A. Mascarenhas. National Renewable Energy Laboratory, 1617 Cole Blvd, Golden, CO 80401, USA

Transcription

1 Growth and Properties of the Dilute Bismide Semiconductor GaAs 1-x Bi x a Complementary Alloy to the Dilute Nitrides T. Tiedje, E. C. Young* Department of Physics and Astronomy, Department of Electrical and Computer Engineering, *Department of Materials Engineering, University of British Columbia, Vancouver, BC, V6T-1Z4, Canada A. Mascarenhas National Renewable Energy Laboratory, 1617 Cole Blvd, Golden, CO 80401, USA Abstract In this review we describe the growth and properties of the dilute bismide semiconductor alloy GaAs 1-x Bi x and show how its properties are in certain respects complementary to the dilute nitride alloy, GaN y As 1-y. Like the dilute nitrides the dilute bismides show a giant band gap bowing effect in which a small concentration of the alloying element has a disproportionate effect on the band gap, however in the case of the bismide the band gap reduction is associated with an increase in the energy of the valence band maximum (VBM) rather than a reduction in the energy of the conduction band minimum (CBM). Under standard GaAs growth conditions Bi acts as a surfactant with associated improvements in surface quality. In order to incorporate Bi, growth temperatures below 400 o C are used with As 2 /Ga flux ratios close to unity. The electron mobility of GaAs is only weakly affected by Bi alloying, in contrast to the dilute nitrides where the electron mobility decreases rapidly with N alloying. Bi alloying also produces a giant bowing effect in the spin orbit splitting in the valence band. Strong room temperature photoluminescence is observed. Prospects for future device applications of this new compound semiconductor alloy are discussed. 1

2 Introduction Molecular beam epitaxy (MBE) is a method for growing single crystals of a wide variety of materials including semiconductors, metals and oxides, by vapour deposition on a crystalline substrate. MBE is an important tool for materials research as well as an established method for production of electronic and optical devices. The thickness of the deposited material can be controlled with mechanical shutters to sub-nanometer accuracy, with the thickness of deposited layers ranging from less than a monolayer, to quantum wells a few nanometers thick, to bulk-like material more than a micron thick. The structure and composition of the substrate surface can be used to make highly strained layers and materials whose properties are otherwise influenced by the structure of the substrate, that are not accessible by bulk synthesis methods. Surface atoms are much more mobile than atoms in the bulk, therefore single crystals can be grown by MBE at relatively low temperatures, typically half the bulk melting temperature. This means that the equilibrium density of thermodynamic defects is much lower than in melt-grown crystals and that metastable alloys can be made from combinations of elements that would be immiscible in thermodynamic equilibrium. In this review we describe the growth and properties of one example of such a metastable material, namely the III-V semiconductor alloy GaAsBi. Semiconductor devices are enormously important in society and there is a continuing need for new semiconductor materials with more advantageous properties for devices. A few examples of applications that require semiconductor materials with new properties are high performance solar cells, light emitting materials with output wavelengths that are independent of temperature, low power consumption transistors for wireless devices, and new kinds of materials that allow the manipulation of electron spins. The dilute bismide GaAs 1-x Bi x can potentially contribute to the solution of these problems. The extension of the GaAs family of semiconductor alloys to include Bi, is a relatively new development. GaAs 1-x Bi x was first synthesized by metalorganic vapour phase epitaxy [1] and later by molecular beam epitaxy [2,3] although small bandgap bismide alloys (InSbBi) were grown by MBE much earlier [4]. Due to the large size and low electronegativity of Bi compared with As, the GaAs 1-x Bi x alloys have anomalous electronic properties and require growth conditions that are rather different than standard GaAs growth processes [2,3,5]. In addition to being the heaviest group V element Bi is also the heaviest non-radioactive element, and unlike the other group V elements As and Sb, and 2

3 neighboring bottom row elements Hg, Th and Pb, bismuth has low toxicity. The large size and low electro-negativity of Bi relative to the As atoms in the GaAs host, is exactly the opposite of the small size and high electronegativity of nitrogen, the lightest group V element, an alloying element that also produces anomalous changes in the semiconducting properties of GaAs. The dilute nitride alloys have been the subject of intense research interest over the past few years targeted at understanding their unusual electronic and optical properties [6-8]. The electronic and optical properties of the dilute bismides have similarities and differences with the dilute nitrides, as we discuss below. The initial work on the dilute GaAs 1- xbi x alloy was motivated by the search for a semiconductor with a temperature independent bandgap [9-11] and by the idea that Bi might co-incorporate with N to minimize the deleterious effects of N on the transport properties and recombination lifetime of InGaAsN [12]. Bertulis et al [13] have explored the possibility of using Bi to extend the spectral response of low temperature GaAs used in fast photoconductive switches, to longer wavelengths. Recent work on the electronic and optical properties of GaAs 1-x Bi x has refined our knowledge of the optical and electronic properties, and led to a better understanding of the potential applications. The relative energies of the valence orbitals for Ga, As and Bi suggests that the Bi 6p bonding orbitals are resonant with the VBM of GaAs as shown schematically in Fig. 1 [14]. This result is once again exactly the opposite of the situation in the case of the dilute nitrides, where the bonding N 2p orbitals are deep in the valence band of GaAs and the empty N 2s antibonding orbital is resonant with the CBM [6]. Physically this difference in behaviour is due to the low electronegativity of Bi and the large electronegativity of N. The resonant interaction between the relatively localized N 2s orbital with the CBM leads to the so-called giant band gap bowing effect in which a small concentration of nitrogen reduces the band gap even though the binary end member of the alloy system, GaN, has a much larger band gap. One would expect a similar resonant interaction in the valence band with the Bi 6p orbital in the case of the dilute bismides [14,15]. In fact the bismide does show a giant band gap reduction for low Bi concentrations, namely 88 mev reduction in band gap for 1% Bi [5]. This experimental value includes a correction for the effect of the compressive strain in the epitaxial film when it is deposited pseudomorphically on GaAs. Although the bandgap reduction is not as big as in the case of the dilute nitride alloys in which the band gap reduction is 200 mev/% N for [N]<1% [16,17], the band gap change is still 7x greater than the 12 mev/% In band gap change due to indium in Ga 1-3

4 win w As alloys [18]. In the case of the dilute nitride the main band gap reduction occurs by a lowering of the CBM, whereas in the case of the bismides we expect the band gap reduction to occur primarily through an increase in the energy of the VBM as illustrated schematically in Fig. 2 for bismuth and nitrogen concentrations equal to 3% [15]. GaAs-nitride-bismide alloys were proposed by Mascarenhas as a means for strain compensating the small N atom by co-incorporating Bi and N as next nearest neighbors [12]. The idea is that if Bi and N are close enough together the resulting strain and charge perturbation in the lattice would be in the form of a dipole rather than a monopole and this would improve transport properties relative to an alloy with isolated N atoms. So far there is no experimental evidence that N and Bi preferentially incorporate as next nearest neighbors in the growth schemes that have been explored to date. Oe and collaborators find that dilute GaAsBi alloys grown by MOCVD and by MBE have an anomalously weak temperature dependence of the band gap [9-11]. A reduction in the temperature dependence of the band gap would have important practical applications in laser light sources for optical communication systems. The temperature dependence of the band gap was inferred from photoluminesence spectra and from photoreflectance and found to be 1/3 that of GaAs for a sample containing 2.6% Bi [11]. However the weak temperature dependence of the gap has not been found in bismide samples made elsewhere [5, 19]. The reason for the discrepancy is not known. In this review we describe the growth, structure, and optical and electronic properties of GaAs 1-x Bi x and GaN y As 1-x-y Bi x alloys grown on GaAs (100) substrates by solid source MBE. MBE Growth with Bismuth All molecular beam epitaxy (MBE) growth experiments were carried out in a V80H growth system with Bi in a standard, single-filament Ga-type K-cell operating at between 600 o C and 800 o C. The Bi flux is a mixture of approximately 50% Bi monomers and 50% dimers. Under typical growth conditions for GaAs, namely a growth temperature of 550 o C and a beam equivalent pressure (BEP) ratio for As 2 to Ga of 5-10, Bi does not incorporate to the sensitivity limits of SIMS (<2x10 17 ) [20]. In fact Bi has such a strong tendency to surface segregate that it can be considered an ideal surfactant [20-23]. Even if trace amounts of Bi did incorporate into GaAs, as long as it incorporates on As sites it is unlikely to produce electronically active defects since Bi is isoelectronic with As. In an early theoretical paper on the possible benefits of surfactant growth [24], the authors expressed optimism about the beneficial effects of a 4

5 surfactant during growth but were skeptical that a suitable surfactant could be found that did not leave residual impurities in the grown layer that would degrade device performance. This pessimism has turned out to be unwarranted in the case of GaAs, as Bi appears to satisfy the requirements of a good surfactant in that it stays on the surface without incorporating any significant density of electronically active impurities. Other heavy group III and V elements also tend to surface segregate including both In, Tl and Sb [23]. Although Sb surface segregates more strongly than In, it does not surface segregate as strongly as Bi and continues to incorporate at low levels under standard GaAs growth conditions [25,26]. The relative Bi surface coverage determined from RHEED as a function of growth temperature is shown in Fig. 3 for a high Bi flux with a Bi cell temperature of 750 o C and BEP of 1.4x10-5 Torr. The Bi coverage measurements were made by measuring the intensity of the specular RHEED spot before and after opening the Bi shutter at a series of different temperatures indicated by the open circles in Fig. 3 [21]. The substrate temperature was measured by band gap thermometry, which was particularly useful in the present experiments as it works well at low substrate temperatures where pyrometry is ineffective [27,28]. The solid line in Fig. 3 is a fit to a Langmiur isotherm with a surface binding energy of ev and a mean field Bi-Bi binding energy of 0.12 ev. The surface binding energy is similar to the binding energy of Bi on the surface of liquid Bi of 1.7 ev [21]. Thus it may be reasonable to think of the Bi on the surface as a liquid Bi layer. The effect of the surfactant layer on the surface morphology is dramatic, as shown in Fig. 4 [20]. In this figure we show AFM images of two GaAs N buffer layers grown at 460 o C at low As overpressure (As 2 /Ga~1), one after the other, under identical conditions except that the sample in Fig. 4(b) was exposed to a simultaneous Bi flux of 1.8 x 10-5 Torr BEP. The surface exposed to Bi is very smooth with an rms surface amplitude of 0.1 nm and large flat terraces indicative of step flow growth. In the sample grown without the surfactant the surface is much rougher with an rms amplitude of 1.1 nm and no individual atomic steps resolvable. The surface grown without the surfactant shows the elongated microstructure characteristic of growth at low As/Ga ratio. The large terraces in the surfactant grown sample indicate that the surfactant facilitates the diffusion of adatoms on the surface and possibly also the transport of deposited Ga atoms from one atomic terrace to the next lower terrace [24]. We speculate that the Ga adatoms on the surface of the GaAs migrate at the interface between the covalently bonded crystal surface and 5

6 the liquid Bi layer that is electron rich with relatively more metallic character to the bonding than the atoms in the substrate. By reducing the activation energy for surface diffusion the surfactant allows growth processes that would normally only occur at high temperatures (step flow, fast surface smoothing) to occur at much lower temperatures. For the high Bi fluxes used in these experiments the Bi coverage data in Fig. 3 suggests that Bi surfactant growth of GaAs can be carried out up to about 500 o C without thermally desorbing the Bi surface layer. Along with the improvements in surface morphology described above we find that surfactant growth also improves the photoluminescence emission efficiency. In Fig. 5 we show photoluminesence spectra (PL) for dilute nitride samples grown with and without the surfactant with the same N concentration, as determined by x-ray diffraction, and all other growth conditions kept fixed as far as possible [20]. The surfactant grown material shows a factor of 3 higher PL intensity before and after annealing, compared with the material grown without the surfactant, as shown in Fig. 5 (a). In Fig. 5(b) we show the PL emission efficiency from 10 nm GaAs quantum wells with 100 nm Al 0.3 Ga 0.7 As cladding layers grown at 600 o C with and without a Bi surfactant (BEP ~ 4x10-7 Torr). The PL efficiency for the quantum well structure shows an increase of almost a factor of two when it is grown with a Bi flux even though at the 600 o C growth temperature and modest Bi flux we expect the Bi coverage to be low [20]. Most likely the improved PL results from an improvement in the quality of the AlGaAs cladding layers since we expect the non-radiative recombination to be dominated by recombination centres in the AlGaAs cladding.. Johnson et al [29] have also observed improvements in the electrical properties of AlGaAs, as well as the surface morphology, when the AlGaAs is grown with an Sb surfactant. We interpret these results as an indication that the surfactant tends to reduce the density of electronically active defects. It is not clear whether the defects that are being suppressed are thermodynamic defects, such as As anti-sites or instead chemical impurities incorporated from the residual gas in the growth chamber such as oxygen, which are blocked from incorporation by the surfactant. In metal organic vapour phase epitaxy with Sb and Bi surfactants a two order of magnitude reduction in carbon contamination was observed [23]. In the case of the dilute nitrides the electronic defects that are being suppressed by the surfactant may be N interstitials [30] or local clustering of N atoms that create bound states in the band gap. The surfactant is found to have no effect on the GaAs growth rate. Further, in the case of the dilute nitride, we find that the surfactant actually increases the N incorporation as shown in 6

7 Fig. 6 for plasma assisted MBE growth [20,21]. The reason for this surprising result is not known. One possible mechanism, namely the suppression of the formation of volatile NAs dimers was explored using line of sight mass spectroscopy during growth [31]. However the partial pressure of NAs was found to increase when the Bi surfactant was turned on, a change in the opposite direction. The fact that the N incorporation increases in the presence of the Bi shows that not all of the reactive nitrogen produced in the low pressure N 2 plasma discharge actually incorporates. Analogous increases in N content for samples exposed to Sb during growth, have also been reported [32]. In MOCVD growth of GaAsN the Bi surfactant has the opposite and intuitively more reasonable effect, namely it reduces the amount of incorporated N [33]. In principle it is possible that the surfactant does not actually increase the N concentration but rather causes the nitrogen to incorporate into a different site in the lattice since we are inferring the nitrogen concentration only from x-ray diffraction measurements of the lattice constant. For example in GaAs 1-y N y grown by metal organic chemical vapour deposition, there is evidence for non-substitutional N when y>3% [30, 34, 35]. However non-substitutional nitrogen is unlikely to be the explanation for the lattice constant changes that occur in the surfactant growth experiments discussed here. In these experiments the nitrogen concentrations are less than 1%, where Vegard s law is found to be valid [34] and in addition the optical bandgaps of the dilute nitrides grown with the surfactant match the bandgaps of the dilute nitrides with the same lattice constants that are grown without the surfactant. Although Bi does not incorporate in GaAs under usual growth conditions for GaAs, by reducing the growth temperature below 400 o C and reducing the arsenic overpressure so that the As 2 /Ga BEP is near 1:1, Bi can be made to incorporate into GaAs [2,3]. The need for growth temperatures below 400 o C to achieve Bi incorporation is consistent with STM-XPS experiments on Bi layers adsorbed on GaNAs, in which Bi was observed to evaporate as the surface was heated from 300 o C to 400 o C [36]. The Bi incorporation into GaAs was determined by x-ray diffraction, calibrated using Rutherford Backscattering Spectrometry (RBS) [2]. Fig. 7 shows the lattice parameter calibration using the RBS data, while Fig. 8 shows typical high resolution x- ray diffraction data for three GaAs 1-x Bi x layers grown on GaAs with Bi contents x, and layer thicknesses as follows: 3%, 20 nm; 1.9%, 125 nm; and 0.5%, 210 nm [31]. The layer thicknesses were designed to be below the critical thickness for structural relaxation through misfit 7

8 dislocations formation. The presence of pendellosung fringes indicates that the films are uniform and smooth. These samples were grown at lower Bi effusion cell temperatures ( o C) and much lower Bi BEP pressure (9x10-8 Torr) than in the higher temperature surfactant growth experiments discussed above. The process window for the growth of bismides with smooth surfaces is rather narrow. If the As 2 overpressure is too low Ga droplets develop on the surface while if the As overpressure is too high Bi droplets form. Light scattering is found to be a useful tool in defining and dealing with the small process window as it is highly sensitive to the presence of small metallic droplets on the surface [37]. In Fig. 9 we show the intensity of the diffusely scattered light as a function of time during growth of a GaAs 1-x Bi x layer at 365 o C. A small reduction in the As 2 pressure, achieved by adjusting the needle valve on the As cracker source, causes the immediate onset of surface roughening, while a subsequent correspondingly small increase in As 2 pressure causes the surface to begin smoothing again. Optical Properties Similar to the situation with N alloying, a small amount of incorporated Bi has a large effect on the bandgap, as shown by the optical band gap as a function of Bi content presented in Fig. 10. Although Bi has a 7x bigger effect on the bandgap of GaAs than In, the increase in the lattice constant with Bi alloying is only 1.7x bigger than with In alloying (0.68% change in lattice constant for 1%Bi compared with 0.41%/1%In) [2,18,38]. Therefore the same bandgap reduction can be obtained with bismide-nitride alloys as with indium-nitride alloys, but with less N, while still maintaining a lattice match to GaAs [14]. Since nitrogen alloying degrades the electronic properties this will be advantageous as long as the Bi alloying does not itself degrade the electronic properties too much. The electronic transport properties and photoluminescence efficiency deteriorate with increasing N content, possibly through the formation of N clusters or other nitrogen related defects which may act as recombination centres. Strained layers whose In/N ratio significantly exceeds the 3/1 lattice match condition to GaAs are acceptable in light emitting devices as long as the active region is a quantum well layer that is only ~10 nm thick. These layers can tolerate high strains without exceeding the critical thickness for dislocation formation. However in other applications, thick low-defect-density layers are required, for which low dislocation densities and low strain are essential. Multi-junction solar cells are an example of an application in which a thick layer is required [39,40]. In a solar cell the semiconductor layer must be ~1 m thick to absorb all of the light at the bandgap. In order to prevent dislocation 8

9 formation the layer must be lattice matched to the substrate. In the case of multi-junction solar cells a preferred substrate is Ge, whose lattice constant is close to GaAs. A mathematical model for the band gap of the quaternary GaAs-bismide-nitride alloys as a function of N and Bi content is of interest for device design. The usual interpolation formulas used for the band gap of conventional ternary and quaternary III-V alloys such as In z Ga 1-z As and In z Ga 1-z As 1-w P w have quadratic bowing terms of the form bz 1 z where b is a bowing parameter. However this mathematical form does not match the composition dependence of the band gap of GaN y As 1-y which changes rapidly for small y [16]. Therefore a new interpolation formula for the composition dependence of the band gap is required for the dilute GaAs-nitridebismide alloys. The following expression has been proposed [41] : where,, D D D D E x y = E x y x y [1] BiN 0 Bi N Bi N EBiN x y is the band gap of the quaternary GaAs bismide-nitride alloy, 0 E is the band gap of GaAs, Bi x D is the band gap reduction due to Bi alloying in the ternary bismide alloy, N y D is the band gap reduction due to N alloying in the ternary nitride alloy and is a generalized bowing parameter that describes the effect of combining N and Bi. For example if >0 then the combined effects of alloying with both Bi and N exceeds the sum of the effect of the two elements independently. Experimental data is available for the composition dependence of the band gap of the ternary semiconductor alloys Ga 1-z InN y As 1-y, GaAs 1-x Bi x and GaN y As 1-y, and for the quaternary alloy Ga 1-w In w N y As 1-y. The analogous formula to Eq. 1 for the quaternary Ga 1- win w N y As 1-y alloy has been found to give excellent agreement with a large body of experimental data, fitting nine experimental values of the band gap widely dispersed in composition with a standard deviation of 28 mev, using = 0.47 ev -1 [41]. The fact that is negative means that the combined effect of In and N alloying on the band gap, is less than the sum of the two elements individually. Experimental data is available for only a small number of quaternary GaN y As 1-x-y Bi x alloy compositions, therefore the value of the correlation parameter and the quality of the fit of the interpolation formula to the experimental data are not as well established. Nevertheless it is interesting that the best fit value = ev -1 for the bismide-nitrides, is positive, meaning that the band gap reduction due to the combined effects of Bi and N is larger than the sum of the two independently. A best fit contour plot for the band gap of the quaternary 9

10 bismide-nitride alloy, using four experimental data points, is shown in Fig. 11. According to this figure, we note for example that an alloy with a band gap of 1.0 ev, lattice matched to Ge, such as that required in four junction solar cells can be obtained with 2.2% Bi and only 1.3% N. This compares with 8% In and 2.8% N in the case of a Ga 1-w In w N y As 1-y alloy. The lower concentration of N in the bismide alloy (<1/2) is significant as the electronic properties degrade with increasing N concentration as discussed above. The room temperature photoluminescence spectrum for two uncapped GaAs 1-x Bi x layers with x=1.7% and 1.9%, and a 30 nm thick quantum well with x=3% are shown in Fig. 12 [31]. For reference in the same figure and on the same scale we show the photoluminesence of a GaAs/Ga 0.8 In 0.2 As multiquantum well sample with ten 5 nm quantum wells. The photoluminesence measurements were all performed at the same time with the same apparatus using a 20 ns pulsed, doubled, diode-pumped YLF laser at 523 nm with average power 0.1 mw. We conclude that the photoluminesence emission for the bismide is rather efficient, since the light output is similar in intensity to GaInAs quantum wells, which are known to be very efficient. This is a remarkable result in light of the fact that the bismide samples are grown at low temperatures ( o C). Low temperature grown GaAs is well known to have high densities of As antisites which form deep levels, cause efficient electron-hole pair recombination, and rapidly quench the photoconductivity [13]. The strong PL emission suggests one of two possibilities; namely that the As antisite defects are suppressed by the Bi surfactant or that the PL efficiency is enhanced by some other mechanism. A possible mechanism for PL enhancement is that Bi clusters trap excitons following a mechanism similar to that proposed for In x Ga 1-x N [42,43]. The bismide emission spectra are broader than the emission from the GaInAs quantum wells, especially for the sample with the largest Bi content. The broadening may be due to local fluctuations in the Bi concentration (alloy broadening) accentuated by the tendency of Bi clusters to form localized states in the band gap, or by non-uniformity in the Bi concentration through the thickness of the samples [44]. The temperature dependence of the PL emission energy was found to be similar to the temperature dependence of the GaAs band gap taking into account the shift in the band gap to lower energies [5, 19]. Using micro-pl in which the excitation light is focused to a micron sized spot, and resonant excitation, Fluegel et al [45] and Tan et al [8] have measured the excited state 10

11 photoluminenscence from the spin-orbit split-off hole band in the dilute bismides and dilute nitrides respectively. The spin-orbit splitting in the valence band is found to be independent of N content in GaN y As 1-y for y<2.8% [8,19,45]. This result is consistent with the bandgap reduction in the dilute nitride being entirely due to the CBM being moved down through interaction with the N orbitals, with no change in the valence band. On the other hand in the case of the dilute bismides, in striking contrast to the dilute nitrides, the spin orbit splitting in the valence band shows a giant bowing effect similar to the band gap [45]. The measurements of the spin orbit splitting by Fluegel are shown in Fig. 13 (a) and (b) [45]. At low Bi concentrations the spin orbit splitting increases by 78 mev/%bi. Since the bandgap is reduced by 88 mev/%bi, this means that the bandgap reduction is almost entirely associated with an increase in the splitting between the VBM and the spin orbit split-off band. Once again this is consistent with Bi alloying primarily affecting the valence band. Transport Properties In the schematic picture of the electronic structure of the bismides presented in Fig. 1, in which the bismuth level is resonant with the VBM, one might expect Bi alloying to lead to stronger scattering of holes than electrons, and the hole mobility to be reduced more than the electron mobility. The opposite behaviour might be expected for the dilute nitrides with electrons in the CBM strongly scattered by nitrogen and the holes less so. Recent terahertz measurements of transient photoconductivity in the dilute nitride and dilute bismide alloys tends to support this picture [47]. In the terahertz experiments the electron scattering time is measured directly, from which the mobility can be inferred if one has an estimate of the effective mass [47]. As illustrated by the frequency dependent conductivity data presented in Fig. 14, the electron scattering rate is only weakly affected by Bi doping whereas the electron transport is strongly affected by N alloying. In a GaAs 1-x Bi x sample with x=1% the electron mobility is reduced to 2800 cm 2 /Vs from 3300 cm 2 /Vs for a reference GaAs sample. In a GaN y As 1-y sample alloyed with a similar concentration of nitrogen (y=0.84%) the electron mobility is significantly lower, namely 920 cm 2 /Vs. In addition the conductivity shows a non-drude frequency dependence, which may indicate that the electrons are localized. Bertulis et al. [13] have reported Hall effect measurements on unintentionally doped p- type dilute GaAsBi with x=3% and 5%. Although these authors do not give a value for the hole mobility one can infer a hole mobility of ~50 cm 2 /Vs from the carrier density and range of 11

12 resistivities presented in the paper. In the case of dilute InGaAs 1-x N x alloys, lattice-matched to GaAs with x=2%, the electron and hole mobilities are thermally activated, suggesting the presence of a mobility edge and localized states near the band edge [7]. The room temperature electron electron and hole mobilities were found to be cm 2 /Vs and cm 2 /Vs respectively. In the case of the dilute nitride GaN x As 1-x without In, Adamcyk [48] and Shan et al [49] find somewhat lower values for the room temperature electron mobility. To summarize, the terahertz measurements show a significant difference in electron transport properties between the dilute nitrides and the dilute bismides, consistent with expectations based on the electronic structure. So far the data on hole transport is not sufficient to draw conclusions about the relative effect of N and Bi alloying on the hole mobility in GaAs. Future Prospects There are many opportunities for further research on the dilute bismides, which can be divided up into the broad areas of materials science, electronic properties and devices. With regard to materials science there is a need for further optimization and better understanding of the growth process and the rather narrow process window for growth of GaAs 1-x Bi x alloys without metal droplets on the surface. GaAs 1-x Bi x is grown in the same temperature range as the low temperature GaAs used in high speed photoconductive switches which contain excess As in the form of As anti-sites. It would be interesting to know how the Bi surfactant layer affects the density of As anti-sites. There is also a need for more work on the growth and electronic properties of the quaternary GaN y As 1-x-y Bi x alloy. In particular it is important to find out if the lower N content possible with the dilute GaAs-bismide-nitride alloys conveys an advantage in terms of superior electronic properties relative to the GaAs-indium-nitride alloys. We expect that similar dilute, immiscible, isoelectronic impurities will also have interesting and useful properties in other semiconductor hosts. Although we anticipate the possibility of states in the band gap close to the VBM associated with Bi clusters, there have been no reports of experimental observations of these states. Furthermore since the Bi states are resonant with the VBM, one would expect the hole mobility to be strongly affected by the Bi doping. Further work is needed on hole transport in the dilute bismides. The large increase in spin orbit splitting in the valence band with Bi doping raises the question of how the Bi alloying affects the conduction band spin-orbit splitting which is non-zero away from the -point. So far there are no theoretical calculations or experimental 12

13 measurements of the electron spin orbit coupling in GaAs 1-x Bi x. A large increase in spin orbit splitting for electrons would be an important development in the field of spintronics where electron spin transport, rather than hole spin transport, is of most interest because of its large coherence length. Dilute bismide alloys have potential for application in transistors, light emitting devices and solar cells. In the case of heterojunction bipolar transistors (HBTs) the smaller band gap of the dilute bismide compared with GaAs, when incorporated into the base of a transistor, will enable lower threshold devices, which in turn can be expect to reduce power consumption [50-52]. Low power consumption is critical for output amplifiers on portable wireless devices. The band alignment of GaAs 1-x Bi x with GaAs is favourable for HBTs in that the CBM of GaAs 1-x Bi x is expected to be approximately aligned with that of GaAs, presenting no barrier to electron injection into the base while the relatively large valence band offset will tend to block the backflow of holes into the emitter, increasing the current gain of the transistor. Furthermore the weak reduction in electron mobility with Bi alloying observed in the terahertz experiments means that the minority carrier transit time in the base will be short, enabling high speed devices. Further, the increase in the energy of the VBM would be expected to increase the maximum hole density that can be achieved with p-type doping since the Fermi level does not have to move down as far [49]. Of course it will be necessary to fabricate devices to test the validity of these concepts. Another potential application of the dilute bismides is as a ev band gap layer in four junction solar cells grown on Ge substrates, as an alternative to Ga 1-w In w As 1-y N y [39,40,53]. As pointed out above, in the case of GaN y As 1-x-y Bi x a factor of two less nitrogen is needed for the same band gap reduction, compared with the GaInNAs alloy, which may convey a benefit in terms of device performance, as discussed above. Conclusions The dilute bismide alloy GaAsBi belongs to a class of semiconductor alloys in which the dilute alloying element Bi, contributes a state that is resonant with the VBM. It is complementary to the dilute nitride GaN y As 1-y, in that in the nitride alloy, nitrogen contributes a state that is resonant with the CBM. Both alloys show a giant bandgap bowing effect in which a small concentration of the alloying element causes a disproportionate decrease in the band gap. The band gap reduction in the case of the nitride comes from a reduction in the CBM whereas the band gap reduction in the bismide comes from an increase in the energy of the VBM. In addition 13

14 to reducing the band gap Bi has a giant bowing effect on the valence band spin orbit splitting. Bismuth has a strong tendency to surface segregate during molecular beam epitaxy growth, making it an excellent surfactant for III-V semiconductor growth. As a surfactant Bi improves both the electronic properties and the surface morphology of the deposited layers. Several promising device applications have been identified. Acknowledgements We thank our numerous collaborators and coworkers for their efforts which we have summarized in this review article, especially Martin Adamcyk, Brian Fluegel, Sebastien Francoeur and Sebastien Tixier. We thank NSERC for financial support of the MBE growth program at UBC. 14

15 References 1. K. Oe, H. Okamoto, Jpn. J. Appl. Phys. 37, L1283 (1998) 2. S. Tixier, M. Adamcyk, T. Tiedje, S. Francoeur, A. Mascarenhas, P. Wei, F. Schiettekatte, Appl. Phys. Lett. 82, 2245 (2003) 3. M. Yoshida, S. Murata, A. Chayahara, Y. Horino, J. Saraie, K. Oe, Jpn. J. Appl. Phys. 42, L1235 (2003) 4. A. J. Noreika, W. J. Takei, M. H. Francombe, C. E. C. Wood, J. Appl. Phys. 53, 4932 (1982) 5. S. Francoeur M.-J. Seong, A. Mascarenhas, S. Tixier, M. Adamcyk, T. Tiedje, Appl. Phys. Lett. 82, 3874 (2003). 6. W. Shan, W. Walukiewicz, J. W. Ager, E. E. Haller, J. F. Geisz, D. J. Friedman, J. M. Olson, S. R. Kurtz, Phys. Rev. Lett. 82,1221 (1999) 7. S. R. Kurtz, N. A. Modine, E. D. Jones, A. A. Allerman, J. F. Klem, Semicond. Sci. Technol. 17, 843 (2002) 8. P. H. Tan, X. D. Luo, Z. Y. Xu, Y. Zhang, A. Mascarenhas, H. P. Xin, C. W. Tu, W. K. Ge, Phys. Rev. B 73, (2006) 9. M. Yoshimoto, W. Huang, G. Feng, K. Oe, Phys. Stat. Sol. (b) 243, 1421 (2006) 10. J. Yoshida, T. Kita, O. Wada, K. Oe, Jpn. J. Appl. Phys. 42, 371 (2003) 11. W. Huang, K. Oe, G. Feng, M. Yoshimoto, J. Appl. Phys. 98, (2006) 12. A. Mascarenhas, Y. Zhang, J. Veerley, M. J. Seong, Superlattices and Microstructures, 29, 395 (2001). 13. K. Bertulis, A. Krotkus, G. Aleksejenko, V. Pacebutas, R. Andromavicius, G. Molis, Appl. Phys. Lett. 88, (2006). 14. A. Janotti, S. H. Wei, S. B. Zhang, Phys. Rev. B 65, (2002) 15. Y. Zhang, A. Mascarenhas, L. W. Wang, Phys. Rev. B 71, (2005) 16. U. Tisch, E. Finkman, J. Salzman, Appl. Phys. Lett. 81, 463 (2002) 17. E. Nodwell, M. Adamcyk, A. Ballestad, T. Tiedje, S. Tixier, S. E. Webster, E. C. Young, A. Moewes, E. Z. Kurmaev, T. van Buuren, Phys. Rev. B 69, (2004) 18. S. Adachi, Physical Properties of III-V Semiconductor Compounds, Wiley (1992) 19. S. Francoeur, PhD Thesis, University of Colorado (2004) 20. S. Tixier, M. Adamcyk, E. C. Young, J. H. Schmid, T. Tiedje, J. Cryst. Growth 251, 449 (2003). 15

16 21. E.C. Young, S. Tixier, T. Tiedje, J. Cryst. Growth 279, 316 (2005). 22. R. R. Wixom, L. W. Rieth, G. B. Stringfellow, J. Cryst. Growth 265, 367 (2004) 23. A. D. Howard, D. C. Chapman, G. B. Stringfellow, J. Appl. Phys. 100, (2006) 24. Z. Zhang, M. G. Lagally, Phys. Rev. Lett. 72, 693 (1994) 25. H. D. Sun, S. Calvez, M. D. Dawson, J. A. Gupta, G. I. Sproule, X. Wu, Z. R. Wasilewski, Appl. Phys. Lett. 87, (2005) 26. J. A. Gupta, G. I. Sproule, X. Wu, Z. R. Wasilewski, J. Cryst. Growth 291, 86 (2006) 27. M. Weilmeier, K. M. Colbow, T. Tiedje, T. van Buuren, L. Xu, Can. J. Phys. 69, 422 (1991) 28. S. R. Johnson, C. Lavoie, T. Tiedje, J. Vac. Sci. Technol. B11, 1007 (1993). 29. S. R. Johnson, Y. G. Sadofeyev, D. Ding, Y. Cao, S. A. Chaparro, K. Franzreb, Y. H. Zhang, J. Vac. Sci. Technol. B 22, 1436 (2004) 30. M. Reason, H. A. McKay, W. Ye, S. Hanson, R. S. Goldman, V. Rotberg, Appl. Phys. Lett. 85, 1692 (2004) 31. E. C. Young, PhD Thesis, University of British Columbia (2006) 32. H. B. Yuen, S. R. Bank, H. Bae, M. A. Wistey, J. S. Harris, J. Appl. Phys. 99, (2006) 33. F. Dimroth, A. D. Howard, J. K. Shurtleff, G. B. Stringfellow, J. Appl. Phys. 91, 3687 (2002) 34. J. N. Beaudry, R. A. Masut, P. Desjardins, R. Wei, M. Chicoine, G. Bentoumi, R. Leonelli, F. Schiettekatte, S. Guillon, J. Vac. Sci. Technol. A 22, 771 (2004) 35. N. Shtinkov, P. Desjardins, R. A. Masut, M. Cote, Phys. Rev. B 74, (2006) 36. P. Laukkanen, J. Pakarinen, M. Ahola-Tuomi, M. Kuzmin, R. E. Perala, I. J. Vayrynen, A. Tukiainen, J. Konttinen, P. Tuomisto, M. Pessa, Phys. Rev. B 74, (2006) 37. E.C. Young, M.B. Whitwick, T. Tiedje, D. Beaton, Phys. Stat. Sol. (to be published 2006). 38. M. Ferhat, A. Zaoui. Phys. Rev. B 73, (2006) 39. F. Dimroth, Phys. Stat. Sol. (c) 3, 373 (2006) 40. T. Takamoto, M. Kaneiwa, M. Imaizumi, M. Yamaguchi, Prog. Photovolt: Res. Appl. 13, 495 (2005) 41. S. Tixier, S.E. Webster, E.C. Young, T. Tiedje, S. Francoeur, A. Mascarenhas, P. Wei, and F. Schiettekatte, Appl. Phys. Lett. 86, (2005). 42. O. Brandt, K. H. Ploog, Nature Mater. 5, 769 (2006) 16

17 43. S. F. Chichibu, A. Uedono, T. Onuma, B. A. Haskell, A. Chakraborty, T. Koyama, P. T. Fini, S. Keller, S. P. DenBaars, J. S. Speck, U. K. Mishra, S. Nakamura, S. Yamaguchi, S. Kamiyama, H. Amano, I. Akasaki, J. Han, T. Sota, Nature Mater. 5, 810 (2006) 44. P. Wei, S. Tixier, M. Chicoine, S. Francoeur, A. Mascarenhas, T. Tiedje, F. Schiettekatte, Nucl. Inst. Meth. Phys. Res. B , 671 (2004). 45. B. Fluegel, S. Francoeur, A. Mascarenhas, S. Tixier, E.C. Young, T. Tiedje, Phys. Rev. Lett. 97, (2006) 46. J. D. Perkins, A. Mascarenhas, Y. Zhang, J. F. Geisz, D. J. Friedman, J. M. Olson, S. R. Kurtz, Phys. Rev. Lett. 82, 3312 (1999) 47. D.G. Cooke, E.C. Young, F.A. Hegmann, T. Tiedje, Appl. Phys. Lett. 89, (2006). 48. M. Adamcyk, PhD Thesis, University of British Columbia (2002) 49. W. Shan, K. M. Yu, W. Walukiewicz, J. Wu, J. W. Ager, E. E. Haller, J. Phys. Cond. Mat 16, S3355 (2004) 50. Hase, US Patent #7,009,225 (2006) 51. P. M. Asbeck, R. J. Welty, C. W. Tu, H. P. Xin, R. E. Welser, Semicond. Sci. Technol. 17, 898 (2002) 52. K. L. Lew, S. F. Yoon, H. Wang, S. Wicaksono, J. A. Gupta, S. P. McAlister, J. Vac. Sci. Technol. B 24, 1308 (2006). 53. Mascarenhas, US Patent #6,815,736 (2004) 17

18 Figure Captions Figure 1 Schematic illustration of band structure of a) GaNAs and b) GaAsBi, showing conduction band (CB), heavy hole valence band (HH), light hole valence band (LH) and split-off band (SO). Bi primarily affects the valence band, due to a resonant interaction with the 6p states of Bi, while the 2s state of N are resonant with the CB. Figure 2 Schematic band alignment for a dilute nitride and a dilute bismide quantum well structure with GaAs barrier layers for x,y=3%. The valence and conduction band offsets in the nitride and bismide respectively, are shown as zero for simplicity. In reality there is likely to be a small band offset for these bands as well. Figure 3 Bi coverage as inferred from RHEED intensity vs. growth temperature for a Bi flux of 1.4 x 10-5 Torr. (ref Young et al JOCG 2005) Figure 4 2 x 1 m AFM images of GaN As grown at 460 o C as follows: (a) no Bi flux and low As2 overpressure As 2 /Ga ~1, (b) a Bi BEP of 10-7 Torr and As 2 /Ga ~ 3 and (c) a high Bi BEP ~1.4x10-5 Torr and As2/Ga~1. The vertical scale in (a) is 10 nm and rms roughness is 1.2 nm, in (b) 2 nm and rms roughness 0.4 nm, and in (c) 0.8 nm and rms roughness 0.1 nm. (ref Tixier et al JOCG 2002) Figure 5 (a) Room temperature PL from as-grown and annealed InGaNAs quantum wells grown with and without Bi surfactant (Bi flux of 10-7 Torr). The peak intensity increased by 2.4x for the sample grown with Bi for both conditions. Dashed lines correspond to samples annealed at 730 o C for 60s (b) Room temperature PL from GaAs quantum wells with AlGaAs cladding grown with (dashed line) and without Bi (solid line) at 600 C. The Bi flux was 4 x 10-7 Torr. (ref Tixier et al JOCG 2002). Figure 6 Nitrogen concentration vs. Bi flux for GaNAs samples grown at 400 and 460 o C. Solid lines show [N] Bi = [N] o ( ), where is the surface Bi coverage given by a modified Langmuir model (ref Young et al JOCG 2005) 18

19 Figure 7 Lattice parameter of the pseudomorphic (open-square) and free standing (filled square) GaAsBi alloys as a function of the Bi concentration. The lattice parameters are from x-ray diffraction measurements and the Bi concentrations are from RBS. The solid line is a linear fit. The calculated lattice parameter for GaBi (d GaBi ) is indicated (ref Tixier APL 2003). Figure 8 High resolution 004 x-ray diffraction data for three GaAs 1-x Bi x films with compositions as shown. The top spectrum corresponds to a 210 nm film, the middle spectrum to a 125 nm film and the bottom spectrum to a 20 nm quantum well with at 290 nm GaAs cap grown at 330 C, 390 C, and 340 C respectively. The presence of pendellosung fringes indicates the high uniformity of the films. Data are offset for clarity (ref EC Young PhD thesis 2006) Figure 9 As 2 pressure and in-situ diffuse light scattering intensity for light with a wavelength of 244 nm, as a function of time during dilute bismide growth at 365 o C. The arrow indicated the opening of the Ga and Bi source shutters. The chamber pressure is a measure of the As 2 flux on the growth surface which is controlled by a mechanical valve on the As cracker (ref Young et al PSSc 2007 in press). Figure 10 Band gap energy of GaAs 1-x Bi x as a function of the Bi composition x (ref Francoeur PhD). Figure 11 Band gap map for strained GaN x As 1-x-y Bi y on GaAs substrate using Eq. 1 as described in the text. Contours and experimental data points are labeled in ev. The lattice matched condition to GaAs is shown by the dashed line (ref Tixier APL 20) Figure 12 Room temperature PL spectra for three GaAs 1-x Bi x films with 1.7, 1.9 and 3.0% Bi grown at 370 C, 390 C, and 340 C. The PL spectrum for an InGaAs multi-quantum well structure with ten 5 nm quantum wells, grown at 450 C is shown for reference (dashed line). (ref EC Young PhD thesis 2006) 19

20 Figure 13 a) Strain-corrected energies of the spin orbit PL (squares) and the band gap PL (circles) of GaAsBi films as a function of Bi concentration taken at 150K and b) the spin orbit splitting obtained by subtracting the band gap energies from the spin-orbit energies in a) (ref Fluegel et al PRL 2006). Figure 14 Measurements of the real and imaginary part of the conductivity as a function of frequency in the Terahertz range for a) GaAs buffer layer, b) GaAsBi (0.84% Bi) c) GaNAs (0.84% N) and d) GaNAsBi (0.85% N, 1.4% Bi) 10 ps after 400 nm excitation at a fluence of 3.7 J/cm 2. The solid and dashed lines are fits to the real and imaginary parts of the conductivity using the Drude model (ref Cooke et al APL 2006) 20

21 Figure 1 GaN x As 1-x GaAs 1-x Bi x Energy Conduction Band CB N 2s CB E g NN 2 E g Bi cluster? Bi 6p Valence Band LH HH D o LH HH SO SO a) b) 21

22 Figure 2 22

23 Figure RHEDdata Langmuirmodel Bi coverage(ml) Substratetemperature( C)

24 Figure 4 24

25 Figure InGaNAsQWs(26%In; 1.1%N) AsGrown Anealedat 730 Cfor60s 0.6 WithBi (BEP~10-7 Tor) RTPLintensity(Arb. units) Wavelength(nm) Room temperature PL (arb. units) Bi BEP ~ 4x10-7 Torr Wavelength (nm) 840 (a) (b) 25

26 Figure C C NitrogenConcentration(%) x10-6 Bismuthflux(Tor) 26

27 Figure d GaBi =6.3±0.06Å GaAsBi laticeparameter(å) BismuthconcentrationfromRBS(%)

28 Figure 8 Diffracted Intensity ( cps ) GaAs 1-x Bi x x = 0.5% GaAs 1-x Bi x x = 1.9% GaAs 1-x Bi x x = 3% θ ( arcsec ) 28

29 Figure 9 Scattered Light Intensity (au) Open Ga Open Bi Scattered light signal Chamber pressure 5x Chamber Pressure (Torr) x10 3 Time (s) 29

30 Figure 10 30

31 Figure 11 31

32 Figure 12 14x10 3 GaAs 1-x Bi x 200 nm, x = 1.7% 12 GaAs 1-x Bi x 125 nm, x = 1.9% RT PL Intensity (au) nm 984 nm GaAs 1-x Bi x QW 20 nm QW, x = 3.0% 10 layer MQW 5 nm In y Ga 1-y As y = 20% nm Wavelength (nm) 32

33 Figure (a) 150 K PL Energy (ev) Spin-Orbit PL Band gap PL 1.30 SO splitting (ev) (b) Bi concentration (%) 33

34 Figure 14 34