Anda di halaman 1dari 21

Home

Search

Collections

Journals

About

Contact us

My IOPscience

Quantum well intermixing

This article has been downloaded from IOPscience. Please scroll down to see the full text article. 1993 Semicond. Sci. Technol. 8 1136 (http://iopscience.iop.org/0268-1242/8/6/022) View the table of contents for this issue, or go to the journal homepage for more

Download details: IP Address: 103.246.106.9 The article was downloaded on 25/08/2013 at 07:22

Please note that terms and conditions apply.

Semlcond. Sci. Technol. 8 (1993) 11361155. Printed in the UK

Quantum well intermixing


John H Marsh Department of Electronics and Electrical Engineering, University o f Glasgow, Glasgow G I 2 8Q0, UK

Received 16 December 1992, accepted for publication


Abstract. Intermixing

February 1993

the wells and barriers o f quantum well structures generally results in an increase in the bandgap and is accompanied by changes in the refractive index. A range of techniques, based on impurity diffusion,dielectric capping and laser annealing, has been developed to enhance the quantum well intermixing (awl) rate in selected areas of a wafer; s u c h processes offerthe prospect of a powerful and relatively simple fabrication route for integrating optoelectronic devices and for forming photonic integrated circuits (PICS). Recent progress in OWI techniques i s reviewed, concentrating on processes which are compatible with PIC atxlications. in Darticular the achievement o f low optical propagation losses.

1. Introduction

An increasing number of optoelectronic devices utilize quantum wells (QWS) in their active regions. In some devices, such as quantum confined Stark effect (QCSE) modulators, quantum confinement is fundamental to the operation of the device, in others, such as semiconductor lasers, device operation does not require quantum confinement but is improved by the reduction in dimensionality. Since the original proposals by Miller [l], monolithic integration of several optoelectronic devices in photonic integrated circuits (PICS) and optoelectronic integrated circuits (OEICS) has been a goal actively pursued worldwide. Until recently, semiconductor optoelectronic devices used bulk materials, and integration usually required many steps of etching and regrowth. As well as resulting in low yields this approach always suffers from a mismatch in the optical propagation coefficient or dimensions of a waveguide at the regrown interfaces. Such a mismatch is an inevitable consequence of using bulk materials because alloys of different bandgaps also have different refractive indices. The use of Qws as the active layers of devices improves this situation considerably: because QW layers are thin they can be removed completely from a waveguide structure leaving the new waveguide with virtually the same refractive index, but a completely different bandgap from the original waveguide. Severalapproaches to integration based on QW active layers are emerging. The first, still based on regrowth, involves growing a structure containing the Qw active layers for all the devices. The active layers are then removed from the regions where they are unwanted and the structure is overgrown with the same upper cladding layers everywhere [2]. In this way a virtually constant optical waveguide propagation coefficient is maintained
$07.50 0 1993 IOP Publishing Ltd 0268-1242/@3/061136+20

across the wafer. Ifmore than one set of active layer QWS is required then they can be grown on top'of each other and the structure designed so that light couples vertically from one layer to the other [3]. This approach does impose limitations on device performance because in some parts of the PIC currents (or electric fields) must be applied to several active layers in series. A second solution is to vary the width of the QWS across the wafer during a single stage of epitaxy [4-61. In this approach the substrate is coated with a dielectric mask in which slots are opened. No growth takes place on top of the mask, but surface migration of the growth species can take place for some distance across the mask to the nearest opening. The growth rate (and hence the QW width) in the opened areas therefore depends on the width of the opening and the patterning of the mask. Finally, quantum well intermixing (QWI) is also emerging as a powerful technique for fabricating PICSand OEICS. In intermixing processes the bandgap of QW structures is modified in selected regions, after growth, by intermixing the wells with the barriers to form an alloy semiconductor. The bandgap of the intermixed alloy is usually larger than that of the original QW structure, thus providing a route to form low-loss optical waveguides, and bandgapshifted QCSE modulators, lasers and detectors using only one epitaxial step. In addition, because the bandgap is increased, the refractive index is modified. In structures containing only a few QWS this refractive index change will have only a small effect on the optical propagation constant, but in MQW (multi-quantum-well) structures the optical overlap between the intermixed well and the optical wave can be large enough to give useful changes in the refractive index. The refractive index changes can be used to provide optical confinement [7, SI, gratings [9], or even laser reflectors [lo, 111. A number of intermixing techniques have been re-

Quantum well intermixing ported, most notably impurity induced disordering (IID), impurity-free vacancy disordering (IFVD) using dielectric caps, and laser-induced disordering (LID). As will be shown, the availability of this range of techniques is important in applying Qwi to different material systems and even in meeting different requirements within a single PIC. In this paper, the three aforementioned techniques will be reviewed, with a strong emphasis on those which retain high crystal quality, low free-carrier concentrations and low optical propagation losses. Three lattice-matched material systems will be discussed-GaAs/ AIGaAs, GaInAslGaInAsPpnP and GaInAs/AlGaInAs/ InP. In section 2 some applications of QWI are described, some targets which QWI processed material must meet are defined and the three different QWI processes are outlimed. In section 3 the physical process underlying QWI processes are described and the thermal stability of the three alloy systems are compared. In section 4, results of neutral IID in the GaAs/AlGaAs system are presented, whilst in section 5 corresponding results for the GaInAs/ GaInAsP and GaInAs/AlGaInAs systems are presented. In section 6 a possible mechanism for the neutral IID process is discussed. In sections 7 and 8 IFVD and photoabsorption-induced disordering (PAID)-a new variation of LlD-ar.2 discussed. In section 9 results from a device application of IID in the GaAs/AlGaAs system, namely an integrated extended cavity laser, are summarized. In section 10 the various points raised in the previous sections are drawn together.
2. Applications and general characteristics of awl processes

Several potential applications of QWI techniques in integrated optoelectronics can be identified: for example, bandgap-tuned modulators [12], bandgap-tuned lasers [13], low-loss waveguides for interconnecting components on an OEIC, integrated extended cavities for linenarrowed lasers, single-frequency DBR (distributed Bragg reflector) lasers and mode-locked lasers, non-absorbing mirrors and either gain or phase gratings for DFB (distributed feedback) lasers. Taking the examples which utilize low-loss waveguides [14], three parameters are of particular importance: the absorption coefficient, the material resistivity and the refractive index change induced by intermixing. An ideal loss target within a PIC is < 1 dB cm-' but 10 dB cm-' would be acceptable in many applications, and values as high as 220 dB cm-l in a DBR grating will still give sufficient finesse for single-mode operation. In integrated extended-cavity lasers, losses can be as much as 40 dB cm-' in the passive section and still give a useful reduction in laser linewidth. A further requirement is that the electrical resistance of waveguides should be sufficiently high to isolate individual components, and studies of an integrated laser/modulator structure [15] have demonstrated that the required isolation resistance between the laser and the modulator might need to be as large as 100 kQ. In IID processes an impurity is introduced into an

epitaxial wafer and the wafer is then annealed. During the annealing step the layers intermix and ion-implantation damage, if present, is to a large extent removed. Current understanding 1161 of the IID process suggests that the role of impurities is to induce the disordering process through the generation of free carriers which, in turn, increase the equilibrium number of vacancies at the annealing temperature. A number of species have been demonstrated to disorder the GaAs/AlGaAs system, the most important of which are Zn (p-type) and Si (n-type). Impurities need to be present in concentrations greater than around lo'* cmF3in order to enhance the interdiffusion rates of the lattice elements. These considerations highlight the problems which arise when electrically active dopants are used as disordering species: the threshold concentration of impurities necessary to induce the IID process is typically > 10'' ~ m -The ~ . most commonly reported impurity for IID is Si and the lowest reported absorption coefficients are around 43 dB cm-' (10 cm-I), which is a consequence of free-carrier absorption. Furthermore, for waveguide dimensions of 3 p n x 1 pm x 0.5 mm long, a carrier density below loL7 cm'-l is needed to provide electrical isolation of 100 kQ. There is clearly a trade-off between the required electrical isolation and the amount of bandgap widening in designing a waveguide for interconnection, but it appears that Si (or Zn) IID is unlikely to give the required performance. As a consequence, the studies reported here have used the electrically neutral dopants F and B introduced by ion implantation as IID species which, as will be shown below, produce low-loss bandgap-widened waveguides. However, traps associated with these species and residual damage from implantation make IID an unsuitable technique for forming bandgaptuned modulators or lasers. These considerations make clear the need for a range for QWI techniques. Like IID, IFVD is also based on creating vacancies in the 111-V semiconductor.Gallium, for example, is known to dissolve in certain dielectric caps, particularly SO,. If a wafer is coated with SiO, then heated, Ga migrates into the dielectric cap creating group 1 1 1vacancies which then diffuse through the QW structure, resulting in intermixing. IFVD can produce low-loss waveguides in both the GaAs/ AlGaAs [17] and the GaInAsP/InP [18] systems, and bandgap-shifted Q C ~ Emodulators [12] and lasers [13]. Good selectivity is a problem with IFVD-it is difficult to prevent intermixing in areas where it is not required. A new cap which completely prevents intermixing in the GaAs/AlGaAs system has been developed and is described below. LID of Qw structures was fust demonstrated using cw radiation from an Ar laser. The beam was focused and scanned across the semiconductor surface. The maximum transient temperature reached by the semiconduo tor was sufficient to melt the surface layer, and the molten material then recrystallized as an alloy semiconductor. Variations of the same technique using pulsed laser irradiation have also been reported. LID therefore differs from the other techniques in that it does not necessarily rely on creating point defects to enhance intermixing
1137

J H Marsh

rates. Because melting needs a high temperature, dielectric capping is vital to preserve the stoichiometry of the semiconductor, but such caps can also act as sourcesof impurities or of vacancies. The quality of recrystallized material is also inferior to that of the original semiconductor. An all-solid-state variation of the LID process has been developed and will be described below.
3. Thermal stability of doped and undoped alloy material systems
QWI takes place when the matrix elements of a semiconductor interdiffuse: in certain material systems this need only take place on one lattice site (e.g. the group 111 site in the GaAs/AlGaAs system) or interdiffusion may need to take place on both lattice sites (e.g. the group ID and group V sites in the GaInAs/GaInAsP system) in order to prevent strain. Because diffusion of matrix elements proceeds through native crystal defects, the self-diffusion rate is determined by the diffusion rate of the defects and by their concentration. Self-diffusioncoefficients of matrix elements are therefore usually small in high-purity 111-V materials in thermal equilibrium at temperatures well below the melting point because the native defect concentration is small. The equilibrium number of defects is determined by the temperature of the wafer, the partial pressure of matrix elements (in particular the group V elements) and the position of the crystal Fermi energy. It is also possible to introduce an excess (nonequilibrium) concentration of defects, for example by introducing ion-implantation damage or by using IFVD. These points can be illustrated by considering selfdiffusion in the much-studied GaAs/AlGaAs system. The thermal stability of the system and the dependence of the QWI rate on the As vapour pressure has been studied by Guido et al[19]. In their experiments undoped single QW (sQw) samples were annealed in sealed quartz ampoules at 825C for 25 h. The As, vapour pressure over the crystal was varied in the range 0.1-10 atm by enclosing measured quantities of excess elemental As in the ampoule. By monitoring the bandgap changes caused by annealing, the authors were able to deduce the A1-Ga self-diffusion coefficient. This lay in the range (7-13) x cmz s-l and showed a clear dependence on the As, vapour pressure, with a minimum occumng at a vapour pressure of around 0.5 atm. This behaviour has been explained by Deppe and Holonyak [16] as follows: the self-diffusion rate is controlled by the concentration of native defects which, in tum, is sensitive to the stoichiometry of the crystal. As the As vapour pressure is increased, so defects associated with an excess of As control the group I11 self-diffusion. These defects are the As interstitial As,, the As antisite defect As,,, and the group 111 vacancy V,,,. In contrast, under a low As overpressure, the As-poor defects-interstitial A1 or Ga, the group 111 antisite defect and the As vacancy-increase in concentration and control the self-diffusion.Of these six defects, the group 111 self-diffusion rate is determined principally by the group I11 vacancy (As-rich

conditions) or by the group 111 interstitial (As-poor conditions). It has been shown by many workers that the presence of a large concentration of impurities in the crystal can increase the group 1 1 1 self-diffusion rate by orders of magnitude. The model of Deppe and Holonyak also explains the effect of the crystal Fermi energy on the QWI rate. In their model, the equilibrium vacancy concentration is found from simple thermodynamicconsiderations. Of the six point defects, only two-the group I11 vacancy V,,, and the group 111 antisite defect-behave like acceptors, whilst the other four are expected to behave as donors. Assuming the native defects are either neutral or singly ionized, the group 111 self-diffusion rate, D,,, is given by
DIU=f;pX{Dv;, -t Dv,eXP[(&

+fip&4{D1:,

-EA)/%I~} -k Di,,,eXP[(& - E F ) / ~ E ~ } .(1)

Here PAS4 is the partial pressure of As,, Dvh, Dv,. D,h, D,,, are the diffusion coefficients of neutral and charged 1 1vacancies, and neutral and charged group 111 group 1 interstitials respectively, f; and f ; are functions which depend on the crystal structure, E , is the vacancy acceptor energy level and EDthe interstitial donor level. From equation (1) it can be seen that the IID rate depends on the type of doping (n-type or p-type) and the As overpressure. Intermixing takes place in p-type QW structures under As-poor conditions and in n-type QW structures under As-rich conditions because the concentration of group III interstitials and group 111vacancies is raised in the respective cases. Although different degrees of ionization have been suggested for the defects [ZO], the behaviour is qualitatively similar. In practice, ampoule annealing is rarely used. Furnace annealing or rapid thermal processing (RTP) is usually camed out with a flowing gas, and a proximity cap of GaAs (or InP) is used to protect the sample and provide a group V overpressure. In addition to the use of a large As overpressure, the use of an SiO, cap also favours group I11 vacancy generation because Ga dissolves in the SiO, layer. SiO, caps can be used to promote QW with or without active dopants. For example, it has been shown that Si-doped material will intermix beneath an SiO, cap but not beneath an Si,N, cap [Zl]. Equally, undoped material beneath dielectric caps will intermix through the IFVD mechanism if the dielectric cap creates point defects. It can be seen from the foregoing discussion that undoped GaAs/AlGaAs QW material is stable up to temperatures well beyond those normally used in epitaxial growth. In contrast, the Q)V material systems used at 1.5 pm are of much more limited thermal stability, and the GaInAs/InGaAsP (P-quaternary) system exhibits considerable intermixing even at temperatures normally encountered in epitaxial growth. Marsh et al [22] have demonstrated that unimplanted P-quaternary samples (capped with either Si,N, or SiO, and in a P-ambient provided by pieces of InP) disorder at annealing temperatures above 5OO0C,with a blue-shift always observed, as

1138

GRiN

60 -

.
a
,

.. . I
0

-5 5 7

. Y

O
Quantum weii intermixing
-10
I

0 .&
U

.c

0 .

40-

20.
0
400

, . - ,
450

.= ." .. ..
.U
,

.3

. J

, , ,
6OO

,
650

, ,
750

-20 550
801)

600

650

700

750

800

500

550

700

Temperature I "C
Figure 1. Exciton peak shifts o f P-quaternary GRIN (gradedindex) and SCH (separate confinement heterostructure) samples annealed for 30 min in a conventional annealing

Anneal IemperdNre ("C)

Figure 2. The exciton peak shift after annealing for 30 min as a function o f annealing temperature for unimpianted

Alquaternary samples.

furnace. shown in figure 1. This deficiency causes serious problems when the bandgap of QW samples alters during overgrowth (as when making DFB lasers) or in the long growth runs needed for vertical cavity surface-emitting lasers when the lower mirror stack intermixes before growth of the structure is complete. It is believed that the blue-shift is caused by the diffusion of phosphorus into and arsenic out of the wells [23]. This is likely to be due to group V vacancies, generated by phosphorus desorption from the surface, diffusing through the structure. To investigate this further, unimplanted samples were annealed for 30 min at 650C in silica ampoules loaded with red phosphorus. It was found that the exciton energy shift could be reduced to 15 meV, compared with 25 meV in the conventional furnace anneal. This result is similar to that found by Komiya et al [24], who found that the thermal degradation of InP/InGaAsP/InP double-heterostructure (DH) structures was less when their samples were annealed in an ampoule containing red phosphorus rather than in an LPE fumace containing InP loaded with tin. They proposed that the degradation of the structures was due to vacancies caused by the desorption of phosphorus from the surface, and concluded that the phosphorus overpressure generated by InP is insufficient to prevent this. Given that most QWI studies of the P-quaternary system have not used sealed ampoules, it is probable that intermixing takes place in samples which always contain a large concentration of group V vacancies. It is therefore not possible to apply the model of Deppe and Holonyak directly to this material system. In contrast, Bryce et al [25] have demonstrated that GaInAs/AlGaInAs (Al-quaternary) samples are stable up to annealing temperatures of 65OoC, as shown in figure 2. Even above this temperature only small red shifts in the low-temperature PL measurements were observed, probably due to Ga diffusing out of the wells and being replaced by In from the barriers. This is unexpected since the In concentration is initially constant but this effect has previously been observed in GaInAs/AlInAs quantum well structures [26, 27l. The effect is probably caused by the low mobility of the Al compared with the Ga and In and the requirement of the material to remain stoichiometric. Since the aluminium does not diffuse significantly the In diffuses to maintain the structure of the material. The full width at halfmaximum (FWHM) of the exciton peak was measured to be 12 meV for unimplanted, unannealed material. It was found that the peak broadened on annealing (figure 3), with higher annealing temperatures leading to broader exciton peaks. This indicates that annealing causes some diffusion at the interface between the wells and the barriers, increasing the interface roughness. The better temperature stability of the Al-quaternary therefore makes it a more attractive material for IID processing. However, the P-quaternary bas considerable advantages including being Al-free and the fact that P-quaternary devices have demonstrated excellent reliability.

'

575

600

625

hSO

675

700

Anneal Temperature / 'C

Figure 3. Width of the exciton peak as a function of annealing temperature for unimplanted AI-quaternary samples. The anneal times were 30 min in all cases. The curve indicates the general behaviour of the data.
1139

J H Marsh

from fully activated silicon doping at the concentrations typically required for QW disordering [30] 4.1. Furnace annealing of GaAs/AlGaAs (> 10" The excess propagation loss associated The impurities boron and fluorine are clcctricdy neuwith boron intermixing is also very small. tral at room temperature in the GaAs/AlGaAs system. QW structures exhibit a number of polarization-sensiO'Neill et aZ[28,29] have made signikant advances by tive effects, most significantly a polarization-dependent demonstrating the potential of these impurities as disordichroism and birefringence, but the polarization dependering species in PIC applications. Several structures dences disappear as the structures are ,intermixed and optical including MQW and double quantum well (DQW) become more like bulk alloys. The dichroic effect arises waveguides have been implanted with Euorine and boron from the selection rules governing optical absorption in a at concentrations between 3 x 10" and 3 x 1019~ m - ~ . QW, with the TE polarization exciting transitions from Annealing temperatures in the 750 to 920 " C range were both the heavy-hole (HH) and^ light-hole (LH) confined used for times up to 4 h. Photoluminescence (PL) specstates into the conduction band states, and the TM troscopy [ZS] at 18 K was used to optimize the implantapolarization exciting transitions only from the LH states. tion and annealing conditions. Features associated with The absorption edge therefore occurs at a longer wavelength-for the TE polarization than for the TM polarizarecombination at the bandgap and those associated with damage were identified and compared in intensity. From tion. Birefringencearises because QW structures consist of these measurements the optimum implant dose was a number of dielectric layers, each layer being much thinner than the wavelength of light, and the effective cm-,) with an found to be around 10" anm3 dielectric constant of the composite structures therefore annealing temperature of 890C. Figure 4 shows the depends on whether the optical electrical field is parallel variation of the energy shift of the bandgap with annealing time at 89O"C, for two different fluorine and boron or perpendicular to the plane of the wells [31]. However, for wavelengths close to the absorption edge, the refracimplantation doses. Using fluorine the energy shift, at times for which the mixing process does not approach tive index spectrum is determined to a substantial extent by the rapidly changing absorption spectrum. Because saturation, is over twice that observed using boron. Preliminary estimates [28] of the propagation loss the absorption spectrum is strongly anisotropic the biindicated that the loss in fluorine- or boron-disordered refringence increases markedly as the absorption edge is material was significantly lower than that in siliconapproached. Hansen et ai [32] have carried out the first disordered material. Accordingly, long (10 mm) Euorinesystematic studies of the effect of disordering on the disordered MQW ridge waveguides were fabricated and refractive index. the loss measured using the sequential cleaving techThe structure investigated was an MQW waveguide nique. Total propagation losses as low as 4.7 dB cm-' at where the MQW consisted of 54 periods of 6.0 nm GaAs a wavelength of 875 nm were obtained in these wavebarriers. Samples were wells and 6.0 nm Al,,,,Ga,,,.Js guides [29] accompanied by a substantial (60 meV) blueuniformly implanted, throughout the depth of the MQW shift in the absorption edge. The process parameters and layer, with boron or fluorine ions and were capped with a annealing conditions used were a dose of 1 0 " ~ m - ~ 120 nm thick layer of plasma-deposited SiOz prior to annealed at 890C for 2 h. The total loss figure o f annealing. The thickness of the capping layer was de4.7 dB cm-' is not the ultimate lower limit, since scattersigned to give a -reasonable output coupling efficiency ing due to rib waveguide top and sidewall roughness when used in the fabrication of an output grating coupler, as well as to give added protection against As makes an important contribution to propagation losses. This figure is, however, much lower than the contribution desorption from the material. Annealing conditions used from free-carrier absorption (e40 dB cm-') expected a temperature of 890C for times up to 4 h. Photoluminescence measurements showed an energy shift of 28 meV for boron after annealing for 120 min and of 40 meV for , 120 I I fluorine after 90 min. After 4 h annealing following fluorine implantation it is believed that the wells are completely disordered. Output grating couplers were then fabricated in the SiOz annealing cap present on top of the slab waveguides. The grating pattern was produced by laser holography and transferred to the SiO, layer by shadow masking and dry etching. The complete process is described in detail elsewhere [33]. The results are shown in figures 5(a) and (b),together with the refractive index results for the MQW waveguide 0.0 1.0 2.0 3.0 4.0 5.0 6.0 before disordering. The largest changes in the refractive Time (hcs) index occur, as expected, at the exciton resonances in the Figure 4. Bandgap increases associated with boron and starting material. At long wavelengths, the implanted fluorine 110in a GaAslAlGaAs MOW using an annealing samples annealed for short times ate observed to have a temperature of 890C.The aggregate implant doseswere higher refractive index than that o f the starting material, 1 x 10'4cm-' ( 0 ,B; 0 F) and 3 x 10'4cm-3 (m. B; 0 , this being particularly evident in the case of boron. After F).
4. IID of GaAslAlGaAs

1140

Quantum well intermixing


(4
3.62
3 . 6
3.58
W '0 3.56 W
X

-ev -

com
F 1.5

+B Z H
+i

4 H

4 B4H

c 0

2 3.54 c
L 5 3.52

LT

3 . 5
3.48
3.46

820

840

860 880 Wavelength (nm)

900

920

CONTROL

F 1.5

B2H
F4H
04H x
c

3S6#

3.46 820

840

860

880

goo

920

Wavelength (nm) Figure 5. Variation of refractive index with wavelength for (a) TE polarization and (6)TM polarization for boron-disordered,fluorine-disordered and control MQW (54-well) sam a1es.

annealing for 4 h following fluorine implantation, the material refractive index in the waveguide core is virtually identical for the two polarizations, confirming that the MQW is completely disordered. The measurements of optical waveguide propagation loss indicate that disordered material has a low freecarrier concentration. This has been confirmed by electrochemical profile measurements performed on the 60 A well/60 8, barrier MQw structure. An electrochemical profile indicated that the unimplanted MQW structure was fully depleted at zero applied voltage and that the residual doping of the AlGaAs cladding layer was 1.8 x 10l6C I I - ~ p-type. Samples were uniformly implanted to

a depth of 1 pm with fluorine or boron to a concentration of 10" cm-3 and annealed at 890C for 90min and 120 min respectively. For these samples electrochemical profiles were as follows: the fluorine sample was p-type with a hole concentration of 1 x 10'6cn-3, and the boron sample was also p-type with a hole concentration of 2.2 x 10'' ~ m - These ~ . carrier densities would give rise to free-carrier absorption losses below 1 dB cm-', significantly lower than those observed in the waveguides. The resistivity of these layers is expected to be around 2C2cm, a factor of approximately 300 times higher than that of material disordered with silicon at a concentration of 10l8~ m - Good ~ . isolation resistances
1141

J H Marsh

should therefore be realized within either fluorine- or boron-disordered waveguides with realistic practical dimensions.
4.2. Diffusion of fluorine and boron

Diffusion of impurities during intermixing leads to two effects: firstly, as the impurities diffuse, unintentional intermixing w i l l take place in regions other than those implanted. Secondly, the volume concentration of the impurity will fall during diffusion and will eventually drop below the threshold concentration at which impurity-enhanced disordering takes place. SIMS analysis of disordered structures has been carried out to determine the extent of diffusion of the fluorine and boron. Figure 6(a) shows the variation of Ga and Al concentration i n the as-grown MQW structure and figure 6(b) shows the same structure after implantation with fluorine, followed by furnace annealing. The oscillations in the Al concentration are, as expected, washed out. (The longer-period

0.0

0.7 0.4 0.6

0.x

1.0

1.1

1.4

1.6 1.8 2.2

Depth1 pm Figure 7.
SIMS analysis showing the diffusiono f fluorine (full curves) during an annealing cycle.

U C

8
104
E
v1

0.0

0.2

0.4

0.6

0.8

oscillations seen in both figures are an artefact of the sampling frequency of the SIMS system.) Figure 7 shows the effect of annealing on the fluorine distribution: rapid diffusion takes place towards the surface and into the substrate. The diffusion to the surface can be explained if it is postulated that fluorine has a high solubility in regions of high defect concentration-a surface is a region of very high defect concentration and so fluorine can migrate against its concentration gradient. In the case of boron, however, negligible diffusion on a macroscopic scale takes place and the position of the three implants can still be seen even after annealing (figure 8). Nevertheless boron still enhances group I11interdiffusion on an atomic scale. Further SIMS scans have demonstrated that only limited diffusion of the grown-in dopants Be and Si takes place during disordering (even

Depth / Lrn
105

SO

IO0

150

200

?SO

.
0.0

.
(1.2

.
(1.4

.
0.6

.
0.8

.
1.0

.
1.2

.
1.4

.
1.6

.
1.8

Cycles
SIMS analysis (a) o f the as-grown GaAsIAIGaAs structure and (b) of a MOW sample after disordering with fluorine. The full curves are for Ga, the broken curves for AI,

Depth / km

Figure 8.
MOW

Figure 8. SIMSanalysis of a boron-implanted sample after disordering. The distribution of the boron is virtually unchanged from that o f as-implanted samples, and the three implantation peaks can still be clearly seen.

1142

Quantum well intermixing

in fluorine-implanted regions) and that the intrinsic region of a DQW laser, for example, remains free of electrically active dopants.
5.

AlGaAs system, Si3N, is used to initiate intermixing-behaviour consistent with that also reported below.

IID

o f GalnAslGalnAsP and GalnAs/AIGalnAs

5.1. Fluorine- and boron-induced disordering of GaInAs/AlGaInAs


Epitaxial material was grown on an InP substrate using atmospheric-pressure metal organic vapour-phase epitaxy (APMOVPE) at a growth temperature of 650C. The structure, intended for photoluminescence measurements, consisted of four 100 8, Ga,,,,In,.,,As wells separated by quaternary bamers 508, thick with a lOOO.& thick quaternary layer below the wells and a 2008, thick quaternary layer above the wells. The 'quaternary composition in all three cases was Alo.zoGa,,2,1n,,,3As. This structure would normally form part of a laser structure which lases at 1.55 pm [46J Photoluminescence measurements, at both room temperature and 10 K, indicated that the variation in exciton energy was li: 2 meV over the area of material used in this study. Samples of approximately 0.25cm2 area were implanted with either fluorine ions or boron ions. In order to assess the relative importance of disordering due to the impurities and disordering due to implantation damage, implantations were made to two different depths. The first implants were deep, which calculations [25] showed deposited a low concentration of impurities in the quantum well layer relative to the amount of damage in the quantum wells. The second, shallow, implants were at lower energy so that the peak of the impurity concentration would be located in the quantum wells. The computer model showed that both implant energies caused a similar amount of damage within the wells, with fluorine giving approximately twice as much damage as boron. The deep implants were made to a depth of 350 nm (ten times deeper than the quantum wells) using fluorine ions at 250 keV or boron ions at 150 keV. Because of limitations imposed by the ion implanter it was not possible to produce ions of low enough energy to be stopped by the quantum well layer alone and it was therefore necessary to cap the samples with a 100 nm layer of plasma-deposited SiO, before carrying out the shallow implantations. The energies used for shallow implants were then 100 keV and 60 keV for fluorine and boron respectively. Calculations [25] showed that the concentration of Si ions withm the wells caused by the implanted ions knocking atoms out of the SiOz film was ~m-~ The . doses were l O I 3 cm-' for all the I implants. The calculated impurity concentrations within the wells were 2 x lot6 cm-3 for the deep implant and 5 x IO" for the shallow implant. Annealing was performed as described previously (section 3), with the samples capped with 120nm of SOz, placed with the top surfaces uppermost, and exposed to an overpressure of either phosphorus, provided by a small melt of indium loaded with indium phosphide and tin, or of arsenic, provided by a gallium melt loaded with gallium arsenide and tin. The addition of the tin to the melt increases the solubility of the group V element
1143

of MQW layers in the P- and Al-quaternary systems has been very much less widely studied than in the GaAs/ AlGaAs system. In the P-quaternary system it has been shown that in-diffusion of zinc [34-361 results in preferential mixing on the group 1 1 1lattice sites. SIMS studies [35, 361 have shown that the wells become increasingly InAs-like which, in turn, causes a narrowing of the bandgap and a corresponding increase in the refractive index. Large changes in the bandgap wavelength can be induced, for example, from 1.4 pm to 1.9 pm [36]. The effects of the resulting strain have recently been modelled [37]. Sulphur 1381 (at a concentration of lo',' and high-concentration protons [39] implants (5 x loL8 to IO" aK3)have been demonstrated to give increases in the bandgap energy, causing intermixing on both the group 1 1 1and group V lattice sites so that the wells become increasingly GaInAsP-like. The amphoteric impurities, germanium [40] and silicon [41], and the isoelectronic impurities, gallium [42] and phosphorus [43], have also been demonstrated to give bandgap. increases if implanted at high doses (~m-~). These processes, however, involve either the use of active dopants or high implantation doses-free-carrier absorption losses are significant in the former case, and residual damage in the latter. Zucker et al [44] have reported losses of 14.7 dB cm-' in an MQW waveguide in which phosphorus was implanted into the MQW at 200C and the sample was then overgrown with InP at 650C for 15min. This loss figure is encouragingly low, although photocurrent measurements showed that residual damage was still present in the structure. Using AlInAs as an alternative to InP and AlGaInAs as an alternative to GaInAsP simplifies the QWI process in that only the group 111 sites need to be intermixed. The improved superior stability of this material system compared with the P-quaternary has already been discussed. IID, again using the active dopants silicon [2n and zinc [27, 451, has been reported, processes which give rise to the associated free-carrier absorption losses. Moreover, the work using silicon demonstrated that a very high concentration was required to induce intermixing, > 3 x
QWI

101'

cm-3.

As in the GaAs/AlGaAs system, the use of either neutral impurity IID or IFVD offers the prospect of low propagation loss in intermixed material. Fluorineand boron-induced disordering of both the material systems used at 1.5 pm has been investigated-GaInAs/ AlGaInAs (Al-quaternary) by Bryce et al[25] and GaInAs/GaInAsP (P-quaternary) by Marsh et al[22], in both cases lattice-matched to InP. These results are summarized below. IND in the P-quaternary system has also been reported and has been used by Miyazawa et al [lS] to form an extended cavity laser. In contrast to the GaAs/

J H Marsh

(i.e. P or As), and so increases the partial pressure of the group V element within the graphite box to a level which prevents decomposition of the substrate or epitaxial layers [47]. No dependence on the overpressure species was apparent in the annealing behaviour. Samples were annealed at 650C for 0.5, 1 , 2 and 4 h and at several temperatures between 600C and 750C for 0.5 h. The samples were investigated using photoluminescence at 10 K. Deep implants of either boron or fluorine resulted in small blue-shifts in the bandgap, of about 10 meV, after annealing at 650 "C for 30 min. Fluorine caused marginally larger shifts in the exciton peak than boron, this being consistent with calculations which show that fluo-. rine causes more damage than boron, but otherwise it was found that the two implant species resulted in similar behaviour. In contrast, disordering induced by the shallow implants did display species-dependent behaviour. Using boron, the effects of either shallow or deep implants were indistinguishable-because the damage ipduced within the quantum wells is similar for either shallow or deep implants. This result suggests that boron is not an active disordering species. When fluorine was present within the QW region there was a large additional disordering effect: for example, for 30 min anneals at 650C the blueshift in the exciton peak was approximately three times greater than for deep implants (figure 9). Finally, it was observed that annealing for longer periods at 650C resulted in complete disordering of the wells (figure 10) with handgap increases of 40-50 meV being obtained. The AI-quaternary system therefore appears to be thermally stable below about 650C. It appears that fluorine is an active disordering species, but boron is not. The effects of ion-implantation damage in disordering the lattice can be observed with either implant, but the use of fluorine results in large additional disordering effects.

Anneal time / h

Figure 10. Shift in the exciton peak after annealing at


65O'C as a function of annealing time for the shallow fluorine-implantedAlquaternary.

5.2 Fluorine- and boron-induced disordering of

GaInAs/GaInAsP Structures were again grown by APMOVPE on n-type InP substrates. First a 1.0pm n-type buffer layer of InP was grown, then the quantum well structure containing wells separated by 1208, four 60 8, Ga,,,,In,.,,As Ga,,,71n,,,,As,,,,P,,63 barriers (bandgap wavelength of 1.16 pm) with cladding layers on both sides of the wells. Two structures were studied-a separate confinement heterostructure (SCH) in which the cladding layers were 0.09 pm thick Ga,,,71n,.,,Aso.3,Po,63 and a graded index (GRIN) structure in which the GaInAsP composition was graded from Gao,,,Ino,~~Aso,3,P,.63 at the MQW boundaries to InP at the outer edges. A further 1pm layer of InP completed both structures. The exciton peak was measured by room-temperature photoluminescence and found to be 1.522 & 0.016 pm for the SCH structure and 1.527 5 0.025 pm for the GRIN structure. Samples of each structure were implanted with either fluorine or boron to give a volume concentration of 2.4 x 10'' in the wells. Some samples were then capped with approximately 100 nm of Si,N, or SiO, and then all samples were annealed for 0.5 to 2 h at temperatures ranging from 475 to 775 "C. The degree of disordering was investigated by measuring the heavy-hole exciton energy using PL at 15 K. The boron-implanted P-quaternary exhibited redshifts after annealing at temperatures above 600 "C (figure l l )which can be explained by an increased rate of diffusion on the group III sublattice. Increasing the group 111 diffusion rate increases the In content of the wells, decreasing the bandgap energy, which causes a redshift in the emission spectra. The red-shift indicates that the group'III diffusion has increased to such a rate that quantum wells become strained [48]. It has been shown previously that an impurity may affect the diffusion rate on the group 111and group V sublattices in a similar way to boron. In the similar GaInAsP/InP system, Zn leads to group 111 disordering only [ 3 6 ] , causing lattice mismatch, whilst Ge [40] and S [38] implantation can lead

0
SS0

6(10

6.50

700

7.50

800

Anneal temperature/ "C

Figure 9. The exciton peak shift after annealing for 30 min as a function o f annealing temperature for the fluorineimplanted AI-quaternary. 0 ,shallow implant; . deep ,

implant.
1144

Quantum well intermixing


100

Conrrol
80
O

Boron

quantum wells and barriers the average composition of the well region would be I n ~ . 6 ~ G a o , 3 1 A s ~ . ~The 6Po,~ corresponding room-temperature bandgap [49] of the Pquatemary alloy would be 903 meV, indicating a maximum possible bandgap shift of 96 meV from the starting material. It can be seen from figure 1 that the enerb shift apparently slows at approximately 90 meV for the GRIN, indicating that the wells are approaching complete intermixing.
5.3. Rapid thermal annealing of P-quaternary MQW

structures
500

6 0 0

700

800

As discussed above. there are serious problems in using


conventional furnace annealing to process the P-quaternary, but Marsh et al [22] and Bradshaw et al [sol have demonstrated that annealing in a rapid thermal processing (RTP) system can overcome some of the limitations associated with furnace annealing. Studies of fluorine-implanted samples have been made by Bradshaw et al [SOJ. Once again both SCH and GRIN samples were implanted to give the following fluorine concentrations 2.4 x 10l8cm-3 and in the QW region: 2.4 x 10'' ~ . samples were then capped with 2.4 x 1019~ m - The 1000 A of either SiO, or Si3N, deposited by plasmaenhanced chemical vapour deposition (PECVD). The samples were placed face down on a high-purity graphite susceptor and enclosed within a cavity completed using two pieces of silicon. Annealing was carried out at temperatures ranging from 650 to 750 "C for times ranging from 20 to 60 s. All anneals were carried out in a nitrogen atmosphere. Firstly we consider the effectof the different dielectric caps. Figure 13 shows the results of annealing samples of the SCH structure with protective dielectric caps at 700C for times between 10 and 60 s. It can be seen that, for short anneals of 30 s or less, disordering of up to 15 meV occurs with the Si3N4-coated samples. The uncapped and SO,-capped samples show only small changes of around 5 meV. It is interesting to contrast this behaviour with GaAs-based QW material, which usually shows large

Temperature I "C
Figure 11. Exciton peak shift of the P-quaternary GRIN samples, annealed for 30 min in the conventional

annealing furnace with boron implants.

to both group I11 and group V interdiffusion, so maintaining approximately lattice-matched conditions. At lower annealing temperatures small blue-shifts were measured, which we attribute to disordering due to the damage from the implantation process. Exciton recovery was not observed in samples of the GRIN annealed at temperatures below '550C in either the boron- or fluorine-implanted cases because the ion-implantation damage is not annealed out at these low temperatures. From figure 12 it can be seen that at low annealing temperatures fluorine-implanted P-quaternary samples have a blue-shift of about 60 meV, approximately twice the size of that in the boron-implanted samples. Calculations show that fluorine induces approximately twice as much damage as boron, consistent with observing a larger shift. For annealing temperatures above 650 "C, the fluorine-implanted samples produce a shift approximately 20meV larger than the control samples (discussed in section 3), suggesting that the intermixing is dominated by disordering due to the thermal instability of the material. Assuming complete disordering of the

. %
E
0
500

"1

600

700

800

I)

IO

10

30

41)

SO

60

70

Temperature f 'C
Figure 12. Exciton peak shift of the P-quaternary GRIN samples, annealed for 30 min in the conventional annealing furnace with fluorine implants.

Time f s
Figure 13. Exciton shift as a function o f anneal time at 700% for P-quaternary scn samples capped with either 1000 A of Si,N, (W) or SiO, (A) or no cap (U).
1145

J H Marsh

shifts for SiO, caps [Sl] and smaller shifts for Si,N,-capped samples. For longer RTP anneal times most samples show significant intermixing (around IOmeV), behaviour consistent with our previous results using conventional annealing for times of 30 min or longer. In the case of conventional annealing, however, no dependence of the degree of intermixing on the cap material used was seen. Annealing at 650C for between IO and 60 s gave a much smaller exciton shift for all samples, as shown in figure 14. The uncapped and SO,-capped samples show an insignificant exciton shift, whereas samples capped with 1000 8, of Si,N, show a maximum shift of IO meV for the 60 s anneal. Secondly, the effect of implanting fluorine into the structures is examined. Figure 15 shows the results of annealing SCH samples at 650C previously implanted with the doses and energies described above. It can be seen that there is an optimum concentration of fluorine implant (around 10l8 ~ m - which ~ ) gives the maximum energy shift. Implanting with fluorine concentrations greater than this inhibits the intermixing processes as shown by the full circles in figure 15.

SCH samples implanted with a fluorine concentration of 10" cm-3 were also annealed at a higher temperature of 700C (figure 16). It was found possible to obtain 40 meV exciton shifts for both anneal temperatures, after 60 s at 650 "C and 20 s at 700 "C. Raising the annealing temperature further to 750C resulted in handgap increases of over 15 meV in unimplanted QWS, consistent with the results for conventionally annealed material. GRIN samples implanted with a fluorine concentration of lO"~m-~ were also annealed at 700C (figure 17). Broadly similar behaviour to that of the SCH material is observed. In this case, however, unimplanted samples showed a 10 meV shift compared with less than 4 meV in the SCH example. This appears to be related to a difference in the etch-pit densities of InP substrates used for the two samples-Glew et al 1521 have reported that samples with high etch-pit densities show less intermixing. In order to explain this result, they propose that dislocations in the substrate are incorporated in the epitaxial layers as they are grown. The dislocations then act as traps for mobile point defects. They have used this

Figure 14. Exciton shift as a function of anneal time at 650C for P-quaternary scn samples capped with either 1000 A o f Si,N, or SiO, or no cap.

Figure 16.

P-quaternary scn annealed at 700C implanted

with a fluorine concentration o f 10"cr1-~.

Exciton shift as a function of anneal time at 650C for PSCH implanted under various conditions to give the fluorine concentrations as indicated in t h e ow structure.
Figure 15.

quaternary

1146

Quantum well intermixing

801

'

'

"

'

'

'

'

"

"

'

'

'

'

'

'

'

i -

Nolmplonr

5 . 4 . Propagation losses in intermixed


GaInAs/GaInAsP waveguides In an attempt to increase the degree of disordering in the fluorine-implanted waveguide structure (fluorine concentration 2 x 10" a n ' ) ) ,whilst leaving unimplanted control samples unchanged, repetitive RTP (RRTP) was used. The conditions for RRTP were 700C for 2 s repeated six times. The sample was left to cool for 20s between anneals and the rise time used during heating was 15 s. Single-mode strip-loaded waveguides were then produced in the GRIN samples described above by dry etching in a methane-hydrogen atmosphere. The waveguide losses were measured by the Fabry-PBrot technique [53] at 1.556 pm using a DFB laser. It was found that for a fluorine-implanted GRIN sample the loss is 8.5 dB cm-', whilst for unimplanted samples, in which the bandgap had been widened from 1.51 pm to 1.47 pm solely by thermal annealing, the loss was around 15 dB cm-'. In one fluorine-implanted and annealed sample the waveguide loss was apparently much lower -around 0.6dBcm-'. These results make IID in the InGaAsPnGaAsP system using fluorine a very important prospective fabrication technique for integrated photonic devices.

2
e .-

60

Flmplanr

1 " 4
c

?(I

0
' ' '

0
'

0
'
1

t
, ,
'

04 20

'

'

'

25

30 Time /s

35

40

Figure 17. P-quaternary GRIN annealed at 7 0 0 ' C implanted with a fluorine concentration of 10''

model to explain qualitatively the intermixing behaviour of a large number of samples. In order to make a direct comparison between furnace annealing and RTP (and so circumvent the problem of variations in the etch-pit density from wafer to wafer), samples from the same wafer were studied in both systems. Figure 18 shows the results of annealing at 750C unimplanted samples and samples implanted with either fluorine or boron with a volume concentration of The control samples from this wafer 2.4 x loi8 gave clear PL spectra with less than 5 meV shifts for all 30 s anneals, and shifts up to 18 meV after 5 min at 750C. Boron-implanted samples gave a red-shift of up to 40 meV after a 5 min anneal at 750C, while fiuorineimplanted samples always produced a blue-shift. A shift of 30 meV was obtained by annealing at 750 "C for 30 s with the corresponding control sample giving a shift of less than 5 meV, indicating that very little disordering had occurred in the control but a high degree of disordering had taken place in the fluorine-implanted material.
100

6. The neutral impurity disordering mechanism

. + Control
~~

% . s
E
X

60
40 -

Fluorine Boron

20:
0-20 -

2 E s .$
U

( . i

-40:
.MI, 0
,

,
5

,
10

,
15

,
20

,
75

30

Time I minutes
Figure 18. Exciton peak shift o f the unimplanted and implanted P-quaternary GRIN samples annealed at 750C where all the samples are taken from the same wafer. The two shorter anneal times were carried out in an RTP system, whilst the long anneals were carried out in a

conventional furnace.

The mechanism by which implanted boron and fluorine atoms disorder 111-V semiconductor QW structures is not clear at present. Disordering rates are ultimately controlled by the local concentration of native defects, so the presence of the impurities must result in an increase in the point defect concentration. It has previously been demonstrated that for the 111-V semiconductors the defect concentrations (and hence disordering rates) are determined by the Fermi energy in the crystal and the group V overpressure [16]. Our results for the GaAs/ AlGaAs system show that although fluorine is an effective disordering species it is essentially an electrically neutral impurity at room temperature, and disordered waveguides do not exhibit significant free-camer absorption. Boron, though less effective, exhibits similar behaviour. In the GaInAs/AlGaInAs system, the comparison of the shallow and deep implants leads to the conclusion that, of the two species studied, only fluorine is an active disordering species. The damage created per unit volume within the wells is similar for the shallow and deep implants, with fluorine creating around twice as much damage as boron. The fact that the observed bandgap shifts are similar for both deep and shallow implants of boron suggests that the presence of boron within the wells does not lead to additional disordering beyond that caused by implantation damage. In the case of fluorine, however, significantly larger bandgap increases are seen for the shallow implants, demonstrating that fluorine is an active disordering species. As fluorine appears to increase the mobility of atoms on the group 1 1 1 lattice

1147

J H Marsh

site, the disordering behaviour of fluorine in the GaInAs/ AlGaInAs system is similar to that in the GaAs/AlGaAs system. It is also possible that fluorine could be an electrically active impurity at the disordering temperature. Boron has been tentatively associated with a variety of deep levels: in GaAs, with deep acceptor levels at 77 meV [54, 551 and possibly at 255 meV [56], and in AlGaAs with levels 23.5 meV and 167.4meV below the band edge [57]. In GaInAs boron is thought to be a deep donor with ionization energies between 0.2 and 0.27 eV [58]. The situation for fluorine is even less well established, but fluorine is also likely to be associated with deep levels and the results from electrochemicalprofiling studies [14] are not inconsistent with this. We therefore postdate that deep levels, associated with boron in the GaAs/AlGaAs system and with fluorine in both systems, become ionized at the annealing temperature. In the case of GaAs/AlGaAs (annealed at 890C) kT is a factor of 4.0 greater than at room temperature and for GaInAs/ AlGaInAs (annealed at 650C) kT is a factor of 3.2 greater than at room temperature. The resulting free carriers would give rise to an increase in the equilibrium point defect concentration, and hence disordering rate, at the annealing temperature.
7. Impurity-freevacancy disordering

beneath the cap. Further complications occur due to chemical reactions between the dielectric cap and the semiconductor. This is a particularly serious problem with Al-containing semiconductor alloys because Al is a very effective reducing agent. When an SiO, cap is in direct contact with an AI-containing alloy then reactions of the type (2) occur, and the resulting free Si can then act as IID species. Any excess Si present in the dielectric film itself, resulting from non-stoichiometric deposition, can also act as an IID species. A dielectric cap is therefore needed which is inert, well matched to GaAs in its thermal and mechanical properties, and which protects the wafer during annealing cycles. Beauvais et al[60] have proposed the use of SrF, for this purpose and have studied IFYD in the GaAs/ AlGaAs system using three different dielectric caps, namely SO,, Si,N, and SrF,. We will show that the use of SrF, as a cap almost completely inhibits QWI under conditions where SO,, and to a lesser extent Si,N,, promote rapid intermixing. Two types of structure were studied to determine the influence of dielectric capping on the quantum well II = 1 (e-hh) exciton energy transition. An MQW structure designed to be a slab waveguide, similar to that described in section 4. was investigated with the purpose of shifting the absorption edge to higher energies in selected areas in order to create low-loss passive waveguiding regions. The second type of structure consisted of a SQW located at a shallow depth below the surface of the material (depth 5 0.1 pm). This material was designed for QW intermixing by IID with high spatial resolution. The energy shift of the absorption edge for both of these structures after annealing is highly sensitive to the presence of surface defects due to the location of quantum wells near the surface [61]. Three different dielectric caps-SO,, Si,N, and SrFJAlF, -were investigated to determine their influence on the excitonic energy transition by photoluminescence at 77 K. The silica and silicon nitride caps were plasma-enhanced chemical vapour deposited (PECVD), while the SrF,/AlF, was thermally evaporated from a tungsten boat. This mixture consists of 92% SrF, with 8% AIF, which is included to reduce the grain size of the evaporated dielectric layer. The processing for intermixing of the quantum wells of all the material discussed here consisted of annealing at 950C for 30 s with a rise time of 15 s. The samples were annealed face down on a piece of GaAs to provide proximity capping. A second piece of GaAs was placed over the back of the.samples. The shallow depth SQW material consisted essentially of an undoped structure with a lower barrier of 20 nm of Al,.,Ga,,,As, 2 nm of AlAs, a 5.0 nm GaAs SQW, 2 nm of AlAs, an upper barrier of Al,,,Ga,.,As and a 5 nm GaAs capping layer. The thin layers of AlAs located on either side of the SQW were inserted to increase the bandgap change caused by intermixing of the quantum well by providing a very high concentration of A1 atoms in close 3Si0,

+ 4Al-

2A1,0,

+ 3%

has mainly been achieved by ion implantation or diffusion followed by IID. As discussed above, only neutral IID can circumvent the large optical propagation losses associated with IID techniques. However, neutral impurities still introduce substantial changes in the material resistivity and trap concentrations, and the residual damage associated with implantation raises concerns about device lifetimes. Impurity-free vacancy disordering (IFVD) can create large bandgap energy shifts without these disadvantages. IFVD has been used in several device applications including bandgap-tuned modulators, particularly planar modulators [12], low-loss waveguide devices [17], and in a tentative demonstration of bandgap tuned lasers [13]. In the GaAs/AlGaAs system, SiO, is generally used to promote disordering whilst Si,N, is used to prevent disordering. In the P-quaternary system, however, Si,N, has been used to promote disordering, and an integrated extended-cavity laser has been formed in this way [l8]. Unfortunately, IFVD has poor reproducibility, especially from run to run [59], and the discrimination between the areas in which disordering is desired and those in which disordering is not required can be poor. In particular, silicon nitride films are rarely pure Si,N, and usually contain a substantial fraction of S O 2 . Also, films of Si,N, are usually highly strained and, although it is possible to control the residual strain at room temperature by changing the deposition conditions or deposition technique, when the samples are heated large differential strain effects occur. It is believed that both the presence of SiO, in the silicon nitride and high interface strain can effect disordering
QWI

1148

Quantum well intermixing proximity to the SQW. The thickness of the upper barrier of Al,,Ga,.,As was varied between 20 nm and 100 nm for the shallow depth SQW structure, and a fourth sample with a 1,um barrier was included in the study to examine the disordering process. at a much greater depth. The thickness of the dielectric caps for the SQW structures were 200 nm for SO2, 240 nm for Si,N, while for SrF,/AlF, the thickness varied from 105 to 185 nm. The PL results obtained after annealing of the shallow depth SQW material are presented in table 1. Large bandgap energy shifts (> 100meV) are observed for uncapped shallow SQW material as well as for samples from the same structures with silica and silicon nitride caps. The influence of the SrFJAIF, cap is quite apparent in restricting this energy shift to values less than 15 meV. A reduction of the energy shift is observed for the SiO, and Si,N, capped samples with the SQW located at an increasing depth below the material surface, down to less than 50 meV for the material with a 1 pm thick Al,,Ga,,,As upper barrier. SIMS studies of silicon implanted into GaAs/AlGaAs material [62] has indicated that silicon shows no significant diffusion when the material is annealed in an RTP at temperatures up to 1000C for 30 s. The results obtained here with energy shifts of 43 meV for the silica capping with the SQw at a depth of 1 pn below the sample surface thus indicate that some vacancy disordering must be occurring. The negligible energy shift in the uncapped sample for this struc; ture also indicates that the dielectric caps play an important role in acting as sources for these vacancies. This may be occurring by Ga diffusing into the dielectric layer for SiO, capping [13]. The out-diffusion of Ga into the Si,N, encapsulant is quite sensitive to the deposition technique used, the presence of oxygen in the dielectric layer [63] and the thickness of the capping layer. The measured index of refraction of 1.76 for the Si,N, caps studied here suggests the presence of some oxygen in the dielectric layer. The SrF,/AlF, cap, on the other hand, is a higher-density material with crystal parameters close to those of GaAs, which appears to make it impermeable to Ga atoms. This encapsulant is also easily evaporated and its capping properties appear fairly insensitive to the layer thickness. Some damage of the SrF,/A1F3 caps is apparent after annealing. This has not affected the luminescence properties of the samples, but it does make the subsequent removal of the cap very difficult. However, a layer of SiO, can be deposited over the dielectric cap to prevent this damage. The dual cap can then he removed by successive wet etching in H F and HCl solutions.
Table 1. sow exciton energy shifts measured by Al&a0.& barrier thickness (nm)
20 40 100 1000
PL

An MQw waveguiding structure was grown by MOVPE, consisting of a GaAs n + substrate, 0.5 pm of undoped 7 8 Aand ~ 54 periods of 6.8 nm GaAs. 2.5 pm of Alo.22Ga0 of GaAs and 6.0 nm Al,,,Ga,,,As. The same dielectric caps were studied for the SQW material, with thicknesses of 90 nm, 135nm and 185 nm for the SO,, Si,N, and SrF,/AIF, caps respectively. Dual-capped material with 185 nm of SrF,/AlF, and 150 nm of sputtered SiO, was also annealed and measured for an exciton energy shift. The observed luminescence peaks are presented in figure 19. The PL peak in the uncapped anneaIed material (curve b) is shifted to 23 meV above the peak in unannealed material (curve a) while the SiO, and Si,N, capped samples (curves c and d) have luminescence peaks which are shifted by 87 meV and 77 meV respectively to higher energies. It is seen that, as in the case of the SQW material, the SrF,/AlF, cap gives the smallest energy shift in the annealed material (7 meV, curve e). Dual-capped material (SrFJAlF, with SO,) exhibited an increase of 4 meV in the PL peak and the cap could be removed easily after annealing by the method described earlier. The PL intensities for all the annealed samples are comparable to that of the un-annealed material. It can also be seen that there is some broadening of the PL peak for the samples capped with SiO, and Si,N4 from 12.5 meV in the un-annealed material to 15.3 meV and 16.6 meV for these two caps respectively. The uncapped sample and the one capped with SrF,/AlF, have slightly narrower linewidths for the PL peak of 12.2 and 11.3 meV respectively. The measured linewidths, which are smaller than that of the un-annealed sample, may be attributable to variations in the uniformity of the material. A slight shoulder is also seen on the low-energy side of the PL peak for the SO,-capped material, indicating that some non-uniform mixing'takes place throughout the MQW structure. No such shoulder is observed in the PL spectra of the sample with the Si,N, cap. In comparing the results for the uncapped material with those from samples with a SrF,/AlF, cap, the
1 1 . 1

at 77 K

A (mev)
No cap

SiO, 260 250 142 43

Si,N,

SrF,/AIF,
7 14 12 17

1.50

1.Y

1.60

'

1.65

I70

1.75

149 192 124 4

286 274 179 38

Energy / eV

Figure 19. PL spectra measured at 77 K for (a) unannealed MOW material and annealed (b) uncapped material, and for samples capped with (c) SiO,, (d) Si,N, and (e)SrF,/AIF,.
1149

J H Marsh

effectiveness of this dielectric layer in reducing the energy shift of the exciton transition in the quantum wells is again clearly demonstrated. The large shifts (> 70 meV) for the Si0,- and Si,N,-capped material indicates that a combination of these two dielectrics with SrF,/AIF, capping could be very effective in performing selectivearea QW intermixing in order to fabricate low-loss passive waveguiding regions in GaAs/AlGaAs material. Both the ease of evaporation of the SrF,/AIF, cap and its excellent masking properties make it a good candidate for this type of device fabrication. In particular, the use of a SiO, layer deposited over the material surface patterned with SrF,/AIF, in the areas where the handgap is to be maintained unchanged should make it simple to obtain large energy shifts in the desired regions, and the entire dielectric double cap can he removed by wet chemical etching. This technique, therefore, overcomes many of the problems associated with IFVD in the GaAs/AlGaAs system. True selective-area intermixing becomes possible using straightforward dielectric deposition techniques. We have recently shown that PECVD or sputter-deposited SiO, arc equally suitable for promoting disordering. Problems with run-to-run reproducibility can be overcome if a short series of calibration tests are carried out for each deposition run. The material quality beneath the SrF, cap appears to he excellent, with processed lasers showing no significant changes in either emission wavelength or threshold current.

8. Photoabsorption-induced disordering

Like IFVD, LID processes do not necessarily change the impurity concentrations in a semiconductor structure. Both annealing of implanted samples [64] and transient melting of multiple layers by pulsed laser irradiation [65] have been shown to he effective. Transient melting using scanning of cw lasers has been utilized to introduce encapsulant Si into the epitaxial layers as a source for impurity-induced disordering which takes place during thermal post-annealing [66]. These techniques and their derivatives, although clearly effective, tend to be complicated by one or more of a number of additional processes, such as the introduction of impurities, various masking steps and high-temperature annealing, and can suffer from the increased free-carrier population introduced as well as from implantation-induced damage. LID processes, although having the potential to he impurityfree and to offer the possibility of direct-write capability, require high power densities to melt the material, can introduce thermal shock damage if used in a pulsed mode and can cause a potentially undesirable redistribution of dopants outside the active region of the device. Furthermore, melting of the semiconductor results in complete intermixing and the process cannot he controlled to give partial bandgap shifts. McLean et al [67] have developed an alternative laser-induced disordering process-photoabsorptioninduced disordering (PAID)-which takes advantage of
1150

the limited thermal stabilities of the GaInAs/GaInAsP and GaInAs/AlGaInAs systems. The method is impurity free, requires only a fraction of the power densities of existing c w techniques ( - 1-20 W mm-, compared [68] with lo5 W mm-') and does not involve a melt phase in the semiconductor processing. Most importantly it is layer composition-selective, which is additionally advantageous in that it is not restricted to near surface layers. We have carried out experiments on a wafer in which a MQW laser structure was grown on an n-type InP substrate by APMOVPE. The layer structure (figure 20) consisted of four InGaAs quantum wells, nominally 10 nm wide and lattice-matched to InP, separated by 12nm InGaAsP lattice-matched barriers and sandwiched between 0.09 pm thick GRIN InGaAsP confinement layers. The growth stage was concluded with a 1 pm thick InP cladding layer. A layer of SO, was then plasmadeposited to act as an antireflection coating. It has already been demonstrated [22] that this 111-V material system exhibits disordering at temperatures above 500 "C when annealed in a conventional furnace. The experimental arrangement for the initial experiments was straightforward. The sample was placed horizontally on an aluminium plate attached to an optical mount. The sample was then irradiated with the 1064 nm line from a Nd:YAG laser operating in cw and multimode. The total incident power density was 5 W mm-, and the total irradiation time was 30 min, including a ramp of approximately 2 min up to the maximum power level, in order to minimize thermal shock, Beam manipulation was effected with gold-coated mirrors and focusing was performed with a long-focallength lens. Photoluminescence measurements at 77 K were performed on the central region of the sample before and after laser irradiation. The longer-wavelength curve in figure 21 shows the quantum well exciton peak prior to laser irradiation, centred around 1.415 pm (876 meV).

Figure 20. Layer composition of t h e sample under investigation. Absorption at t h e irradiating wavelength of 1.064pm occurs in t h e active layer and t h e confining layers only, t h e cap layer of InP and the substrate being

effectively transparent.

Quantum well intermixing tion of heat. This heating then has the effect of enhancing the interdiffusion between the wells and the barriers. The extent of the heating depends on many factors including the total energy absorbed by the active region, which is in turn determined by the length of the absorbing region, the power density and the period of illumination, as well as the degree of heat loss due to heat transport through the substrate and due to radiation. In the sample structure investigated, absorption at the laser wavelength also occurred in about 30% of the GRIN confining layers, which led, assuming an optical absorption coefficient of 20000 cm-', to approximately 23% of the incident power being absorbed, compared with about 14% in the active region alone. This combination served to enhance the beating effect, while use of a filament pyrometer indicated that a temperature of 8W "C was typically obtained with the specified incident power density. Although such pyrometer measurements are subject to considerable uncertainty, this estimate appears reasonable when compared with conventional furnace annealing data for the same structure [22], which indicates that total disordering, with a shift 90-100 meV (less than observed here), occurs at of temperatures around 750C. Improved selectivity in the disordering process should be obtainable by choosing the laser wavelength so that absorption occurs only in the quantum wells, although a structure containing a larger number of wells would be required to provide the same total absorption as the structure investigated. Furthermore, the disordering mechanism should, in principle, be self-limiting since substantial absorption, and therefore continued beating, occurs only as long as the irradiating laser photon energy is larger than the bandgap energy of the material. Once equivalence is obtained, band-to-band absorption ceases, and heat generation in the active region is greatly reduced. Furthermore, the process is not restricted to disordering of structures with surface or near-surface multilayers, as it is not limited by the extent of diffusion of surface impurities into the material or by ion-implantation ranges, although it does rely on the upper cladding layers of the structure being transparent at the laser wavelength. In conclusion, the PAID displays significant advantages over existing LID techniques. So far, our experiments have been restricted to large-area processing. Studies are currently under way to determine the optimum irradiation conditions ,of power density, substrate temperature, period of illumination and wavelength, for various multilayer structures. Demonstration of the selflimiting nature of the process will also he of major importance, as well as the ability to obtain predefined optical characteristics. It is expected that optical monitoring of the bandgap of the sample during laser irradiation will provide a means for accurate control of the intermixing process, although complications may arise due to the elevated temperatures involved, producing shifts and broadening of the luminescence peaks. As the thermal conductivity of the InP substrate is higher (by more than an order of magnitude) than that of the active layers, heat transport is likely to be normal to rather than

Figure

structure, showing the exciton peak prior to (1.415pm) and following (1.241 p m ) laser irradiation at 1.064pm ( 25 W mm-' for 30 min). The dip in t h e short-wavelength edge is a feature of the luminescence apparatus and can be attributed to absorption by water vapour within the spectrometer.

*,,

PhotolumineScencespectra at ?? K for the

MoW

The dip in the short-wavelength side of the spectrum results from absorption by water vapour within the spectrometer. After processing, the peak had shifted to 1.241 pm (999 meV), also shown in figure 21. The energy shift of 123 meV is 15% of the initial room-temperature peak energy. At room temperature, the final luminescence wavelength was 1.299 pm (955 meV), which is significant since, assuming that the four wells intermix only with the equivalent of four harriers, the bandgap would he 1.358 pm (913 meV) for complete intermixing of the active region. This result suggests that complete disordering of the active layer did indeed occur. Furthermore the bandgap discrepancy suggests that substantial diffusion from the graded confining regions must also have occurred. The peak luminescence intensity apparently increases considerably, though this is likely to he due to absorption by water vapour in the air. No visible surface damage was evident after processing, although the sample was clearly glowing during the course of the experiment. The laser-induced disordering process which we have demonstrated relies, in principle, upon the differing bandgaps of the various semiconductor layers comprising the sample and on the choice of laser wavelength used for irradiation. The bandgaps of InP, In,_xGa&P,, (lattice-matched with x = 0.19 and y = 0.37) and In,Ga,-,As (x = 0.53) are calculated [36] to be 1.35, 1.084 and 0.72 eV respectively and hence InP is effectively transparent to the laser at 1064 nm (1.165 eV). However, band-to-band absorption does occur in the wells and barriers, where subsequent carrier relaxation and non-radiative recombination results in the genera-

1151

J H Marsh

in the plane of the layers, thus offering realistic prospects for lateral patterning either by direct-write or masking. We have demonstrated that an Au mask, deposited on top of the SO, cap, acts as an effectivemask, reflecting the laser radiation from the areas where intermixing is not required. By using pulsed-laser illumination to prevent steady-state thermal conditions arising, we anticipate being able to pattern on a micrometre scale. Improvements are currently being made to the experimental arrangement, most importantly the incorporation of more effective heat sinking of the sample, in order to dissipate excess laser power and heat transported through the substrate, and raising the heat-sink temperature to reduce the required laser power.

9. Extended-cavity GaAs quantum well lasers

To illustrate the use of one QW technique i n a photonic integrated circuit (PIC) application Andrew et a I [69] have fabricated and studied extended-cavity GaAs/ AlGaAs lasers containing an integrated passive optical waveguide formed using neutral IID. Semiconductor lasers are often line-narrowed by operating them in external cavities which are both bulky and subject to alignment problems. Integrated cavities are mechanically stable and around a factor of four shorter than the equivalent air cavity. Estimates of the linewidth reduction can be made using the Schawlow-Townes formula for the laser linewidth:

where the cavity linewidth

cg, and La are the absorption coefficientsand length of the active region, a, and L, are the absorption coefficients and length of the passive cavity region, nnponis the spontaneous emission factor, and a,, is the Henry alpha parameter. In a QW laser the gain coefficient at threshold is given by [70].
NJr where N is the number of wells, r, is the overlap between the optical wave and a single well, J a is the threshold current density, and yo = 840 cm-' and J , = 66 A cmd2 are theoretical values for 10 nm wells in GaAs. Double quantum well (DQW) metal clad ridge waveguide (MCRW) lasers were fabricated with extended passive cavities. The material structure was grown by molecular beam epitaxy (MBE) and consisted of a 0.23 p thick Al,,,Ga,,,As active core, containing two 1Onm GaAs QWS separated by a 10 nm barrier, surrounded by Al,,Ga,.,As cladding layers. Devices were fabricated with a variety of permutations of active and passive len&h, up to 600 pm in each case. The mask used during
1152

ion implantation was a 3 pm thick layer of SiO,. Fluorine was implanted into the unmasked regions with a dose of loL4 at an energy of 1 MeV, giving an implant depth into the semiconductor of 1 pm. Devices were placed beneath a GaAs proximity cap and were fumace-annealed at 890Cfor 90 min to produce a blueshift in the energy of the absorption edge of 40 & 5 meV in the implanted regions. Details of the complete structure of the MCRW laser and fabrication route have been given elsewhere [68]. The devices were pulse-tested using 400 ns pulses at a 1 kHz repetition rate. Figure 22 shows light/current curves for a normal 600 pm Fabry-PBrot laser and for an integrated device with a 600 jim long active section and a 600 pm long passive section. As a control, part of the wafer was masked completely to allow comparison of the compositecavity integrated devices with noma1 lasers of the same total length. The threshold current of the 1 2 0 0 p unimplanted laser was 160mA, much higher than the 60mA measured for the composite-cavity device. This indicates that fluorine passivates the disordered region electrically and that current is injected preferentiallyinto the active region, a result confirmed by making current/voltage measurements on an implanted waveguide section. The resistance of such a waveguide section, 300 pm long, measured across the epitaxial structure was in excess of 1 0 k n Integrated devices with 500pm long active and 300pm long passive sections similarly showed an insignificant increase in threshold current when compared with 500 pn long conventional lasers. The lowest measured threshold current of a 600 pm active/600 pm passive integrated device was 60 mA. The threshold of 600 pm Fabry-PBrot lasers fabricated from the same wafer was 55 mA. By a simple extension of the theory applied to active MQW lasers by McIlroy et al[70] the ratio of the threshold current of extended/normal devices can be shown to be

Gth= NT,y, In

(5)

(5)

I1

so 100 Current / mA

Figure 22. Pulsed lightlcurrent characteristics of 600 pm active and integrated 600 pm active/600jim passive cavity now MCRW lasers operating at 2 0 ' C .

Quantum well intermixing The threshold current of the lasers could be determined with an error of 2 2 mA, and so the experimental values of the above ratio gave a passive loss of 19 dB cm-' (9.5 dB m-' per well) with an uncertainty of less than rt 8.4 dB cm-l at the lasing wavelength of 0.87 pm. The lowest loss reported for comparable DQW devices fabricated by silicon disordering is 50.4 dB cm-' (25.2 dB cm-' per well) [71]. The lowest propagation loss we have obtained using fluorine is 4.7 dB an-' in an intermixed MQW waveguide and 10.3 dB cm-' in an intermixed DQW waveguide. Although the loss obtained in the extended cavity region is somewhat higher than these values, this particular structure contains active p and n dopants with which the guided mode overlaps and which contribute free-carrier absorption losses to the propagation loss. The expected contribution from free-carrier losses in the cladding layers is around 4-9 dB cn-', consistent with the total loss observed. Despite being operated under pulsed conditions, the lasers operated in a single longitudinal mode, as shown in This is possibly the result of figure 23, up to about 21th. reflections arising from the refractive index change between the active and passive sections giving rise to selection of a single longitudinal mode through coupled cavity effects. At higher currents additional longitudinal modes could be seen. Broad-area DQW lasers with a cavity length of 300 pm had a threshold current density of 420Acm-' (consistent with the values of yo and J , given above if an intemal quantum efficiency of 0.8 is assumed). The propagation loss of ridge devices was l o a n - ' when lasing (arising principally from free-carrier absorption from the carfiers injected into the active region). Using these parameters the estimated reduction in linewidth for the 600 pm active region integrated with a 600 pm extended cavity laser is only by a factor of 3.0. However, if it is assumed that the laser can be driven at around three times ihreshold (corresponding to 1200 A m - ' , the value of threshold current density in a typical doubleheterostructure laser), then the correspondingincrease in gain (equation (3)) implies that the maximum length of passive waveguide which can be used is 0.94cm. The linewidth will then be reduced by a factor of 62 In practice larger reductions than this would be expected because the extended cavity laser effectively dilutes the Henry alpha parameter.
10. Conclusions

Wavelength (nm)
Figure 23. Emission spectrum of an integrated 600 pm adive/600 pm passive laser just above threshold.

It has been shown that a variety of QWI intermixing techniques can be used to modify the bandgap and refractive index of QW structures formed in 111-V semiconductors. In this paper particular attention has been focused on techniques suitable for the fabrication of PICS in which optical propagation losses in intermixed material are ofmajor concern. Pioneering work by Thornton et al [30] has demonstrated that QWI can be used to fabricate moderately complex PIcs-in their case consisting of a gain region, an interconnecting waveguide and a electroabsorption modulator. The QWI technique used was IID using silicon. Although the resulting propagation losses were high (17m-'), the integrated devices all functioned individually as expected. The use of QWI techniques associated with low optical absorption would overcome many of the limitations experienced by these workers. By using gratings etched into the semiconductor surface to provide filtering and feedback, a true planar fabrication route to PICS can be envisaged. A variety of QWI techniques are available and each has characteristics which are desirable in different situations. It is likely that more than one technique will be used in fabricating a particular PIC. Taking, for example, the P-quatemary system, low-loss waveguides might be defined using fluorine implantation, whilst the bandgap of modulators might be tuned using the PAID process. In this way implantation damage can be kept remote from the electrically active devices by using only the PAID process in electrically active regions, but the excellent spatial resolution of implantation would be used to, define the position of the passive optical waveguides. Many issues remain to be resolved before QWI can be used routinely in the production of PICS. The reproducibility of certain processes, particularly IFVD, is relatively poor. Although the use of SrF, caps in the GaAs/AlGaAs system, as reported here, overcomes the serious problem of unintentional bandgap widening occurring where it not required, the magnitude of the shift beneath the SiO, capped regions varies from deposition run to deposition run. This suggests that the surface condition of the semiconductor is very important. Lateral control of the extent of intermixing is also desirable-recently Beauvais et al [72] have shown that this can be achieved by patterning the SrF, into small squares (using electron beam lithography) before depositing the SiO,. If the dimensions of the squares are small compared with the depth of the QWS then uniform intermixing appears to take place, while the degree of intermixing is controlled by the fraction of the area masked by the SrF,. This process, called selective intermixing in selected areas (SISA), allows a complete range of bandgaps to be created aaoss a wafer in a single processing stage.
1153

J H Marsh

Similarly, effects related to growth are important. Zinc, for example, is the most widely used p-dopant in
MOVPE, but zinc diffuses very quickly at elevated temperatures. Beryllium, on the other hand, is widely used i n MBE, and has a small diffusion coefficient compared with zinc. T h i s may give MBE-grown material an advantage in QWI processing. There appear to be other, more subtle, effects related to growth. We have found that, in the GaAs/AlGaAs system, MBE-grown material shows larger bandgap increases than comparable MovPE-grown structures when uncapped samples are subjected to thermal processing. On the other hand, when capped with SrF,, the stability of both MBE- and MOVPE-grown structures is excellent. This again suggests that the surface condition is of considerable importance in determining intermixing behaviour. In the P-quaternary system, the effect of the substrate etch pit density is significant [52] -as discussed in section 5.3, the intermixing rate of undoped structures is slower in samples with a high etch-pit density. In conclusion, Q W processes offer a simple and potentially low-cost route for fabricating PICS. Optimized discrete devices will always possess the ultimate performance-the compromises inherent in a n y integration process make it impossible to optimize every device in a PIC-but the advantages gained from integration in terms of reliability, cost and functionality should more than offset any loss in individual device performance. QWI techniques have the potential to be high-yield and have now been demonstrated to meet the performance requirements of PICS in terms of such parameters as propagation loss, low residual carrier density, retention of quantum confined Stark effect, lateral patterning and even lateral control of the degree of intermixing.

Acknowledgments

I am very grateful to all of my colleagues who have contributed to this programme of work. Financial support was provided by BNR Europe Ltd, by SERC under the DTI/SERC Optoelectronic Systems LINK Programme (GR/F/93913), and by SERC under grants GR/ F/b5248 and GRIGI13488.
References
[l] Miller S E 1969 Bell Syst. Tech.,J. 48 2059

[2] Alferness R C, Koren U, Buhl L L, Miller B I, Young M G, Koch T L,Raybon G and Burrus C A 1992 Appl. Phys. Lett. 60 3209 [3] Zucker J E, Jones K L, Newkirk M A, GnaU R P, Miller B I. Youne M G. Koren U. Burrus C A and Tell B 1992 Elecllon. Lett. 28 1888 [4] Aoki M, Takahashi M, Suzuki M, Sano y Uomi K, Kawano T and Takai A 1992Photon Technol. Lett. 4 580-2 [SI Colas E, Caneat C, Frei M, Clausen E M Jr, Quinn W E and Kim M S 1991 Appl. Phys. Lett. 59 2019-21 Demeester P, Buydens L and Van Daele P 1990 Appl. Phys. Lett. 57 168

[a

[7l Kapon E, Stoffel N G, Dobisz E A and Bhat R 1988 Appl. Phys. Lett. 52 351-3 [XI Wolf T, Shieh C-L, Engelmann R, Alavi K and Mattz I 1989 Appl. Phys. Lett. 55 1412-14 [9] Ralston J D, Camnitz L H, Wicks G W and Eastman L F 1987 Gallium Arsenide and Related Compounds I986 (Inst. Phys. Con$ Ser. 83) ed W T Lindley (Bristol: Institute of Physics) pp 367-72 [IO] Holonyak N Jr, Laidig W D, Camras M D, Coleman J J and Dapkus P D 1981 Appl. Phys. Lett. 39 102-4 [ll] Laidig W D, Lee J Wand Caldwell P J 1984 Appl. Phys. Lett. 45 485-7 [12] Ralston J D, SchaffW J, Bour D P and Eastman L F 1990 Appl. Phys. Lett. 54 534 [13] Deppe D G, Guido L J, Holonyak N Jr, Hsieh K C, Burnham R D, Thornton R L and Paoli T L 1986 Appl. Phys. Lett. 49 510-12 Marsh J H. Hansen S I. Brvce A C and De La Rue R M 1991 Opt. Quantum Electron. 23 S941 Suzuki M, Tanaka H, Akiba S and Kushiro Y 1988 J. Lightwave Technol. 6 779 Deppe D G and Holonyak N Jr 1988 J. Appl. Phys. 64 R93-R113 Suzuki Y, Iwamura H and Mikami 0 1990 Appl. Phys. Lett. 56 19-20 Miyazawa T, Iwamura H and Naganuma M 1991 Photonics Technol. Lett. 3 421 Guido L J, Holonyak N Jr, Hsei K C, Kaliski R W, Plano W E, Bumham R D, Thornton R L, Epler J E and Paoli T L 1987 J. Appl. Phys. 61 1372 Tan T Y and Gosele U 1988 Appl. Phys. Lett. 52 1240 Guido L J, Hsei K C, Holonyak N Jr, Kalliski R W, Eu V, Feng M and Bumham R D 1987 J. Appl. Phys. 61 1329 Marsh J H, Bradshaw S A, Bryce A C, Gwilliam R and Glew R W 1991 J. Electron. Mater. 20 973-8 Nakashma K, Kawaguchi Y, Kawamura Y, Ashasi H and Imamura Y 1987 Japan. J. Appl. Phys. 26 L1620 Komiya S, Tanahashi T and Umebu I 1985 Japan. J. Appl. Phys. 24 1053 Bryce A C, Marsh J H, Gwilliam R and Glew R W 1991 IEE Proc. Pt. J: Optoelectronics 138 87-90 Baird R J . Potter T J. Kothival G P and Battacharva P K 1988 Appl. Phys. Leir. 52 2055 T271 Baud R J. Potter T J. Lai R. Kothival G P and Battacharya P K 1988 Appl. Phis. Lett. 53 2302 [ZS] ONeill M, Bryce A C, Marsh J H, De La Rue R M, Roberts J S and Jeynes C 1989 AppL Phys. Let!. 55 1373 1291 ONeill M, Marsh J H, De La Rue R M, Roberts J S and Gwilliam R 1990 Electron. Lett. 26 1613-5 [30] Thornton R L, Mosby W J and Paoli T L 1988 IEEE J. Lightwaue Technol. 6 786 1311 van der Ziel J P, Ilegems M and Mikulyak R M 1976 Appl. Phys. Leir. 67 735 [32] Hansen S I, Marsh J H, Robeas J S and Gwilliam R 1991 Appl. Phys. Lett. 58 1398400 [33] Hansen S I, Marsh J H and Roberts J S 1991 IEE Proc. P!. J : Optoelectronics 138 309-12 [34] Razeghi M, Archer 0 and Launay F 1987 Semicond. Sei. Technol. 2 793 [3q Nakashima K, Kawaguchi Y, Kawamura Y and Imamura Y 1988 Appl. Phys. Lett. 52 1383-5 [36l Pape I J, Li Kam Wa P, David J P R, Claxton P A, Robson P Nand Sykes D 1988 Electron. Lett. 24 910-11 [37] Micallef J, Li E H and Weiss B L 1992 Appl. Phys. Lett. 61 435-1 [38] Pape I J, Li Kam Wa P, David J P R, Claxton P A and Robson P N 1988 Hectron. Len. 24 1217-18 [39] Pape I J, Li Kam Wa P, Roberts D A, David J P R,

- _

1154

Quantum well intermixing Claxton P A and Robson P N 1989 Gallium Arsenide and Related Compounds 1988 (Inst. Phys. Con$ Ser. 96) ed J Harris (Bristol: Institute of Physics) p 397 [401 Bradley M A, Julien F H, Gillies J P, Gao Y, Rao E V K, Razeghi M and Omnes F 1990 Electron. Left. 26 209 [41] Tell B, Johnson B C, Zyzkind J L, Brown J M, Sulhoff J W, Brown-Goebeler K F, Miller B I and Koren U 1988 Appl. Phys. Lett. 52 1428-30 L421 . .Sumida H, Asahi H, Jae Yu S, Asami K. Gonda S and Tanoue H 1989 Appl. Phys. Lett. 54 520-2 C431 Tell B, Shah J, Thomas P M. Brown-Goebeler K F. . _ Giovanni A D, Miller B I and Koren U 1989 Appl. Phvs. Lett. 54 1570 [44] Zucker J E, Tell B, Jones K L. Divino M D, BrownGoebeler K F, Joyner C H, Miller B I and Young M G 1992 Appl. Phys. Lett. 60 3036 [45] Kawamura Y, Asahi H, Kohyen A and Wakita K 1985 Electron. Lett. 21 219 r461 Glew R W. Garrett B and Greene P D 1989 Electron. Lerr. 25 1103 r471 Antvuas G A 1980 ADD[. Phvs. Lett. 37 64 c48j Nadory R E, Pollack' M A, johnston W D and Adam R L 1978 Appl. Phys. Lett. 33 659 [49] Pearsall T P 1982 GalnAsP Semiconductors (Chichester: Wiley-Interscience) [SO] Bradshaw S A, Marsh J H and Glew R W 1992 Proc 4th Int. Conf:on InP and Reluted Muteriuls (New York: IEEE) pp 604-7 r511 Ralston J D. OBrien S. Wicks G W and Eastman L S 1988 Appl. Phys. L e k 52 1511-13 [52] Glew R W, Briggs A T R, Greene P D and Allen E M 1992 Proc. 4th Int. Conf on InP and Related Mulerids (New York: IEEE) p 234 [53] Walker R G 1985 Electron. Lett. 21 581-3 [54] Dansas P 1985 J. Appl. Phys. 58 2212

[ S q Moore W J, Hawkins R Land Shanabrook B V 1987


Physicu 146B 65 [5q Fischer D W and Yu P W 1986 J. Appl. Phys. 59 1952 [57l Makita Y and Gonda S 1976 Appl. Phys. Lett. 17 333 [58] Tell B and Brown-Goebeler K F 1987 J. Appl. Phys. 62 813 [59] Koteles E S, Elman B, Melman P, Chi J Y and Armiento C A 1991 Opr. Quantum Electron. 23 S779-S787 [60] Beauvais J, Marsh J H. Kean A H, Bryce A C and Button C 1992 Electron. Lett. 28 1670-2 [61] Guido L J, Holonyak N Jr, Hsieh K C and Baker J E 1989 ADD[. Phvs. Lett. 54 262-4 [62] Uematsu M and Yanagawa F 1988 Jupun. J. Appl. Phvs. 27 L1734-5 [63] Kuzuiara M, N o d i T and Kamejima T 1989 J. Appl. Phys. 66 5833-6 [64] Rys A, Shieh Y,Compaan A, Yao H and Bhat A 1990 Opt. Eng. 29 329 [65] Ralston J D, Moretti A L, Jain R K and Chambers F A 1987 Appl. Phys. Lett. 50 1817 [66] Epler J E, Burnham R D, Thornton R L, Paoli T L and Bashaw M C 1986 Appl. Phys. Lett 49 1447 [67] McLean C J, Marsh J H, De La Rue R M, Bryce A C, Garrett B and Glew R W 1992 Electron. Lett. 28 1117-19 r681 EDler J E. Burnham R D. Tbornton R L and Paoli T L 1987 Appl. ,Phys. Lert.'51 731 r691 Andrew S R Marsh J H. Holland M C and Kean A H 1992 Photonics Technol. Lett. 4 426-8 [70] McIlroy P W A, Kurohe A and Uematsu Y 1985 IEEE J. Quunrum Electron. 21 1958 [71] Werner J, Lee T P, Kapon E, Colas E. Stoffel N G, Schwarz S A, Schwarz L C and Andreadakis N C 1990 Appl. Phys. Lett. 57 810 r721 Beanvais J, Avline. S G and Marsh J H 1992 Electron. Lett. 28 2i40

_ _

- _

- _
_ _

- -

~~~

1155

Anda mungkin juga menyukai