GaAs heterostructures

CRYSTAL GROWTH

Journal of Crystal Growth 127 (1993) 536—540 North-Holland

Monolayer scale study of segregation effects in InAs/GaAs heterostructures Jean-Michel Gerard

a,1

Cécile d’Anterroches

b

and Jean-Yves Marzin

a

“Centre National d’Etudes des Télécornmunicatiopts, 196 Auenue H. Racera, F-92220 Bagneux, France b Centre National d’Etudes des Téléco,n,nunications, CNS, Chemin du Vieux Che~ne,BP 98, F-38243 Meylan Cedex, France

We grow by molecular beam epitaxy and study by low temperature photoluminescence and high resolution electron microscopy pairs of identical AlAs/GaAs quantum wells, in which an InAs monolayer is inserted at nominally symmetric positions in the well, close to one or the other interface. The absolute displacement of indium atoms due to surface segregation processes is studied with respect to the well-defined GaAs/AlAs interfaces. Both techniques show that the InAs-on-GaAs interface lies at its nominal position, which gives the first clear example of a kinetical freezing of segregation processes in semiconductor heterostructures. PL data allow us to refine previous estimates of the segregation efficiency and to compare quantitatively various growth techniques.

The perfection of semiconductor heterostructures is ultimately limited by surface segregation processes, as shown for numerous systems, including InAs/GaAs [1], GaAs/AlAs [1,21 and Si/Ge/Sn [3]. The amplitude of these processes has been initially studied by conventional surface analysis techniques (AES, XPS): for instance, the analysis of nominally 1 monolayer (ML) thick InAs quantum wells (QWs) in GaAs reveal the surface segregation of In atoms, and their progressive incorporation in the GaAs matrix on a nanometre scale [1]. This results in a broadening of the GaAs-on-InAs interface, clearly visible on high resolution electron micrographs (HREMs) [4]. Both techniques suffer, however, from a rather limited precision when quantitative studies are undertaken. Accurate estimates of the segregation efficiency are, however, of paramount interest, in order to test growth conditions or growth techniques aiming at its reduction.

InAs is inserted at nominally symmetric positions in the well, close to one or the other interface, are grown by molecular beam epitaxy (MBE) and studied by low temperature photoluminescence (PL) and HREM. The absolute displacement of indium atoms due to the surface segregation is studied by HREM for the first time by using the well-defined GaAs/AlAs interfaces as markers. Due to the asymmetric broadening of the indium-containing layer, such QWs nominally symmetric of each other with respect to the centre of the well (NS-QWs) display marked bandgap differences: this effect allows one to study with great detail the actual composition profile of a nominal InAs ML [51,as demonstrated hereafter for samples grown by MBE at 530°C. Comparative estimates of the segregation efficiency for various growth techniques are finally given. We study in a first experiment two multi-OW samples, S1 and S2, grown at 530°C (i.e. just

We propose a novel approach allowing an accurate probing of interfacial compositional gradings to be made. Pairs of identical AlAs/GaAs quantum wells (QWs), in which a monolayer of

below the onset of thermal desorption of indium atoms in MBE). Each multi-QW contains three InAs/GaAs/AIAs asymmetric QWs; the I ML thick InAs layer is nominally inserted at a given distance a of the (first grown) GaAs-on-AlAs interface in S1 or of the AlAs-on-GaAs interface in S2. The distance a is respectively equal to 3, 5

Member of the Direction des Recherches Etudes et Techniques, French Ministry of Defence, 0022-0248/93/$06.00 © 1993

—

Elsevier Science Publishers B.V. All rights reserved

J.-M. Gerard et a!.

/ ML scale study of segregation effects in InAs / GaAs HSs

or 7 ML for the three QWs, and the well width is in all cases 28 ML. AlAs barriers are nominally 3 ML thick, and the QWs are embedded in thick Ga0 7A103As layers. These three pairs of NS-QWs have been studied by HREM. The study is performed in a 400 keV transmission electron microscope on crosssectional specimens prepared by ion milling. In the thinner areas of the sample, the contrast between GaAs layers and indium-containing layers is rather poor: we thus worked on a relatively thick specimen (about 200 nm). Fig. 1 corresponds to sample ~ The three InAs/ GaAs/ AlAs asymmetric QWs with a = 3, 5 or 7 ML are clearly identified. The projected AlAs barrier layer appears as a bright domain in the grey GaAs or Ga0 7A103As matrix, whereas indiumcontaining layers correspond to darker areas. The GaAs/AlAs interfaces look abrupt on the ML scale, with possibly few ML steps. Although Raman scattering experiments [2] have revealed some broadening of GaAs/AlAs interfaces, the composition is very different for the last Al-rich layer and the first Ga-rich layer, so that a welldefined interface is observed in HREM. The “InAs”-on-GaAs interface is well defined too, although it appeared even flatter, with few ML

a~3ML

537

steps, in previous experiments performed on thinner specimens, [4]. The contrast related to indium presence clearly extends over several MLs, with a decreasing intensity. It is, however, clear for a = 7 ML that the contrast associated to GaAs is not recovered before AlAs, i.e. indium atoms are spread during the growth all over this layer. This experiment clearly confirms that this asymmetric redistribution of indium atoms is due to a segregation process, and not to a bulk diffusion process at growth temperature (which would lead also to a spreading of indium atoms in the layer below the nominal InAs layer). In order to study more precisely the location of In atoms, and in particular of the “InAs”-onGaAs interface, we extract brightness profiles from this micrograph for each OW. These profiles, plotted in fig. 2, have been obtained on a small area (corresponding typically to 15 nm x 15 nm) for which the first (GaAs-on-AlAs) interface looks perfectly flat. Each spike corresponds to 1 ML. The 4 ML thick AlAs layers are rather well defined; we can therefore locate in the structure the nominal position of the InAs ML, labelled as n. Hereafter, n 1 and n + 1 will refer to the MLs surrounding n, and, respectively, deposited before and after n. For each quantum well, the —

a~5ML

a=7ML

Uttt

AlAs GaAs ALAs

Ga0 7AL0 3As

growth

Ga0 7AL0 3As

t

ft

AtAs GaAs ALAs

Fig. 1. High resolution electron micrograph obtained on InAs/GaAs/A1As multi-OW S2 the nominal value of a is indicated in ML for each OW.

538

J.-M. Gerard et a!.

AlAs

/ ML scale study of segregation effects in InAs / GaAs HSs r~AlAs

AlAs

AlAs~~

r—,A[As AlAs

growth

growth

growth

________

_______

a~3ML

a 5ML

a=7ML

Fig. 2. Profile along z of the brightness on the micrograph of fig. 1 (averaged in the direction perpendicular to z) for each QW.

darker ML is n + 1; n has a similar, though slightly weaker, darkness. A regular variation of the brightness for the subsequently grown MLs

each of the three (intrinsic [6]) PL peaks to a given value of a, since the bandgap energy of InAs/GaAs/Ga1 ~Al~As QWs decreases mono-

reveals, as expected, a gradual (decreasing) indium composition of the layer grown after InAs. Finally, n 1 looks darker than pure GaAs layers, and much brighter than n. Equilibrium models of the segregation process [1,5] predict, however, that the indium composition x1~is maximum for n 1. Our brightness profiles clearly contradict this prediction: x1~is much smaller for n 1 than for n. We can condude that the exchange reaction between In and Ga atoms does not reach its equilibrium when 1 ML InAs is deposited on GaAs. This result constitutes the first clear example of a breakdown of the equilibrium description of the segregation process (which, on the opposite, satisfyingly describes the surface enrichment of Ill—V alloys in one of their constituents, or the composition profile of GaAs-on-”InAs” interfaces). The quantitative analysis of the observed contrasts is, however, not obvious. PL data on such asymmetric QWs give, on the contrary, some reliable and precise information on the composition profile of the InAs layer [5].We can in particular study the composition of ML n 1 (i.e. how incomplete the exchange reaction is for a deposition of InAs on GaAs) and the shape of the composition profile for the GaAs-on-”InAs” interface. The PL spectra obtained at 4 K on samples, ~1 and ~2 are plotted in fig. 3. We easily attribute

tonically when the location of InAs varies from the edge to the centre of the well [6]. For a given a, the PL peak energy is clearly lower for S~than for ~2• Bandgap differences G2 G1 as large as 29 meV are observed for a = 3 ML. A non-equivalence of the GaAs-on-AlAs and AlAs-on-GaAs interfaces, whose abruptness look very similar on the micrograph of fig. 1 cannot explain these large bandgaps differences for a given a. On the contrary, In segregation can account for this behaviour: In atoms are (on average) moved away from the GaAs-on-AlAs interface, and brought closer to the AlAs-on-

—

—

—

—

41<

E GaAs

•

LIAIAs ~

S1

_____

_____

—

1.48

1.52 1.56 ENERGY leVI Fig. 3. PL spectra obtained at 4 K on InAs/GaAs/AIAs multi-QWs S1 and S2 the PL peaks are labelled by the value of a in ML. The nominal structure of one given OW (a = 3, 5 or 7 ML) is recalled in the inset. The growth direction is z for S~and ~ for ~2’

J.-M. Gerard et a!.

/ ML scale study of segregation

GaAs one. As a result, the PL peak is displaced toward lower energies for the S1 QWs, and higher energies for the S2 QWs, with respect to its expected energy for the nominal composition profile. The PL data obtained for six pairs of (GaA1)As/GaAs/InAs asymmetric QWs (including the three pairs described here) have been analysed. Two parameters are used to model the composition profile of 1 ML InAs in GaAs. First of all, we consider as unknown the composition ~ of the last nominal GaAs ML before InAs. We then assume that the equilibrium scheme accurately describes the segregation process for a deposition of GaAs on InAs. A single phenomenological parameter, the exchange energy E~,enters this calculation [1,5] (E~describes the propensity of In atoms for being segregated at the surface). For a given couple (x~ Es), the composition profile of the structure is perfectly defined; G2 G1 can then be calculated in the effective mass approximation for the six pairs of QWs. A fit of the experimental PL data is then performed by varying x,~ and E5. The variety of the pairs of QWs (corresponding to various a, barrier heights AlAs or Ga07Al03As and well width, 28 or 56 ML) allow to extract accurately and independently x,~ and E5 [5]. The best fit is obtained for x,~ = 0, i.e. if no exchange reaction occurs when InAs is deposited on GaAs. As a result, the “IriAs”-on-GaAs interface lies at its nominal position. More precisely, x,~ is unlikely to exceed 5%. This result thus confirms and refines our HREM data. ML fluctuations of the width of the GaAs layer grown before InAs can account for the residual discrepancies of the HREM results with our model profile. Starting from an ideal In composition profile (x~ = 0), such fluctuations increase the average In composition of ML n 1, and reduce that of ML n. As such, their occurrence might explain the darkness of ML n 1, and the slightly meversed contrast observed for MLs n and n + 1. We also extract an estimate for E5 (0.15 E~ 0.2 eV), which refines results previously obtained from extensive surface analysis studies (E~ 0.15 ±0.1 eV). The extreme composition profiles corresponding to our uncertainty range (calculated .

.

~ o.~j-. -‘

—

—

—

—

—

—

—

—

=

~

539

growtir~

o~1I

£srzO.1SOV I

.

— ~,

—

effects in InAs / GaAs HSs

0

E~~O.2 eV ., ,.

4

8

12

I

0

•

4

8

12

monolayer index Fig. 4. Calculated In composition profiles for 1 ML InAs in GaAs grown at 530°C, including segregation (and kinetic freezing) effects. 0 labels the nominal position of the InAs ML. E~= 0.15 and 0.20 eV correspond to extreme values of our uncertainty range on E,.

for an InAs ML grown at 530°C)are plotted in fig. 4. This rather surprising breakdown of the equilibrium model for the “lnAs-on-GaAs” interface alone is explained by the very different activation energies of exchange processes for a deposition of GaAs on InAs and InAs on GaAs. Firstly, In/Ga exchange reactions require, for a deposition of InAs on GaAs, the breaking of stable Ga—As bonds of the substrate; when, on the opposite, GaAs is deposited on an InGaAs layer, In segregation at the surface only needs to break less stable In—As bonds. Secondly, the surface of GaAs has few steps under standard growth conditions, whereas a short-scale roughness appears as soon as the growth of (InGa)As is initiated on GaAs [7]. Since exchange reactions have a lower activation energy at surface steps or kinks (i.e. less bonds have to be broken), the flatness of the GaAs surface can possibly lead, as well as the stability of Ga—As bonds, to a kinetic freezing of the exchange reactions at the “InAs”-on-GaAs interface. This result furthermore suggests that kinetics might be used to limitate the segregation efficiency. The most attractive approaches, a growth at low temperature 7 and migration enhanced epitaxy (MEE) (which improves the surface flatness), have been tested. We have compared G2 G1 for pairs of 28 ML thick AlAs/GaAs/InAs QWs grown for a = 5 ML under various conditions. No reduction of the segregation efficiency.. is detected for Tl7~ down to 450°C in standard MBE and lower I~are too detrimental for the optical quality. For a growth by MEE at 400°Cof —

540

J. -M. Gerard et a!.

/ ML sca!e study of segregation effects in InAs / GaAs HSs

InAs aws

8K

—

check the position of the GaAs-on-”InAs” interface for MEE grown InAs MLs. To conclude, the optical study of InAs/ GaAs/ GaAlAs QWs allows us to refine previous GaAs, andoftothe estimates compare amplitude it quantitatively of In segregation for vanin

az5ML

JLAs/GaAs/AIAsIS 1.48

1.50 1.52 ENERIII teV)

1.54

ous growth techniques. By probing the position of an InAs-on-GaAs interface on the ML scale, we demonstrate that an off-equilibrium surface is obtained when 1 ML InAs is deposited on GaAs. tion process, by implementing for the instance MEE Kinetics can thus be used to limit In segregaat low temperature.

Fig. 5. PL spectrum obtained at 8 K on a pair of InAs/GaAs/ AlAs NS-QWs (a = 5 ML) grown by MEE at 400°C.

References the whole structure (OW and AlAs barrier), however, the PL efficiency is perfectly preserved. The PL spectrum obtained at 8 K for this pair of QW5 is plotted in fig. 5. The lower energy peak stems from a 1 ML InAs/30 nm GaAs multi-OW, which has been inserted in the sample so as to test the deposited In quantity (here 0.95 ML) by X-ray diffraction and PL. We note that G2 G1 is close —

to 9 meV here instead of 24 meV for standard MBE. This effect probably results from a clear reduction of the segregation efficiency. Since G2 G1 strongly decreases for x,~_1> 0 [5], complementary experiments must be conducted so as to —

[1] J.M. Moison, C. Guille, F. Houzay, F. Barthe and M. Van Rompay, Phys. Rev. B 40 (1989) 6i49, and references therein. [2] B. Jusserand, F. Mollot, J.M. Moison and G. Le Roux, AppI. Phys. Letters 57 (1990) 560. [3] G. Abstreiter, K. Eberl, F. Friess, U. Menczigar and W. Wegscheider, in: NATO ASI Series B 253 (Plenum, New York, 1991) p. 471. [4] C. d’Anterroches, J.M. Gerard and J.Y. Marzin, in: NATO AS! Series B 203 (Plenum, New York, 1989) p. 47. [5] J.M. Gerard and J.Y. Marzin, Phys. Rev. B 45 (1992) 6313. 161 J.Y. 2172.Marzin and J.M. Gerard, Phys. Rev. Letters 62 (1989) [7] P.R. Berger, K. Chang, P.K. Bhattacharya and J. Singh, J. Vacuum Sci. Technol. B 5 (1987) i182.