Materials Science and Engineering B82 (2001) 151–155

Optical properties of self-assembled InGaN/GaN quantum dots

T. Taliercio a, P. Lefebvre a,*, A. Morel a, M. Gallart a, J. Allegre a, B. Gil a,H. Mathieu a, N. Grandjean b, J. Massies b

a Groupe d’Etude des Semiconducteurs-CNRS-Uni�ersite Montpellier II, Case courrier 074, 34095 Montpellier Cedex 5, Franceb Centre de Recherche sur l’Hetero-Epitaxie et ses Applications-CNRS-Rue Bernard Gregory, Sophia Antipolis, 06560 Valbonne, France

Abstract

Optical spectroscopy under varying temperature is used to investigate samples containing planes of self-assembled Ga1−xInxNquantum dots (0.15�x�0.20), embedded in a GaN matrix. The samples have been grown by molecular beam epitaxy onsapphire substrates and the nano-islands have been obtained by the Stranski-Krastanov growth mode transition. Half-widths athalf maximum as small as 0.05 eV are obtained for photoluminescence (PL) lines at T=2 K. Increasing the growth time decreasesthe PL energy and drastically increases the PL decay time, as a result of the increasing of the average dot height. Time-resolvedPL measurements with variable temperature allow us to observe the competitive influence of several mechanisms, namely the usualradiative and nonradiative recombination processes, plus the carrier feeding from random fluctuations, which plays a crucial rolein the case of the larger dots. © 2001 Elsevier Science B.V. All rights reserved.

www.elsevier.com/locate/mseb

1. Introduction

In recent years, group-III nitride semiconductorshave demonstrated their large potential, especially forthe production of light-emitting diodes covering theentire visible spectrum [1]. Besides, it is now widelyadmitted that several superior characteristics of GaInN/GaN quantum wells (QWs) are essentially induced bythe presence of indium-rich nanoclusters [2]. The sizeand compositions of these clusters, often referred to asquantum dots (QDs), are hardly controlled. On theother hand, recent progress in molecular beam epitaxy(MBE) has made it possible to monitor the growth ofnano-islands of variable sizes, with a one-monolayer(ML) accuracy. For instance, in GaN/AlN QDs, due tostrong internal polarization fields, the room-tempera-ture photoluminescence (PL) of these nano-objects isextremely dependent of the dot height and covers wave-lengths ranging from the ultra-violet to the orange[3–6]. Electric fields as large as 5 MV cm−1 [4,5] arepresent in such GaN/AlN systems, due to both sponta-neous and piezoelectric polarization effects in thesehexagonal compounds [7]. For systems using a ternary

alloy as the barrier material (e.g. AlxGa1−xN), theelectric field increases linearly with x [3]. For instance,we have used the Stark shift of excitonic transitions toestimate a field of �0.8 MV cm−1 in isolated GaN/Al0.17Ga0.83N QWs. The electric field also increasesdrastically exciton lifetimes [8–11] by separating elec-tron and hole wave-functions. But we have shown thatthe overall recombination dynamics strongly depends,too, on inter-well carrier transfers, in case of multipleQW structures [11].

In this paper, we present optical studies of GaInN/GaN QDs. We particularly focus our attention on PLdecay dynamics versus temperature, T. We interpretour results in terms of the competition between (i)radiative recombination, (ii) nonradiative carrier lossesand (iii) slow carrier feeding from disorder-inducedpotential fluctuation in the ternary.

2. Samples and experimental details

The GaInN/GaN QD samples were grown by MBEon c-plane sapphire substrates. After a short nitridationat 950°C, a 25 nm GaN buffer layer is deposited at500°C. It is then annealed at 900°C for a few minutesbefore the growth at 800°C and 1 �m h−1 of a fewmicron thick GaN epilayer. The substrate temperature

* Corresponding author. Tel.: +33-467-143756; fax: +33-467-143760.

E-mail address: lefebvre@ges.univ-montp2.fr (P. Lefebvre).

0921-5107/01/$ - see front matter © 2001 Elsevier Science B.V. All rights reserved.PII: S0921 -5107 (00 )00788 -1

T. Taliercio et al. / Materials Science and Engineering B82 (2001) 151–155152

is decreased to 500–550°C for growing the GaInN/GaN QDs. Then the growth rate is also reduced to lessthan 0.3 �m h−1. The In composition deduced fromX-ray diffraction and reflection high-energy electrondiffraction is about 15–20%. The nominal GaInNthickness ranges between 1 nm (QDs emitting near 3eV) to 3 nm (QDs emitting near 2.4 eV).

In this communication, we will restrict ourselves to aseries of 3 samples with narrow PL lines, shown in Fig.1, lying in the range of 2.7–3.0 eV. On these continu-ous-wave spectra, obtained at T=2 K, we observe thatthe larger the growth time, the larger the dot heightalong the (0001) growth axis and the lower the transi-

tion energy. This is due to both quantum-size andquantum-confined Stark effects in the QD. The PLlinewidths are very small, compared to what is obtainedfrom GaInN QWs with comparable emission energies.This allows us to measure a rich sub-structure in termsof phonon replica, corresponding to the LO-phononenergy of bulk GaN (91 meV). This point and therelative intensities of the replicas will be discussed else-where. The apparent splitting of some lines is due, infact, to optical interference effects in the multi-layeredsystem. The latter effect is also visible on the ratherbroad line, at �3.2 eV, which is common to all sam-ples. We assign this line to the so-called ‘‘wetting layer’’on which the 3D growth of islands has occurred. Thiswetting layer has a thickness of �1 nm and its compo-sition is a priori the same as that of the QDs, althoughwe cannot discard In enrichment of the QDs by straininduced migration.

For time-resolved PL (TRPL) experiments, we havechosen to generate photo-excited carriers only into thelayer of interest and not into the GaN barriers. This iswhy we have excited the samples by the frequency-dou-bled laser pulses from a Ti-sapphire cavity, with typicalpulse width of 2 ps. The photon energy of the excitationlaser was set to 3.1 eV, which lies on the low-energyside of the PL from the wetting layer. We took specialcare, also, of the repetition rate of the excitation pulses,which we could reduce by using an acousto-opticalmodulator. By doing so, we allow the all photo-carriersto recombine between two pulses, thus eliminating nu-merous artifacts. In particular, we suppress the back-ground quasi-continuous emission, which generallychanges the apparent decay dynamics in cases wherethe PL decay rate would be much smaller than thenormal repetition rate of the laser (82 MHz).

3. Low-temperature experiments

Fig. 2 displays the time decays of the PL intensity foreach of the 3 samples. The decays of the correspondingphonon replica were similar, as expected. The presentnonexponential decays are actually a general feature ofall our QD samples, including those, not shown here,with larger QD heights and emitting at much lowerenergies. These complex decays could be, in fact, multi-exponential decays, which would result from the super-position, within the present linewidth, of several,independent, emitting channels, each having its owndynamics. In favor of this explanation, which is notdefinitive, we note the extreme sensitivity of the decayrate with the PL peak energy. We have chosen tocharacterize these decays by the delay, denoted by �10,after which the maximum intensity is divided by afactor of 10. From the values reported in Fig. 2, wenote that �10 is multiplied by a factor of �30 whereas

Fig. 1. Continuous-wave PL spectra of the 3 InGaN/GaN QDsamples. The larger the nominal thickness deposited, the lower the PLenergy. ‘‘WL’’ stands for ‘‘wetting layer’’.

Fig. 2. PL decays obtained from the three samples. The lower the PLenergy, the slower the decay.

T. Taliercio et al. / Materials Science and Engineering B82 (2001) 151–155 153

Fig. 3. Temperature dependence of the characteristic time �10. Dottedand dashed lines show the variations assumed for the radiative andnonradiative lifetimes, respectively, in order to achieve the fittingshown by the solid curve.

measured for GaN/AlGaN QWs [6,8]. Several explana-tions can be invoked for these long lifetimes. First ofall, the quasi zero-dimensional character of excitonsenhances the components of their wave function awayfrom the photon wave vector, thus increasing the life-time [17,18]. Second, in small spherical QDs, the elec-tron-hole (e-h) exchange interaction is increased, beingextremely sensitive to the e-h envelope function overlap.This is capable to push the J=2 forbidden state tolower energy than the radiative state, yielding a long-lived luminescence [19]. In the present case, we discardthe latter explanation because the e-h overlap isstrongly reduced by the internal electric field whichseparates the electron from the hole, along the growthaxis. Precisely, this spatial separation, a third effect, isthe most obvious cause of increasing of the radiativelifetime in nitride-based QWs [20]. Provided that thepertinent parameters are known, it is rather easy tomodel the subsequent increasing of the radiative life-time when increasing the width of a nitride-based QW,within the envelope function approximation [11,20–22].This approach could be reasonably extended to thepresent QDs, by using some equivalent, ‘‘effective’’QW, for instance. But many material parameters (effec-tive masses, bowing parameter, band offsets, strains,electric fields…) are really poorly known and the com-position and height of the QDs have not been deter-mined with a good accuracy. Consequently, it isimpossible to ascertain that the present well widthdependence is purely an effect of the electric field.Nevertheless, an element in favor of this interpretationwould be the fact that, at such low temperatures, thenonradiative mechanisms are extremely slow, comparedto radiative recombinations.

In other words, we need to check whether the low-temperature PL dynamics is controlled by radiative orby nonradiative processes. This is why we have per-formed temperature dependent TRPL experiments, de-scribed below, which have allowed us to measure theoverall PL decay time and the time-integrated PL inten-sity versus T.

4. TRPL experiments under variable temperature

Fig. 3 shows the evolution versus T of the quantity�10, for each of the three samples investigated. Asexpected, the general trend is a reduction of the PLdecay time versus T, as a consequence of the domina-tion, at high T, of nonradiative processes. This trend isconfirmed by the collapse of the PL intensity, IPL,which indicates a reduction of the radiative efficiency.This result is shown in Fig. 4. Going into details,however, we remark several surprising facts.

(1) First of all, the higher the transition energy, theslower the collapse of �10 and IPL versus T. This result

Fig. 4. Temperature dependence of time-integrated PL intensities.The solid curve has been obtained by using the �R and �NR variationsextracted by the fitting in Fig. 3, with no extra parameter.

the PL line is red-shifted by only 0.3 eV. Then, consid-ering the inhomogeneous linewidths of �0.05 eV, it istempting to deduce that several contributions, with arather wide spanning of decay times, are present withineach PL line. We have, indeed, verified that the decayrate, within a given line, is slightly slower on the lowerenergy side. Nevertheless, even for a restricted range ofwavelengths, the decays always appeared nonexponen-tial. Rather than reducing the wavelength range, itwould be more efficient to try and isolate a single QDor a few of them, e.g. by near-field techniques [12,13].Then we may be able to check whether or not one givenQD really shows a mono-exponential decay, as usuallyassumed in theoretical approaches of radiative emissionin collections of QDs [14].

The present decay rates are similar to those measuredin the past few years for InGaN/GaN QWs grown byMOCVD [15–17], and they are much slower than those

T. Taliercio et al. / Materials Science and Engineering B82 (2001) 151–155154

is the opposite of what we would expect from simplearguments based on the quantum confinement in QDs:the larger the dot, the deeper the carrier confinementand thus the smaller the probability of carrier escapingtowards nonradiative recombination centers. On thecontrary, for our samples, the collapses of �10 and IPL

occur at lower T and much more steeply for large dotsthan for small ones.

(2) Second surprise: except for the QDs in sample 1,emitting at 3.0 eV, both �10 and IPL exhibit a slightincrease at low temperatures, before the steep collapsescommented above. At first sight, this could either bedue to an increase of the nonradiative lifetime, �NR, orto a decrease of the radiative lifetime, �R. But, theprobability of nonradiative events can be safely pre-sumed to increase steadily with T. Then �NR can in noway be increased. Moreover, the increase of IPL cannotresult from a decrease of �R: in this case, the overall PLdecay time would become smaller.

(3) The QDs in sample 1 behave really differentlyfrom the other ones. In their case only, we can fit thechange of �10 with T by assuming a constant �R and athermally activated �NR, as shown in Fig. 3. The activa-tion energy is of 59 meV. A temperature-independent�R is compatible with the zero-dimensional character ofexcitons [15,18,23,24]. By assuming that IPL varies like�NR/(�NR+�R), and by using the values determinedfrom Fig. 3, we obtain the variation of IPL shown by acontinuous curve in Fig. 4, in very good agreement withexperimental data, without any further adjustableparameter.

(4) For the two other samples, with larger decaytimes, it is clear that no thermal activation law can beused to fit the high-T collapses of �10 and IPL. Rather,these collapses seem to follow a simple and fast decay-ing exponential variation. In any case, by using only aconstant �R and such an exponentially decaying �NR, itis impossible to fit the behaviors of �10 and IPL forlow-temperatures. In fact, we cannot describe entirelyour TRPL results by involving only the usual radiativeand nonradiative processes. This is why we considerbelow the possible role played by disorder-inducedpotential fluctuations on the overall radiative emissiondynamics in this type of sample.

5. Discussion

Fig. 5 sketches the possibility that potential fluctua-tions (e.g. in the wetting layer or in the QDs them-selves) may retain a fairly large amount ofphoto-excited carriers just after excitation. Then, withsome characteristic time constant, �F, these traps can‘‘feed’’ the QDs, more or less efficiently, depending e.g.on T, thus altering the apparent decay dynamics ofthese dots. For instance, if the QDs have a naturalradiative lifetime �R=33 ns, but are fed by a secondarysource with �F=50 ns, the decay will show an apparentcharacteristic time of �A= (1/�R−1/�F)−1�100 ns.

Obviously, if the feeding time, �F, is much larger thanthe emptying time, �R, the apparent decay dynamics isvery close to a purely radiative system and the feedingdynamics almost changes nothing to the apparent be-havior. We believe that this is the case for the dotsemitting at 3.0 eV (sample 1), with short radiativelifetimes. On the other hand, if �F and �R have com-parable magnitudes, like in the above example, then theapparent dynamics is slower than the purely radiativeone. In this latter case, which we believe applies tosamples 2 and 3, with slow decays, the first consequenceof increasing T is to reduce �F, thus slightly increasingthe apparent decay time. Indeed, if some carriers aretrapped on disorder induced potential fluctuations, in-creasing T should efficiently empty these traps, begin-ning with the shallower ones. This can also increaseslightly the PL intensity integrated over a finite timewindow, as for the results shown in Fig. 4. We remarkthat this feeding process is essentially a percolationmechanism, which is known to induce a stretched-expo-nential like decay dynamics, which may also explain thepresent nonexponential decay pattern.

Concerning the fast collapse of IPL and �10 at hightemperatures, for samples 2 and 3, we propose twocomplementary explanations:

(1) First, the scattering of carriers outside the QDs�ia absorption of phonons has a small probability,especially for optical phonons, since ��LO=91 meV, asshown by phonon replica in Fig. 1. Up to room temper-ature, the characteristic times of these processes arefairly long, if compared, with �R values of �3 ns(sample 1). Nevertheless, these times can be comparableor shorter than radiative lifetimes of several tens ofnanoseconds (samples 2 and 3). This is a first reason forthe fastest decrease of IPL and �10 versus T, in the lattersamples: by increasing T, �R and �NR rapidly reachcomparable magnitudes, which is not the case for sam-ple 1.

(2) The second explanation lies on the feeding mech-anisms described in Fig. 5. We have explained why theoverall dynamics for sample 1 is rather insensitive tothese mechanisms. On the other hand, increasing Tsurely favors the loss of carriers from the potential

Fig. 5. Schematic view of the secondary feeding process �ia potentialfluctuations in the ternary alloy. The transfer time, �F, as well as thecarrier loss times �NR and � �, are reduced when T is raised.

T. Taliercio et al. / Materials Science and Engineering B82 (2001) 151–155 155

fluctuations, with a characteristic time � ’, as shown inFig. 5. We may state that, at some temperature, theoverall loss of efficiency of the QDs is double, sincesome carriers are lost directly from the QDs and othersare also lost from the feeding potential fluctuations.

6. Conclusion

We have analyzed TRPL results obtained on assem-blies of InGaN/GaN QDs, grown by MBE on sapphiresubstrates. We have found a fast decrease of the recom-bination rate when the recombination energy decreases,in qualitative agreement with expectations based on thepresence of internal electric fields. However, our tem-perature dependent TRPL results can only be explainedby including the dynamics of feeding from some reser-voir of carriers, presumably related to disorder inducedpotential fluctuation in the ternary alloy.

Acknowledgements

This work is supported by the French Ministry ofEducation, Research and Technology within the ‘‘BO-QUANI’’ and ‘‘NANILUB’’ Research Programs. Wealso acknowledge support of the EEC under contractN° HPRN-CT-1999-000132.

References

[1] B. Gil (Ed.), Group III Nitride semiconductor Compounds,Clarendon Press, Oxford (1998) ISBN 0-19-850159-5.

[2] K.P. O’Donnell, R.W. Martin, P.G. Middleton, Phys. Rev. Lett.82 (1999) 237.

[3] B. Damilano, N. Grandjean, F. Semond, J. Massies, M. Leroux,Appl. Phys. Lett. 75 (1999) 962.

[4] N. Grandjean, B. Damilano, S. Dalmasson, M. Leroux, M.Laugt, J. Massies, J. Appl. Phys. 86 (1999) 3714.

[5] F. Widmann, J. Simon, B. Daudin, G. Feuillet, J.L. Rouviere,N.T. Pelekanos, G. Fishman, Phys. Rev. B. 58 (1998) R15989.

[6] M.A. Reshchikov, J. Cui, F. Yun, A. Barski, M.I. Nathan, R.Molnar, Hadis Morkoc, Proc. MRS 2000 Spring Meeting, Mat.Res. Soc. Proceedings edited by R.J. Shul, F. Ren, M.Marukami, W. Pletschen, ISBN: 1-55899-530-7 (in press).

[7] F. Bernardini, V. Fiorentini, D. Vanderbilt, Phys. Rev. B 56(1997) R10026.

[8] M. Leroux, N. Grandjean, M. Laugt, J. Massies, B. Gil, P.Lefebvre, P. Bigenwald, Phys. Rev. B 58 (1998) R13371.

[9] P. Lefebvre, J. Allegre, B. Gil, H. Mathieu, P. Bigenwald, N.Grandjean, M. Leroux, J. Massies, Phys. Rev. B 59 (1999) 15363and references cited therein.

[10] M. Leroux, N. Grandjean, J. Massies, B. Gil, P. Lefebvre, P.Bigenwald, Phys. Rev. B 60 (1999) 1496 and references citedtherein.

[11] P. Lefebvre, M. Gallart, T. Taliercio, B. Gil, J. Allegre, H.Mathieu, N. Grandjean, M. Leroux, J. Massies, P. Bigenwald,Phys. Stat. Solidi (b) 216 (1999) 361.

[12] H.F. Hess, E. Betzig, T.D. Harris, L.N. Peiffer, K.W. West,Science 264 (1994) 1740.

[13] R. Grousson, V. Voliotis, N. Grandjean, J. Massies, M. Leroux,C. Deparis, Phys. Rev. B 55 (1997) 5253.

[14] M. Grundmann, R. Heitz, D. Bimberg, J.H.H. Sandmann, J.Feldmann, Phys. Stat. Sol. (b) 203 (1997) 121.

[15] Y. Narukawa, Y. Kawakami, S. Fujita, S. Fujita, S. Nakamura,Phys. Rev. B 55 (R1938) 1997.

[16] M.S. Minsky, S.B. Fleischer, A.C. Abare, J.E. Bowers, E.L. Hu,S. Keller, S.P. DenBaars, Appl. Phys. Lett. 72 (1998) 1066.

[17] A. Sohmer, J. Off, H. Bolay, V. Harle, V. Syganow, J.S. Im, V.Wagner, F. Adler, A. Hangleiter, A. Dornen, F. Scholtz, D.Brunner, O. Ambacher, H. Lakner, MRS Internet J. NitrideSemicond. Res. 2 (1997) 14 (http://nsr.mij.mrs.org/2/14/).

[18] L.C. Andreani, F. Tassone, F. Bassani, Solid State Commun. 77(1991) 641.

[19] M. Nirmal, D.J. Norris, M. Kuno, M.G. Bawendi, Al.L. Efros,M. Rosen, Phys. Rev. Lett. 75 (1995) 3728.

[20] J.S. Im, H. Kollmer, J. Off, A. Sohmer, F. Scholtz, A.Hangleiter, Phys. Rev. B 57 (1998) R9435.

[21] P. Bigenwald, P. Lefebvre, T. Bretagnon, B. Gil, Phys. Stat. Sol.(b) 216 (1999) 271.

[22] P. Bigenwald, A. Kavokin, P. Christol, B. Gil, P. Lefebvre, Phys.Stat. Sol (b) 216 (1999) 481.

[23] M. Colocci, M. Gurioli, J. Martinez-Pastor, J. Phys. France C5(3) (1993) 3.

[24] D.S. Citrin, Phys. Rev. B 47 (1993) 3832.

.