Semiconductor Heterojunction Topics: Introduction and Overview

Embed Size (px)

Citation preview

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    1/23

    003X-1101/X6 3.00 + 00

    ‘c 1986 Pcrgamon Press Ltd.

    SEMICONDUCTOR HETEROJUNCTION TOPICS:

    INTRODUCTION AND OVERVIEW

    A. G. MILNES

    Carnegie-Mellon University, Pittsburgh, PA 15213, U.S.A.

    (Received 17

    June

    1985; in

    reuisedform

    31

    Ju v

    1985)

    Abstract-Semiconductor heterojunctions with ideal lattice matching, well-controlled in fabrication, yield

    devices that cannot be achieved in any other way. These devices include modulated-doped high-speed

    field-effect transistors, ultra-high-gain and high-speed bipolar transistors, efficient injection lasers and

    light-emitting diodes and sensitive photo-detecting structures. Atomic reconstructions take place at

    heterojunction interfaces and are process-fabrication-dependent and not adequately understood. The

    barrier discontinuities observed are therefore scattered in value and also somewhat dependent on the

    determination method. Many papers in this Special Issue contain review aspects of these matters. Others,

    however, are specific contributions on very particular heterojunction topics. Not all aspects of heterojunc-

    tions are dealt with by the papers that follow, and the present article is intended for newcomers to the

    field as a brief commentary on topics that are not adequately represented.

    1. INTRODUCTION

    Developments in

    semiconductor heterojunction

    structures up to fifteen years ago are described in the

    book,

    “Heterojunctions and Metal-Semiconductor

    Junctions,” by Milnes and Feucht[l], and in “Semi-

    conductor Heterojunctions,” by Sharma and

    Purohit[2]. Various aspects have also been reviewed

    in other volumes [3,4].

    Up to the early 1970’s, the only heterojunction

    successes that had been achieved were those with

    injection lasers [3]. The GaAs/Al,Ga, _-xAs double-

    heterojunction laser fabricated by liquid-phase epi-

    taxy was exhibiting threshold-current densities of a

    few hundred amperes/cm* and ability to function

    above room temperature because of the carrier inver-

    sion-confinement and optical confinement made pos-

    sible by the band discontinuities. Other devices in-

    volving heterojunctions were of marginal value, such

    as the nCdS/pCu,S solar cell[5], or even existed

    only as concepts that had not been realized in any

    meaningful way.

    With the development of epitaxy methods, particu-

    larly organometallic CVD and molecular-beam epi-

    taxy in the mid 1970’s, growth technology improved

    in precision and ability to handle ternary and

    quatemary III-V semiconductors[4] and a new era

    began for semiconductor heterojunction devices.

    Control became possible in compositionOand in dop-

    ing over distances as small as 50 or 100 A and lattice

    matching was improved to 10~3-10~4. The devices

    that followed included high-quality lasers and optical

    detectors, negative-electron-affinity structures, mod-

    ulated-doped FET’s and high-gain high-speed bi-

    polar transistors. Also in laboratories across the world

    many structures are being studied involving hetero-

    junctions that are yet to find significant applications.

    The present limitations in the experimental control

    of heterojunction spike barriers and in the ability of

    theoretical treatments to match these barriers are

    briefly discussed in the section that follows. This

    issue contains papers by Margaritondo and by Wang

    dealing with these matters in detail. Section 2, also,

    touches on lattice matching and factors determining

    interface states. Three important uses of heterojunc-

    tions namely, FETs, bipolar transistors and light

    detectors and emitters are then mentioned and fin-

    ally, there is a brief discussion of quantum-well

    structures and strained-layer superlattices and

    speculative bandgap-engineering concepts.

    2. ENERGY BAND STEPS CREATED BY

    HETJZROJUNCTIONS

    In a homojunction the energy gap does not vary

    across the junction (except for possible bandgap

    narrowing produced by heavy doping) and so con-

    duction and valence-band-edge steps are similar and

    controlled by Poisson’s equation. However, in a het-

    erojunction between two different semiconductors

    the variations can be quite different since they may

    include energy steps AE, and AE,. that are abrupt

    to within a few tens of angstroms. These variations

    allow the designer a degree of freedom in indepen-

    dent control of majority- and minority-carrier flow

    (1).

    In metal-semiconductor junctions, the simple

    work-function electron-affinity model does not usu-

    ally predict successfully the observed Schottky-bar-

    rier heights[7-91. The failure to do so has been

    attributed to interface-state pinning and more re-

    cently the role of interface reactions has been

    evoked[lO, 111. Attempts to devise new simple mod-

    els, such as the common-anion model[7], useful

    though this is, have not stood the test of detailed

    experimental examination. In heterojunctions there

    is a similar failure of simple models. The first-order

    primitive model is that of Anderson-Shockley in

    99

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    2/23

    100

    A. G. I

    1I.Nt.S

    which AE, is postulated to be the difference of the

    semiconductor electron affinities [l, 21. This intuitive

    model is useful as an introductory concept for stu-

    dents, but the model is not well representative of

    experimental results. Quantum-mechanical modeling

    has been attempted with various degrees of success

    as discussed by Margaritondo and by Wang in this

    issue. The problem of obtaining such models in a

    form matching experimental results is frustrated by

    atomic reconstructions that take place at heterojunc-

    tion interfaces, even the most abrupt and carefully

    prepared [12]. Complicated situations cannot be ex-

    pected to lead to very simple models. The Ander-

    son-Shockley model, and the recent dipole-minimi-

    zation model of Tersoff[l4] will probably remain the

    main vehicles for introducing students to the subject.

    Some measured values of AE,, for a number of

    binary semiconductors with Si or Ge applied[l5] are

    presented in Table 1 and compared with the values

    expected from the electron-affinity model (EA). The

    EA predictions are often far from these measured

    values (16). Departures of i0.15 eV, or even more,

    are common.

    Tersoffs model postulates that each semiconductor

    has states in the bandgap at or near the interface and

    at some energy level EB - E,,, often near the mid-

    gap, the states change from acceptor to donor in

    nature. Tersoff proposes that when a heterojunction

    is formed the EB values on the two sides of the

    junction line up and this determines the band offset

    A E,,. If the band-diagram did not line up in this way

    then charge transfer would occur between the states

    on either side of the interface and this would create a

    dipole field: so from energy considerations the

    system favors dipole minimization and therefore

    equalization of the

    EB

    levels. Values of

    EB

    may be

    estimated from A E,, measurements for a set of het-

    erojunction pairs and a transitivity table prepared,

    something like the table presented by Margaritondo

    in this issue. The state-transition-energy EB may be

    Table 1. Experimental valence-band discontinuitiea

    compared with the values expected from the

    electron-affinity model[l5]

    Substrate

    Gc

    Si

    GAS

    GaP

    GaSb

    InAs

    InP

    InSb

    CdS

    CdSe

    CdTe

    ZnSe

    ZnTe

    Electron-Affinity Model

    Experimental

    Predictions

    (eW

    (eV)

    Si

    Ge

    Si

    Ge

    -0.17 -0.31

    0.17

    0.31

    0.05 0.35 0.27 0.70

    0.95 0.80 0.33 0.64

    0.05 0.20 -0.37

    ~ 0.07

    0.15 0.33

    0.15 0.46

    0.57 0.64 0.55

    0 85

    0.00 0.00 - 0.34

    ~ 0.03

    1.55 1.75 1.30

    1.61

    1.20 1.30 0.49 0.80

    0.75 0.85

    0.64 0.94

    1.25

    1.40 1.68

    1.99

    0.85 0.95 0.64

    0.96

    expected to play a similar role in Schottky-barrier

    formation if E, is a fairly intrinsic property of the

    semiconductor. Therefore the model provides a

    bridge between Schottky-barriers and heterojunc-

    tions since it implies that if Schottky barrier heights

    are known for two semiconductors for a common

    reference material such as Au, these results may

    perhaps be used directly to construct the heterojunc-

    tion barrier diagram for these two semiconductors.

    Further discussion of the Tersoff model appears in

    the papers by Margaritondo and Wang in this issue.

    The precise value of the model is something that

    time and further experimentation will establish, but

    at the very least it is useful as an academic tool in

    developing heterojunction concepts for students.

    Literature searches of the 2000 or more papers that

    have been published on heterojunction topics in the

    last decade show that experimental values of band

    discontinuities are scattered widely by perhaps f 0.1

    eV. Not all measurement techniques yield the same

    band discontinuities for specimens of the same semi-

    conductor compositions[17], and the nature of the

    preparation method may contribute to the values of

    discontinuities that are present. For instance. the

    A E,, values for Ge grown on (111) Ga. (100) Ga.

    (110,100) As and (111) As (111) As surfaces of GaAs

    have been reported as 0.48, 0.55, 0.56, 0.60. and 0.66

    eV, respectively[lS]. On the other hand, other

    workers have not seen orientation dependencies in

    AlGaAs/GaAs heterojunctions[l9] and the weight

    of the evidence is for no orientation dependence in

    well prepared junctions.

    Methods for measuring heterojunction barriers in-

    clude: (a) photo-emission spectroscopy relative to

    core bands (this involves taking the difference be-

    tween two large numbers that have corrections that

    are of order iO.l eV and are imperfectly known);

    (b) photoluminescence of superlattices with para-

    bolic wells (this rather critically depends on knowl-

    edge of effective mass); (c) CV apparent doping

    profiles (this is only really useful for nh’ and pP

    structures); and (d) measurement of A E,, by study of

    thermal emission of holes in a PIP structure [a tech-

    nique developed by S. L. Wright and J. Batey. J.

    Appl. Phys. 57(2), 484(1985)].

    The conclusions of such studies with the

    GaAs/AlGaAs/Ge system are that barriers of

    GaAs/AlGaAs are near-enough commutative,

    namely A E,?/’ = A EIfjA so the order of growth has

    relatively little effect. Secondly it is concluded that in

    this three-component system the barriers are fairly

    closely transitive, namely A

    E;:‘/’ +

    A

    E,f’< +

    A

    E,“’ 4

    = 0. The values are typically, according to Katnani

    and Bauer (1985 Electronic Materials Conference,

    Boulder, CO) AlAs/Ge A E,, = 0.78 k .07, AlAs/

    GaAs A E, = 0.39 of: 07 and GaAs/Ge A E,, = 0.44 f

    .06 eV.

    The variation of the barrier A

    E,,

    as a function of

    Al content (x) in the system Al,Ga, ~, As/GaAs

    has been reported as a linear relationship over the

    whole range x = O-l and is given by A E,, = 0.551(Al)

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    3/23

    Introduction and overview 101

    eV in PIP structures. This is not a fixed fraction of

    either the direct or indirect bandgap difference. In

    another study (M. Heiblum, 1985 IEEE Device Re-

    search Conference) structures that were metal

    (molybdenum)/iGaAs/nAl,Ga,_,As were ex-

    amined by internal photoemission for x in the range

    O-0.4, and it was concluded that A E,/A E, was

    about 0.60-0.64 in agreement with other work [20,21]

    and that A E,, was about 0.47x(Al) eV. For a AER of

    about 0.5 eV at x about 0.4, the 0.64 factor corre-

    sponds to a barrier for electrons of 0.32 eV.

    Several of the papers in this issue discuss hetero-

    junction-band-discontinuities. All measured values

    however are not presented and a transitivity table

    must be used or the general literature must be

    searched for AE,., AE,, values once a potential user

    has selected the heterojunction-pair of interest. As a

    guide to heterojunction-pair selection the following

    rules may be suggested.

    (a) Select the prime semiconductor of the active

    region with due regard for the bandgap and the

    mobilities required.

    (b) Select the paired semiconductor(s) on the basis

    of very close lattice match (say 10e3) and probable

    A E, , A E,, values from the literature.

    (c) This will commonly involve a binary substrate

    and a transition to ternary or quaternary com-

    pounds [22]. Select a fabrication process that is com-

    patible with this and is not likely to have severe

    cross-doping problems (such as may occur in

    Ge-GaAs structures if As enters the Ge). Further-

    more, thermal-expansion coefficient differences must

    be considered if a substrate such as Si is used. For

    lattice-mismatched heterojunctions, with mismatch

    greater that 0.5%, the odds are strongly against

    achieving useful performance if minority-carrier ac-

    tions are involved or if trapping actions are likely to

    adversely affect the performance of the device that is

    under consideration. Similar remarks apply to het-

    erojunctions involving polycrystalline layers.

    Strained-layer superlattices are presently under

    study as discussed briefly in Section 8 of this intro-

    ductory paper, as a way of tailoring effective lattice

    dimensions and bandgap modulation.

    3. LATTICE MATCHING AND INTERFACE STATES

    Lattice matching is a very important factor in

    determining the quality of epilayers in heterojunc-

    tion devices. A diagram of energy gaps versus lattice

    constants for a wide range of ternary three-five and

    two-six semiconductors is given as Fig. 1. Available

    substrates are normally binary in nature, such as

    GaAs or InP, since direct attempts to melt-grow

    ingots of ternaries usually result in phase-separation

    effects. The heterojunction pairs GaAs/Al ,.Ga, _ , As

    are naturally close in lattice constants, but can be

    further refined in lattice matching by the addition of

    small amounts of an element such as phosphorus.

    Growth can be by liquid phase epitaxy (LPE),

    molecular-beam epitaxy (MBE) or chemical-vapor-

    deposition (CVD). There is evidence that with LPE

    the interfaces are graded over 100 A or more because

    of back-etching of the melt[23] and may not be

    uniform on a scale of lo-50 pm in composition

    along the plane of the junction[24]. The LPE tech-

    nology is less frequently used these days, partly for

    this reason but mainly because CVD technology

    appears more interesting as a manufacturing process.

    In

    0 S,Ga, 47A~ grown on InP is another techno-

    logically important pair since the devices made pos-

    sible by this heterojunction are of use in optical-fiber

    communications. The conduction band discontinuity

    between these materials has been reported as about

    40% of the bandgap difference[2.5]. Stress and dislo-

    cations in such structures have been measured by

    Chu et a1.(26).

    Other alloy semiconductors may be grown lattice-

    matched to InP, as may be seen from Fig. 2 the

    quaternary alloy diagram for In, _.,Ga,P, _zAsz [27].

    Further data of this nature useful for planning

    lattice-matched bandgap-varying structures are found

    in Kressel and Butler[3]. Other useful references

    include Nakajima et al. [28] on In, ,Ga r As, )’PYon

    InP; Williams et al.[29] also on III-V quatemary

    alloys: Neuse 1301 on (AlGa) (AsSb) quatemaries;

    Osamara

    et al.

    [31] on Al-Ga-Sb; and Masu

    et al. [32] on (Al YGa, _ ,)In,,, As.

    Films of GaSb, SAs, 5

    and Al ,Ga, .~Sb,.As ~,

    have been grown by MBE, lattice-matched to InP[33]

    but with some evidence of compositional modulation

    due to spinoidal decomposition. CdS is also ap-

    proximately a lattice match to InP and a A E, of 0.56

    eV has been reported[34].

    In general if the device requires that a lattice

    mismatched film be grown, it is customary to provide

    a lattice-graded interface layer. For instance,

    In,, zGa,,

    As

    might be grown on GaAs by providing

    a sequence of thin layers stepped in composition

    between 0 and 20% In. The stepped composition of

    the buffer layer is cooled between each step to allow

    strain relief. By this process threading dislocations

    tend to be turned into the surface plane instead of

    continuing into subsequent layers.

    This kind of approach has been applied to the

    growth of GaAs on Si with the aid of a thin buffer

    layer of Ge. The Ge nucleation layer grown on the Si

    is heavily dislocated (1012 cme2) because of the 4%

    lattice mismatch, and the first layer of GaAs grown

    is similarly dislocated. Growth of the GaAs by a

    sequence of 10 growth interrupts, and thermal cycles,

    reduces the dislocation density by one or two orders

    of magnitude relative to GaAs grown without inter-

    ruption, which shows a 10’ cme2 surface dislocation

    density[35,36]. It must be noted however that the

    thermal-expansion coefficient of Si is less than that

    of GaAs and the grown layer is in tension and liable

    to cracking and there also may be bending of the

    wafer.

    Recently, however, it has been concluded that the

    provision of a thin buffer layer of Ge is not an

    advantage and that surprisingly good quality GaAs

    can be grown directly on (100) or 4” off (100) Si

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    4/23

    102

    A. G. MILNES

    3.6

    2.8

    2

    w” 2.4

    Y-

    8 2.0

    %

    $ 1.6

    &

    [L 1.2

    :

    0.8

    I

    I

    I

    I

    I

    I

    I I I I

    I

    I , I ,

    I ,

    I

    1

    I , I ,

    I

    5.40 5.46 5.56 5.64 5.72 5.00 5.88 5.96 6.04 6.12 6.20 6.28 6.36 6.44 6.52

    LATTICE CONSTANT, a0 % (300 K)

    Fig. 1. Energy gap versus lattice constant for three-t&z and two-six semiconductors.

    GaAs

    .Y

    GaP

    120

    1 10

    220ev

    210

    100

    200

    090

    1.90

    180

    f

    1.70

    160

    Fig. 2. Lattice constant and energy gap (77 K) as a function

    of composition for the quaternary alloy In, _ ,Ga,P, ;As,.

    The solid lines are constant energy gap curves; the dashed

    lines are lattice constant curves obtained by application of

    Vegard’s law. The intersection of the energy gap curves with

    the lattice constant curves makes it apparent that the energy

    gap may be varied while a fixed lattice constant is main-

    tained, indicating a variety of heterojunction possibili ties

    from the infrared to the yellow for the quaternary alloy.

    (After Coleman er al. Reprinted with permission from J.

    App. Phvs. 47, 2016 1976.j

    wafers by molecular-beam epitaxy. The Si wafers are

    subjected to a cleaning process at 900 or 1000” C and

    growth is begun at a low temperature such as

    400-450°C and at a slow rate. This presumably

    results in small closely spaced nucleation sites that

    interact with each other in such a way that by the

    time 1200-1500 A of layer has been grown the

    surface tends to be single crystal that is not exces-

    sively dislocated and a further thickness can be grown

    at a more normal temperature 570°C or higher, and

    at a normal growth rate (- 1 pm/hr). Quite good

    MESFET and MODFET performances have been

    achieved on such material (R. Fischer

    et cd.,

    1985,

    IEEE Device Research Conference, Boulder, CO).

    Furthermore it has been shown that GaAs grown by

    MOCVD on Si gives similar device performance (T.

    Nonaka et al., 1985, IEEE Device Research Con-

    ference, Boulder, CO).

    This is a promising technology since it may allow a

    mixture of III-V and Si devices on large commer-

    cially convenient Si substrates. The III-V devices

    offer optoelectronic performance possibilities that the

    Si devices cannot deliver. The GaAs layers on Si

    wafers have not been widely available for enough

    examination to determine their limitations. It may be

    surmised that they will be subject to having strain

    effects that might influence long-term stability and

    that the minority-carrier performance will be unim-

    pressive.

    Time will tell whether this technology

    survives as viable and worthy of transfer from the

    laboratory to production.

    In another approach GaAs of FET quality has

    been grown by MBE on Ge by the provision of a 1

    pm quantum-well-structure interface of GaAs-

    (AlGa)As [37].

    The quality of the interface region in a high-grade

    heterojunction such as Al rGa, _ 1As/GaAs is of spe-

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    5/23

    Introduction and overview

    103

    cial interest.

    When grown by MBE the interface

    width (lo-90% Alp-p height) has been measured as

    about 15 A by Auger profiling[23]. In general the

    interface of Al.Ga,-,As grown on GaAs (termed

    the “normal” interface) is superior in electrical prop-

    erties to the interface of GaAs grown on Al,Ga, _ .As

    (termed the “inverted” interface) [38]. The valence-

    band discontinuities have also been found to exhibit

    dependence on the growth sequence by some

    workers[39] and not by others. The reason for this is

    not properly known and the effect is usually dismiss-

    ed as a consequence of greater interface roughness

    and possibly greater contamination. In a quantum-

    well structure, photoluminescence possibly associ-

    ated with carbon has been detected in the first few

    tens of A of GaAs grown (40).

    Transport of electrons and holes across an ideal-

    ized (100) interface of GaA-GaAlAs has been mod-

    eled by Osbourn and Smith[41] who predict, for

    example, that an electron near the X minimum

    normal to the interface in Ga, _

    \

    Al,

    As

    should

    transmit into the X valley of GaAs with much

    greater probability than it transmits into the r

    minimum of GaAs.

    Continuing the discussion of the perfection of a

    heterojunction, consider now the interface states at,

    for instance, an n-GaAs/N-Al, 25Ga,,,, As inter-

    face (the large N is by convention associated with

    the material of larger bandgap). Measurement is

    usually made by causing a depletion region to spread

    through the interface as reverse bias voltage is ap-

    plied with the aid of a Schottky barrier on the

    nGaAs or in a

    p+

    nN

    structure (42). If there is a A

    E,

    at the interface and no interface electron traps, there

    will usually be an accumulation region and depletion

    region on the two sides of the inteiface. If in ad-

    dition to a A EC there is an interfacial sheet of traps

    the C-V profile of apparent concentration versus

    depletion width (42) becomes as shown in Fig. 3.

    Kroemer et

    al

    [43,44] have shown how to calculate

    the interface trap density and the band discontinuity

    A

    E,

    from such data and the method is valid even if

    the junction is graded. From Fig. 3 the apparent

    sheet density of electron traps may be determined to

    be 3 X 10” cmm2. Application of deep-level transient

    spectroscopy (DLTS) suggests that

    ~

    he traps are

    spread over a distance of about 140 A on the GaAs

    side of the interface (the GaAs having been grown

    after the AlGaAs in this specimen) with a concentra-

    tion of about 2 X 1016 cmm3. The trap-level average

    energy was estimated as about E,, - 0.66 eV. From

    the threshold-current density (about 3 KA/cm2) for

    an injection laser containing this interface a non-

    radiative interfacial recombination velocity S( =

    auN,) of about lo4 cm/s was inferred. For 3 x

    10”

    states cm-’

    this would represent a capture cross

    section u of 8 X lo-l5 cm2 in rough agreement with

    an estimate of capture cross sections from the DLTS

    measurements. The physical nature of the states

    is

    not known.

    18

    “; lo -

    P-n JUNCTION

    I

    I

    n-N INTERFACE

    2 ,+n-GoAs-----

    Y- -

    0 _

    F

    E

    Y

    s

    Y ‘7

    G 10

    FOR PROFILE

    g

    i 1 k, ( Jg,,21

    0

    1000

    2000

    3000

    DEPLETION WIDTH,W(V)(i)

    Fig. 3. C-V doping profile of P+- n-N DH laser structure

    used to study interface states. The position of the n-N

    interface at 0.17 pm was determined by direct measure-

    ments in a scanning electron microscope. The tic marks on

    the profile indicate the values of the reverse bias voltage in

    the C-V measurement. The Debye length of 100 A indi-

    cates the lower limit on spatial resolution[42].

    Consider other measurements of this nature. In a

    study of an Al,,,,Ga, xsAs/A10,4, Ga,,, As double

    heterostructure as a function of active-region thick-

    ness the value of S obtained was close to lo3

    cm/s[45]. Such a low value lends to have little

    adverse affect on device performance.

    Since 1.55 pm is the wavelength of minimum at-

    tenuation in optical fibers, attention has focussed on

    In

    o,sXGao,4,As which has a suitable bandgap (equiv-

    alent to 1.62 pm) and is a lattice match to InP. A

    confining material is needed for laser action and

    In

    o 52Al, 4RAs (bandgap 1.47 eV and a match to InP)

    has been used. Capacitance profiling applied to a

    Au/Ti N-In,,,Al,,,,As/n-In,,,Ga,,,As/n

    InP

    structure yielded 0.50 eV for the conduction band

    discontinuity (corresponding to about 70% of AE,)

    [46]. An interface charge density of 4 X 10” cm-*

    was obtained which is high enough to suggest that

    the junction was somewhat imperfect.

    A study has been made of the recombination prop-

    erties of LPE grown In,Ga,_,P/GaAs heterojunc-

    tions as a function of the degree of lattice

    mismatch[47]. This and other related studies have

    been discussed by Aspnes[48] with the results pre-

    sented in Fig. 4. The measured interface surface

    recombination velocity is seen to vary linearly with

    the lattice mismatch. The density of dangling bonds

    at the interface may be expected to vary as 8Aa,/a,‘,

    and so to be about 2.5 x 1013 cm-2 for a 1% lattice

    mismatch [49] and 2.3 x 10” cm- 2 for a 0.01% match.

    The relationship between the dangling-bond density

    and the number of misfit dislocations in the plane of

    the interface and threading dislocations developing

    from the interface depends on strain effects and

    impurity-related defects perhaps present at the inter-

    face. The relationship between the interface recombi-

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    6/23

    104

    A. G. MILNES

    Fig. 4. Interface recombination velocities for lattice-mis-

    matched heterojunctions as a function of relative mismatch

    c = Au,,/u,,. Data are as follows:

    a,~),

    In,Ga, _ ,P/GaAs.

    determined from measurements of photoinduced short-cir-

    cuit currents in p-n junctions; (O,O), In.Ga, ,P/GaAs.

    from heterojunction solar cells; (m) Pb,

    ,Sn

    ,Te/PbTe,

    minority-carrier lifetime measurements on double hetero-

    structure lasers; (+), Al,,,Ga,,As/GaAs. Filled and open

    symbols represent epitaxial layers in compression and ten-

    sion, respectively (after Aspnes (481).

    nation density NR and the dangling-bond N,, and

    dislocation densities is unclear. The recombination

    velocity is given by uv,NR where u is the recombina-

    tion cross section and U, is the thermal velocity

    - lo7 cm s-‘. If NR is taken equal to NDB, as

    assumed by Kressel[49], then

    s = olo7SAa,/a;

    = 2.5 x 10220Aa,/a,i.

    This leads to the theoretical lines shown in Fig. 4 for

    two assumed values lo-l4 and lo-l5 cm2 of capture

    cross section.

    If a film of thickness d is cladded on both faces by

    a heterojunction interface of recombination velocity

    S, and if the film thickness is much less than the

    minority-carrier diffusion length as in a laser, then

    the relationship that applies for the effective lifetime

    7 is

    where 7R is the bulk lifetime. If 7B is lo-’ corre-

    sponding to a GaAs minority-carrier diffusion length

    of about 10 pm and if

    d

    is 10m4 cm, then it is seen

    that a value of recombination velocity of lo3 cm/s

    or less makes 2S/d an acceptably small term in its

    effect on 7.

    If the results of Fig. 4 for

    In ,Ga, ,

    P/GaAs are generally valid, then lattice

    matching to closer than 10m4 may be desirable in

    most heterojunctions. However, the need for lattice

    matching may not be quite as severe as suggested by

    Fig. 4. There are studies in progress with lattice-

    mismatched heterojunctions that are not thin enough

    to be pseudomorphic that suggest that other effects

    enter, and that equating of recombination centers to

    the dangling-bond density of the simple mismatch

    model is open to question, perhaps by as much as

    two orders of magnitude. However, there is no doubt

    that the better the lattice match, the better the chance

    of the device being good.

    4. HETEROJ UNCTION FIELD EFFEfl

    TRANSISTORS

    This Special Issue on Heterojunctions has five

    papers dealing with field-effect transistors but these

    involve special topics. Therefore, some of the funda-

    mentals are discussed below and some key references

    given. The May 1986 issue of IEEE Transactions on

    Electron evices

    is to be devoted entirely to hetero-

    junction FETs so the present discussion is brief.

    GaAs because of its high electron mobility has the

    potential of outperforming Si FET’s. GaAs FET

    principles are discussed in a volume edited by

    DiLorenzo and Khandelwal[50]. There is further dis-

    cussion in the papers by Baliga

    et

    u/.[Sl] and by

    Wieder [52]. A major problem with GaAs is that it is

    not easy to obtain a low enough density of states

    at the interface between GaAs and an insulator

    (SiO,, Si,N,, Al,O,

    or native oxide) to make

    MISFETs that are effective. Hence CMOS operation

    is not easy to achieve.

    For low-power-dissipation circuits, where normally

    off devices are needed, it is necessary to use MESFET

    structures that are fully depleted by the inherent

    built-in Schottky barrier voltage and to bias these

    with a voltage of about 0.5 V to achieve turn-on.

    With these enhancement MESFETs the GaAs logic-

    gate-delay versus power-dissipation parameters tend

    to be in the region of the performance graph shown

    in Fig. 5. Thus the GaAs MESFET technology is

    superior in performance to Si technologies (except at

    present in yield and price). GaAs-heterojunction bi-

    Fig. 5. Power-delay regiona of various transistor technolo-

    gies. The diagonal lines are lines of constant power-delak

    product (from IEEE Spectrrtv~. p 2X

    (Feb. 19X4))

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    7/23

    Introduction and overview

    105

    polar transistors are seen to be significantly higher

    in

    power dissipation, with a slight edge

    in speed, but

    are capable of delivering larger output currents or

    powers for heavier loads. The Josephson-junction

    technology has reliability and ultra-low temperature

    system-integration problems and seems unlikely to

    be pursued at present.

    The technology labeled MODFET 77 K in Fig. 5 is

    of special interest. Modulation-doped field-effect

    transistors have very high electron mobility in the

    GaAs channel since this is maintained relatively free

    of ionized-impurity scattering because the dopants

    that supply the channel electrons are confined to an

    (AlGa)As region parallel to the channel that is sep-

    arated from it by a heterojunction conduction-band

    barrier. This effect is useful even at 300” K but

    becomes impressive at 77 K. The structures are also

    termed selectively doped FETs (SD FETs) or

    high-electron-mobility transistors (HEMT) or two-

    dimensional electron-gas FETs (TEGFETs). The

    concept dates from Bell Laboratories work in 1978

    and the history to 1984 has been reviewed by Morkoc

    and Solomon [53] and the materials aspects discussed

    by DiLorenzo et al. [54].

    A typical MODFET energy diagram is that shown

    in Fig. 6. The A1,,,SGa,,,,As layer (0.1 pm) is doped

    1.5

    x

    101scm~3. Thus electrons pass over a 100 A

    undoped region into a heterojunction notch in a 0.2

    pm GaAs layer to form an almost two-dimensional

    surface-channel layer of lo’* electrons cm * [55].

    The Hall mobility in the surface layer is about 9000

    cm*/Vs at 300 K and perhaps as high as 2

    x

    10’

    cm*/Vs at 77 K (56). Typical transconductances are

    100~200 mS/mm at 300 K and 300 mS/mm at 77 K

    for devices of gate lengths l-3 pm[57-591. Matters

    that have been of concern in the fabrication of

    MODFETs include the following.

    N

    5

    = 1.14 X td2cm-2 (300K)

    Depleted

    I

    I

    Fig. 6. Typical MODFET (AlGa)As-Ga.As energy diagram

    (after Morkoc).

    (a) Imperfections of the interfaces, particularly the

    inverted (GaAs grown on AlGaAs) interface [60-621.

    (b) The proportioning of the device, including the

    role of the undoped (AlGa)As region, in improving

    the channel mobility and aPfecting the available sheet

    charge and so the device transconductance[63]. The

    selective placing of the Si doping impurities in the

    (AlGa)As effectively makes the devices insulated-gate

    field-effect transistors with the undoped (AlGa)As

    acting as the gate insulator[65,66].

    (c) The discovery that Si doping of the (AlGa)As

    causes deep donor vacancy complexes (termed DX

    centers) to develop in the doped layer and the

    MODFETs then exhibit collapse of I V characteris-

    tics in the dark and threshold voltage Vr instabilities

    associated with change of channel electron con-

    centration with light illumination and temper-

    ature [59,64]. The effect may be studied by examina-

    tion of a persistent-photoconductivity effect [68] at

    77 K and is the subject of papers in this issue.

    Si doping of GaAs does not produce deep donors

    and GaAs does not normally exhibit persistent pho-

    toconductivity. For improved threshold stability the

    doped (AlGa)As layer in a MODFET may be re-

    placed by an AlAs/n-GaAs superlattice (for in-

    stance an active 0.5 pm undoped GaAs layer fol-

    lowed by a 450 k superlattice of alternating 20 A

    layers of undoped AlAs and Si doped GaAs with

    an equivalent superlattice doping level of 2

    x

    10IXcm ’ [59].

    (d) Understanding the noise-level-determining fac-

    tors in MODFETs. Laviron et d. [69] find room-tem-

    perature noise figures of 1.07 dB with associated

    gains of 10.5 dB at 10 GHz for 0.5 pm gate length

    structures. At 100 K and 17.5 GHz the noise figure

    becomes 0.34 dB with associated gain 9.6 dB.

    (e) Attention must be given to minimizing the

    parasitic resistances and capacitances of the slruc-

    tures if the full advantages of potential high-speed

    performance are to be obtained[70-721. Dual-

    gate-heterojunction FETs of master-slave flip-flops

    formed by four cross-coupled

    AND/NOR

    gates where

    each four-input AND/ NOR gate consists of two dual

    gate SDHTs of 1 pm gate length[73] have operated

    as a divider at 5.5 GHz at room temperature and

    10 GHz at 77 K. One-kilobit RAMS based on

    MODFETs have been demonstrated with access

    times of less than 1 ns.

    (f) Another aspect of MODFETs that has been

    explored with some success is the feasibility of p-

    channel devices. Sheet-carrier concentrations of 1 x

    10’” cm’ have been obtained with 77 K hole mobili-

    ties of 3650-5000 cm’ V--’ s ’ in Be-doped

    (AlGa)As/GaAs structures[74]. Transconductances

    of about 30 mS/mm at 77 K have been obtained for

    1.5-2 pm gate lengths[75] and p-channel devices

    exhibit very little shift of threshold voltage with

    light [76].

    Structures that consist of (1nGa)As grown on InP

    have been giving interesting performances. n-

    In

    o &a,, 47

    As doped 10” cm

    ’ exhibits a 300 K

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    8/23

    1 6

    A. G. MILNES

    electron mobility of 8000 cm*/Vs, which is almost

    twice that for GaAs of comparable doping. The gap

    between the r central valley and the L satellite

    valley is larger than for GaAs and therefore an

    electron overshoot velocity that is higher by a factor

    of two is perhaps achievable[77].

    A number of junction or insulated gate FET

    structures have been made in (InGa)As[78-821

    and selectively doped FETs have been made

    with Ga,,,In, 53As/Alo4XIn,,,,As [83]. Recently,

    Rosenberg et al. (1985 IEEE Device Research Con-

    ference, Boulder, CO) have demonstrated a high

    electron mobility transistor that has a GaAs sub-

    strate and a thin (300 A) pseudomorphic-strained

    active layer of In,,,Ga, ssAs followed by a GaAs

    spacer layer and a GaAs

    (

    n = 2-8 x 10’%mP

    s

    layer

    to provide the carriers for the 6 X 10” cm -’ electron

    gas. For 1 pm gate lengths the g,, observed was 170

    mS/mm at room temperature and the 77” K elec-

    tron mobility was 40,000 cm*/Vs. An important

    feature of the device is that it does not exhibit

    collapse of the IV characteristics in the absence of

    light since AlGaAs and its D-X center is not part of

    the structure.

    n+

    Substrate

    b Collector

    (a)

    n+

    1

    ,i

    Substrote

    Emitter

    (b)

    Much remains to be studied in high speed FET

    GaAs structures[84-881 including the performance

    of various vertical- and permeable-gate structures

    [89-921.

    Fig. 7. The two single heterojunction bipolar transistor

    structures discussed in the text. (a) The conventional

    emitter-up/collector-down configuration. (b) The inverted,

    emitter-down/collector-up configuration. Note the relative

    positions of the narrow and wide bandgap layers in the two

    structures; note also the position of the heterojunction. The

    devices illustrated are single heterojunction devices but the

    5. BIPOLAR HETEROJUNCIION TRANSISTORS

    In conventional (homojunction) n+pn bipolar

    concept is equally valid for double heterojunction tran

    sistors[95].

    transistors the emitter is more heavily doped than

    the base to control the reverse injection from base to

    emitter that otherwise would reduce the current gain.

    The valence-band barrier AE,, for heterojunction

    transistors, such as the N(AlGa)As/pGaAs/nGaAs

    structure, eliminates this problem and the base re-

    gion may be more heavily doped than the emitter[l].

    This results in a base resistance that is reasonably

    low even if the base is made very thin to lower the

    base transit time. The low base resistance therefore

    reduces the base-emitter and base-collector RC

    time-constant terms that also limit the frequency

    response of the device provided the associated capa-

    citances can be held small by emitter and collector

    doping control.

    One of the important applications of heterojunc-

    tion bipolar transistors is likely to be in analog to

    digital converter circuits where very high sampling

    rates are desirable and the circuits require the high

    driving power (large transconductance) of an HBT.

    use of an inverted (collector-up) design in which the

    current flow is restricted to the small collector region

    of the structure by the barriers of the wide-gap npi

    junction near the base in Fig. 7(b) if this is of low

    leakage current. With such a structure the relative

    frequency performance of bipolar heterojunction and

    homojunction-transistors[95] has been estimated to

    be about 1.7:1 although this is yet to be demon-

    strated. An (AlGa)As/GaAs heterojunction tran-

    sistor should have a much higher current gain than a

    GaAs homojunction transistor if the emitter-base

    interface is well prepared[96] but this is not always

    seen for reasons that are not fully examined, never-

    theless serviceable current gains of lo-100 are com-

    mon. The modeling of (AlGa)As/GaAs devices pre-

    sents special problems that have not yet been fully

    resolved[97]. The papers by Ankri et al. and Fischer

    et al.

    in this issue discuss matters relevant to

    emitter-injection action in the (AlGa)As system.

    The achievement of very-high-frequency perfor-

    In a typical heterojunction bipolar transistor the

    mance in a bipolar heterojunction transistor depends emitter may be nAl,,,Ga,,As, the base pGaAs (0.1

    critically on minimizing device parasitic time pm 10” cm 3, and the collector nGaAs on a semi-

    delays[93]. In a normal transistor configuration the

    insulating substrate. The two emitter fingers may be

    collector area is larger than the emitter area, as 4.5 pm wide and 10 pm long. In such a design[98]

    shown in Fig. 7(a), so the collector capacitance-base

    the current gain

    i,/i,,

    observed was 90 and the

    resistance time constant in a heterojunction tran-

    transconductance per emitter width was 4750

    sistor may be a factor limiting the frequency re-

    mS/mm. The common-emitter current-gain cut-off

    sponse unless this is addressed in the design.

    frequency fr was 25 GHz corresponding to an

    Krocmer [94] has shown how this may be reduced by

    emitter-to-collector transit time of 6.4 ps of which

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    9/23

    Introduction and overview 107

    5.2 ps was emitter charging time. It was considered

    that this could be reduced to obtain an fr of 50

    GHz. Other research groups have obtained compar-

    able performances [99-1021, so heterojunction bi-

    polar transistors are promising devices for ultra-

    high-speed integrated circuits. Propagation delay

    times of about 50 ps per gate have been achieved in

    ring oscillators. This issue of

    Solid-State Electronics

    contains a study of current-mode-logic HBT circuit

    performance by Katoh et al. that predicts switching

    times between 30 and 10 ps. The d.c. characteristics

    of double heterojunction bipolar transistors are dis-

    cussed in the paper by Ankri et al.

    Grading of the base region of an HBT to improve

    base transit time by provision of a quasi-electric-field

    in the base has been examined by Hayes et al.[103].

    Heterojunction bipolar transistors tend to exhibit a

    significant collector/emitter offset voltage because of

    the difference in base-emitter and base-collector

    turn-on voltages unless a second heterojunction is

    provided at the base-collector junction. Composi-

    tional grading of the emitter to base region however

    smooths out the conduction band discontinuity and

    so reduces the emitter-base junction turn-on

    voltage[l04] and reduces the need for a double het-

    erojunction structure.

    Numerical modeling of heterojunction bipolar

    transistors with graded bases suggest that a current-

    gain cut-off frequency of 100 GHz should be ob-

    tainable [105]. Monte Carlo simulations [106,107]

    suggest that in a base region of 1000 A with a

    quasi-electric field of about 20 kV/cm there is near-

    ballistic transport of the electrons and a current-gain

    cut-off frequency of even 150 GHz might be ex-

    pected. Ballistic transport has also been considered

    in 0.25 pm gate-length FETs[lOS] with prediction of

    fr up to 160 GHz. Hot carriers may cross barriers in

    real-space transfer and become trapped [109], as in Si

    MOSFETs, so such predictions must be viewed with

    caution.

    The basic concept that electrons in short devices

    may attain high velocities and result in fast charge

    transfer is however a matter of continued inter-

    est [llO-1121. Ballistic transport implies collision-free

    transit over distances shorter than a mean free

    path for the particular electron energy. The presence

    of a conduction band spike (energy discontinuity) at

    a heterojunction emitter has some influence on the

    energy distribution of the electrons injected in the

    base and the mean free paths may be affected by this

    and contribute to the degree of ballistic transport in

    narrow base transistors. A GaAs planar-doped ver-

    sion of this has been reported by Hollis et al. but

    with limited performance[ll3].

    Although the discussion has centered around

    (AlGa)As/GaAs structures, heterojunction bipolar

    transistors have been fabricated in the (AlIn)As/

    (GaIn)As/InP system [114,115]. More studies with

    this and other materials systems are in progress. It is

    noteworthy that

    nA1

    a 35Ga, 65P/pGaP/n GaP

    bi-

    polar transistors have demonstrated useful transistor

    operation at temperatures up to 550” C and might be

    of value for geothermal and other specialized appli-

    cations[ll7,118].

    In recent years Si bipolar transistors have been

    studied with emitters of SIPOS or n doped hydro-

    genated amorphous silicon [119,120]. The low mobili-

    ties and localized energy states that exist in such

    emitters are performance-limiting factors. Recently

    Sasaki et al. (1985 IEEE Device Research Con-

    ference, Boulder, CO) have reported an amorphous

    SiC:H emitter for a Si base-collector heterobipolar

    transistor. The AE . and AE values observed are

    0.16 and 0.54 eV, respectively, with the 1.8 eV emitter

    material.

    Also recently there has been progress with the

    growth of strained Ge,,,Si,, layers on Si and bipolar

    transistor action may be expected from these.

    6. LIGHT DETECI ION WITH HJXEROJUNCIION

    DIODES AND TRANSISTORS

    Silica fibers are of low loss in the 1.3-1.6 pm

    wavelength range and therefore there have been ex-

    tensive studies of light-detecting diodes and tran-

    sistors in this range involving ternary or quaternary

    III-V semiconductor heterojunction structures. The

    ability to sense low light levels with fast response

    time and high gains is of prime importance in optical

    communication systems.

    6.1.

    Diode Detectors

    Conventional photodiode structures have been well

    reviewed by Melchior [121] and others [122]. If a PIN

    form is used the photons are absorbed in a depleted

    region of high field and the gain is increased by

    avalanche multiplication and nanosecond electron

    transit times are readily achieved. However care in

    design and fabrication is needed to obtain uniform

    avalanche action [123,124] and low dark currents.

    The performance of a heterojunction avalanche

    photodiode with nIn, s,Ga,,,,As as the light absorb-

    ing layer and npInP as the avalanche multiplication

    junction is shown in Fig. 8. Separate absorption and

    multiplication regions tend to reduce tunneling com-

    ponents of dark current and in this structure 1 x 10m4

    A cmm2 at 0.9V, is achieved, where V, is the junc-

    tion breakdown voltage[l23]. When illuminated with

    1.15 pm light, the diode has a maximum multiplica-

    tion gain of 880 and an external quantum efficiency

    of 40%.

    Structures of this kind tend not to perform well at

    high bit rates because the bandwidth is restricted by

    hole trapping at the valence-band discontinuity at

    the InP/InGaAs interface. These holes must escape

    by thermal emission in the recovery phase and this is

    a relatively slow process. The incorporation of a 0.3

    pm intermediate layer of n In, ,Ga,, As,, 65PO 5 be-

    tween the

    nIn

    o.ssGa,47As (10 pm) absorbing layer

    and the npInP multiplication junction solves this

    problem and allows a gain of 40 or 50 at 0.9 VB and

    a dark current density of 5

    x

    10m4 Acm’ (13 nA)

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    10/23

    108

    A. G.

    MILNES

    10 -I-

    /l\zn-InGaAc

    . .

    - AuSn

    -11

    KI

    L I 1 , I

    0 40 20 30 40 50

    REVERSE BIAS VOLTAGE V,(V)

    I O3

    1 4

    J

    IO’

    si

    5 -E

    IO

    &

    5

    g -7

    5 10

    II

    -E

    10

    -E

    10

    -1c

    10

    (a)

    I

    I

    I

    I

    I

    I

    I

    I I

    I

    I

    10 20 30 40 50 60

    REVERSE BIAS VOLTAGE (VI

    (b)

    Fig. X. A hetcrojunction avalanche photodiode with

    In

    .)

    57Cra,,47As absorbing layer and nInP avalanche multi-

    plication layer and window layer [123]. (a) Dark currents as

    a function of reverse bias voltage for several diodes made

    from the same wafer. Inset shows a schematic cross section

    of the HAPD with buffer layer. (b) Typical

    against reverse bias voltage as a parameter

    length of incident light.

    coupled with excellent performance at

    420 Mb/s and 1 Crb/s[125].

    photocurrent

    of the wave-

    bit rates of

    In other studies with separate absorber and multi-

    plication regions, Al, 48 n, 52 n/Ga, 47In, 53As

    avalanche photodiodes have exhibited good perfor-

    mance [126]. A typical structure and the correspond-

    ing energy diagram are shown in Fig. 9. The ad-

    vantage claimed for this particular version of an

    SAM APD is that the A

    E

    value is low. This reduces

    the pile-up effect of holes at the notch and so the

    detector response time is shorter than for other

    material combinations.

    Hole trapping at a heterojunction interface is the

    essential feature of operation of the modulated bar-

    rier photodiode of Fig. 10. The device although not

    superfast is sensitive to low light powers and may

    have a d.c. optical gain of 1000 in the nanowatt

    illumination range[127,128]. The gain is caused by

    barrier lowering produced by the accumulation of

    the photogenerated holes in the valence band notch

    region at the source-gate interface. The rise time is

    about 50 ps and the fall time about 600 ps for 0.83

    pm pulse illumination. This may be compared with 2

    ns for a phototransistor in the low-gain base-con-

    trolled condition or lo-100 ns in the high-gain Roat-

    ing-base condition.

    A typical avalanche photodiode of 50-100 ps re-

    sponse time needs a bias voltage of 20 V or more

    A

    + *‘0.48I”0.52 As

    GY : : % ’ ’ ’ . _’ ’

    (a)

    nEc =0.55ev

    (b)

    Fig. 9. Avalanche photodiode of (GaIn)As/(AlIn)As. (a)

    Schematic of the 1.3 Frn avalanche detector with separate

    absorber and multiplication regions. (b) Energy-band di-

    . 1.

    agram unaer reverse-mas conaltlom

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    11/23

    Introduction and overview

    109

    m Ge-Au CONTACT

    (a)

    v

    (b)

    Fig. 10. An (AlGa)As/GaAs low-light photodetector struc-

    ture. (a) Schematic diagram of the majority-electron photo-

    detector. Incident photons are absorbed partly in the gate

    but most in the drain region. (b) Schematic representation

    of the energy band diagram of the majority-electron photo-

    detector under normal bias condition.

    although if avalanche gain is sacrificed a few volts is

    adequate.

    A low

    voltage may be generated by trans-

    mission of light energy through a fiber to what is

    essentially a miniature solar cell structure providing

    milliwatts of power

    [129].

    In a few applications there may be interest in

    high-speed photodetectors that require no external

    bias voltage. Such a structure is shown in Fig. 11.

    The light pulse passes through the (AlGa)As window

    layer and generates electron-hole pairs in the

    p- GaAs layer where they are separated by the built-

    in electric field of the junction region[l30]. Electrons

    drift towards the heterojunction interface of the

    selectively doped Al,,,Ga,,,As/GaAs structure and

    are collected by the Au-Ge electrodes. The photo-

    generated holes drift towards the pm GaAs and semi-

    insulating GaAs substrate and induce electrons in

    the conductive epoxy and a transient signal is cou-

    pled into the stripline. Pulses with rise times of 30

    ps, and 60 ps width at half maximum, have been

    detected. This detector appears suitable for integra-

    tion with modulation-doped field-effect transistors.

    Photoconductive detectors can in some respects be

    comparable to photodiode PIN detectors in high

    performance in high-data-rate long-wavelength

    lightwave communications systems. Photoconductive

    detectors can exhibit gains of a few hundred at low

    frequencies and gains of the order of 10 at high bit

    rates. The gain is less sensitive to temperature than

    for an avalanche photodiode. On the other hand,

    the detectors with such photoconductive gain show

    long (nanosecond) fall times. In a study of a

    Ga,, ,,Ine 53

    As photoconductive detector grown on

    an Fe doped semiinsulating InP, the received optical

    (b)

    To

    SAMPLING SCOPE

    2 DEG

    (cl

    Fig.

    11. Bias-free Al,Ga, _xAs/GaAs photodetector. (a)

    Cross-sectional view of the bias-free photodetector (not

    drawn to scale). The dashed line indicates the existence of

    the two-dimensional electron gas. Note that the Ge-Au

    contacts penetrate the n-Al,,,Ga,,As layer. (b) Schematic

    diagram showing the mounting scheme of the detector. (c)

    Energy-band diagram of the selectively doped

    Al, aGa, ,As-GaAs structure. The built-in electric field

    separates the photogenerated electron-hole pairs, as in-

    dicated by the straight arrows. The inserted circuit diagram

    illustrates the electron flows (indicated by arrows). The

    capacitance C is associated with the semiinsulating GaAs

    substrate. Resistance R, (or

    R2)

    represents the series resis-

    tance associated with the electrode No. 1 (or No. 2).

    power necessary for a bit error rate of lo-’ at 1.3

    pm was - 34.4 dBm at 1 Gb/s[l31,132].

    6.2. New

    Photodetector Device Concepts

    Although avalanche multiplication provides desir-

    able gain it also contributes to signal degeneration

    and noise. If the ionization coefficient for holes is

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    12/23

    110 A. G.

    MILNES

    comparable with that for electrons the holes created

    in the first ionizing action travel backwards and

    create further hole-electron pairs that have the effect

    of deteriorating the response to a pulse of light.

    Ideally then in avalanche photodiodes a high ioniza-

    tion rate for electrons (a) and low rate for holes

    (p) is desirable. In Si the ratio a/p is about 20

    and reasonably acceptable excess noise factors are

    obtained for Si avalanche photodiodes (APDs) oper-

    ating at photon energies above the Si bandgap (cor-

    responding to wavelengths shorter than 1.1 pm).

    However for smaller bandgap materials the ratio of

    a/b is not so favorable. Ideas have been proposed

    for overcoming this problem that involve gradedLgap .,

    APDs, superlattice and staircase APDs and channel-

    ing APD’s[133-1371.

    (a)

    The concept of the graded-staircase avalanche-

    multiplier photodiode is illustrated in Fig. 12. At

    E,

    each A E step the electric field is high and the

    electrons ionize hole electron pairs as suggested by

    the arrows in Fig. 12(b). The holes do not ionize

    L

    in their reverse flow to the cathode because the

    A E steps are small. In progress towards construc-

    tion of such a photodiode a superlattice of

    Al

    o 45Ga,, 55

    As/GaAs has been studied and an effec-

    tive a//3 ratio of 8 demonstrated. However graded

    structures are needed to eliminate electron trapping

    at AE< notches and there are problems still to be

    overcome. (b)

    The concept for a channeling avalanche photodi-

    Fig. 13. A proposed channeling structure

    for

    an avalanche

    ode is shown in Fig.

    13. The structure consists of

    photodiode. (a) Schematic of the channeling APD. (b) Band

    alternate widegap p-

    and low-gap n-layers and

    diagram of the channeling APD ( ER1 > E 2). c is the paral-

    lel field causing carriers to ionize. AE,. ias been assumed

    negligible with respect to A

    E,

    Fig. 12. Proposed staircase avalanche photodiode. (a) Un-

    biased graded multilayer region. (b) The complete staircase

    detector under bias. The arrows in the valence band indi-

    cate that holes do not impact ionize.

    voltage is applied so that all the layers are depleted.

    The explanation of device operation that follows is

    taken directly from the Capasso-Tsang-Williams

    paper [134].

    “Suppose that radiation of suitable

    wavelength is absorbed in the lower gap

    layers thus creating electron-hole pairs.

    The two p-n heterojunctions formed at

    the interfaces between the relatively nar-

    row bandgap and the surrounding higher

    bandgap layers serve to confine elec-

    trons to the narrow bandgap layers while

    sweeping holes out into the contiguous

    wider bandgap p-layers where they are

    confined by the potential. The parallel

    electric field c causes electrons confined

    to the narrow bandgap layers to impact

    ionize. Holes generated in this way are

    swept out in the surrounding higher gap

    layers before undergoing ionizing colli-

    sions in the narrower gap layers since

    the layer thickness is made much smaller

    than the average hole ionization distance

    l/p. In conclusion, electrons and holes

    impact ionize in spatially separated re-

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    13/23

    Introduction and overview

    111

    gions of different bandgap. Holes in the

    wider gap layers impact ionize at a much

    smaller rate compared to the electron

    ionization rate in the relatively lowgap

    material, due to the exponential depen-

    dence of

    a /3

    on the bandgap, so that

    a/@ ca n

    be made extremely large. Note

    that this structure has the advantage of

    providing a high

    a/P

    ratio at very high

    gains (> 100) because electrons and

    holes impact ionize in different materi-

    als. A typical design for 1.3-1.6 pm

    detectors would have

    p-

    and n-layer

    thicknesses of 0.5-1.0 pm and doping

    levels p = n = 10’6/cms. The

    p

    layers

    could be of InP or Al,,,In, 52A~ and

    the n layers of In,,,Ga,,,As. These

    materials can be grown lattice matched

    to a semiinsulating Fe doped InP sub-

    strate. The estimated a//3 ratio is = 350

    for a parallel field of = 2

    X

    lo5 V/cm

    at a gain of = 150 for a layer length of

    = 25 pm.”

    There are technological difficulties in fabricating

    such structures and it remains to be seen whether

    good performance can be achieved.

    6.3. Phototransistors

    Bipolar-junction and field-effect transistors may be

    used as light detectors. In the bipolar device the

    transistor is designed so that illumination creates

    carriers in the base region. The supply voltage is

    applied between the collector and the emitter, and

    the base floats at a potential VEB that suits the

    current flow and photoinduced carrier conditions as

    shown in Fig. 14. There is a buildup of excess holes

    in the base and so development of the voltage I’,,

    that allows a small hole current related to the photon

    flux to flow into the emitter. However, the voltage

    &.a causes a much larger current of electrons to flow

    from emitter to collector. Thus the photon-induced

    carriers are multiplied by the injection gain of the

    transistor, provided the base is free to find its own

    potential. A heterojunction transistor with wide-gap

    emitter provides an optically transparent emitter re-

    gion for photons within a certain energy range, and

    also the high gain associated with the A E,, barrier at

    the heterojunction interface. Interface recombination

    may reduce this gain and it is not unusual to find

    that the transistor gain is low at low currents or low

    light levels. At nW power levels the current gain

    (h,,) or optical gain may be 30 or 40 and rise to

    many hundreds at high signal power levels. The

    performance of heterojunction phototransistors in

    the material system InGaAs/InP has been studied

    by Chand et al. [138] who conclude that many of the

    recent reports of very high optical and current gains

    may involve avalanche multiplication as in Fig. 14(c)

    enhancing the gain of the HPTs. The gain depen-

    dence (on the base-collector voltage) reported by

    (b)

    Cc)

    Fig. 14. Heterojunction phototransistor action. (a) Oper-

    ation with floating base. (b) Energy-band diagram with hole

    accumulation in the base causing emitter-base bias and

    electron injection and collection. (c) Optical-input power

    and voltage dependence of optical gain.

    Chand et al. and attributed to avalanche multiplica-

    tion is not generally observed to such a degree. If

    holes are being generated by avalanche at a high rate

    and must escape from the base by emission into the

    emitter, then an unstable switching or looping-action

    might be expected. Since this is not seen then it is

    possible that heavy recombination is taking place in

    the base of Chand’s structure or at the emitter-base

    interface. Narrowing of the base width with increase

    of &.a is another factor that could contribute to the

    dependence of the gain on I&. There is more to be

    examined here in respect of HPT gain and frequency

    response. The response time of a graded-base photo-

    transistor may be a few tens of picoseconds[l39].

    Wavelength-selective photodetectors are of poten-

    tial interest in wavelength-division multiplexing

    transmission systems and in heterojunction tran-

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    14/23

    112

    A. Ci. bfII.NES

    sisters some degree of selectivity can be achieved by

    choice of a suitable absorbing layer in the optical

    path as shown in Fig. 15 [140]. The power dependent

    gain-bandwidth performance of heteroJunction bi-

    polar phototransistors for communication systema

    has been considered by Milano et ul.1141].

    Modulation-doped FETs may be used as photodc-

    SOURCE

    SEMI-INSULATING

    n+-GoAs12008,2x,d* cn?r

    UNDOPED Al, 3 Gao7As( 75iil

    tectors. The structure shown in Fig. 16(a) has a

    gate-drain spacing of > X pm and a gate length of

    >

    20 pm, and in spite of these large dimensions it

    exhibited (for GaAs laser pulses) a rise time of 12 pb

    with a full width at half maximum of 27 ps[140].

    The a.c.

    (> 20

    MHz) external quantum efficiency

    was nine times more than for a PIN diode.

    MODFET detectors therefore offer comiderable pcr-

    formance promise.

    In general a PIN detector must be followed by a

    0 Ge-AU

    m Al

    (a)

    A”FeN,,Au

    Ln DlFFlJSlON

    IllP

    InGaAs

    tl+-IllP

    transistor to provide increased gain. Fig. 16(b) shows

    s I

    IrlP

    the integration of a PIN InGaAs detector with an

    (b)

    InP MISFET transistor [143]. PIN-FET integration

    has also been based on the AlGaAs/GaAs

    Fig. 16.

    FET photodetection. (a) With modulated-doped

    FET[142]. (b) With PIN detection and integrated InP MIS

    system[144].

    FET[143].

    6.4. Photocwltuic Solur Cells

    There is extensive literature on heterojunctions

    applied to solar power generation [145~~147]. The May

    1984 IEEE Trunsuctlons on Electron Dec$ices is an

    issue devoted entirely to photovoltaics and a broader

    review appears in IEEE Spectrum for March 19X4.

    In general polycrystalline heterojunction solar cells

    show degradation effects and do not exhibit

    3

    AU-5

    hL

    , OHMIC CONTACT

    I

    SUbSTFcATt

    I

    \

    Au-9

    OHMIC CC,NTA:CT

    (3)

    the efficiencies needed to be cost effective (perhaps

    > lo-12%’ for air-mass-one conditions is needed in

    large arrays).

    Single-crystal solar cells based on the III-V system

    are capable of performances that exceed those of Si

    cells. Single junction efficiencies of 20% or more are

    possible with well designed and fabricated heteroface

    cells of AlGaAs/GaAs at high solar-concentration

    conditions. In multijunction cascade structures

    (tandem-cells) the predictions are for 35% efficiency,

    but for various technological reasons the present

    performance is below that of single crystal cells.

    7. HE~‘ER~.JUNCTIONLIGH~‘ EMI~IN~;OIODES

    AND

    INJEffION LASERS

    Heterojunctions have made significant contribu-

    tions to light-emitting diodes by allowing ternary

    and quaternary structures to be grown effectively,

    and by providing carrier confinement and low inter-

    facial-recombination and window action. Typical

    structures are shown in Fig. 17(a),(b),(c). The GaAs

    structure of Fig. 17(b) provides a CW output of 5.X

    mW, or a radiance of 92 W/sr cm’, at 150 mA [14X].

    The InGaAsP structure has a total hemispherical

    light output of 1.2 mW at 200 mA (10 KA/cm’) and

    is capable of coupling 40 PW of optical power into a

    63 PW core 0.21 NA optical fiber. The proJected

    lifetime at room temperature is estimated to be over

    10” h[149]. This issue of Solid-State Electronics con-

    tains a paper by Komiya et ul. that is concerned with

    luminescence from InGaAsP.

    Very often an LED has a hemispherical emitting

    structure to focus the light into an approximately

    I .,,

    Fig.

    15. A wacclength

    xlective heterojuncrion

    parallel beam by minimizing total internal reflection.

    InGaAaP/InP phototransistor-. (a) Structure \*ith selective

    However another approach by Thornton et (11.(1985

    ahsorbing layx (h) Band diagram.

    IEEE Device Research Conference, Boulder, CO) is

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    15/23

    Introduction and overview

    113

    LIGHT EMITTING REGION METAL CONTACT

    GaAs SUBST

    GoAs(

    1225

    GoAs

    (~ ) CONTACT

    GoAs SUBSTRATEtn

    I

    GaAs (n

    GaAs(p

    -2pm

    ‘Zn DIFFUSION

    (b)

    n-lnGaAsP(Eg =lOZ/~rn)

    (Eg =l 2711”)

    p-lnGoAsP(Eg =l 02pm)

    (c)

    AuSn

    Fig. 17. Typical light emitting diode structures. (a) Double

    heterojunction structure with GaAs light emitting region

    and Burrus-type geometry for coupling to an optical

    fiber.

    (b) Structure with Zn diffusion to reduce series resistance.

    (c) Structure with InGaAsP active layer that emits at 1.27

    to take a GaAs laser structure and almost entirely

    eliminate the reflection at one emitting surface by

    providing an antireflection coating of ZrO,. This

    suppresses laser action at high current densities (per-

    haps even

    to 6-8

    times the current density typical

    for laser action in a similar structure with a complete

    FabryPerot cavity). In this way 200-350 mW of

    CW light has been obtained from an edge-emitting

    device at room temperature with no active cooling

    and power conversion efficiencies obtained of about

    3%. The structureless far-field pattern so obtained,

    with absence of laser speckle, is of value in certain

    applications.

    The general principles of light-emitting diodes have

    been discussed by Bergh and Dean [150] and by

    Pilkuhn [151]. Extensive use of visible-light-emitting

    diodes already exists and may be expected to grow as

    high-brightness high-efficiency low-cost diodes are

    developed [152]. Diodes are available in red, orange,

    yellow and green. Blue emitting diodes of SIC are

    expected to come to market soon.

    Turning now to semiconductor injection-lasers, the

    role of heterojunctions has been of the essence al-

    most from their first inception. No attempt will be

    made here to review or highlight this role[3,153].

    The June 1983 and June 1985 issues of the IEEE

    Journul of Quuntum Electronics are given to semicon-

    ductor lasers. The

    IEEE Journal of Quuntum Elec-

    tronics also publishes many of the papers given at

    the International Semiconductor Laser Conferences.

    The IEEE Spectrum for December 1983 contains a

    review of single-frequency semiconductor lasers for

    fiber-optic systems. These references are sufficient to

    introduce newcomers to the field. Contributions on

    injection lasers were not invited for this issue of

    Solid-State Electronics in view of the very specialist

    nature of present developments and the coverage

    available elsewhere.

    8. QUANTUM-WELL STRUCTURES AND

    STRAINED LAYER SUPERLATTICES AND

    BANDGAP ENGINEERING

    Semiconductor superlattice and quantum-well

    structure studies were initiated by Esaki and co-

    workers about 1970. One kind of superlattice con-

    sists of periodic layers, a few hundred angstroms

    thick, of a homogeneous single-crystal semiconduc-

    tor with large doping swings to form n-i-p-i struc-

    tures. Such structures may be expected to have un-

    usual conductive, capacitance and optical properties.

    More usually quantum-well superlattice structures

    involve periodic thin layers of two (or occasionally

    more) different semiconductors and the semiconduc-

    tors are often selected to be closely lattice-matched.

    Sometimes they may be lattice-mismatched so that

    alternate layers are in elastic tension or compression

    and average in lattice constant to the lattice of the

    substrate on which they are grown. A journal Super-

    luttices and Micro-structures, published by Academic

    Press began in 1985.

    In a recent review of superlattices[l54] Esaki char-

    acterized them as types I-III as shown in Fig. 18.

    Type I occurs for systems such as GaA-AlAs and

    GaSb-AlSb, or the strained layer structure of

    GaA-Gap. The sum of A E, and A E,, is seen to be

    equal to the bandgap difference

    Eg2 - Eg,

    of the

    two semiconductors. The type II staggered structure

    is found in certain superlattices of ternary and

    quatemary IIIIV’s. Here it is seen that AE, - AE,,

    ELECTRONS

    Ec2

    TYPE

    in

    5

    Ev2

    Fig. 1X. Discontinuities of bandedge energies at four kinds

    of superlattice heterointerfaces: band offsets (left), band

    bending and carrier confinement (middle), and superlattices

    (right) [154].

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    16/23

    114

    A. G.

    MILNES

    equals the bandgap difference Egz - Eg,. The type

    II misaligned structure is an extension of this in

    which the conduction band states of semiconductor 1

    overlap the valence band states of semiconductor 2.

    This has been established as occurring for

    InA-GaSb. Electrons from the GaSb valence band

    enter the InAs conduction band and produce a di-

    pole layer of electron and hole gas shown in Fig.

    18(c). The zero gap HgTe/CdTe interface also has

    unusual properties since interactions between the

    hole bands of the CdTe and the HgTe occur.

    Semiconductor superlattices offer several perfor-

    mance features of interest for device physicists and

    engineers. The value of selective doping for

    MODFETs is one property that has already been

    amply demonstrated. A second property of value is

    in injection-laser structures where the provision of a

    quantum-well active region results in modification of

    the density of states from a parabolic distribution as

    in bulk material to a staircase distribution. This

    results in fewer electrons being needed to achieve the

    same cavity gain and therefore a lower threshold

    current density[l55] as illustrated in Fig. 19.

    Another feature of interest of superlattices is that

    defect and dislocation densities tend to be reduced in

    layers grown above superlattices [156,157]. Impuri-

    ties coming from the substrate are intercepted by the

    superlattice [158,159]. Device quality GaAs and

    AlGaAs can be grown on Ge or Ge/Si substrates for

    Al,Ga,_, As BARRIERS

    (a)

    BARRIER HEIGHT OF THE

    MULTILAYERS L m&l

    GaAs/Al,Ga,_,As

    AlAs MOLE FRACTION X IN Al,Ga,_,

    BARRIER LAYERS

    (b)

    Fig. 19. Quantum well injection laser structure and perfor-

    mance[155]. (a) The schematic energy band diagram from

    the modified multi-quantum well laser. (b) Shows the varia-

    tion of the averaged Jlh of several wafers as a function of

    their respective AlAs composition x (and barrier height) in

    the Al ,Ga,

    _ ,

    As barrier layers.

    MODFETs with the aid of an intermediate super-

    lattice layer. Interdiffusion at Ge/GaAs interfaces

    without a quantum-well superlattice has been studied

    by Sarma et al.[160]. Certain alloy systems such as

    Ga(As, Sb) have miscibility gaps but superlattices of

    them can be grown and matched to InP[161].

    The Ge-GaAs superlattice has excellent lattice

    match but may exhibit planar defects in the GaAs

    layers and these have been attributed to the anti-

    phase boundaries expected from localized nucleation

    of the GaAs on the Ge[162]. Antiphase disorder may

    be reduced by careful control of nucleation condi-

    tions as found for GaAs and AlGaAs grown on

    Si[163]. Control of nucleation has been studied by

    Beam

    et al.

    [164,165] in the growth of Ge, Si,

    ,/Si

    strained-layer superlattices and conditions for avoid-

    ance of island nucleation established. Modulation

    doping resulting in a two-dimensional hole gas has

    been demonstrated for such structures at 4.2 K with

    a h& of about 0.1 eV[165,166]. The combination of

    (GeSi) and Si is expected to be of device interest in

    the next few years.

    One feature of superlattices that is usually unde-

    sirable is that disordering is produced by impurities

    such as Zn at quite moderate temperatures

    (5755615°C) [167,168] and by Si at 850°C [169]. This

    tends to limit the formation of junctions to in situ

    doping during growth. The intermixing phenomenon

    induced by diffusion in superlattices has been dis-

    cussed by Van Vechten [170].

    Tunable below-gap radiation can be obtained from

    staggered line-up heterojunctions and quantum-well

    structures[l71]. Perhaps of greater device interest is

    the demonstration that a long wavelength multi-

    quantum-well laser with Ga,,,In,, 53A~ wells and

    Al

    o.4x n,, s2

    As barriers can be made to cover the

    range from 1.7 to 1.5 p,rn by adjustment to the well

    width from 1000 to 80 A[172].

    The thickness of the layers and the composition of

    alternate layers may be graded in a superlattice, as

    shown in Fig. 20 for a transition from InP to

    Ga,, ,,In, 53

    As. Such structures have been used in

    the fabrication of avalanche photodiodes[l73].

    A 50 period multiple-quantum-well superlattice of

    GaAs/AlGaAs in a PIN diode structure as shown in

    Fig. 21 has an optical absorption edge that is abrupt

    because of excitation resonances. Application of an

    electric held causes changes of carrier confinement in

    the wells and shifts the absorption edge to longer

    wavelengths. For 857 nm applied light a factor of 2

    change of transmission can be achieved with an 8 V

    bias on the diode and the switching time is suitable

    for high speed, ns fast modulation[l74]. Optically

    bistable light transmission has been demonstrated in

    such a structure[l75]. Waveguide action in super-

    lattices is discussed in the paper by Bhattacharya

    et al. in this issue.

    Another bandgap engineering idea that is emerging

    is the concept of selective mass tunneling filters.

    Tunneling probability depends exponentially on the

    barrier E, and carrier mass m as exp( - nr’/“E~/’ )

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    17/23

    Introduction and overview 115

    I

    I

    Goo.47’“0.53As

    I

    I

    I

    DISTANCE

    Fig. 20.

    Energy band gap of a graded gap psuedo-quaternary

    GaInAsP. The thicknesses of the InP and of the

    Gap ,,Inc 5~

    As are gradually varied between 5 and 55 A

    while keeping constant the period of the superlattice (= 60

    A). The dashed lines in (a) represent the average band gap.

    (b) Schematics of the HI-LO heterojunction avalanche pho-

    todiode incorporating the superlattice. (c) Electric field

    pro -

    tile[173].

    Fig. 21. Schematic view of an AlGaAs/GaAs superlattice

    optical modulator. The semiconductor layers are grown via

    MBE on a GaAs substrate, and then the diode is defined

    lithographically. The lower portion of the figure shows the

    calculated electric field strength, S, as a function of position

    within the device for two applied voltages[l74].

    Xl

    GoAS

    r3

    2

    x3

    hAS

    =1

    a)

    b)

    Fig. 22. Concept for a heterostructure light demultiplexing

    device[l76]. (a) Structure. (b) Energy diagram.

    Thus it is possible to consider a superlattice in which

    there will be selective tunneling of electrons but not

    of holes and this may have high photosensitive gain

    because the holes are essentially trapped while elec-

    trons make many transits of the device. Furthermore

    the electron band state formation in the quantum

    well insures that tunneling action selects electrons of

    particular energies hence giving an energy-resonant

    tunneling action. So it is possible to envisage an npn

    bipolar transistor that contains quantum wells in the

    base and that transmits to the collector only elec-

    trons in selected energy ranges that correspond to

    certain values of emitter base voltage. So the device

    when ultimately refined and developed to have an

    undulating output as the emitter-base voltage is

    increased may have potential as a logic device with

    multiple logic levels or possibly as providing a crude

    analog to digital action.

    Optical communication systems may in some

    applications be multiplexed with light of several

    wavelengths. Fig. 22 shows a heterostructure demul-

    tiplexer concept in which a detector has a well-struc-

    ture graded from GaAs to InAs. Selected absorption

    of light of different wavelengths in sequential wells

    may allow signal separation [176].

    For another potential application it has been sug-

    gested that strained layer superlattices in the InAsSb

    system may be suitable for long wavelength detector

    applications [177]. Suitable absorption must be

    achievable in a superlattice of reasonable length and

    this may be a problem.

    Finally in this examination of bandgap-engineer-

    ing a sawtooth superlattice structure proposed by

    Capasso et ~I.[1781 is shown in Fig. 23. The action

    of the device depends on the lack of reflection sym-

    metry and it is suggested that the structure may lead

    to high-speed displacement-current photodetectors.

    The explanation of the action given by the authors

    follows:

    . electron-hole pairs are excited by

    a very short pulse as shown in Fig. 23a.

    Electrons experience a high quasielectric

    field (typically > lo5 V/cm) due to the

    grading whereas the total force acting on

    holes is virtually negligible because of

    the valence band-edge lineup in p-type

    materials. Therefore electrons separate

  • 8/18/2019 Semiconductor Heterojunction Topics: Introduction and Overview

    18/23

    116

    A. G. ibflLNtS

    Fig. 23. Illustration of the formation and decay of the

    macroscopic electrical polarization in a superlattice strut-

    ture [178].

    from holes and reach the low-gap side in

    a sub-picosecond time (< 10 ” s).

    This sets up an electrical polarization in

    the sawtooth structure which results in

    the appearance of a voltage across the

    device terminals [Fig 23(b)]. This mac-

    roscopic dipole moment and its associ-

    ated voltage subsequently decay in time

    by a combination of (a) dielectric re-

    laxation and (b) hole drift under the

    action of the internal electric field pro-

    duced by the separation of electrons and

    holes.

    The excess hole density decays by

    dielectric relaxation to restore a flat

    valence band (equipotential) condition,

    as illustrated in Fig. 23(c). Note that in

    this final configuration holes have redis-

    tributed to neutralize the electrons at

    the bottom of the wells. Thus also the

    net negative charge density on the low-

    gap side of the wells has decreased with

    the same time constant as the positive

    charge packet (the dielectric relaxation

    time).”

    When excited with 4 ps laser pulses (X = 6400 A)

    at 86 MHz repetition rates, a structure graded from

    GaAs to Al,,Ga,,As exhibited a sharp rise-time

    output pulse with a 200 ps decay tail. The structure

    does not respond to a constant dc light signal. Un-

    like conventional semiconductor photodiode and

    photoconductive detectors, the current carried in this

    photodetector is of displacement rather than conduc-

    tion nature since it is associated with a time-varying

    polarization.

    9. CONCLUSIONS AND OVERVIEW

    Heterojunction devices envisaged many years ago

    have reached new levels of performance because of

    improvements in growth technologies and in physical

    understanding and examination of interfaces. This

    has led to revitalization of the concept of bandgap

    engineering and the free-thinking produced by this

    has suggested a number of interesting new device

    concepts. Some will survive and others no doubt will

    fall by the wayside, but progress is certainly being

    made.

    The preceding review has attempted to set the

    stage for the papers that follow in this special issue.

    Basic heterojunction barrier studies are represented

    by the papers of Margaritondo and Wang. Bipolar

    transistor studies follow with the papers of Ankri

    et al. and Katoh et al. Modulated-doped FET struc-

    tures are represented by the papers of Look and

    Norris, and Schubert

    et al.

    and Nathan. The first

    deals with channel mobility and the others with

    undesirable trapping effects. Contacts to such struc-

    tures are considered by Mukhergee et 01. and by

    Christou and Papanicolaou.

    Graded heterojunctions are discussed in the contri-

    butions of Fischer et al. and Petrosyan. Then quan-

    tum-well effects are examined in the papers by

    Masselink

    et al.

    and Bajaj

    et ul.

    and superlattices by

    Bhattacharya et al.

    Photoeffects in InGaAsP are studied in the contri-

    butions of Diadiuk and Groves, and Komiya et ul.

    and photoeffects in ZnSe-GaAs junctions by Zhuk

    et al. Trap levels in heterojunctions are often im-

    portant and a comparative study of admittance and

    DLTS spectroscopy for CdTe-ZnTe heterojunctions

    is offered by Khan and Saji. The growth of

    CdTe-InSb heterostructures is discussed by Blat

    et

    II.

    The issue concludes with a paper by the Morkoc

    group at the University of Illinois on the very prom-

    ising performances achieved for GaAs grown directly

    on Si.

    Ack~lo,l’led~~emerzts-Contributors are thanked for their

    papers and for allowing their manuscripts to be held while

    the complete set could be assembled for this special issue.

    NSF Grant ECS 82-14859 is acknowledged for partial

    support during the preparation of my contribution.

    1.

    2.

    3

    4.

    5.

    6.

    I.

    8.

    REFERENCES

    A. G. Mimes and D. L. Feucht, Hetero~u,rc~rro~zs crtrd

    Mad Semrconductor Junctrons,

    Academic, New York

    (1972).

    B. L. Sharma and R. K. Purohit, Senrrconducror Iler-

    eropnctions,