum efficiency 7 of a photodetector is defined as the probability that a single photon incident on the device generates an electron hole pair which contributes to the detector current.t4 When many photons are present, which almost always is the case, 17is defined as the current flux/photon flux ratio. Absorption z=o Region / z=L G Incident Light - DBRt--L-( k--L2-‘Ge cl DBR Ri=r? Reflector ,Rs=rg FIG. 1. Analysis model of a RCE photodetector. The active region of thickness d is a small bandgap semiconductor. The top and bottom distributed Bragg reflection mirrors consist of alternating layers of non-absorbing larger bandgap materials. Appl. Phys. Rev.: M. Selim finlij and S. Strite 9 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp For the derivation, a generalized RCE photodetector structure is chosen (Fig. 1). In practice, the end mirrors are made of quarter-wave stacks of GaAslAlAs, InGaAsfInAlAs or any of a number of other semiconductor or insulator materials which are capable of providing the desired refractive index contrast. In simplified designs, the top mirror can be the native semiconductor to air interface which, due to the large refractive index difference at the boundary, provides a reflectivity of approximately 30%. The active layer, where absorption occurs, is placed between the two mirror structures and is defined by its thickness d and absorption coefficient a. The separations between the active layer and the top and the bottom mirrors respectively are indicated by L1 and Lz. The absorption coefficient outside the active region is denoted (Lag. The field reflection coefficients of the top and the bottom mirrors are rre-i@r and rae-j@z, respectively, where fir and & denote phase shifts due to light penetration into the mirrors. The transmitted component of the incident light wave electrical field component (Ei) equals tl. Ei . In the cavity, the forward traveling field Ef is composed of this transmitted field and the feedback as a result of internal reflections at the mirrors. Therefore, for a propagation constant p, the forward traveling wave Ef at z=O (Fig. 1) can be obtained through a self-consistent consideration, i.e., Ef is the sum of the transmitted field and the feedback after a round trip in the cavity: -ad-cu,,(Ll+L~!~-j(2PL+91+~2i)E Ef=tlEf+rlrze -ad-a,,(L1+L2)e-j(2pL+@,+~2) Eie t2) The backward traveling wave (i.e., Eb at z = L) is found by propagating Ef (2) through the detector region: Ehzr,e-“dt2e-i”ex’2) - ~ (L,+Lz!~ -j@L+&i)+ (3) The optical power inside the resonant cavity is given by: Pse& )ES12 (~=f or b), (4) (5) Under the assumption that all the photogenerated carriers contribute to the detector current, 17 is the ratio of the absorbed power to the incident optical power, i.e., J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 0.90 0.92 Wavelength 0.94 0.96 A Old FIG. 2. Wavelength dependence of 17for RCE detectors having various top mirror reflectivities for fixed L=2 pm, Rz=0.9 and &=O.l (after Ref. 8). 7~ PI I Pi. Hence: (e-‘ae.&l+e-%.&2~2e@) 77= l-2$&-@ cos(2pL-f JI1+ qb2)fR1R2e -2”% I X(1-Rt)(l-Pd), (6) act= (rx& + a,,L2 + ad)lL. In a practical detector design, the material around the active layer absorbs negligibly ( a,,-5 - 10 cm-‘) compared to the active layer (aa IO4 cm-‘), so acx in (6) can be neglected, except when active layer is extremely thin. Thus, TJcan be written: (1 +R2eWed) ‘= i 1 -2aeWad X(1-Rr)(l-eAad). where r/O and n are the vacuum characteristic impedance of electromagnetic waves and the refractive index of the detector material, respectively. In this case, the light power absorbed in the active layer (Pl) (neglecting the standing wave effect to be considered below) can be obtained from the incident power Pi in the form: --e-ad) (1 -rT)(e -ru,Ll+r2e-“exL~e-neL>(l 2 = 1-2~-,r~a-~~~ cos(2~L+$1+~~)+(rlrz)2e-2”cL XPi* FSR F ?I (1) tl 1 - r1r2e - where LY, is: f- Solving for Ef gives: Ef= ud = 0.1 , R2 = 0.9 0.6 COS(~~L+(CI~~~~)+R~R~~ -2nd . C8> Since the propagation constant p (/3=2nr/Xo, where X0 is the vacuum wavelength and y1 is the refractive index) has a wavelength dependence, 17is a periodic function of the inverse wavelength. This is easily seen in Fig. 2, which illustrates the calculated wavelength dependency of 0. The three curves correspond to the cases of RI =0.9, 0.3, and 0.05 while Rz=0.9, crd=O.l, and L=2 ,um are fixed. 7 is enhanced periodically at the resonant wavelengths which occur when 2/?L+ fir + G2=2 mrr (m= 1,2,3...). The spacing of the cavity modes, i.e., resonant wavelengths, is defined as the free spectral range FSR.” For the given cavity parameters, the FSR is around 500 nm. On the right hand side of (8), the term inside the braces represents the cavity enhancement effect. This term becomes + unity when R2=0 giving 77 for a conventional detector. The flat dotted line in Fig. 2 indicates the maximum r,~attainable for conventional photodetectors for the same active layer thickness (ad = 0.1). The contrast between the two types of detectors is clear. Conventional photodetectors provide Y roughly constant ‘17across a broad wavelength range while .%;.? . ;” Appl. Phys. Rev.: M. Selim klti and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp RCE photodetectors can be designed to have significantly improved 77at specific wavelengths. The peak p at the resonant wavelengths can be derived by imposing the resonant condition in (8) (1 fR$+) v nmx= (l-me-adj2 (l-R1)(l--e-nd). (9) i I In the limit of a thin active layer ad+i 1, (9) reduces to (l-I-Rz(l-cud)) (1 -R&d. (10) I Proper selection of the device parameters allow for RCE detectors having nearly unity 77to be designed. Smx’ (1 - &-&l B. Standing -ad))’ wave effect In the above derivation of (8), the spatial distribution of the optical field inside the cavity was neglected. A spatial distribution arises from the standing wave formed by the two counter propagating waves [Eqs. (2) and (3)]. It follows that g, which was derived from the power absorbed in the active region, is a function of the placement of the active region in the optical field. We refer to this as the standing wave effect (SWE). When detectors with thick active layers which span several periods of the standing wave are considered, the SWF can be neglected. For very thin active layers, which are necessary for strained layer absorbers, the SWE must be considered. The SWE is conveniently included in the formulation of absorption coefficient, i.e., 37 as an effective aeff = S WE. a, which is either enhanced or decreased by the placement of the active region. The effective absorption coefficient aeff is the normalized integral of the product of (Y and the field intensity across the absorption region. Using a perturbation analysis of Maxwell’s equations, including the loss factors and assuming uniformity along the transverse direction, the effective absorption coefficient can be expressed as? a@ff= l/dJ&(z)lE/2(z,X)dz UhJ:‘21E12(z,X)dz (11) ’ where X=X0/n, E(z,X) is the total electrical field in the cavity at a given wavelength and the denominator is the average of the standing wave. Taking a to be negligjble outside the active region and constant within lldJ;:+dlE/2(z)dz SWE= y zz 21XJ-;R\E12(z)dz ‘.? .. I i -. . (12) To apply (12), the standing wave distribution must be calculated. Several assumptions simplify the analysis. The cavity is assumed to be lossless; reasonable for a thin active layer which absorbs only a small fraction of the total power density. The change in p in the active layer is neglected since the real permittivity is much larger than the imaginary (absorptive) part for low loss dielectrics. Finally, reflections from the active region interfaces are also neglected. This approximation is valid when heterostructures having subtle permittivity changes are considered. An exact numerical -> . . . 20 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 FIG. 3. Optical field distribution in a RCE detector as a function of wavelength and position. Top and bottom mirrors are 5 and 15 periods C&As/ AlAs, respectively, with a center wavelength of 0.9 pm (after Ref. 11). method for the more general case is described in the following section. The forward (Ef) and backward (Eb) components of the standing wave are given by (1) and (2). The total electric field E and intensity 1El” are: E= Ef(0)exp( -jPz) t Eb(L)exp[jp!z- L)] (13) and IE~2=~~.~(0)~2+~E~(L>~2+2~e {-J$Yz)Edz)~. (14) Substituting (1) and (2) into (14) and assuming a=O: P-4 IEiZ=[Ilvrlr2&3L+uil+v4 (15) In (15), the term in braces represents the enhancement effect of the resonant cavity. Note that under the resonance condition with an ideal bottom mirror ( r2 = 1) this term reduces to [ 1 - r$l 1 - r i / 2 > 1. It increases rapidly with r , representing the resonant cavity field enhancement. The position dependence represents the influence of the standing wave. Figure 3 shows the wavelength dependent cavity field distribution calculated from (15) for a GaAs based RCE photodetector. Substituting (15) into (12), we obtain the dependence of the SWE on the cavity parameters:” 27-Z SWE-l+pdtl+r;)[sin@d cos(2/?L2+pd+qk2)]. 06) The SWE is an explicit function of both the magnitude r2 and phase $2 of the bottom mirror reflectivity, and it is implicitly dependent on the top mirror phase +i through the resonance condition (2pL-k qil + t,b2=2m7rj. The SWE varies with the h not only through /I, but also due to the strong wavelength dependence of the phase of the quarter wave stack mirrors ($i , 4V2).The wavelength dependencies of the mirror reflectivity and phase are demonstrated in Fig. 4 for several different mirror retlectivities. The reflectivity maxiAppl. Phys. Rev.: M. Selim cnlti and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp (4 0.8 ffi SO.6 2 4 g 0.4 d !s I\! ,s.t’,,,,‘,,.,“,.“.,,,“.‘, 8250 8500 8750 9000 9250 Wavelength (A) 9500 9750 (b) 1.57 region center is at the standing wave maximum and minimum, respectively. These extremes of the SWE are shown in Fig. 5(b) which highlights the decreasing importance-of the SWE for thicker absorption regions. The above approximation (16) can be used to calculate the photosensitivity of RCE detectors using (17). Excellent agreement between experiment and theory was realized for a structure having a small mole fraction In,Ga,-,As (zKO.1) absorbing layer in a GaAs cavity.8Y’7 This success was due to the fact that the imaginary part of the propagation constant p was much larger than the real part (cz), making the overall change in the refractive index negligibly small despite drastic changes in the absorption spectrum. However, for a more detailed analysis of the detector response, such as crosstalk calculations, it is desirable to have a method which accounts for any refractive index change in the active region and the resulting reflections at these interfaces. C. Exact numerical efficiency z ;I calculation of RCE quantum To more accurately evaluate the spectral dependence of the total cavity reflection, a simple transmission line model can be applied” in which wavelength dependent impedance is calculated at every interface, beginning at the substrate and a, 0.00 2 2 -1.57 -3.14’,‘.““‘,‘,“,‘,,,~‘,,~.1,,,,1 8250 8500 8750 9000 Wavelength 9250 (A) 9500 9750 FIG. 4. (a) Magnitude of mirror reflectivity for a GaAs/AlAs (643 A/776 A) quarter wave stack mirror for 10 (solid), 15 (dashed), and 20 (dottedj periods. ibj Phase of the same mirrors as a function of wavelength (after Ref. 11). mum increases and sharpens with additional periods, saturating near unity above fifteen AlAsJGaAs periods. The phase shows a strong wavelength dependence which sharpens as the number of periods increase. Knowledge and control of the phase behavior is particularly important for proper positioning of very thin absorbing layers in high 77 photodetectors. Figure 5(a) shows the wavelength dependence of the SWE for various active layer thicknesses d. At d=XO/4n (solid line), the SWE ranges between 0.35 and 1.7, resulting in drastic variations in the device photosensitivity at different wavelengths. For d=XO/2n (dotted line), the SWE is weak, again due to the active region spanning an entire half period. For thicker active regions, the maximum deviation occurs at d=3XO/4n (dashed line) where the SWE ranges from 0.8 to 1.2, giving a photosensitivity variation at resonance below 210% for n=lO” cm-‘. For an ideal bottom mirror (r2 = 1,q2= 0)) a real top mirror reflectivity ( $t = O), and L r = L2 (centered active region), the SWE reduces to a simple form:t6 sin /Sd sin ,Bd SWE= 1 + cos(mw) pd = 12 W ’ where + and - correspond to the cases when the active J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 0.0I 8500 10 -= I 9250 9000 8750 Wavelength X,, (A) 9500 10 -’ 1 Normalized active layer thickness 10 FIG. 5. (a) The SWE as a function of wavelength for three different active layer thicknesses: d=650 %, (solid), d=1300 %, (dotted), and d=2000 8, (dashed) for a cavity with 20 pairs GaAs/AlAs bottom mirror and native GaAs surface as the top mirror (L,=L,=2 pm). (b) Dependence of the SWE on the active layer thickness. The extremes of the SWET are shown (after Ref. 11). Appl. Phys. Rev.: M. Selim cnlti and S. Strite 611 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp ending at the epilayer surface. Every semiconductor layer is considered as a transmission line segment in which the characteristic impedance is given as:t9 I Vi= d P ~Et -jE” 3 8 ’ where E’ and E” are the real and imaginary parts of the dielectric constant. For a low loss dielectric (E’ % e”), the propagation constant y can be expressed as:t9 I, a y=-+jp-jw 2 i )I 1 d’ 2 pur’ 1-js+- 8 E’ J-7 . Equating the imaginary parts yields p-&zj 1+;( ;)‘I. (21) For a lossless dielectric (a = 0, E is real), the characteristic impedance is real and the propagation constant is purely imaginary. For practical detectors, a is small enough at the relevant wavelengths to satisfy these assumptions. The physical parameters, such as E and a, are well known for common semiconductors and can be found for all pertinent wavelengths.20 The substrate (approximated as infinitely thick) impedance (Z+t) is first computed from (18). The impedance is then transformed for each semiconductor layer in accordance to its thickness Zi and characteristic impedance Zai by’8,‘9: i=l to N (22) until the surface is reached. N is the total number of contrasting layers including those in the quarterwave mirror stacks. Although this formulation can yield analytical results, we preferred simple computational methods allowing for analysis for a variety of wavelengths and device parameters. Once the input impedance (Zi,) of the entire detector structure is known, the overall complex reflection coefficient can be evaluated as:l’ (23) Q~uantumefficiency, 7, is approximated as the difference between the reelection coefficients for a lossless cavity (assuming (Y=O) and a cavity with a finite absorption coefficient in the active layer. This approximation is valid when the bottom mirror has high reflectivity, a requisite for functional RCE designs. This method automatically accounts for the position of the active layer with respect to the standing wave. Figure 6 compares the SWE approximation and the more accurate calculations by the above described transmission line analogy for d=h0/4n. The low mole fraction active region was assumed to have the same refractive index (n=3.5, GaAs) as the spacer regions (~5, ,L2= 2 pm) and a constant 612 5 2 0.2 3 CY 0.0 i2@ Zi; 1+ Zni tanh YZi Zaf+ Zi- t tanh YZi ’ 4 (19) Equating the real parts of (19) leads to Zi’ Zoi 0.4 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 8500 8750 9000 Wavelength 9250 9500 Xo (A) FIG. 6. Comparison of 17 calculated by different methods. Dashed line shows the case in which the SWE is neglected. Solid lines (nearly coincident) compare the first order SWB approximation (A) with the numerical method through transmission line analogy (B) (after Ref. 11). (a= lo4 cm-‘, InCaAs) for the entire spectrum. A 20 period Al.As/GaAs quarter-wave stack and the GaAs/air interface comprised the bottom and top mirrors. The 7’s calculated by the two methods are nearly identical over the entire wavelength range as a result of the relatively low contrast InGaAs active layer. The importance of the SWE at small d is illustrated by the deviation of the dashed line [plotted from Eq. (S)], which does not take the SWE into account. D. Resonant cavity enhancement efficiency of quantum The origin of the drastic enhancement in v is the greatly increased amplitude of the electric field inside a high Q resonant cavity which causes more energy to be absorbed in the active region. An equivalent interpretation is that an individual photon is multiply reflected at the mirrors and therefore makes many passes through the absorption region. The field enhancement is apparent when one calculates the increase in the electric field intensity (i.e., the optical power) in the cavity (]Ef]2+]Eb]2)/]Ei]2. The result is identical to the right hand term in the expression for ~7 (8), except for ( I- e - nd). Therefore, the enhancement of 7, with respect to an idealized conventional detector with equal active layer thickness, is simply the increase in the internal optical power. Figure 7 presents the calculated internal power enhancement factor at resonance for varying mirror refiectivities as a function of c&. For small ad, i.e., a low loss cavity, the factor exceeds 10. At higher mirror reflectivities, the enhancement can exceed 50, yielding a correspondingly high ~7despite the thin active layer. As ad increasespthe enhancement falls off rapidly, especially in high Q cavities (e.g., RI = 0.9 and R2 = 0.99)) eventually becoming less than unity. This is because most of the light is absorbed in the thicker active layer before it reaches the bottom mirror negating the RCE feedback mechanism. Appl. Phys. Rev.: M. Selim finlij and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp (Fig. 7), 17 is maximized at a certain ad before decreasing asymptotically to the conventional RI limited value v~+,= 1 - R, . The maximum occurs at - -j : Ra = 0.99 R’=R2e-2’ud 1 0.5 0 0.1 Normalized 0.2 0.3 Absorption 0.4 0.5 Coefficient crd 0.6 FIG. 7. Internal power enhancement factor in resonant cavities with various end minor refkctivities as a function of the normalized absorption coefficient (after Ref. 8). Multiplying the internal optical power enhancement factor by (1 -E -“d) yields 7 (8) for an RCE photodetector. Figure 8 plots 7 at resonance as a function of ad for R2=0.9 as a function of RI. The conventional case (Rz=O) is given by a dotted-dashed line and 7 for R1=0.7, R2=0.99 is shown by a dashed line. For a native semiconductor surface (R, =0.3), the R2=0 and 0.9 curves contrast the conventional and RCE cases. RCE improves 77a factor of 6.5 for a 0.1 pm thick absorption layer ( UY=lo4 cm- *) . Quantum efficiency can be further enhanced by more reflective mirrors. The R2=0.99, R1=0.7 curve reaches a maximum 7 in excess of 98%. Because the absorption term ( 1 -e Iad) increases with ad, while the internal power enhancement factor decreases (24) as obtained by differentiating (9) with respect to R, . Figure 9 plots vrnax as a function of ad for various R, values. Conventional photodetectors (R, =0) exceed 90% only for active layers in excess of 2 pm. In contrast, RCE photodetectors having R2=0.99 exceed 90% for very thin active layers (0.05 pm thickness at (Y= 10” cm-‘). A 2 JUIYI thick conventional detector will be bandwidth limited by carrier transit times. Increased recombination also erodes 17 in such thick detectors. On the other hand, because of their thin absorption regions, RCE photodetectors offer increased speed along with enhanced v. When ad is very small, o+, , neglected up to this point, becomes significant. The dashed line in Fig. 9 shows 7 for aex=5 cm-’ and LI = L2= 1 pm. The effect of external absorption can only by seen for very large reflectivities (R2=0.99) at small ad. For practical ad values, the discrepancy is negligibly small and LK~, can continue to be neglected. E. Near unity quantum efficiency photodetection Many applications, such as background limited astronomical observations” and high sensitivity interferometry for squeezed-light measurements,‘l require photodetectors with 7 approaching unity. Farhoomand and McMurray” suggested that RCE photodetectors were capable of providing near unity 77 detection at the expense of wide spectral photosensitivity. Kishino et al.* presented a detailed analysis of near unity detection. The critical requirements were 8r 0.8 r" 0.8 c E! '3 0.6 Ff? E t R2 =\".gg ,' *" 1 El .Ei 0.2 3 2 R1 = 0.3 , Ra = 0 I1II.II 0.02 0.1 Normalized Absorption 0.3 Coefficient 1 3 cud FIG. 8. Calculated 77as a function of the normalized absorption coefficient. Solid lines show 7 at the resonance mode peaks for R2=0.9, dashed lines at R2=0.99 and RI=0.7, and dotted-dashed fines the conventional detector case (Rz=O, RI =0.3) (after Ref. 8). J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 0.4 0.0 10-3 10-z 10-l. 1 LO Normalized absorption coefficient, ord FIG. 9. Maximum 7 (R1=R2ezcd)as a function of normalized absorption coefficient. Dashed line includes parasitic cavity absorption (~4~75 cm-‘) (after Ref. 8). Appl. Phys. Rev.: M. Selim blti and S. Strite 613 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp 2.5 3 3 2.0 f uk3 1.5 'S e 5! n * % 1.0 1 0.5 25001 z cl 0.2 0.4 Top Mirror Reflectivity FIG. 10. Constant 77contours as a function of normalized absorption coefficient and top mirror reikctivity for fixed R2=0.99. 7 in excess of 0.99 (meshed region) can be obtained over a fairly large parameter space (after Ref. 8). shown to be very high bottom mirror reflectivity and a moderate absorption layer thickness. To attain the highest 7, near unity Rz is desirable. Mirrors with near unity reflection coefficients require a large number of periods and become impractical to grow. The use of metal films as a unity reflection mirror has been proposed.” However, at optical frequencies, even noble metals such as Au are not ideal reflectors. The reflectivity of metals on different materials can be evaluated using their wavelength dependent optical constants.22Y”3For example, Au is 98% reflective in vacuum at wavelengths around 1 pm which reduces to about 94% on GaAs. Therefore, metal films alone are not ideal mirrors. For near unity v detectors, a hybrid approach of a quarter-wave stack and a metal film is necessary for the bottom mirror. For example, a 6 period AlAs/GaAs mirror with a deposited Au film will have a reflectivity in excess of 99% at ho= 1 pm. Figure 10 shows the constant 77contours as a function of RI and ad for R2=0.99. The meshed region defines the parameter space of RI and ad which yields 77BO.99. 4t RI =0.2, a range of active layer thicknesses from d = 0.7- 0.95 pm at LY= lo4 cm’ is available. Similarly, high 77can be obtained for a range of RI values. For example, at cud=0.7, the ~>0.99 region extends from RI =0.2 to R 1=0.3. In contrast, a conventional photodetector with a perfect anti-reflection (AR) coating (R 1=0) requires an absorption layer in excess of 5 ,um to absorb 99% of the incoming light. Such a detector design with very thick absorption region would suffer from secondary recombination of photogenerated carriers which, in turn, limits the maximum attainable quantum efficiency. F. Wavelength selectivity of RCE photodetection At the off-resonance wavelengths, i.e., 2pL + $1 + qb2 =(2m+l)r (m=1,2,3...), the cavity field amplitude decreases due to the destructive interference of the forward and 614 SO2508500 8750 9000 9250 Q500 9750 Wavelength (A) 0.6 FL1 I FIG. 11. Effective optical length of GaAs/AlAs (643 A1776 A) quarterwave mirrors designed for a 0.9 pm center wavelength calculated from the mirror phase [Fig. 4(b)] for 10 (solid line), 15 (dashed line), and 20 (dotted line) mirror periods. backward traveling waves, resulting in suppressed 117 (Fig. 2). The wavelength spacing between neighboring resonant peaks, called the free spectral range (FSR), can be expressed as. 15- x2 FSR= 125) 2neff(L+ k?ff,l f L,ff,2) ’ where ncff is the effective refractive index, and L,ff,i are the effective optical lengths of the mirrors. Generally, L,ff,l can be given byz4: L eff,i=y l a+i ap (i= 1,2). For quarter-wave stack mirrors, Lcff reflects the strong wavelength dependence of the mirror phase. Fig. 11 shows the wavelength dependent Lcff of the mirror from Fig. 4. Experimentally, a typical Leff value for two GaAslAlGaAs mirrors is 0.7 pm.= The finesse F, the ratio of the FSR to the wavelength F;wHM A~,,T,. measures the wavelength selectivity of the cavity. F can be derived from (8) as: T(R~R,)“~ FSR F=G = epndi2 l--Ji?&emad . (27) Ecluation (27) clearly indicates that F increases at higher mirror reflectivity and thin absorption regions. For example, when cud=O.Ol and cu,,=O, an F as large as 100 can be expected for R2=0.99 and RI =0.985. G. Formulation of crosstalk The narrow spectral range of RCE devices can be viewed as a potential disadvantage. Fortunately, it is quite easy to vary the spectral response of RCE detectors by altering the effective cavity length. This can be accomplished through a variety of means and permits the fabrication of arrays from a single epitaxial layer which combine the high sensitivity of RCE with a broad spectral range. ----.-- J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 Appl. Phys. Rev.: M. Selim cnlti and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp 0 0.2 0.4 0.6 0.6 1 Top Mirror Reflectivity Ri FIG. 12. Crosstalk attenuation and v of the demultiplexer for &=O.l as a function of top mirror retlectivity for R?=0.99 (solid) and 0.9 (dashed) [after Ref. 8). In RCE structures in which the top mirror is a simple semiconductor/air interface, the effective cavity length can be altered by recessing the surface using conventional wet chemical etching.a6 Surface recessing can be followed by the deposition of a dielectric or metallic top mirror if a high finesse cavity is desired. It is also possible to grow a several period quarter wave top mirror beneath a final surface recessing layer. After etching and the deposition of a high reflectivity metal, a very high reflectivity hybrid top mirror can be realized.” A critical parameter for any ‘wavelength demultiplexer is its crosstalk, the measure of the spectral sensitivity overlap between adjacent detector channels. Using the wavelength dependent expression for 17 (8), the interchannel crosstalk attenuation (C) can be estimated as a function of F (26) and the number of channels N: c=20 log (NB3). (28) 50 s 40 u & 30 From (28), it is clear that a very low crosstalk is obtained for high F cavities, i.e., large mirror reflectivities and thin absorption regions. Figure 12 shows the dependence of C and ~7 on RI at a fixed ad=O.l. Crosstalk attenuation rises monotonically with increasing RI , reflecting the higher F. However, v is optimized by the condition (R 1= R, emzad) and degrades rapidly once RI exceeds that value. An optimized WDM design must balance the requirement of high C with high 7. A suitable measure of WDM detector performance reelecting the above tradeoff is the product q. C. Figure 13 shows the product as a function of ad at optimized 7 (R,=R, e -2ryd) in a 4 channel demultiplexer. Peak performance is achieved near ad=0.07 (less as R2 increases). As ad increases, performance suffers due to the lower F (28). At smaller ad, the reduced performance level reflects the decrease of 77. Wavelength selectivity of a RCE photodetector is similar to a device comprised of a Fabry-Perot filter placed in front of a conventional detector. In this case, the filter and detector can be designed and optimized separately. Since the filter can be fabricated out of transparent materials very high finesse (low crosstalk) can be achieved. However, the design of the overall structure will be more complicated and a different filter for each wavelength will be required. Besides, for the filter detector combination, the overall efficiency is simply the product of the. filter transmission coefficient and the quantum efficiency of the conventional detector. The advantages of the RCE approach are the simplified design and the drastic quantum efficiency enhancement. While the simple design allows for monolithic fabrication of WDM detector arrays,” quantum efficiency enhancement enables the use of thin absorbers, and thus high-speed devices. H. Angle dependence ;(:)‘X. (29) 2. \ I‘ , Returning to normal incidence (8) and inserting the angular dependence of 7 as a function of F (27), h, L, neff, and n gives 20 5 2 efficiency Earlier, we restricted our derivation of 77to normal incidence. This can easily be expanded to the general case by replacing ,3 with /3 cos 8 in the equations. Here, 6, is related to the angle of incidence 0, by Snell’s law, n sin B=sin ei. The phases r++i(i = 1,2) are functions of p and are also angle dependent. The resonant wavelength is also shifted, since the optical path lengths increase. The new resonance wavelength condition becomes 2pL cos 0+r,$(0)+&(s>=2m~. At small Bi, the wavelength shift AX can be approximated as: [AX/- *g “0 ot quantum 10 A?- Q-v- 1%?ffw-&-f,l+~ 77 0 0 0.1 Normalized 0.2 Absorption 0.3 Coefficient 0.4 crd FIG. 13. Dependence of 77.C product versus normalized absorption coefEcient for four-channel demultiplexing (after Ref. 8). J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 v 1 eff.2) A pi\‘]’ _ \nl 1 (30) Equation (30) indicates that 7 is most sensitive to the angle of incidence in a high F cavity at shorter A. This be havior may be very useful in selected applications (such as high response angle photodetection). In the general case, the Appl. Phys. Rev.: M. Selim fit& and S. Strite 615 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp g 9 a” I$4 j$ 3 ti 3 length by as much as a factor of ten. As a result of RCE, in an optimized structure as much as 90% of the spontaneous emission can be coupled into the resonant wavelength. For LEDs, ISE serves to significantly improve the spectral purity and directivity of the emission. While conventional LEDs have their spectral linewidths limited by thermal effects to - 1.8kT, RCE LED linewidths are determined solely by the Q of the optical cavity35: -10 AX h-1/Q=-20 0 5 10 15 Incident Light Angle 6, (deg) FIG. 14. Degradation of 7 for off-angle incidence at X = 1.0 and 1.5 pm for F=lO and 20 (after Ref. 8). angular dependence of the detector performance is not so large that it presents significant alignment difficulties (Fig. 14). Below 3”, 17degrades by less than 1% even for a very high F cavity at 1 ,um. For lower F and longer A, the angular constraints can be relaxed while mainta.ining the benefits of RCE. I. Formulation of inhibited spontaneous emission Above derivation of the quantum efficiency and its dependence on cavity parameters was carried out for detector structures. A similar formulation for light emitters can also be considered simply by replacing the absorption coefficient with a gain coefficient. Such a formulation is valid as long as the optical emission properties of the device/material is not altered by the presence of the cavity. Additional considerations are required for systems where the cavity influences the emission characteristics, such as onset of lasing in cavities. Spontaneous emission characteristics are also effected by the cavity. Purce112* realized many years ago that the spontaneous emission spectrum of a system could be altered by modifying its external environment. It has since been shown that the spontaneous emission of trapped electronsZ9 and atoms3’ can be inhibited if the system is placed inside a microwave cavity which prohibits electromagnetic modes at the transition frequency. Yablonovitch31 realized that this concept could be profitably applied to semiconductor light emitters. He showed that if an active region was placed in a Fabry-Perot cavity, the spontaneous emission at the resonant wavelength of the cavity could be greatly enhanced while off-resonance emission would be inhibited. This is essentially the RCE effect discussed thus far for detectors, but applied in reverse for emitters. Yokoyama3’ has published a review of the potential applications. A detailed analysis of inhibited spontaneous emission @SE) has been presented by Bjijrk et al. 33,34They considered a QW in a thin Fabry-Perot cavity and showed that the SWE could be used to enhance the spontaneous emission of the QW at the resonant wave616 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 A 26 l-\IRIR2 ‘?? (31) ’ where L, is the total effective cavity length including the field penetration of the mirrors. The magnitude of the intensity enhancement for a single dipole at the resonant wavelength is:35 B.h.nrrment=2( :” E) (t--g:;;) R,>R,. (32) For their RCE LED design, Hunt et aZ.35estimated a maximum intensity enhancement factor of 73. In order to apply this result to a practical LED, the cavity lifetimes and substrate/air reflectivity would have to be considered. III. DESIGN CRITERIA FOR RCE PHOTODETECTORS The RCE scheme is capable of operation over a large and continuous wavelength range, either by tuning within a material system, or by moving to a complementary material system. Having derived the basic equations describing RCE in the previous section, we move onto the basic design criteria which satisfy the above equations and yield optimal device performance in common semiconductor material systems. A number of semiconductor material systems are evaluated below for RCE applications. A. Material requirements for RCE photodetection The superior performance of the RCE photodetection scheme depends critically on the realization of a low loss cavity. This dictates that both the mirror and cavity material must be non-absorbing at the detection wavelength and that the mirrors have high reflectivity. The multiple mirror periods, each h/4 thick, have a combined thickness on the order of microns. Therefore, the materials composing the mirror must be well lattice matched to avoid the introduction of defects into the active layer. To minimize the number of mirror periods, thereby simplifying growth and reducing the device series resistance, it is desirable to have as large a refractive index difference as possible between the mirror materials. The active layer material must have a smaller bandgap than the mirror and cavity materials, but not so much smaller that large heterojunction band offsets hinder the extraction of photogenerated carriers. The active layer absorption coefticient should be moderate o! (i.e., 1 X 103Ca<5 X lo4 cm-“) within the operation spectrum. If (Y is too small, a thick active layer is required, resulting in increased transit time and degraded high speed performance. For large LY, RCE remains beneficial only for very thin active layers leadAppl. Phys. Rev.: M. Selim blij and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp TABLE I. Pertinent properties of various material systems for RCE devices. Refractive index and absorption coefficient are given for energies at which the RCE application is most viable for the specific material system. S: Substrate, Ml: mirror material 1, M,: mirror material 2, A: Active region material, negl.: negligible, ind: indirect gap. Material E,(eV)(300 K) G~WMI S AlO+lGaAs(M,) ~-LO-o.zGa-MA) 1.42 1.42-2.16 1.42-1.18 Inw3 hmGao.4MW Tn0.szA1o.wWMd h~s-+o.,G~sW 1.34 0.81 1.46 0.81-+0.64 GaMM, ALWM,, Ge(A) 3) Absorption.coeff. (cm-‘) at 1.35 eV 5.660 5.660-5.653 5.6645.74 3.5 3.5+2.97 3.5-13.52 negl. negl. -lo4 5.87 5.87 5.87 58745.94 at 0.75 eV 3.2 3.45 3.21 3.45-+3.48 at 0.75 eV negl. negl. negl. -104 1.42 2.16’“’ 0.66’“” 5.660 5.653 5.646 1.12’“d. 5.431 5.47 5.646 SW, 3) %.sG~dW 1 opd. Ge(A) 0 @ji”d. GWW AWW Si(A,S) Refractive index at 1.35 eV a(&(300 K) 2 2ind. 2.4 1’“” 1.12 5.451 5.463 5.43 1 ing to SWE complications. Furthermore, a thin absorption regions may limit the breakdown voltage for some photodetector structures. The desirability of a moderate active layer thickness also constrains the material to be nearly lattice matched to the cavity material so that the absorption region remains below the pseudomorphic thickness limit. The electronic properties of the material system should also be considered. For high speed and low noise operation, high carrier saturation velocities and low bulk and surface generation/recombination rates are important. Large carrier velocities directly reduce transit time delays. However, the device must be designed to withstand the necessary biases which produce high carrier velocities. Low recombination rates reduce the dark current and noise of a detector. As the device area is reduced to seek higher bandwidths, the surface recombination component increases in relative importance reflecting the larger surface area to device volume ratio. Parasitic resistances can also limit device bandwidth. It is important to choose materials on which low resistance ohmic contacts can be easily formed. The heterojunction band offsets and dopability of the mirror materials need to be considered for their effect on the device series resistance. If the mirror resistance cannot be suitably minimized, then an etched mesa contact can be used to circumvent conduction through the mirror stack, but only at the expense of added fabrication complexity and the introduction of a lateral access resistance. B. Material system photodetection combinations for RCE Different material combinations satisfy all of the above criteria and are therefore amenable to the RCE scheme. By J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 at 1.3 eV 3.48 2.96 4.2 at 0.8 eV 3.33 2.87 4.2 at 1.3 eV negl. negl. 2.4X lo4 at 0.8 eV negl. negl. 3x103 at 0.8 eV 3.48 3.6 4.2 at 2.4 eV 3.55 2.97 4.15 at 1.4 eV 3.1 2.77 3.6 at 0.8 eV negl. negl. 3x103 at 2.4 eV <loo negl. -l.5X104 at 1.4 eV negl. negl. -103 using different material combinations, RCE photodetection can be extended to wavelengths ranging from the visible to well beyond 1.55 pm. Table I lists the material properties of each material relevant to RCE design. The material properties are tabulated for energies at which the fundamental requirement for RCE detectors, i.e., low loss cavity formation, can be satisfied. Below, we discuss the capabilities and limitations of each combination. c 1. AIGaAdGaAsAnGaAs To date, this has been the most commonly studied material system for RCE detection because of the ease with which GaAs based alloys and heterostructures can be grown by MBE. GaAs has good electronic properties, reasonably low carrier recombination rates, and excellent lattice matching to AlAs. Because of the fine lattice match, AlGaAs alloys across the entire range are easily incorporated as wide bandgap contact layers and graded heterojunctions. GaAs and AlAs also have a good refractive index contrast which allows a mirror of nearly unity reflectivity to be realized with a twenty period quarter wave superlattice. InGaAs as the active material allows the device spectrum to be extended to wavelengths at which GaAs does not absorb. While GaAs/AlAs mirrors can function easily out to 1.55 pm,36 beyond about 1 pm the attainable pseudomorphic InGaAs active layer thickness begins to limit the potential detector performance. For this reason, GaAs based RCE devices are expected to remain important for testing concepts and prototype devices, but find limited commercial applications. Appl. Phys. Rev.: M. Selim finki and S. Strite 617 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp semiconductor material near a heterojunction because of autodoping caused by interdiffusion. It is also a challenge to grow GaAs on Ge.37 Thus, a workable detector must be designed as shown in Fig 15. The growth of such a structure requires two separate deposition systems for Ge and the arsenides. The heavily doped upper GaAs layer minimizes the lateral access resistance without increasing the thickness of the heavily doped (neutral) Ge contact region. Furthermore, the resonant modes of the optical cavity can be tuned by varying the thickness of the GaAs contact layer before Ge growth. Ge can be grown at low temperature with very high doping levels, so interdiffusion effects at the heterojunction should not be important. A top mirror could in principle be added. A disadvantage of this design is that absorbing Ge contact layers are necessary to avoid charge trapping. Light in I undoped Ge GaAs contact layer n xN GaAs Substrate FIG. 15. Structure of a Ge photodetector grown onto a GaMAlAs (after Ref. 11). mirror 2lnP/hGaAdlnAIAs The Ino,53Gao.47As/Ino52Alo.48Asalloy system grown lattice matched to InP substrates is of great interest for practical applications due to its outstanding electrical properties and ability to function between 1.3 ,um and 1.55 ,um While lattice matching to the substrate must be maintained, this is easily accomplished. One disadvantage of this system is the poor refractive index contrast between InGaAs and InAlAs. Roughly 35 periods are required for a near unity reflectivity mirror. However, the continuing development of gas source MBE has greatly improved the quality of epitaxial phosphides. Another potential difficulty is that the bandgap of lattice matched InGaAs is very close to the desired 1.55 pm~wavelength range. The use of InGaAs/InAlAs mirrors is, therefore, limited to longer wavelengths. For the detectors operating at 1.3 to 1.55 pm range, quaternary materials (InGaAlAs or InGaAsP) are required for mirror formation. Researchers are developing Ina525Gau475-xA1,As alloys which are lattice matched to InP across a large wavelength range. The introduction of.these alloys will further increase the flexibility of this material system Andyallow graded heterojunctions to be introduced. While more development work is required, the InP/InGaAlAs system has the greatest potential to make an impact in the commercial arena. 3. AIAs/GaAdGe Ge active layers have been suggested” to extend the sensitivity of GaAs based detectors to 1.55 pm and beyond, making them applicable to fiber optic systems. The good lattice matching (0.25%) enables dislocation free growth of Ge on GaAs. Ge also has very low recombination rates and the desired moderate absorption coefficient above 1 pm. However, the lack of an intermediate bandgap alloy would lead to charge trapping at the heterojunctions. Some difficulties arise when outstanding materid- quality is required. It is very difficult to control the carrier concentration in either 618 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 4. WSiGe Si-based photodetectors are widely used for operating wavelengths below 1 pm. Recently, the ability to grow strained SiGe alloys has extended the usefulness of Si-based detectors to longer wavelengths.38 WSiGe quarter-wave reflectors have also been suggested as a candidate for wavelengths below 1 pm39 although they are absorbing at those wavelengths. The basic advantage of Si based photodetection is the potential for integration with Si electronics. Kuchibhotla et aL4’ extended the RCE scheme to this material system by demonstrating Si/SiasGeoZ reflectors optimized for 1.5 pm. WSiGe mirrors are limited by the allowable strain, refractive index contrast and absorption of the SiGe alloy at the wavelength of interest. A Ge mole fraction of 0.2, as chosen by these authors, remained pseudomorphic and transparent at the required thicknesses of 0.1 ,um, but only provided a refractive index contrast of 0.1, a factor of 4-5 less than provided by the III-V materials. As a result, a reflectivity of only 50% at 1.5’pm was achieved. However, because of its other inherent advantages, SiGe based detectors are highly desirable, even if it proves difficult to fully exploit RCE in this material system. I 5. Si/AIP/GaP For shorter wavelengths, above the GaAs band edge, the nearly lattice matched Si/AlP/GaP material system has excellent potential but requires considerable development. Although the bandgap of GaP is 2.2 eV, its absorption coefficient remains sufficiently low (i.e., 9Si) to permit RCE detection of blue light at 2.4 eV. Therefore, RCE detectors with GaP/AlP mirrors and Si active regions can be fabricated to function across the entire visible spectrum. The refractive index difference between GaP and AlP is less than that of GaAslAlAs so more mirror periods are required. However, for short wavelength detection, this is compensated for by thinner mirror periods. These detectors share many of the disadvantages of the Ge based detectors necessitating a design similar to Fig. 15. The need for two separate deposition chambers and absorbing Si contact layers, the lack of an intermediate bandgap alloy, and potential for autodoping, all complicate the design Appl. Phys. Rev.: M. Selim finlij and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp I of high performance devices. However, unlike the material systems discussed above, this system spans the entire visible range. Fairly conservative device designs could easily provide higher 77detection than the conventional Si photodetectors presently in wide use today. The structure can be grown on Si substrates allowing integration with electronic devices. Growth of GaP on nearly lattice matched Si has already been demonstrated.41 Once the structure is grown, well established Si processing technology can be utilized for fabrication. The combination of long Si carrier lifetimes and high 17RCE detection could yield high performance optical devices. IV. HIGH SPEED PHOTODIODES PROPERTIES 200 150 OF RCE Intuitively, RCE photodetectors are expected to have improved speed compared to conventional photodetectors because of the reduced active layer thicknesses which can be used. Experimental reports confirming this are beginning to appear.42s43 A relatively simple analysis comparing the bandwidth of RCE and conventional heterojunction p-i-n photodiodes (PD) provides insight into high speed performance limitations of each. More complex simulations support the intuitive predictions.44 A. High speed capabilities photodetectors (a) of conventional The most important response speed limitations in heterojunction p-i-n PDs are drift time across the depleted region, charging and discharging times of inherent and parasitic capacitances, the diffusion time for the carriers generated in the undepIeted regions, and charge trapping at the heterojunctions. Good detector design minimizes the latter two by incorporating non-absorbing contact regions, grading the absorbing region heterojunctions, and placing the active layer in the depleted region. Therefore, we neglect those speed limitations and focus on intrinsic limitations imposed by the transit and capacitance-charging times. The transit time depends on the electron u, and hole uh velocities, and is dominated by the slower holes. In a conventional PD [Fig. 16(a)], photogenerated carriers have to traverse the entire depletion region. The transit time limited 3 dB bandwidth of a thin detector is:45 1.5 1.0 0.5 0.0 (b) depletion 2.0 width (pm) FIG. 16. (a) Schematic respresentation of a conventional heterojunction p-i-n photodetector illustrating the distances that the photogenerated carriers carriers have to transit. (b) The bandwidth-depletion width dependence (solid lines) are shown for 10X 10 @rns and 5X5 pm’ area conventional p-i-n detectors. The transit time limited bandwidth (dashed line) is common for different size devices and capacitance limited bandwidth (dotted lines for a total output resistance of 50 0) varies with the device area (after Ref. 44). (33) and (34), an optimum L exists for a given A and RT. Figure 16(b) shows the bandwidth of 10X 10 pm2 and 5 X 5 ,um’ area conventional detectors as a function of depleted layer thickness for a total resistance of 50 s1. The maximum bandwidth for this structure is reached at a 0.3 ,um and 0.15 pm depleted layer thickness and are about 43 GHz an 86 GHz for the larger and smaller device dimensions, respectively. For a high speed detector, in which the depletion width is small, i.e., crLe1, with absorption occurring only in the depleted region, 7 is: a=(l-R)(l-e -=+(l -R)aL. (35) For an ideal AR coating, (35) reduces to f*,=O.45i ) (33) where L is the depletion width, and uzaAs= 6 X lo6 cm/s. For a 1 ps response, a depletion width of 0.06pm is needed. However, as the intrinsic layer thickness is decreased, the capacitance increases and becomes the fundamental bandwidth limitation. The capacitance limited bandwidth is*45 1 fRCD-.--~ 2?rR,C L 2'rrlilTe,+4' 04) where E, is the relative permittivity of the semiconductor, A is the area of the device, and RT is the total resistance (load resistance + contact resistance). As can be seen from Eqs. J. Appt. Phys., Vol. 78, No. 2, 15 July 1995 77mffL. (36) Note that 7 increases with L in contrast to the transit time limited bandwidth. At the optimum bandwidth of the larger device shown in Fig. 16(b) (L=O.3 ,um), vsO.26. The increasing dependence of the sensitivity on the depletion width results in an optimum bandwidth efficiency product shifted towards larger thicknesses as shown in Fig. 17. The maximum value of the bandwidth-efficiency product for a 10X 10 pm2 GaAs detector with R,=50 n is 16.2 GHz and occurs at L=O.7 pm. The maximum bandwidthefficiency product of the smaller device is only slightly higher than the larger device despite the significant difference in their maximum bandwidths. Appl. Phys. Rev.: M. Selim &la and S. Strite 619 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp 150 3 5 100 % f ‘3 'G %T i B .C 2 5 A2 +--I&++ (4 i L: I 50 \ \\ \\ : .’/ Ll 4 8’ I~;~\ 1 150 0 0.5 0.0 1.0 1.5 2.0 3 depletion width (pm) FIG. 17. Bandwidth-efficiency product versus the depletion (absorption) region width for conventional detectors as depicted in Fig. 16(a). The transit time and capacitance limitations are shown in dashed and dotted lines, respectively (after Ref. 44). The inverse relation of transit time limited bandwidth with L (33) suggests that infinitely large bandwidths can be achieved by decreasing the depletion width so long as the capacitance limitation is simultaneously relaxed by shrinking the device size or reducing the total resistance. However, if we consider the bandwidth. efficiency product for a thin detector with no capacitance limitation, we obtain from (33) and (35): SW- v=ftr- 77=0.45auJl(l -R.l of RCE photodetectors The p-i-n RCE PD structure can incorporate a smaller bandgap absorption region of thickness d placed in a depletion region of width L [Fig. 18(a)]. In this case, the carriers do not have to traverse the entire depletion region since they are generated only at the active layer. The transit times, re and rh, for electrons and holes, respectively, are given by the distances they must travel: L2 Ll 3-<=and rh=* V, U@ The transit time is optimized when r,= rh : 620 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 100 E 0 50 0 0.0 0.5 1.0 1.5 2.0 depletion width (pm) 03 FIG. 18. (a) Schematic respresentation of a RCE p-i-n photodetector illustrating the distances that the photogenerated carriers carriers have to transit. (b) The bandwidth-depletion width dependence (solid lines) are shown for 10X 10 p m ’ and SXS ,um2 area RCE p-i-n detectors. The transit time limited bandwidth (dashed line) is common for different size devices and capacitance limited bandwidth (dotted lines for a total output resistance of 50 Cl) varies with the device area (after Ref. 44). (37) which is independent of the design parameters and can be perceived as the y-intercept of the dashed line in Fig. 17. For AR coated (R=O) GaAs detectors (~~-10~ cm-‘, uh=6 X lo6 cm/s) this product is 27 GHz. The 77of thin p-i-n PDs has been improved by collecting the tight through the mesa edge of the device, perpendicular to the electrical current.‘6 This configuration provides a long (albeit narrow) absorption region and a short current path simultaneously, enabling, the independent optimization of speed and 17. In this waveguide configuration however, the insertion loss of the incoming light limits the overall 7, while the waveguide design creates charge trapping at the heterojunctions. 6. High speed performance H 2 (38) L+d=L,+L,. (39) Therefore, the optimized transit time limited bandwidth for the RCE detector is j+o.45= for L>d. This represents a drastic improvement over conventional detectors given by (33), since in most compound semiconductors u,>uh (e.g., ucg I X lo7 cm/s in GaAs). The capacitance limited bandwidth (34) remains unchanged in the RCE design. Figure 18(b) shows the dependence of bandwidth on the depletion width L for IOX 10 pm” and 5X5 ,um’ area GaAs RCE-detectors with a 0.1 pm thick absorption region. When compared with the bandwidth of the conventional detector (Fig. 16), we see that the optimum bandwidth moves toward thicker depletion regions and a maximum of 64 GHz is reached at L= 0.5 ,um for the larger device, and maximum bandwidth approaches 120 GHz (at L= 0.25 pm) for the smaller device. In the RCE scheme, large 77can be maintained independently of L. Applying the peak v relation (9), we obtain ?7=0.9 for a 0.1 ,um thick absorber (ad=O.l) with R2=0.99 and RI =0.7. 17 remains unchanged as long as L>d = 0.1 F. When L<d, 17must be evaluated by replacing d with L in (9). The bandwidth efficiency product evaluated under these constraints for the RCE p-i-n is shown in Appl. Phys. Rev.: M. Selim blij and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp L/V, 0 depletion width (pm) FIG. 19. Bandwidth-efficiency product versus the depletion (absorption) region width for RCE detectors as depicted in Fig. 18(a). The transit time and capacitance liiitations are shown in dashed and dotted lines, respectively (after Ref. 44). Fig. 19. The maximum bandwidth-efficiency product of 58 GHz for the 10X 10 pm2 RCE p-i-n represents a more than 3.5-fold improvement over the conventional detector (16.2 GHz) For the 5X5 pm’ device this predicted improvement is more drastic, over a 5-fold increase from 19 GHz to 106 GHz. In the above discussion, we presented the bandwidth considerations for 10X 10 pm” and 5X5 pm” device sizes. As the deirice size is reduced, the capacitance limitation shift towards thinner depletion regions; At this extreme, the influence of the RCE-scheme is even more prominent since the intrinsic limitation on the bandwidth-efficiency product is relaxed by removing the direct dependence of the quantum efficiency on the depletion layer thickness. To treat fully the high speed response of the RCE PDs, we must also consider the photon lifetime in the optical cavity as one of the limiting factors. The photon lifetime rp can be viewed as the time required to build or decay the optical fields inside the cavity and is given by15: 7=- rRT p Loss 3 where TRT is the time required for photons to make one round trip in the optical cavity, and Loss is the total decay they encounter during this round trip. For a 1 pm long GaAs cavity, ?-RT=23 fs. For a Fabry-Perot cavity of mirror reflectivities RI and R, and absorption region ad, the loss term can be expressed as Loss=[l--R1R2 exp(-2ad)]. IA/ t I& 0 W(v,+ v<) t t FIG. 20. The theoretical current response of a photodetector with a depletion region width of L when excited by an optical impulse under constant drift velocity and no diffusion assumption. The vertical scales are arbitrary, and the horizontal scales are intended for a qualitative comparison of different cases. (a) The optical impulse ariving at t=O. (b) The cmrent for a single electron-hole pair generated in the middle of the depletion region. The response continues until the hole reaches the contact. (c) The current for a conventional structure with uniform generation across the depletion region. The response lasts until the holes generated at the opposite end traverse the entire depletion region. (d) The current for an optimized RCE detector. The electrons and holes arrive at the contacts at the same time (after Ref. 44). C. Impulse photodiodes response of RCE and conventional In the previous two sections, we presented a simplistic bandwidth picture for conventional and RCE p-i-n PDs which neglects diffusion, field dependence of the mobilities, carrier recombination, and the influence of heterojunctions. Under these assumptions, we can further predict the impulse response of each PD design. The current i(t) due to a carrier of charge Q moving with a velocity of u(t) in a semiconductor material of thickness L is given byI i(t)= - $ u(t). (43) Figure 20 shows a comparison of the theoretical current responses for conventional and RCE p-i-n PDs. For a single electron-hole pair generated in the middle of the depletion region, the detector current is as shown in Fig. 20(b). For a conventional detector, under uniform illumination, the transit time spread is more severe since the full width is determined by the slower holes traversing the farthest distance [Fig. 20(c)]. By equalizing the electron and hole transit times in the RCE device, a uniform response with a small time spread tian be realized [Fig. 20(d)]. (42) For typical parameters (R, = 0.7, Rz = 0.99, ad = 0.1) , ~~-50 fs, which is much smaller than the carrier drift time. Even for RCE PDs with thinner active regions (d-0.05 prnj and higher reflectivity mirrors (R,>0.9), ~~ remains on the order of 100 fs. Therefore, the photon lifetime will be an important factor only in RCE detectors designed for THz operation. D. Simulated high speed response photodetectors of RCE When photodetectors with picosecond response times are considered, a more accurate representation of the transient response is desirable. Efforts have been underway towards obtaining a better understanding of the transient behavior of p-i-n PDs.“~-~“,~’ J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 Appl. Phys. Rev.: M. Selim finlti and S. Striie 621 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp AlGaAs AlGaAs ‘n?qiyA ‘=- l2ce #2 ‘S 2 8 D FIG. 21. The schematic flat band diagram (applied and built-in fields are not shown) of the modeled p-i-n photodiodes and the qualitative representation of the photogeneration terms. (a) Device #l: the conventional AlGaAslGaAs p-i-n heterojunction photodiode. The photogeneration occurs in the depleted GaAs region on an exponentially decaying profile. (b) Device #2: the GaAs RCE photodiode with an InGaAs absorbing region. The generation is localized into the InGaAs region (after Ref. 44). &~lii et aZ.51,52 have deveIoped a one dimensional simulation method for the transient analysis of PDs and used it to compare the high speed response of RCE and conventional heterojunction p-i-n PDs. The structures, shown in Figs. 21(a) and 21(b), were chosen to follow the design rules for optimal bandwidth. For the conventional design, a depletion width L = 0.72 pm was selected, corresponding to the maximum bandwidth-efficiency product for a 10X 10 pm’ detector. A 0.64 pm thick, normally depleted n--GaAs absorbing region and transparent Ala.s~Gass4As contact layers completed the conventional PD structure. The small Al mole fraction avoided large band discontinuities. The heterojunctions were graded and positioned inside the depletion region to prevent charge trapping. W ithin the absorption edges of the AlGaAs and GaAs, photogeneration occurs only in the depleted GaAs region. As a result, the device speed is limited solely by the transit time of photogenerated carriers across the depletion region. Finally, a nearly ideal AR coating (Ri = 0.05) was assumed, yielding 77=0.45 from (35), within the detection range. The RCE PD structure was assigned the same depletion width L= 0.72 pm. The absorption region was a graded heterojunction 0.08 pm n--In0.07Gac,9sAslayer optimally positioned in the GaAs depletion region. The remainder of the detector consisted of n- and p-type GaAs contact regions. The resonant cavity was defined by an ideal bottom mirror (R2= 1.0) and a high reflectivity (R, = 0.7) top mirror. 17for this structure was 0.9 near 900 nm.’ The model allowed for the exact computation of the transient photocurrent under arbitrary optical illumination. 622 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 15 Time(ps) FIG. 22. The transient short circuit photocurrent under a 5 ps E W H M optical pulse for AlGaAs/GaAs conventional p-i-n (Device #l, dashed line), and GaAs/InGaAs RCE detector (Device #2, solid line) (after Ref. 44). W ithin the convergence capabilities of the simulator, it was possible to calculate the output current for an optical pulse of less than 1 ps FWHM. Such a short pulse is a good approximation of an impulse excitation, and provided the best estimate of the speed of each device. Since the estimated transit time for conventional and RCE device designs are about 5 to 10 ps, the transient response of the two devices are compared under a 5 ps optical pulse. Figure 22 shows the simulated short circuit currents for the conventional (dashed line) and RCE (solid line) p-i-n detectors under a 5 ps F W H M optical pulse. At first glance, we observe a larger (magnitude) and sharper (less time spread) current for the RCE-detector compared to the conventional case. A close inspection of the individual current components reveals the expected transient features and superiority of the RCE-detector. (i) The rise and fall times of the total current are comparable for the RCE design. For the conventional detector, the fall time is much larger than the rise time as a result of long transit time for holes. (ii) The RCE-detector not only has larger current under identical optical excitation owing to the enhanced quantum efficiency, but also the response is faster than the conventional counterpart. Therefore, the bandwidth-efficiency product is doubly improved by the RCE-detection scheme. (iii) Due to the simultaneous arrival of both kinds of carriers at the contacts, the time spread of the conduction current for the RCE design is much smaher than the conventional p-i-n in which the carriers reach the contacts at different times determined by the position of the photogeneration. For a comparison of the RCE and conventional detectors in the frequency domain, we used Fast-Fourier-Transform (FFT). The simulated temporal responses depicted in Fig. 22 were converted into frequency domain by FFT and shown in Fig. 23(a). Fourier analysis of the optical excitation term gives a flat spectrum (within 1 dB) up to approximately 100 GHz verifying the accuracy of the frequency domain repreAppl. Phys. Rev.: M. Selim ijnlti and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp ,o” ,..... .d...~~~_.L~..~.;.~ ......i..;:;,:~..::i::::,:::::i:::: :::::::::::. tr.I--~~~~~:-~~~~~~~~~....’ 1 R,. ; : :: : .:::: . . .. . . . . . . . . . . ;......: i...:..A.:..:.:.: I: . . .. . . ..: y; ._:. *..;.i ..: . . : .::::: ; .): i I.. : .. . . .. . . . .. . ..._.......... ;...:...i..;..;.:.; .....Y.....::. . .. . . .: :...:.I.:..>.:.:. : Ii::‘Z” : : 1 :j >;.:: :: \.\::: : : ~ : ::..:. P \ .;<,I ::::.; :::: ::, 5 ,A . . . ..................... . . . ..I...... L .:...i...Liii . . . . i ,....,. i i...;...i....: /. ...... ...... . ........ ;..:..l.; .. .... :...; ..; .., * ., ............ is P 6 j 14- ........ ........ ........ ........ ,” I’ .... ............... ........... .......... .. ........ a.’ .““” \ ... ... 1 ..~..............~....~.!.~............!.......~....!...~...~..~.~...~ . ....... :‘..I. :“.:“.).‘:“!.<‘: ..!. .. . ... :“7.:.<.: ............ . ........... .......... .i...:...~..:..;..:.i..~. . .... ..~.....,......~.~..j.~.~............~ ... ..:....;...!...r..i..~.~.~ ............ . .................................... .... .{......\’1 ...... ..... .......... .:. ..... . ..... ;...;...;..;..;.z.;. ........... j.. ...........i.. ..:...;...:..i.i \ : : ::::: .................... L....;...;...j.;..;.: ..... .:. ........... i.. ..... . ........ :...:..:..:.^.:. ... .\I.. ..... I : :“‘z:::c :::::: : m:!!:/:i ..m..i..I..i.i.l....... 1 . . . . . . . . . . . .. . . . . ..~....~.......~..~..~........................~..... : : :::::: :: ‘::’ \ .i::. : : : :;::: i \I\ I - : ‘::jf : 16" : : .::.: 10’ Frequency 10" . ..... . ... . ..... !...+..+.! . .... . ...... . .... ,: ..... . .., .......... ......... ..... ..; ... )...:...; ....... ..> . ...... (...... :_ ..... . .......... .... ‘” .:. (GHz) FIG. 23. (a) Frequency response of conventional (dashed) and RCE (solid line) p-i-n photodetectors obtained by Fourier transform of pulse responses shown in Fig. 22. The optical excitation term was deconvolved to extract the impulse response. (b) Detector response (peak value of the normalized detector current) versus the inverse excitation pulse width for conventional (dashed) RCE (solid line) p-i-n detectors (after Ref. 44). sentation of the current pulses. Furthermore, to eliminate the influence of the finite width of the optical excitation pulse, it was deconvolved from the simulated current responses. From Fig. 23(a), the transit time limited 3-dB bandwidth of the conventional p-i-n detector can be calculated as 52 GHz in comparison with 70 GHz for the RCE p-i-n, corresponding to a 35% improvement. For a direct comparison of the bandwidths for RCE and conventional p-i-n structures, we plot the peak value of the normalized short circuit current (or detector response) as a function of the inverse pulse FWHM [Fig. 23(b)]. When the pulse duration is large, the output current reaches its maximum value determined by the internal quantum efficiency, i.e., r)= 0.45 for conventional and o= 0.9 for RCE detectors. As the optical pulse width becomes smaller, the electrical current is unable to reach the steady state value. Although the ratio of electron hole pairs to the number of incident photons J. Appt. Phys., Vol. 78, No. 2, 15 July 1995 i? 4- ! a-. 0 5 10 15 20 Time(ps: FIG. 24. The transient short circuit photocurrent under a long optical pulse with 0.01 ps fall time for conventional (dashed) and RCE (solid) p-i-n detectors. The excitations is long enough to allow both devices to reach their steady state current levels (after Ref. 44). . ....... .... 10’ i/(2 pulse FWHM) i I.... Es 3 0 ! ... ............... . i.<.; E -2 ld ..... <. _ ‘:::’ (GHz) .p ..........y. -3 ;‘O .1::: ;.i.:. .... ........... 12- remains the same, the peak current decreases since the current is spread over time. The peak short circuit current drops to half of its maximum at a pulse FWHM of 4.5 ps and 3.5 ps for conventional and RCE p-i-n detectors, respectively. Therefore, the bandwidth of the RCE p-i-n is approximately 30% larger than that of the conventional diode conlirming our findings through Fourier analysis. Note that, both of these comparisons are made for a single optical pulse of varying duration. If we consider repetitive pulses, or sinusoidal excitation, the bandwidth difference will be even larger due to the large fall time of the conventional p-i-n. To this end, we also compare the fall times for RCE and conventional p-i-n diodes. We calculated the response of these devices (Devices #l and #2) under a long optical pulse such that both devices can reach the steady state current levels. The optical pulse is terminated abruptly (fall time=O.Ol ps) and we observe the decay of the short circuit current for both cases as shown in Fig. 24. Despite the higher steady state level,‘the current from the RCE detector decays faster and drops to lower values than the current for the conventional p-i-n. We have computed the fall time rf as the time it takes the current output to drop from 90% to 10% of the steady state value. The fall time Q-~=7.1 ps for the RCE p-i-n is 60% of that for the conventional structure ( rf= 11.7 ps) corresponding to 65% faster response. The simulation results demonstrated a 35% bandwidth improvement along with a two-fold enhancement in quantum efficiency over conventional p-i-n photodiodes optimized for a 10X 10 ,um2 device area. For smaller area devices, even more drastic improvements exceeding a three fold increase in the bandwidth-efficiency product can be achieved. The simulation results suggest that high quantum efficiency RCEphotodetectors can be realized with bandwidth-efficiency products approaching 100 GHz. Appl. Phys. Rev.: M. Selim fin1C.iand S. Strite 623 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp 1 ix t 5 E s !z z f 8 5 E r? - GaAs substrate FIG. 26. Schematic cross section of the RCE APD (after Ref. 57). I 00 1 1200 (a) 1400 WAVELENGTH 1600 1800 2000 Figure 25 compares the photoresponse of the RCE device with that of a similar diode having no bottom mirror. At the resonance wavelength, a 50% enhancement in detected photocurrent was observed. The moderately thick absorption region in these devices prohibited the full exploitation of the RCE concept. The design for bottom illumination causes losses due to reflection at the substrate surface and limits the highest attainable quantum efficiency. This problem can be circumvented by a AR coating on the substrate surface. (nm) ct J 2 5 s 2 5 8 6 E B. Avalanche IO 1200 (b) 1400 WAVELENGTH 1600 -I 1800 2000 (nm) FIG. 25. Photoresponse of (a) RCE Schottky photodiode and (b) a similar detector wihout resonant cavity (after Ref. 53). V. EXPERIMENTAL DEVICES RESULTS ON DISCRETE RCE photodiodes Avalanche PDs constitute another device in which the v. bandwidth product can be optimized by RCE. Conventional APDs in the low gain regime are bandwidth limited by the transit time of secondary electrons through the depletion region.54-56 The ability to obtain high 17detection with a thin absorption region allows the transit time limitation to be greatly relaxed. Smaller depletion regions also translate into higher electric fields in the gain region at a given external bias permitting reduced power operation. Kuchibhotla et aLS7 demonstrated the first RCE APD (Fig. 26) which consisted of a thin (900 A) InaosGa,-,~sAs absorbing layer (doubling as the multiplication region) situated in an optical cavity. The bottom mirror was a 15 period Most of the theoretical relations derived in the previous sections have been experimentally verified by the many c workers in the field. The RCE scheme has been ,profitably applied to many different types of photodetectors, modulators and emitters in order to enhance the efficiency, realize wavelength selectivity, or both. A. Schottky photodiodes An InGaAlAs Schottky PD was among the first reports of a RCE device.9*53Chin and Chang realized that by placing a conventional Schottky PD inside of a Fabry-Perot resonant cavity, a thinner absorption region could be employed, reducing transit time bandwidth limitations while increasing 77 at the resonant wavelength. Their structure was a relatively conservative one in which an 8 period InAlAs/InGaAlAs mirror was grown onto an InP substrate. This was followed by a 475 nm In,,5sGaa.47As absorbing region capped by a 50 nm A&As Schottky contact layer. The cavity was completed by the deposition of a high reflectivity Al Schottky contact. 624 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 Wavelength (nm) FIG. 27. Measured photoresponse of the RCE APD depicted in Fig. 26 for vario.us mirror reflectivities (after Ref. 57). Appl. Phys. Rev.: M. Selim blij and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp , , , Dielectric optical Stack -lnGaAs Absorbing Layer GaAschamlel -n:lnP > pulses buffer InPilnGaAsP Bragg Reflector t, layer K m -n t* MlyGa,, As) t, UU,Ga., As) tm (Al Ga Contact Y l-Y As) FIG. 28. Schematic layer structure of InP/InGaAsP/GaAs RCE photodiode (after Ref. 62). quarter wave AlAs/GaAs stack having an estimated reflectivity of &90% in the wavelength range of interest. In order to investigate the RCE effect, RI was varied by depositing a CaFs cap layer on some samples. Figure 27 shows the photoresponse of three RCE APD devices having RI = 15%, 30%, and 73%. As the Q of the cavity is increased, the response sharpens around the 840 nm resonance wavelength of the optical cavity and the corresponding 77at that wavelength increases. The maximum 17 of 49% achieved for RI ==73% was limited by the low reflectivity of the bottom mirror. A further limitation was the design of the cavity resonance at 840 nm were the AlAs/GaAs mirror is lossy. Because of the rapid increase of the electric field in the active layer with external bias, gain in the RCE APD is achievable at relatively low applied voltage. Near breakdown at 9 V, an internal gain of 200 was measured. This operating bias is the lowest reported to date for APDs. C. P-I-N photodiodes P-i-n PDs are leading candidates for high speed photodetection.46*58T59 P-i-n PDs should be capable of speeds exceeding 100 GHz.““~~ For applications requiring faster response, photoconductive detector8’ are currently favored, but suffer from poor sensitivity. RCE p-i-n PDs are ideal for overcoming the bandwidth/sensitivity tradeoff and have received the majority of the attention directed towards the high speed RCE photodetectors. The first RCE p-i-n PD was reported by Dentai et ~2.~~ who studied an InGaAsZInGaAsPZInP structure (Fig. 28) designed to operate near 1.55 pm. An 77of 82% was achieved using a 2000 A InGaAs absorbing layer with Rt=0.7 and Rz=O.95. Kuchibhotla et aL4’ have demonstrated an RCE p-i-n PD in the SiGe system. Their 77was severely limited by the reflectivity obtainable with Si/Sia,sGec,, mirror stacks which suffer from a relatively small refractive index contrast. Several refinements and variations of the basic structure have since been reported. RCE p-i-n photodetectors have been integrated with HBT drivers.63 Huang et ~1.~ reported a 30% improvement in their p-i-n PDs by optimally positioning the absorption regions at cavity antinodes. Lai and Campbell65 FIG. 29. Conceptual diagram of a RCE MSM photodetector with a buried DBR mirror (after Ref. 70). predicted that an RCE p-i-n PD with a MQW absorption region could be actively tuned over a 20 nm range due to the quantum confined Stark effect but their approach is limited by large crosstalk. Hunt et aZ.66introduced wavelength selectivity to a back illuminated InGaAs p-i-n PD by growing a Si/SiOz resonant cavity onto the backside of the InP substrate while Corbett et LzZ.~~reproduced the RCE effect by using epitaxial lift-off to physically place a conventional p-i-n PD inside a cavity consisting of two Ag mirrors. Finally, Sverdlov et aL6* analyzed InGaAsZInAlAs p-i-n PDs and observed that an added advantage of the RCE scheme is reduced dark current resulting from the thinness of the active region. D. Metal-semiconductor-metal photodetectors The MSM photodetector design is conceptually similar to photodiodes in so far as incoming photons create an electron-hole pair that is separated by applied electrical field and collected at contacts causing an electrical current. However, the implementation of the MSM approach differs from the others in being a planar rather than vertical device structure.69 This is an advantage in applications where the photodetector must be integrated with amplifiers43 or an emitter. Nevertheless, the 7. bandwidth product remains governed by similar fundamental tradeoffs as p-i-n PDs and APDs, and is therefore amenable to RCE. The response speed of MSM detectors can be increased by reducing the finger spacing, and thus the carrier transit times. However, for very small finger spacing, the thickness of the absorption layer becomes comparable to the finger spacing and limits the high speed performance. The RCE design allows for the use of very thin absorption layers while maintaining large quantum efficiency, thus eliminates the fundamental speed limitation for MSM photodetectors. Li et aL7’ theoretically investigated the properties of an AlGaAs/GaAs MSM photodetector situated on top of a buried quarter wave mirror (Fig. 29) based on a comparison with experimental results on conventional devices with simi- J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 Appt. Phys. Rev.: M. Selim cnlti and S. Strite 625 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp Incident Light 10 t w F z mV 6 0 100 200 * * 300 400 ps 500 time ---+ FIG. 30. Measured temporal response of AlGaAs/GaAs RCE M S M photodetectors showing a E W H M of 19 ps (after Ref. 42). FIG. 31. Structure of the AlGaAs/GaAs/InGaAs RCE HPT (after Ref. 17). lar layer structures. As a result of the reduced active layer thickness, a 1.5 pm channel RCE MSM photodetector was capable of operation at twice the speed of a similar conventional MSM structure. Prank et .1.a2171have reported substantially improved AlGaAs/GaAs RCE MSM photodetectors by increasing the cavity finesse and fabrication technique. A device incorporating 20 and 7 period A~A~IA&,&zz~~As lower and upper mirror stacks exhibited a photoresponsivity peak at 846 nm having a F W l!&I of only 1 nm. High frequency measurements indicated that these devices were capable of operation up to at least 35 GHz as a result of the photoresponse rise and fall times on the order of 14 ps (Fig. 30). Litvin et aL43 have also reported high frequency measurements of an RCE MSM photodetector. Their device consisted of a GaAs absorption layer sandwiched between a high reflectivity distributed Bragg reflector (DBR) and a transparent AlGaAs cap layer. After the growth of the detector, a high electron mobility transistor structure was deposited. This allowed the MSM photodetectors to be integrated with a low noise amplifier. A time Iimited bandwidth in excess of 40 GHz was measured for devices having a 0.5 pm finger spacing. E. Heterojunction phototransistors Heterojunction phototransistors (HPT) are often employed to combine photodetection and amplification with less noise than APDs.‘~-‘~ Their combination of excellent sensitivity, large gain, low noise and high bandwidth places HPTs among the most promising RCE devices. AlGaAs/GaAs HPTs having InGaAs active layers (Fig. 31) are among the first RCE devices demonstrated.” The resonant cavity was defined by a 10 pair AlAslGaAs bottom mirror (R,=0.9) and the GaAs top surface (R1 0.3). The 0.1 pm InGaAs active layer extended the photosensitivity spectrum beyond the GaAs absorption edge (>900 nm) where absorption losses in the heavily doped GaAs base and collector were negligible. The RCE effect was evaluated by comparison to a conventional HPT having an identical structure except for the bottom mirror. Figure 32 compares the 626 photosensitivity of the conventional and RCE HPTs. The spectral responses were normalized at 750 nm where the cavity effect was negligible. The conventional HPT response decreased sharply beyond the GaAs band edge near 870 nm. The small increase near 900 nm reflected absorption by the InGaAs layer, which was too thin to have a large effect in a conventional structure. Under pulsed laser illumination, the optical gain of the HPT was measured to be above 500 at 850 nm (~=25%) and 150 at 900 nm (~=6.7%). As expected, the photosensitivity of the RCE HPT was greatly enhanced at resonance. At 900 nm, where the cavity was nearly lossless, the measured photocurrent was 7 times that of the conventional HPT. The enhancement at 850 nm was only a factor of 1.7, reflecting absorption losses. Both values agreed well with theory. Higher F cavities are desirable for improving 77 and wavelength selectivity. Bryan et a1.77have described an RCE HPT capable of operation at 930 nm in which InGaAs QWs placed optimally within a high F cavity benefit from the SWE. As a result of the high Q cavity design (R2=99%, R 1=70%), a very sharp photoresponse of 3 nm l?#HM with J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 34 '3 i? $3 3 .!? $2 5 f' V n 750 800 850 900 950 1000 Wavelength (urn) FIG. 32. Spectral response of conventional (solid line) and RCE (dashed line) floating base HPTS illustrating the enhanced v resulting from the resonant cavity (after Ref. 17). Appl. Phys. Rev.: M. Selim &dii and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp Cr/AuZn c%ml 3 QUANTUM WELL ABSORBING Lens SI GaAa FIG. 33. Device cross section of the fabricated BICFJ5T/HFET detector (after Ref. 87). an optical gain of 600 was achieved. In this design, the resonant wavelength is set during the layer growth requiring precise control of growth rates. High finesse RCE-HPTs with a thin 200 A semitransparent Au top mirror and a conventional AlAslGaAs bottom mirror (R,=95%) have also been demonstrated.78An advantage of a metal top mirror is that the cavity length can be pre-tuned by surface recessing before the top mirror is evaporatedT9 The FSR and F W H M of the devices were 78 mn and <6 nm respectively, resulting in F> 13, a 3.5 fold improvement over the same devices without the Au film. Dodabalapur and Chang” reported an InGaAs/InAlAs/ InP HPT capable of operation at the technologically important 1.3-1.6 p m wavelength range. Because of the relatively poor contrast attainable with InAlAs/InGaAs mirrors, a high reflectivity Au top mirror was used and the device was back illuminated. Two different designs optimized for 1.3 and 1.55 pm were demonstrated having optical gains as large as 2500. F. Bipolar inversion phototransistors channel field effect Bipolar inversion channel field effect transistors (BICFET)81 have been promoted for their suitability for integration with laser structures.82-84 Like HBTs, BICFETs function as HPTs if a narrow bandgap absorbing region is inserted into the structure, and they benefit in a similar manner from RCE. BICFET HPTs are of interest for their ability to be integrated along with an edge emitting or vertical cavity (VC) lasers to form a completely optical switch.85’86 Daryanani et a1.87-gohave studied RCE BICFET HPTs designed to operate at 940 run. In their structure (Fig. 33), three InGaAs QWs doubled as both the inversion channel and the absorbing region. The devices exhibited a sharp photoresponse with lYWHM=2 nm and an optical gain of 25. The measured ~=80% represented a factor of 26 improvement over a single pass BICFET phototransistor structure. G. Wavelength demultiplexing and phototransistors with RCE photodiodes Wavelength Division Multiplexing (WDM) is a key technology for increasing the transmission capacity of optical fiber communication systems.24 High-speed, high gain J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 FIG. 34. Conceptual diagram of a W D M application for the detector composed of four RCE HPTs having shifted resonance wavelengths (after Ref. 26). demultiplexing receivers are a crucial component of these systems.‘l-” RCE detectors are ideal for this application be cause they combine very high 77with low crosstalk and high speed. Each RCE detector in an F&ay can function as a channel discriminator, detecting only a specific band from the common incoming light. Sharing of the optical input signal can be compensated by the enhanced 17 and gain of each detector. RCE W D M was first demonstrated in a simple array of 4 RCE HPTs having complementary spectral responses (Fig. 34).26 Recessing the epitaxial layer surface with standard wet etching tuned the response wavelengths of each device by changing its overall cavity length. Three RCE HPTs were recessed to produce three equally spaced modes such that the resonance of the most deeply etched device coincided with that of the unetched device. Figure 35(a) shows the spectral responses of three de vices under equal illumination at a collector emitter bias of 3 V. The FSR and F W H M were 55 mn (at 900 mn center wavelength) and 15 nm, yielding a finesse F=3.6. Maximum crosstalk attenuation of 15 dB and - 12 dB were realized for 2 and 3 channel demultiplexing, respectively. The F was greatly improved by the addition of semi-transparent Au top mirrors78 allowing an experimental crosstalk attenuation of 16 dB for four channel demultiplexing to be realized. Spectral response of this four channel W D M detector is shown in Figure 35(b). The measured finesse of more than 10 suggests the crosstalk attenuations better than 28 dB and 14 dB for 4and lo-channel demultiplexing, respectively. The discrepancy between the experimental and theoretical results was attributed to fabrication imperfections. An elegant RCE based W D M scheme has been demonstrated by Pezeshki et a1.96*97(Fig. 36) in which a waveguide is coupled into a Fabry-Perot cavity with a spatially varying cavity length. The light propagates along the waveguide, until it reaches the region with which it is resonant where it is transmitted to a dedicated photodetector. The tapered cavity spatially demultiplexes the incoming light by selecting only the band which is in resonance with the cavity at each position. Unlike the approach of Unlii et al., the incoming. light does not have to be divided between detectors. Appl. Phys. Rev.: M. Selim cnlij and S. Strite 627 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp Mirror RI iTIR) FIG. 36. Schematic of WDM device application for a tapered RCE structure (after Ref. 96). I I I 0.0 I. 880 (a) - 900 Wavelength As-grown 920 (nm) 4, ,, ‘340 use of fewer QWs thus lowering the operating voltage. In addition, at the cavity resonance, nearly all of the incident light is transmitted into the device which allows very low insertion losses to be realized. 1988 saw the first reports”9*‘oo of RCE MQW reflection modulators. The prototype GaAs/AlGaAs device of Simes et ~1.~~exhibited an ON/OFF contrast ratio of 8:l at 873 nm. Whitehead and Perry82*o* realized the potential for reduced voltage operation. The reflectivity of their device dropped from 43% to nearly zero when the applied bias was varied from 0 to 9 V corresponding to a contrast ratio of 100: 1 (Fig. 37). A modulation in excess of 10 dB was maintained over a bandwidth of 4 nm. Fritz et al.“* extended the bandwidth to 20 nm by using two cavities in the same device structure. Yan et uZ.‘~~~‘~~reported a higher F reflection modulator incorporating both top and bottom mirrors. The improved F allowed the number of QWs to be reduced to 24. As a result, the device exhibited a 47% change in reflectivity for an operating voltage swing of only 2 V, Several other groups’05-108 have reported RCE modulator structures along the same lines. In an effort to increase epitaxial yield, Karim et aL7’ combined surface recessing with an externally deposited top mirror so that the cavity resonance could always be tuned exactly to the exciton line. This approach is applicable over 350h etched i*_:j;__7ooh .-&a-a. losoh Exp. 950 975 (b) 1000 WAVELENGTH 1025 1050 (nm) FIG. 35. (a) Experimentally observed photoresponses of three devices whose resonances have been shifted by surface recessing plotted over one FSR (after Ref. 26) (b) Spectral response of similar photodetectors with semi-transparent Au top mirrors (after Ref. 78). H. Optical modulators As fiber communication bandwidth increases, the need for devices capable of very high speed optical modulation has arisen. Typical devices, such as the self electro-optic effect device (SEED)Y8 are based on the quantum confined Stark effect in which an applied electric field red shifts the exciton absorption band. Because the extent of optical modulation provided by SEED devices is limited by the amount of absorption, their contrast ratio can be improved by RCE. Increased absorption resulting from RCE also permits the 628 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 840 850 860 Wavelength 870 680 (nm) FIG. 37. Reflectivity spectra for 0 and 9 V applied bias showing high ON/ OFF ratio (after Ref. 101). Appl. Phys. Rev.: M. Selim &dii and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp 8, ;i I TRANSPARENT CONTACT I I I -I LAYER ACTIVE REGION PHASE MATCHING LAYER IJ-AIvG 1 I 0 5 DISTRIBUTED BRAGG REFLECTOR FIG. 38. Schematic layer structure of AlGaAs/GaAs resonant cavity LED (after Ref. 114). the entire family of RCE devices and offers the advantage of post-growth cavity tuning. Whitehead et a1.to9 pointed out that a high F cavity has undesirable elements including reduced optical bandwidth, higher insertion losses, and greater dependence on layer thickness and temperature fluctuations. They achieved the desired low voltage modulation by increasing the thicknesses of the 15 GaAs QWs to 150 A. At this width, the dominant electro-optic effect is ionization of the lirst exciton level which is more strongly dependent on the applied electric field. A 6.7 dB reflection contrast was obtained for a voltage swing of 3.5 V. Grindle et al.“’ and Choi et al.“’ have since reported devices of similar design and performance. Modulators have also been integrated with HPTs to perform more complicated functions.tt2 I. Light emitting I CONTACT diodes Light emitting diodes (LED) are used widely in short and medium distance optical fiber communication because of their low cost, excellent reliability and temperature insensitivity.“3 Since they produce spontaneous emission, LEDs have a zero threshold current and therefore can be more efficient than lasers in low power applications. Through the use of an optical cavity, more of the spontaneous radiation can be coupled into a narrow wavelength band whose width is solely a function of the cavity finesse. Also, the cavity increases the directivity, allowing more efficient coupling of the electroluminescence (EL) into an optical fiber. The first RCE LED (Fig. 38) was reported by Schubert et aL114 who designed a conventional LED placed in a high Q (Rt=90%, R,=99%) optical cavity. The optical cavity enhanced the spontaneous emission at its resonant wavelength by as much as a factor of 10, channeling more EL into a narrow spectral band. In addition, instead of the usual isotropic emission, RCE LEDs emit their light largely through the lower reflectivity mirror allowing the useful EL to be doubled. Care had to be taken to place the active region at a cavity antinode to benefit from the SWE,“’ and also to tune the cavity resonance to the active region exciton wavelength for optimal efficiency. 2 0 1.0 1.5 CWIENT fmd) l?lOO 1150 2 5 1250 1200 1300 WAVELENGTH 1350 1400 (nm) FIG. 39. Emission spectra of LEDs having similar designs with the exception of RCE (Device A). The sharper spectra of the RCE LED leads to a factor of 12 enhancement of the power density at resonance and a linewidth of 3 nm (0.11 kT) (after Ref. 116). After the initial demonstration, there have been several reports of RCE LEDs having improved properties.*t6-‘I9 Blondelle et al. ‘I9 noted a peak external quantum efficiency in excess of 6% in InGaAs/AlGaAs RCE LEDs with optimized Fabry-Perot cavity parameters. Hunt et al. ‘I6 reported the narrowest EL linewidth 2.8 meV (3 nmj at 1.3 ,um which was achieved by integrating InGaAsP LEDs into a cavity having R1=92% and R,=95%. Figure 39 compares the EL spectra of an RCE LED and a conventional LED with the same structure. While the total emitted power of the devices was similar, the spectral purity of the RCE device yielded a factor of 12 enhancement of the power density at resonance. The RCE linewidth of 0.11 kT, was much smaller than the 3.3 kT linewidth of the conventional LED. Because the emission mechanism of the RCE LED is spontaneous and determined only by the external cavity, the EL FWHM and wavelength were independent of the forward bias over a broad range. This is an advantage over devices relying on optica gain which have variable spectral purity and mode shifting, I”“‘.“.‘ (I) z3 x z! B 0' h r E I E ).“(‘.‘I” Ah=17nmm+' 1 +Aht0.9nm : " .a. .:' ..-. L.. .~ 600 I 620 _... _... .- ,/ ; ., JL 9.. 640 ~~~.___ .*,. *, 660 6.90 wavelength.nm I’.’ 700 -I -I 720 FIG. 40. Normalized LED output spectra highlighting the extremely narrow spontaneous emission linewidths obtainable in RCE LEDs situated in high finesse cavities (after Ref. 120). Appl. Phys. Rev.: M. Selim cnlti and S. Strite 629 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp 0, lo w& 8 G s 6 8 4 I. 400 2 450 500 550 Wavelength 600 650 700 750 (nm) FIG. 41. EL spectrum of an RCE LED (Solid line) having an organic dye active region versus a similar LED without the cavity (dotted line, scaled up by a factor of 3.5). The multi-mode resonant cavity dictates the three EL wavelengths permitted by ISE (after Ref. 122). Lott et al. 120~‘2*demonstrated red RCE LEDs fabricated from AlInGaP heterostructures grown on GaAs substrates. Devices incorporating AlAs/AlGaAs quarter wave mirrors and an Ino,s~Gaa~P QW surrounded by InGaAlP barriers emitted at 665 nm with linewidths as sharp as 0.9 nm,r2o a factor of 19 better than a comparable conventional LED (Fig. 40). Comparable performance was obtained in a structure incorporating AlInP/AlInGaP mirrors, although significantly more mirror periods were required to realize a high Q cavity.“r An organic semiconductor RCE LED has been reported to emit simultaneously at three wavelengths through the use of a multi-mode Fabry-Perot cavity.“’ The organic dye is a suitable choice for the active region since it provides gain across a wide wavelength range. The cavity enhanced the spontaneous emission at its three resonant wavelengths overlapping the gain spectrum of the active layer, producing the three primary colors together (Fig. 41). It is eventually desirable that some independent control mechanism over the relative intensities of the colors be developed. Nevertheless, this is an exciting result for color dispIay applications. Hunt et al. 123investigated the performance of InAlGaAs RCE LEDs for optical communications. An RCE LED emitting at 940 nm was compared to a conventional GaAs LED (presently used in 50 Mbit/s systems) by coupling both into a multimode, graded-index optical fiber. As a result of the narrow EL spectrum, chromatic dispersion of the RCE device signal was reduced yielding sharper rise and fall times in the output signal. After transmission through 3.37 km of fiber, the RCE LED signal had a rise/fall time of -5 ns compared to a value of = 16 ns for the conventional GaAs LED (Fig. 42). The enhanced directivity in the RCE LED resulted in an intensity which was far greater than the conventional devices, and even exceeded the theoretical maximum of an ideal isotropic LED by as much as a factor of two.‘a4 Since the EL is collapsed both into a narrow wavelength band and a narrow light cone, RCE LEDs are ideal for coupling into and transmission through optical fibers. The authors pre630 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 Time (ns) FIG. 42. Comparison of rising and falling edges of a standard ODL.50 LED and an RCE LED when driven by a square wave. Dashed-dotted and solid lines are the transition edges after transmission through 5 m and 3.37 km of 62.5 um core graded-index fiber, respectively. The sharper rise and fall times of the RCE LED refiect decreased chromatic dispersion as a result of its sharper spectrum (aktel Ref. 123). dieted that the RCE LED communication system could function at a maximum bandwidth of 200 Mbit/s compared to a limit of 66 Mbit/s for the standard device. Further bandwidth gains are possible if multiple discrete wavelength channels are transmitted and demultiplexed through a single fiber link. J. Optical amplifiers Er-doped optical amplifiers are important elements for long-haul soliton fiber optic communication systems.‘25 Generally, several meters of E-doped fiber is spliced into the transmission line. One or more semiconductor diode lasers pump the Er3+ ions at their resonant absorption wavelengths of either 0.98 or 1.48 pm. The pumping allows the signal propagating at the dispersion free 1.55 pm wavelength to be regenerated by inducing stimulated emission of the excited 4f Ers + electrons. 1267127 These pump sources are quite expensive and are only used in long haul systems where their expense comprises a small fraction of the whole. For short haul LED systems, devices based on the spontaneous emission of Er are attractive alternatives. Schubert et al.‘28-‘30 have studied both theoretically and experimentally the properties of Er-doped SiOZ situated in a resonant cavity formed by Si/Si02 mirrors. The essential device structure, grown by rf-magnetron sputtering is shown in Fig. 43 along with the distribution of the ion implanted Er. When the resonant cavity wavelength was tuned to the 1.55 pm Er3+ emission,128 a factor of 50 enhancement of the photoluminescence (PLj at that wavelength was observed compared to a sample in.which the top mirror was removed (Figure 44). Similarly, when the resonant wavelength of the cavity matched that of the 980 nm excitation source,tz9 the overall E?+ PL intensity increased by a factor of 28, although it was no longer confined within so narrow a spectral range. The enhancement of the PL in both cases was in agreement with theoretical calculations based on the cavity parameters. Appl. Phys. Rev.: M. Selim &-dii and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp (4 VI. RCE OPTICAL LOGIC AND SYSTEMS (b) Er-DOSE: 7.7x10’5cm” lqyIIEz?j T SUB&ATE STRAGGLE: 2450A 1z T FIG. 43. (a) Schematic of the Si/SiO~ VC structure and (b) the corresponding &-implantation profile (after Ref. 128). While the performance of discrete RCE devices is impressive compared to their conventional counterparts, it is expected that RCE devices- will be most useful when integrated with other RCE and conventional devices for use in photo& interconnect and Iogic systems. Practical optical logic systems wihcapitalize on the inherent advantages of light: i.e., rapid propagation over typical off-chip distances (>lO cm) and the ability to couple multiple wavelengths into a single fiber creating a whole new dimension of parallelism. The wavelength selectivity of RCE devices make them highly attractive as low crosstalk optical interconnects. RCE detectors integrated with LEDs or lasers can function as wavelength selective optoelectronic switches or logic gates. The most elaborate schemes proposed to date describe cascadable, optical input/output, optoelectronic logic circuits in which the RCE components permit the wavelength to function as an additional variable. A. integration The advantage of this optical amplifier is that the RCE effect, by greatly increasing the absorption of the Er3’ ions, offers the potential of a much more compact and lower cost design since conventional pumps require several meters of B-doped fiber. Furthermore, the pump light has to be coupled into the fiber (typically with a 50% loss) which must then be spliced into the transmission line. RCE Er-doped amplifiers allow the pump light to be coupled into the gain region perpendicularly and with nearly 100% efficiency so that the amplification occurs over a much smaller volume. The challenge of implementing this scheme is coupling the signal into and out of the relatively narrow gain region which would also involve losses. T-- iz I 7- t 5 - z F z 4 8s 3- In order to have an optical logic element, three functions must be satisfied, namely detection, amplification and emission. One of the simplest ways to accomplish this is by integrating an HPT and LED in a vertical n-p-n-p configuration.61*‘31-135 In such a device (Fig. 45) the upper two heterojunctions form the HPT whose collector current drives the LED in the bottom heterojunction. The resulting structure is a photothyristor (or a Shockley diode) which is also referred to as a light amplifying optical switch (LAOS).131 The LAOS is designed so that some light from the LED is re-absorbed by the HPT. When this positive feedback mechanism exceeds a threshold, the LAOS switches to a low impedance state in which a large optical output is generated. Depending on the external circuit and adjacent device interconnections, one or more LAOS devices can function as bistable optical switches, optical inverters, AND, NAND, and NOR gates. The LAOS device structure is suitable for RCE,‘36 making these devices ideal for WDM optical interconnects. SilSiOgEr I--L WITH CAVITY -‘- -.-. WITHOUT 5- with LEDs T=300K 3” c- of detectors CAVITY - InP Substrate, Semi-Insulating ,-“C -.“ 1450 __ . ..f 2.’ 1500 WAVELENGTH y..-n V..“‘.. 1550 k(nm) FIG. 44. Room temperature photoluminescence spectra of Er-implanted Si/SiO, structures with and without resonant cavities (after Ref. 128). 1 Output Light FIG. 45. Schematic of the LAOS structure. Input light (top) turns on the HPT which drives the LED (below) (after Ref. 135). J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 Appl. Phys. Rev.: M. Selim lklij and S. Strite 631 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp W gate/emitter vo R P v SEL hvout R ” AlGaAsl AlAs subcollector 1 e HPT hVin HPT i FIG. 46. Schematic of the DOES (after Ref. 141). band A similar bistable optical device named the double heterostructure optical switch (DOES)‘372’38 (Fig. 46) is an extension of the BICFET (Section V F). The key to the device operation is the band bending introduced by the p-type charge sheet and the n-type AlGaAsIGaAs heterojunction. Under increasing optical or electrical forward bias, the DOES behaves much like the LAOS in that it switches from a high to a low impedance, tight emitting state. In general, the broadband output of the LED limits the usefulness of these devices by limiting the cascadability and increasing the crosstalk. A monochromatic laser output is much more desirable as an input signal for a next level of optical logic. As a result, attention has been redirected towards incarnations of these devices incorporating laser OUtPUtS84-86,88,139-143 which can be achieved by placing the active device structure in either a vertical or horizontal cavity. band B .- hVin Ei A FIG. 47. Band diagram of a vertically integrated VCSIZLJHIT. Incoming light (bottomj is absorbed, turning on the HPT, whose output drives the VCSEL (above) (after Ref. 146). with an intermediate selective etch (Fig. 48). The goal was to integrate a low threshold, high efficiency VCSEL and a high 7 HPT with a broader spectral bandwidth. The etch step enabled the RCE HPT to be situated in a double VC which provided a broadened response spectrum, independent from that of the laser. Figure 49(a) shows the VCSEL output linewidth of 1.5 A with a differential efficiency of 0.25 W/A. Due to its separate cavity design, the RCE HPT had a 50 A FWHM response [Fig. 49(b)]. Numai et ~2.‘~~ reported a more compact optical logic element consisting of a thyristor situated in a VC which they B. Vertical cavity lasers integrated with named VSTEP (Fig. 50). The VSTEP functions as a wavephotodetectors length selective photodetector and VCSEL emitter. Optical or electrical excitation switches the thyristor into its low imVertical cavity surface emitting lasers (VCSEL) offer pedance state which results in lasing.‘51p’52 Under external several important advantages over edge emitters which make them attractive for integration into photonics systems.27*144 electrical bias, the VSTEP is capable of amplification or switching within a single wavelength band. A VSTEP elecMany of the WDM approaches applied to RCE detectors can be extended to VCSELs. Also, VCSELs are structurally similar to RCE detectors and modulators and can therefore be fabricated alongside one another or a single device can perform both functions.145 Such a combination provides both wavelength selective photodetection and narrow band emission, ideal for multiple wavelength photonics. The first all optical logic element incorporating a VCSEL’46 was realized by growing the laser on top of an HPT engineered to detect wavelengths at which the VCSEL was transparent (Fig. 47). When illuminated, the HPT collecSI-GaAs Substrate tor current served to drive the laser. This device is cascadable and is therefore suitable for optical logic. Two other groups 147~148 have reported similar structures. However, for most applications, a planar integration rather than a vertical one is desirable. Kosaka et a1.‘49 realized a planar integration of a RCE FIG. 48. Horizontally integrated H!?T and VCSBL. The HPT is situated in a HPT and thyristor VCSEL using a two stage growth process double VC in order to broaden its response spectrum (after Ref. 149). 632 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 Appl. Phys. Rev.: M. Selim l&Iii and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp preach is cascadable, dynamically reconfigurable, and can in principle perform arbitrarily complex logic operations. It is also complicated due to the need to fabricate two separate types of devices in a non-planar geometry. The PTs, being outside of the VCSEL, are not highly wavelength selective leading to crosstalk concerns. C. Wavelength selective optical logic What is required for a practical, low crosstalk W D M WAVELENGTH (nm) WAVELENGTH (nm) system is a scheme in which both the emitters and detectors FIG. 49. (a) Emission spectrum of the VCSEL from Fig. 47, and (b) photocurrent response of the HPT together with the calculated absorption spectrum (after Ref. 149). trical to optical conversion efficiency as large as 1 1.4%“53 was achieved by coating the device sidewalls with a reflective Au layer to reduce spontaneous emission losses.‘54,‘55 A similar device consisting of a photothyristor (PT) and VCSEL has been reported by Z~OU.‘~~-*‘~ The PT is grown on top of the VCSEL, not inside the optical cavity. Individual devices are fabricated in a horizontal array (Fig. 51) and interconnected to perform various logic operations. This ap- i -Ak.2sGao.7sAs 1ooaxs i - Ino.zGao.sAs i -AIor25Gaa75As p-Alo.25Gao.75A.s n-Alo.sGao.sAs n-AIAs/Ga L- ~- n-GaAs As mirror Sub. i:? FIG. 50. Schematic of the VSTEP device structure. The thyristor situated within the VC functions both as a detector and VCSEL, leading to a compact, highly wavelength selective optical logic element (after Ref. 150). operate in very narrow wavelength bands which will permit a large number of channels with low crosstalk. Willner et al. Is9 have proposed such a scheme in the form of a two dimensional optical interconnects composed of arrays of VCSELs cascaded through wavelength selective detector arrays (Fig. 52). The pixels are defined by a VCSEL array in which one device emits at each of the different system wavelengths. Each successive detector array detects one of the VCSEL wavelengths while remaining transparent to the remaining wavelengths. This scheme realizes both wavelength selective emission and detection, but it is complex and bulky requiring N+ 1 precisely aligned cascaded array planes for N channel WDM. A simpler wavelength selective optical logic (WSOL)r2 can be realized from the basic structure of Zhou et al.‘58 if the PT is also placed inside of an optical cavity. The major advantage of this scheme is that the top mirrors are formed after surface recessing which permits each laser and PT to be tuned during the fabrication process to an individual narrow wavelength band. ‘2*27,79,90~‘VCSELs 61 tuned by surface recessing before top mirror deposition have recently been demonstrated by Wipiejewski et al. 162 The WSOL device is well suited for optical interconnect applications. The low crosstalk enables a closely packed multiple wavelength array to be fabricated whose light is coupled in and out of a single optical fiber or waveguide. Such a cor@uration greatly reduces the total number of optical fibers in and out of the interconnect chip, thereby reducing the complexity and increasing the reliability of optoelectronic signal conversion. The enhanced 7 and gain of each device can compensate for the light intensity lost from sharing. Different horizontal integrations of the WSOL VCSELs and PTs function as completely optical OR (Fig. 53) or AND (Fig. 54) gates. An XOR gate can be implemented in a six mesa structure coupled to a single fiber. The XOR output wavelength can be chosen as any within the range of the tunable VCSEL. This structure brings about a considerable simplification to a previously described two level single wavelength optical input/output XOR gate requiring nine mesas at each level.*60 WSOL devices can also be designed to perform different logic functions for different sets of input wavelengths.‘” Another important aspect of the WSOL device is its cascadability which enables sequential logic operations by placing wafers in series. In the cascaded configuration, wavelength can be used as an independent variable which greatly increases the flexibility of WSOL circuits. Figure 55 illustrates how a FULL ADDER can be fabricated from only two J. Appt. Phys., Vol. 78, No. 2, 15 July 1995 Appl. Phys. Rev.: M. Selim finki and S. Strite 633 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp LIGHT IN PHOTOTHYRISTOR VERTICALCAVITY SURFACE-EMITTING FIG. 51. Horizontal integration of a PT and VCSEL (after Ref. 156). layers of WSOL and three input wavelengths. The low crosstalk resulting from wavelength selectivity relaxes the requirements for accurate alignment between the wafers forming the different levels. In wavelength selective logic each local and global variable can be represented by a different wavelength. in a synchronous system, a global clock wavelength can be broadcast over the entire system eliminating deleterious clock skew effects. Instead of forming logic circuits by means of directly configuring the wavelength selective optoelectronic switches, a hybrid approach can also be employed. In this approach, the optical input variables are coupled into an array of WDM detectors through a single fiber. Once the input is demultiplexed, arbitrarily complex logic operations of many variables can be performed by conventional electronic circuits and the optical output can be provided by a multiple wavelength VCSEL array as depicted in Fig. 56.163 VII. TOWARDS A PRACTICAL RCE WDM SYSTEM RCE is a conceptually simple technology which is vastly preferable in potential price and performance to today’s com- 2-D WDM Optical Interconnects hDependent Detector Arrays VCSEL Multi-?. VCSEL Pixel c hD Plane A B c D FIG. 52. Optical plane interconnect scheme incorporating a twodimensional pixel array of identical three wavelength VCSEL mini-arrays and sequential wavelength selective detector arrays (after Ref. 159). 634 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 mercial WDM systems.‘@ It is relatively easy to understand how RCE affects common device structures and to predict how more complicated, integrated systems might perform. However, the complexity of implementing RCE has hindered its development. The highly precise vertical structures require sophisticated and expensive crystal growth techniques which nevertheless do not offer high yield. VCSELs continue to suffer from high series resistance and other difficulties which have thus far kept them from realizing their potential. The horizontal and vertical integration schemes discussed in the previous section require fairly elaborate fabrication techniques. To realize a practical, low cost prototype RCE photonic system technology, it will probably be necessary to simplify the structures and performance goals to decrease the complexity of the growth and fabrication. For this reason, RCE will likely only find applications in high performance systems in which improved speed and/or WDM are required. A number of simplifications arising from the nature of RCE can lend themselves towards realizing a practical RCE WDM architecture. MBE growth yield may be increased by monitoring the layer refiectivity and using high energy electron diffraction during growth permitting in situ corrections.t6’ Devices can be designed with sacrificial top layers for surface recessing. Epilayers then can be pre-tuned before device fabrication and top mirror deposition.‘61*79 Fabrication might also be considerably simplified if a simple VCSEL structure were used as both detector and emitter’45 at the expense of some device performance. These technologies together are realizable, suitable for WDM, and potentially high yield and low cost. WDM systems would be greatly improved by an innovation which enables dynamic tuning of the RCE component devices. An architecture can be envisioned where an optical data stream carries several multimedia channels which the user can access by tuning an RCE detector. A tunable VCSEL and RCE photodetector would allow a single PC user to easily send and receive video, data and voice over the wavelength channels dedicated to each. Appl. Phys. Rev.: Pvl.Selim cnlti and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp Lightin( 1, A,,) n-substrate FIG. 53. Structure of a WSOL OR gate with an equivalent circuit representation. The input and output wavelengths are set during the fabrication (after Ref. 12). With such applications in mind, efforts towards achieving dynamic tuning by modulating the optical length of the VC through refractive index variations have been undertaken. Simes et LzZ.‘~observed that an electrical bias applied across the QW active regions of their RCE reflection modulators was sufficient to provide a modest index shift leading to a 10 a shift of the cavity resonance. Blum et aZ.1667’67 applied an electrical bias across QWs positioned inside of the DBR mirrors. They succeeded in altering the mirror reflectivity but have not applied their technique towards actual device tuning. Lai and Campbell65 have predicted that this effect can provide as much as 20 nm of tuning in the GaAs/ AlGaAs system. The largest continuous tuning range in a VC device attained to date is 2.2 nm at 970 nm by Wipiejewski et ~1.‘~’This was achieved in a three terminal VCSEL structure in which a tuning current flowed over one DBR mirror causing ohmic heating and a corresponding shift in the refractive index. None of the above approaches is really satisfactory for a practical WDM system which will require multiple channels, preferably over the entire 1.3 - 1.55 pm low dispersion band. A dynamic, wide range tuning mechanism for RCE remains a major goal of device researchers. Pezeshki and Harrislc9 have proposed combining VC lasers and detectors with micro-machined, movable Al mirrors (Fig. 57). In this approach, conventional RCE structures with a DBR bottom mirror are grown. A freely suspended Al top mirror is then fabricated using standard micro-machining techniques. The distance between the Al mirror and the episurface, and therefore the cavity length, can then be controlled by electrostatic deflection of the mirror. One difficulty of the proposed struc- ----_ 9 ,’ - - - - - - - - - - - -* ACAt) + HalfAdder r------____---_________ Light In ( Cl”(Al) 4 t A B Ci, h,) :I,) FIG. 54. Structure of a WSOL AND gate (after Ref. 12). J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 B (Id =AQB@C c0.t Q 3) FIG. 55. Conceptual representation of a FULL ADDER implemented in WSOL by cascading two layers (after Ref. 12). Appl. Phys. Rev.: M. Selim cnlij and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp 635 Multiple wavelength digital optical input Multiple wavelength digital optical output 5 d multi-h VCSELs multi-l switches .ptlcal(iber\ ru-fl-&: lI!z‘t h 01P J-11- Ti;i, km= j;;,+k,, i FIG. 56. Conceptual implementation of multipIe wavelength optical interconnects (after Ref. 163). ture is maintaining the cavity Q which requires keeping the mirror face flat and perpendicular to the cavity axis. The most impressive demonstration to date of this concept was an optical reflection modulator consisting of a SiN, membrane fabricated onto a doped Si substrate.‘70 A fairly weak RCE effect was in evidence since no mirrors were incorporated into the structure. Nevertheless, it was shown that the membrane could be modulated (and therefore, the cavity response could be tuned) on a 1 ,US time scale. A more robust implementation of the electrostatic tuning scheme is a Au mirror fabricated on the underside of a SiN, membrane. Figure 58 is an illustration of an optimized LED/PD structure in the InGaAs/AlGaAs material system. The device is based on an extended vertical cavity (EVC) defined below by the DBR mirror and above by a hybrid mirror consisting of a several period DBR, a sacrificial GaAs spacer layer, and the Au/air interface. Due to the SWE, two or more QWs are required to maintain a flat device response over a large wavelength range. The asymmetric positioning of the QWs equalizes the electron and hole transit times to improve the device bandwidth. By making the arms of the membrane thin, mirror tilt and buckling should be minimized. Simulations show that both AR coatings are critical. n-G& substrate FIG. 58. Tunable RCE emitter/detector EVC device structure (after Ref. 171). The lower coating insures that the maximum amount of light is coupled into or out of the cavity. The upper AR coating suppresses multiple reflections between in the air gap. The device structure is tunable across 100 run (Fig. 59) with a fiat response ( 77=90+ 5%) across the entire range. The tuning range is limited by the DBR mirror stop band width, not the Q W or membrane tuning mechanism. Eight channel operation within this range can be realized with only a 125 crosstalk. The EVC scheme is promising because the resonant cavity can be tuned across its full FSR inside of a ,us. Once tuned, the bandwidth is dictated by the RCE LED or PD and will be in the 200 MHz on up to GHz range. A single EVC device (Fig. 60) would replace the entire tapered waveguide array of Pezeshki et aLg6 (Fig. 36). Since only a single wave- rdy suspenaea #8 Al mirror i #7 #6 # #4 II ii Ii I i ! i ! ’ I i ! 1 f I , ( I t t I i i I I I 1 , L t i, active cavity region II - substrate YUO FIG. 57. Proposed RCE laser/detector in which the cavity width is modulated by electrostatic deflection of a micro-machined AI top mirror (after Ref. 169). 636 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 YlLI 920 930 940 950 960 wavelength(nm) 970 980 990 1000 FIG. 59. Spectral response of an EVC device designed for 8 channel operation in the 900-1000 n m band. They axis corresponds to EL intensity (LED case) or generated photocurrent (PD case) (titer Ref. 171). Appl. Phys. Rev.: M. Selim &lli and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp I substrate I FIG. 60. Coupled EVC-waveguide device. Gnly a single wavelength is selected from the optical data bus. The non-resonant wavelengths are free to propagate to other users. length is detected, the remainder can continue to propagate. The permits multiple users to share the same optical bus, even at remote locations, without expensive and time consuming optical switching or signal regeneration between nodes. The EVC scheme would also increase the flexibility of the multi-wavelength LED of Dodabalapur et al. lz2 (Fig. 31) which electroluminesced at three wavelengths, but simultaneously and without control of the relative intensities. By seIecting the EL wavelength with an electrostatically tunable membrane, an organic dye LED could be tuned continuously across the visible band. LEDs with narrower gain bands could be switched on and off, or between several emission lines, by membrane modulation. In these implementations, the wavelength and the brightness would be independent. VIII. CONCLUSIONS This review article has attempted to describe both the physics and the applications of semiconductor devices situated within Fabry-Perot microcavities. Placing a device inside of a microcavity was shown to have two major advantages. Due to the wavelength selectivity of the microcavity, only resonant wavelengths are admitted or emitted. Resonant radiation undergoes multiple reflections within the cavity, causing its intensity to by amplified. The wavelength selectivity and field amplification of the resonant cavity has some important favorable effects on many common device structures, and we therefore refer to such devices as resonant cavity enhanced (RCE). RCE photodetectors adopt the highly wavelength selective response of the cavity in which they are situated. They also benefit greatly from the increased amplitude of the resonant field (or in an equivalent picture, the multiple passes of each resonant photon), allowing much thinner absorption re gions to achieve nearly unity quantum efficiency. Thinner absorption regions allow higher transit time limited bandwidths due both to shorter transit distances and sharper rise/ fall characteristics. The wavelength selectivity makes RCE photodetectors attractive for low crosstalk wavelength demultiplexing. LED structures can also be integrated into a VC which radically alters their spontaneous emission spectra. The overall radiative efficiency is improved, and the linewidth of the emission becomes limited by the cavity finesse rather than thermal effects. RCE LEDs have radically sharper spectra, increased directivity and spectral power density compared to conventional designs which permit more optical power to be coupled into optical fibers and higher bandwidth due to reduced dispersion. RCE detectors and phototransistors have vertical structures which are quite similar to VCSELs, simplifying integration. Several optical logic schemes have been proposed which combine the sharp response spectra of RCE detectors with the nearly monochromatic output of VCSELs, resulting in cascadable optical input/output circuits which are capable of incorporating the wavelength as an additional logical variable. Present day optical logic schemes will increase dramatically in power and flexibility when dynamic tuning of individual elements is fully incorporated. For these reasons, RCE devices can be expected to play a growing role in optoelectronics over the coming years. The continuing push towards higher bandwidth plays right into the natural capabilities of RCE devices for greater speed and wavelength demultiplexing. ACKNOWLEDGMENTS This work was supported in part by the National Science Foundation under the grant .No. 9309607. The authors wish to acknowledge the contributions of Professor K. Kishino and Professor A. MorkoG to the development of this work and all of our colleagues who have graciously provided preprints and original figures. ‘A. Perot and C. Fabry, Astrophys. J. 9, 87 (1899). ‘G. 8. Airy, Philos. Mag. 2, 20 (1833). ‘Goedbloed and Joosten, Electron. Len. 12, 363 (1978). 4R. ti. Hunsperger, Integrated Optics: Theoq and Technology ispringer, New York, 1991). 5P. L. Gourley and T. J. Drummond, Appl. Phys. Lett. 49, 489 (1986). “B. S. Ryvkin, Sov. Phys. - J. Technol. Phys. Lett. 5, 2.5 (1979). 7B. S. Wherrett, IEEE J. Quantum Electron. 20, 646 (1984). *K. Kishino. M. S. &Iii, J. I. Chyi, J. Reed, L. Arsenault, and H. Morkoc, IEEE J. Quantum Electron. 27, 2025 (1991). ‘A. Chin and T. Y Chang, J. Lightwave Technol. 9, 321 (1991). “J. Farhoomand and R. E. Murray, App 1. Phys. Lett. 58, 622 (1991). “M. S. I&ii, K. Kishino, H. J. Liaw, and H. Morkoc, J. Appl. Phys. 71, 4049 (1992). “M. S. enhi, S. Suite, A. L. Demirel, S. Tagusn, A. Salvador, and H. Morkoc, IEEE J. Quantum Electron. 29, 411 (1993). 13S. L. Daryanani and G. W. Taylor, Opt. Quantum Electron. 25, 123 (1993). I4 R. E. A. Saleh and M. C. Teich, Fundamentals of Photonics (Wiley, New York, 1991). “J. T. Verdeyen, Laser Electronics (Prentice Hall, Englewood Cliffs, 1981). t6S. C. Corzine, R. S. Geels, J. W. Scott, R. H. Yan, and L. Coldren, IEEE J. Quantum Electron. 25, 1513 (1989). 17M. S. ~nlii, K. Kishino, J. I. Chyi, J. Reed, S. Noor Mohammad, and H. Morkoc, Appl. Phys. Lett. 57, 750 (1990). ‘*H Haus Waves and Fields in Optoelectronics (Prentice Hall, London, England;, pp. 41-46. 19D K Cheng, Field and Wave Electromagnetics (Addison Wesley, New York,’ 1989), pp. 368-369, 451. Appl. Phys. Rev.: M. Selim finlti and S. Strite 637 J. Appt. Phys., Vol. 78, No. 2, 15 July 1995 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp “0. Madelung 1Editor 7Physics of Group NElements and III-V Compounds, edited by Landolt-Biirnstein new series vol. III/17a (Springer, New York, 1982). ‘IO. Aytik and F? Kumar, Opt. Lett. 15, 390 (1990). “P. B. Johnson and R W. Christy, Phys. Rev. B 6,437O (1972). ‘3A. V. Sokolov, Optical Propetiies of Metals (American Elsevier, New York, 1967), pp. 9-17. “Y. Suematsu, S. Arai, and K. Kishino, IEEE J. Lightwave Technol. 1, 161 (1983). z5R. H. Yan, R. I. Simes and L. A. Coldren, IEEE J. Quantum Electron. 25, 2272 (1989). 26M. S. Unlii, K. Kishino, J. I. Chyi, L. Arsenault, J. Reed, and H. Morkoc, Electron. Lett. 26, 1857 (1990). “7C. Chang-Hasnain, J. P Harbison, C. Zah, M. W. Maeda, L. T. Florez, N. G. Stoffel, and T.-I? Lee, IEEE J. Quantum Electron. 27, 1368 (199 1). ‘sE. M Purcell, Phys. Rev. 69, 681 (1946). “G. Gabrlelse and H. Dehmelt, Phys. Rev. Lett. 55, 67 (1985). 30R. G. Hulet, E. S. Hilfer, and D. Kleppner, Phys. Rev. Lett. 55, 2137 (1985). 3’E. Yablonovitch, Phys. Rev. L&t. 58, 2059 (1987). 32H. Yokoyama, Science 256, 66 (1992). 33G. Bjiirk, S. Machida, Y. Yamamoto, and K. Igeta, Phys. Rev. A 44, 669 (1991). 34Y. Yamamoto, S. Machida, K. Igeta, and G. Bjijrk, Coherence, Amplijication, und Quantum Effects in Semiconductor L.asers, edited by Y. Yamamoto (Wiley, New York, 1991). sSN. E. J. Hunt, E. F. Schubert, R. A. Logan, and G. J. Zydzik, Appl. Phys. Let. 61,X287 (1992). 36S. S. Mmtaza, A. Srinivasan, Y. C. Sbih, J. C. Campbell, and B. G.-Streetman, Electron. Lett. 30,643 (1994). 37S. Suite, M. S. &hi, K. Adomi, G.-B. Gao, A. Agarwal, A. Rocket& H. Morko9, D. Li, Y. Nakamura, and N. Otsuka, J. Vat. Sci. Technol. B 8, 1131 (1990). 38H. Ten&in, T. P. Pearsall, J. C. Bean, R. A. Logan, and S. Luryi, Appl. Phys. Lett. 48, 963 (1986). 39S. Murtaza, J. Campbell, J. C. Bean, and L. J. Peticolas, Electron. Lett. 30, 315 (1994). “R. Kuchibhotla, J. C. Campbell, J. C. Bean, L. Peticolas, and R. Hull, Appl. Phys. Lett. 62, 2215 (1993). ’ “Y Kohama K. U&da, T. Soga, T’. Jimbo, and M. Umeno, Appl. Phys. Lit. 53, 862 (1988). “U. Prank, M. Mikulla, and W. Kowalsky, Appl. Pbys. Lett. 62, 129 (1993). j3K. Litvin, J. Burm, D. Woodward, W. Schaff, and L. F. Eastman, lEEE Microwave and Millimeter-wave Monolithic Circuits Symposium, Atlanta, Georgia, June 1993. 44M. S. &hi, B. M. Onat, and Y. Leblebici, IEEE J. Lightwave Technol. 13, 406 (1995). 45S. M . Sze , Semiconductor Device Physics and Technology (Wiley, New York. 1985). 46J E. Bowers and C. A. Burrus, Jr., IEEE J. Lightwave Technol. 5, 1339 (i987). j7G. Lucovksy, R. F. Schwarz, and R. B. Emmons, J. Appl. Phys. 35, 622 (1964). ‘sM. Dentan and B. Cremoux, IEEE J. Lightwave Technol. 8, 1137 (1990). 49J. Hollenhorst, IEEE J. Lightwave Technol. 8, 531 (1990). “Y. Leblebici, M. S. Unlil, H. Morkoq, and S. M. Rang, Proceedings IEEE of the International Symposium Circuits and Systems, 1992, pp. 895-898. ‘lM. S. &hi, Y. Leblebici, S. M. Kang, and H. Morkop, IEEE Photon. Technol. Lett. 4, 1366 (1992). ‘aY. Leblebici, M. S. &hi, S. M. Kang, and B. M. Onat, J. Lightwave Technol. 13, 396 (1995). s3A. Chin and T. Y. Chang, J. Vat. Sci. Tecbnol. B 8, 339 (1990). s4H D. Law K. Nakano, and L. R. Tomasetta, lEEE J. Quantum Electron. 15, 549 (1979). ssT. P. Pearsall, IEEE J. Quantum Electron. 16,709 (1980). 56N Susa H. Nakagome, H. Ando, and H. Kanabe, IEEE J. Quantum Electron. 17: 243 (1981). 57R Kuchibhotla, A. Srinivasan, J. C. Campbell, C. Lei, D. G. Deppe, Y. S. He, and B. G. Streetman, IEEE Photon. Technol. Lett. 3, 354 (1991). “T P Lee C. A. Burms, Jr., and A. G. Dentai, IEEE J. Quantum Electron. l;, 232 (1981). 59M Brain and T. P. Lee, IEEE J. Lightwave Technol. 3, 1281 (1985). 60D.. H. Auston, IEEE J. Quantum Electron. 19, 639 (1983). 638 J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 *‘M. S. Unlii, S. St&e, A. Salvador, A. L. Demirel, and H. Morko9, IEEE Photon. Technol. Lett. 3, 1126 (1991). ‘*A. G. Dentai, R. Kuchibhotla, J. C Campbell, C Tsai, and C. Lei, Electron. Lett. 27, 2125 (1991). ‘s A. Dodabalapur, T. Y. Chang, and S. Chandrasekhar, J. Vat. Sci. Technol. B 11, 919 (1993). MF. Y. Huang, A. Salvador, X. Gui, N. Teraguchi, and H. Morkop, Appl. Phys. Lett. 63, 141 (1993). 6SK Lai and J. C. Campbell, IEEE J. Quantum Electron. 30, 108 (1994). 66N’ E. J. Hunt, E. F. Schubert, and G. J. Zydzik, Appl. Phys. Lett. 63,391 (1.993). 67B. Corbett, L. Considine, S. Walsh, and W. M. Kelly, Electron. Lett. 29, 2148 (1993). @B. N. Sverdlov, k E. Botchkarev, N. Teraguchi, A. Salvador, and H. Morko9, Electron. Lett. 29, 1019 (1993). 6y0. Wada, H. Nobuhara, H. Hamaguchi, T. Mikawa, A. Tackeuchi, and T. Fujii, Appl. Phys. Lett 54, 1617 (1989). 7oZ.-M. Li, D. Landheer, M. Veilleux, D. R. Conn, R. Surridge, J. M. Xu, and R. I. McDonald, IEEE Photon. Technol. Lett. 4,473 (1992). 7’U. Prank and W. Kowalsky, Jpn. .I. Appl. Phys. 32, 574 (1993). 72J. C. Campbell, A. G. Dentai, C. A. Burrus, Jr., and J. F. Ferguson, IEEE J. Quantum Electron. 17, 264 (1981). 73F. Capasso, W. T. Tsang, C. G. Bethea, A. L. Hutchinson, and B. F. Levine, Appl. Phys. Lett. 42, 93 (1983). 74N. Chand, J. Klem, and H. Morkoq, Appl. Phys. Lett. 48, 484 (1986). 75J. K. Twyham, P. A. Claxton, R. C. Woods, and D. R. Wright, Electron Lett. 25, 86 (1989). 76J. C. Campbell, W. T. Tsang, and G. J. Qua, IEEE Electron Device Lett. 8, 171 (1987). 77R P Bryan G. R. Olbright, W. S. Fu, T. M. Brennan, and J. Y. Tsao, Appl. Phys. Lett. ;9, 1600 (1991). ‘*M. S. Unlii, S. Tagtran, A. L. Demirel, and H. Morkop, Proceedings of the 1992 Bilkent International Conference on Lightwave Communication, Ankara, Turkey, 1992. “2 Karim, C. Kyriakakis, A. R. Tanguay, Jr., K. Hu, L. Chen, and A. Madhukar, Appl. Phys. Lett. 64, 2913 (1994). soA. Dodabalapur and T. Y. Chang, Appl. Phys. Lett. 60, 929 (1992). s’G. W. Taylor and J. G. Simmons, IEEE Trans. Electron Devices 32,2345 (1985). x2M. Whitehead and G. Parry, Electron. J.&t. 25, 566 (1989). s3P Cooke, G. W. Taylor, and P. Claisse, IEEE Photon. Technol. Lett. 2, 537 (i990))). 84P. Cooke, P. A. Evaldsson, G. W. Taylor, and B. Tell, Electron. Lett. 27, 1095 (1991). ssG. W. Taylor, P. R. Claisse, and P. Cooke, Appl. Phys. Lett. 58, 2957 (1991). K6P Cooke, P. Evaldsson, and G. W. Taylor, IEEE Photon. Technol. Lett. 4, 6b5 (1992’). “S. Daryanani, G. W. Taylor, P. Cooke, I? Evaldsson, and T. Vang, Appl. Phys. Lett. 59, 3464 (1991). ‘*l? Evaldsson, S. Daryanani, P. Cooke, and G. W. Taylor, Optical and Quantum Electron. 24, S133 (1992). “S. L. Daryanani, G. W. Taylor, S. K. Sargood, T. Vang, and B. Tell, IBEE Photon. Technol. Lett. 5, 677 (1993). 9oS. L. Daryanani, G. W. Taylor, T. Vang, S. &good, B. Tell, D. Wendling, and L. Harriet, Proc. SPIE 2022, 248 (1993). “H. Uetsuka, N. Kurosawa, and K. Imoto, Electron. Lett. 26, 251 (1990). ‘“T. Baba, S. Tamura, Y. Kokubun, and S. Watanabe, IEEE Photon. Technol. Lat. 2, 191 (1990). a3A. Larsson, P. A. Anderkson, P. Andersson, S. T. Eng, J. Salzman, and A. Yariv, Appl. Phys. Lett. 49, 233 (1986). a4H. Takahashi, S. Suzuki, K. Kato, and I. Nishi, Electron. Lett. 26, 8788 (1990). “F? E. Green, Fiber Optic Networks (Prentice Hall, Englewood Cliffs, NJ, 1992). 96B. Pezeshki, F. K. Tong, J. A. Kash, D. W. Kisker, and R. M. Potemski, IEEE Photon. Technol. Lett. 5, 1082 (1993). 97B. Pezeshki, E E Tong, 5. A. Kash, and D. W. Kisker, IEEE J. Lightwave Technol. 12, 1791 (1994). “A. L. Lentine and D. A. B. Miller, IEEE J. Quantum Electron. 29, 655 (1993). “R. J. Simes, R. H. Yan, R S. Geels, L. A. Coldren, J. H. English, A. C. Gossard, and D. G. Lishan, Appl. Phys. Lea. 53,637 (1988). Appl. Phys. Rev.: M. Selim lhlij and S. Strite Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp ‘O”Y H. Lee. J. L. JewelI, S. J. Walker, C. W. Tu, J. P. Harbison, and L. T. Florez, Appl. Phys. Lett. 53, 1684 (1988). “‘M. Whitehead, A. Rivers, G. Parry, J. S. Roberts, and C. Button, Electron. Lett. 25, 985 (1989). “‘L J. Fritz, J. F. KIem, and I. R. Wendt, Appl. Phys. Lett. 59, 753 (1991). lo3R.-H. Yan R. J. Simes, and L. A. Co&en, IEEE Photon. Technol. Lett. 2, 118 (199Oj. ‘“R.-H. Yan, R. J. Simes, and L. A. Coldren, IEEE J. Quantum Electron. 27, 1922 (1991). “‘J. Woodhead, I? A. Claxton, R. Grey, T E. Sale, J. P R. David, L. Liu, M. A. Pate, G. Hill, and I’. N. Robson, Electron. Lett. 26, 2118 (1990). IMA. J. Moseley, J. Thompson, M. Q. Kearley, D. J. Robbins, and M. J. Goodwin, Electron. Lett. 26, 913 (1990). “‘3. E Heffeman, M H. Moloney, I. Hegarty, and J. S. Roberts, Electron. Lat. 27, 659 (1991). “*I. J. Fritz, D. R. Myers, 0. A. Vawter, T. M. Brennan, and B. E. Hammons, Appl. Phys. Lett. 58, 1608 (1991). ‘OPM. Wbitehead, A. Rivers, and G. Parry, Electron. I&. 26, 1588 (1990). “OR J Grindle, J. E. Midwinter, and J. S. Roberts, Electron. Lett. 27, 2329 (lb&). “‘Y-W. Choi, 0. Kwon, and E. Lee, IEEE Photon. Technol. Lett. 5, 1406 (l-993). “‘C. Amano, S. Matsuo, and T. Kmokawa, IEEE Photon. Technol. Lett. 3, L 736 (1991). “‘R H Saul, T. P. Lee, and C. A. Bum’s, Semiconductors and Semimetals, (Academic, London, 1985), Vol. 22, pp. 193-237. ‘14E.F. Schubert, Y.-H. Wang, A.Y. Cho, L.W. Tu, and G.J. Zydik, Appl. Phys. Lett. 60, 921 (1992). *15N. E. J. Hunt, E. E Schubert, D. L. Sivko, A. Y. Cho, and G. J. Zydzik, Electron. Lett. 28, 2169 (1992). “‘N. E. J. Hunt, E F. Schubert, R A. Logan, and G. J. Zydzik, I992 IEDM Technology Digest (IEEE, Piscataway, NJ, 1992), pp. 651-654. “‘U. Keller, G. R. Jacobovitz-Veselka, J. E. Cunningham, W. Y. Jan, B. Tell, and K. E Brown-Goebeler, Appl. Phys. Lett. 62, 3085 (1993). ‘**B. Corbett, L. Considine, S. Walsh, and W. M. Kelly, IEEE Photon. Technol. Lett. 5, 1041 (1993). “‘5 Blondehe, H. De Neve, P. Demeester, P. Van Daele, G. Berghs, and R. Baets, Electron Lett. 30, 1787 (1994). “‘5 A. Lott, R. P. Schneider, Jr., G. A. Vawter, J. C. Zolper, and K. J, klloy, Electron. Lett. 29, 328 (1993). 12’J. A. Lott, R P. Schneider, Jr.. J. C. Zolper, and K. J. Malloy, Photon. Teclmol. Lett. 5, 631 (1993). ‘=A. Dodabalapur, L. J. Rothberg, and T. M. Miller, Appl. Phys. Lett. 65, 2308 (1994). ‘=N. E. J. Hunt, E. F. Schubert, R. E Kopf, D. L. Sivko, A. Y. Cho, and G. J. Zydzik, Appl. Phys. Lett. 63, 2600 (1993). IwE. E Schubert, N. E. J. Hunt, M. Micovic, R. J. Malik, D. L. Sivco, A. Y. Cho, and G. J. Zydik, Science 265, 943 (1994). ‘=P. K. Runge, AT&T Technol. J. 71, 5 (1992). ‘%R. G. Smart, J. L. Zyskind, and D. J. DiGiovanni, IEEE Photon. Technol. Lett. 5, 1041 (1993). *“H. Horikawa and A I&ii, J. Lightwave Technol. 11, 167 (1993). ‘=E. F. Schubert, A. M. Vredenberg, N. E. J. Hunt, Y. H. Wong, P. C. Becker, J. M. Poate, D. C. Jacobson, L. C. Feldman, and G. J. Zydzik, Appl. Phys. Lett. 61, 1381 (1992). “‘E. F. Schubert, N. E. J. Hunt, A. M. Vredenberg, T. D. Harris, J. M. Poate, D. C. Jacobson, Y. H. Wong, and G. J. Zydzik, Appl. Phys. L&t. 63,2603 (1993). 13’A. M. Vredenberg, N. E. J. Hunt, E. F Schubert, D. C. Jacobson, J. M. Poate, and G. I. Zydzik, Phys. Rev. Lett. 71, 517 (1993). 13’H. Beneking, N. Grote, W. Roth, and M. N. Svilans, Electron. Lett. 16, 602 (1980). “‘A. Sasaki and M. Kuzuhara, Jpn. J. Appl. Phys. 20, L283 (1981). ‘33F R. Beyette, Jr., S. A. Feld, X. An, K. M. Geib, M. J. Hafich, G. Y. Robinson, and C. W. Wilmsen, Electron. Lett. 27, 497 (1991). ‘34F. R . Beyette, Jr., K. M. Geib, S. A. Feld, X. An, M. J. Hafich, G. Y. Robinson, and C. W. Wihnsen, IEEB Photon. Technol. Lett. 4, 390 (1992). 13’C. W. Wilmsen, E R. Beyette, Jr., X. An, S. A. Feld, and K. M. Geib, IEEE J. Quantum Electron. 29, 769 (1993). “‘M. S. &hi, A. L.~Demirel, S. Strite, S. Taslran, A. Salvador, and H. Morko9, Appl. Phys. Lett. 60, 1797 (1992). J. Appl. Phys., Vol. 78, No. 2, 15 July 1995 13’G. W. Taylor, J. ii. Simmons, A. Y. Cho, and R. S. Mand, J. Appl. Phys. 59, 596 (1986). ‘38J G. Simmons and G. W. Taylor, IEEE Trans. Electron. Dev. 33, 973 (i987). “‘G. W. Taylor, P. R. Claisse, and P. Cooke, Appl. Phys. Len. 58, 666 (1991). 14’l? A. Kiely, G. W. Taylor, D. P. Dotter, P. R. Claisse, T. Vang, P. A. Evaldsson, S. K. Sargood, S. Daryanani, P. Cooke, and K. F. BrownGoebeler, IEE Proc. 139, 208 (1992). 14’G. W. Taylor, P. A. Evaldsson, P A. Kiely, T. Vang, P. R. Claisse, S. L. Daryanani, D. l? Dotter, S. K. Sargood, and P W. Cooke, IEEE J. Quantum Electron. 29,785 (1993). ‘42F R Beyette, Jr., K. M. Geib, C. M. St. Clair, S. A. Feld, and C. W. &n&en, IEEE Photon. Technol. Lett. 5, 1322 (1993). ‘43S. Noda, K. Shibata, and A. Sasaki, Sensors and Actuators A 40, 125 (1994). ‘@K. Iga, F. Koyama, and S. Kinoshita, IEEE J. Quantum Electron. 24, 1845 (1988). ‘45H. Kosaka, K. Kmihara, M. Sugimoto, and K. Kasahara, Jpn. J. Appl. Phys. 30, Ll172 (1991). ltiG. R Olbright, R. P. Bryan, K. Lear, T. M. Brennan, G. Poirier, Y. H. Lee, and J. L. Jewell, Electron. Lett. 27, 216 (1991). ‘47W. K. Ghan, J. P. Harbison, A. C von Lehmen, L. T. Florez, C. K. Nguyen, and S. A. Schwarz, Appl. Phys. Lett. 58, 2342 (1991). ‘48J. I. Song Y H. Lee, J. Y. Yoo, J. H. Shin, A. Scherer, and R. E. Leibenguth, IEEE Photon. Technol. Lett. 5, 902 (1993). la9H. Kosaka, I. Ogura, H. Saito, M. Sugimoto, K. Kurihara, T. Numai, and K. Kasahara, IEEB Photon. Tecbnol. I.&. 5, 1409 (1993). “‘T Numai, M. Sugimoto, I. Ogura, H. Kosaka, and K. Kasahara, Appl. Pilys. Lea. 58, 1250 (1991). 151D. G. Deppe, C. Lei, T. J. Rogers, and B. G. Streetman, Appl. Phys. Lett. 58, 2616 (1991). ‘52D. L Huffaker, W. D. Lee, D. G. Deppe, C. Lei, T. J. Rogers, J. C. Campbell, and B. G. Streetman, IEEE Photon. Technol. Lett. 3, 1064 (1991). 153T.Numai K. Kurihara, I. Ogura, H. Kosaka, M. Sugimoto, and K. Kasahara, IEEE Photon. Technol. Lett. 5, 136 (1993). ls4T. Numai, H. Kosaka, I. Ogura, K. Kurihara, M. Sugimoto, and K. Kasahara, IEEE J. Quantum Electron. 29, 403 (1993). “‘T. Numai. H. Kosaka. I. Onura. K. Kmihara. M. Suuimoto. and K. Kasahara, IEEE J. Quantum El&r&. 29, 2006 (1993). ‘56P. Zhou, J. Cheng, C. E Schaus, S. Z. Sun, C. Hains, K. Zheng, E. Armour, W. Hsin, D. R Myers, and G. A. Vawter, IEEE Photon. Teohnol. Lett. 3, 1009 (1991) 15’P. Zhou, J. Cheng, C. E Schaus, S. Z. Sun, C. Hains, E. Armour, D. R. Myers, and G. A. Vawter, IEEE Photon. TechnoL Lett. 4, 157 (1992). “‘J. Cheng, P. Zhou, S. Z. Sun, S. Hersee, D. R. Myers, J. Zolper, and G. A. Vawter, IEEE J. Quantum Electron. 29, 741 (1993). “‘A E. Willner, C. J. Chang-Hasnain, and J. E. Leight, IEEE Photon. Technol. Lett. 5, 838 (1993). 16’C W. Wilmsen, S. A. Feld, F. R. Beyette, Jr., and X. An, Ciic. Dev. Mag., p. 21 (November 1991). “*P. A. Evaldsson, G. W. Taylor, P. W. Cooke, S. K. Sargood, P. A. Kiely, and D. P. Dotter, IEEE Photon. Tecbnol. Lett. 5, 634 (1993). 162T. Wipiejewski, M. G. Peters, and L. A. Co&en, Proceedings of IEEE LEOS Annual meeting, Boston, 1994. ‘63M. S. Unlii and H. Morko9, Optics and Photonics News, p. 30 (December 1992). ‘“IBM Res. Mag. 8, 1 (1993). Ifi5D. B. Young, J. W. Scott, F. H. Peters, B. J. Tbibeault, S. W. Corzine, M. G. Peters, S. L. Lee, and L. A. Coldren, IEEE Photon. Technol. Lett. 5, 129 (1993). ‘66G Blum, J. E. Zucker, T. H. Chiu, M. D. Divino, K. L. Jones, S. N. G. Chu, and T. K. Gustafson, Appl. Phys. Lett. 59, 2971 (1991). 16’0. Blum J E. Zucker, X. Wu, K. H. Gulden, H. Sohn, T. K. Gustafson, and J. S.‘Smitb, IEEE Photon. Technol. I&t. 5, 695 (1993). ‘68T Wipiejewski, K. Pam&&f, E. Zeeb, and K. J. Ebeling, IEEE Photon. Technol. Lett. 5, 889 (1993). 16’B. Pezeshki and I. S. Harris, Jr., US Patent No. 5,291,502. “OK. W Goossen, J. A. Walker, and S. C. Arney, IEEE Photon. Technol. Lett. 6, 1119 (1994). “IS. Strite and M. S. &Iii, Electron. Lett. 31, 672 (1995). Appt. Phys. Rev.: M. Selim finki and S. Strite 639 Downloaded 18 Jun 2002 to 128.197.180.119. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp