Investigation of a short-period (001) HgTe-Hg0.6Cd0.4Te superlattice by transmission electron microscopy

A molecular beam epitaxially grown short-period (001) HgTe-Hg0.6Cd0.4Te superlattice was studied by means of transmission electron microscopy and high-resolution electron microscopy (HREM). Cross sections of the as-grown samples revealed good epitaxial features and surprisingly strong contrast between HgTe and Hg0.6Cd0.4Te layers in spite of the small difference in Cd concentration between the wells and the barriers. The fact that the variation in Cd concentration is so small is due in part to interdiffusion between the very narrow wells and barriers, which are only four and six monolayers wide, respectively. Precipitates of dimensions 1 by 12 nm have been identified in HREM images as the monoclinic, high-pressure phase of elemental Te.


Introduction
The HgTe-Hgl-,Cd,Te superlattice (SL) was first proposed as a potential infrared material in 1979 [I] and it has received a good deal of attention over the last few years. It has been pointed out that the bandgap of this SL could be varied throughout the entire infrared spectral region by varying the thickness of the HgTe well and to a lesser extent by varying the Hgl-,Cd,Te barrier thickness. Molecular beam epitaxy (MBE) or variations thereof is perhaps the best technique to prepare multilayers of this type. After the first HgTe-Hgl-,Cd,Te SL was grown by means of MBE   Intermixing of the HgTe and Hg,-,Cd,Te layers has attracted much attention because this would cause a change in the composition of the wells and barriers, and consequently would affect the properties of this SL system such as the bandgap, as proposed by Schulman and Chang [16]. In spite of many previous diffusion investigations [3,4,7,[17][18][19][20] knowledge of the microscopic structure of the diffused HgTe-Hg,_,Cd,Te SL is not complete an,d the results in the literature are not always consistent.
Defects such as precipitates and dislocations in Hg,-,Cd,Te and CdTe have been the subject of numerous investigations [15, 21-26], but to our knowledge no HREM results on precipitates with atomic resolution have been published. Therefore an investigation of precipitates with atomic resolution was undertaken. In addition, misfit dislocations could be present in the SL in order to accommodate the strain fields caused by the lattice mismatch between heterolayers. Furthermore, knowledge of the crystal structure and composition in the vicinity of the filmsubstrate interface is also very important in improving the quality of the SL films.
It is of interest to note that most of the published work to date on the HgTe-Hg,-,Cd,Te SL is concentrated on superlattices whose periods are appreciably larger than 3 nm. In this article, we report on TEM studies of an MBE-grown (001) HgTe-Hg,-,Cd,Te SL with the extremely short period of 3.2 nm.

Experimental details
Four (001) HgTe-Hgl_,Cd,Te superlattices with 900 periods were grown simultaneously by MBE on (001) Cdo.96Zno.04Te substrates. Epitaxial growth was carried out in a four-chamber RIBER 2300 MBE system. The substrates were degreased, chemo-mechanically polished and etched as has been described previously [27]. A thin CdTe buffer of about 30 nm was grown at 270 "C and the SL was grown at 180 "C. The thickness of the HgTe well is 1.1 nm and that of the Hgi_,Cd,Te harrier is 2.0 nm, as determined directly with a five-crystal x-ray diffractometer [27,28], i.e. the period is 3.14 nm. Details of the corresponding x-ray analyses of this SL and its growth have been published elsewhere [27]. The substrate temperature was measured with an accuracy of f 2 "C by means of a thermocouple which was in physical contact with the molybdenum substrate holder. The thermocouple was carefully calibrated at the melting point of indium.
In order to prepare the cross-section samples for TEM examination, an as-grown sample was cut into small square pieces with {IlOj-type edges. The small pieces were then glued face to face with epoxy resin and thinned so as to ensure transparency to electrons first by mechanical methods and thereafter by means of Ar ion milling on a liquid-nitrogen-cooled stage. TEM experiments were performed in a JEOL 4000EX transmission electron microscope operated at 400 kV.

Results
The morphology of a cross section is shown in figure 1 . The buffer layer between the SL and the substrate is 30 to 40 nm thick and is indicated by BF in figure 1. An ambiguous interface is visible between the substrate and the buffer while, as may be expected, a sharp and smooth interface is present between the SL and the buffer. The latter interface is indicated by a pair of arrows. Figure 2 is a micrograph which shows a region close to the %-buffer interface. This interface can be readily recognized by comparing the alternately stacked darker and brighter layer structure of the SL with the homogeneous brightness of the buffer. From the deposition sequence, the darker and brighter layers are known to be Hgl-.Cd,Te and HgTe respectively. The quality of the very first layers of the SL is good in spite of the somewhat irregular thicknesses of these Hgl_,Cd,Te and HgTe layers. Moreover, after the first five periods the thicknesses of these HgTe and Hgl-,Cd,Te layers are constant within experimental errror. No misfit dislocations or specific structural defects were resolved in the vicinity of the SL-buffer interface.
A [lIO] bright-field image and a corresponding selected-area electron diffraction pattern are shown in figures 3(a) and ( b ) respectively. Somewhat irregular interfaces between HgTe and Hgl_,Cd,Te layers can be seen and a defect is indicated by a bold arrow. The period of the SL in the region displayed in figure 3(a) is 3.3 nm. This was confirmed by means of the electron diffraction pattern shown in figure 3 (b). The main reflections arising from HgTe and Hgl_,Cd,Te coincide because of the extremely small difference between their lattice parameters: the lattice parameter for HgTe is 0.6461 nm and that for CdTe is 0.6482 nm [29]. The first-and second-order satellites, which are due to the periodicity of the superlattice, are indicated by black arrowheads and a white arrow respectively. These   distortions. Our computer simulations suggest that there should be no discernible contrast between HgTe and CdTe layers when the defocus of the objective lens is changed from -30 to -70 nm if the crystal thickness is less than 1 1 nm or more than 16 nm. However, an observable image contrast between HgTe and CdTe can be achieved under the conditions of a -40 or -50 nm defocus together with a crystal thickness of 4000EX electron microscope is about -50 nm). Under these imaging conditions, the HgTe layers are brighter than the CdTe layers. In addition, the simulations predict a weaker but observable contrast between HgTe and Hgl-,Cd,Te for these imaging conditions when 0.3 6 x < 1. Even though the resulting contrast for x = 0.4 fits that of the experimental observations in figure 4 reasonably well, the perception of this contrast is difficult due to the extremely high resolution of the simulation. Hence, for demonstration purposes, the simulation of HgTe and CdTe layers is shown in figure 5. The reduction or even the disappearance of the contrast between H~T~ and ~~~-, c d ,~~ in Some regions can be attributed to these fairly strict imaging conditions. A defect region near the SL-buffer interface is shown in figure 6(a), A defect can be seen about 20 nm above the buffer surface. An image of this defect with a higher magnification is shown in figure 6 (b). This defect can be thought of as a small mosaic crystalline slab whose extended plane is parallel to the ( I IO) plane of the Hgl-,Cd,Te matrix. Its size is about 12 nm in the ( I 10) plane and 0.9 nm along a direction perpendicular to the (1 10) plane. A spherically symmetric strain field with a diameter of about 12 nm is apparent from the contrast variation in this region. The lattice match between this slab and the Hgl-,Cd,Te matrix is good along (001)  defocus of -40 to -50 nm, a defocus spread of 10 nm and an illumination half-angle of 1.0 mad. The brighter HgTe layers and the darker Hgl-,Cd,Te layers can be better distinguished along the [ 1 IO] direction (horizontal) by viewing from the side at a glancing angle. Close inspection of this local region shows that an average period consists of four HgTe monolayers (brighter) and six H g l -J X T e monolayers (darker) along the [ooll direction. Thus the average thicknesses of the HgTe and H g l d X T e layers are 1.3 and 1.9 nm respectively, and the period is 3.2 nm in agreement with that concluded from figure 3. The average thicknesses Of the wells and the barriers according to x-ray diffraction, 1.1 and 2.0 nm, corroborate the HREM results.
The ( I I I)-type lattice fringes in figure 4 run across the interfaces between HgTe and Hgl-,Cd,Te layers without any major disturbance, albeit with appreciable waviness which may be due to interdiffusion between the layers. I~ order to teSt the feasibility of the imaging conditions necessary for the observed contrast in HgTe-Hgl-,Cd,Te heterostmctures, image simulations were carried out for mentioned above, the SL thickness necessary for the near In order to characterize the structure and composition absence of contraSt between the H~T~ and ~~~_ , c d , T~ of this small crystalline slab, computer simulations were layers (< 12 nm or , 17 nm) shown in figure 6 is performed using the Program described above [30l. Pure consistent with the 6 nm thickness used in this computer Hg could be excluded after a series of calculations with simulation of monoclinic Te. Te precipitates can be various values for the defocus and crystal thicknesses caused by inappropriate local growth parameters [321. as well as for different zone-axis directions. PUre Te-rich compounds that have nearly the same structure Cd is not stable and is also unlikely according to the as that of pure Te are also possible. In addition to image simulation. However, good agreement can be these defects, dislocations have also been observed in achieved with a simulation of monoclinic Te, whose our Hg,-.Cd,Te films, lattice parameters [31] are a = 0.3104 nm, h = Small circular or elliptical loops are displayed in 0.7513 nm, c = 0.4766 nm and +3 = 92.71", but figure 8. Some are indicated by arrowheads. The not with hexagonal Te. The [010] direction of the size of these loops varies between 10 and 50 nm.
monoclinic Te is parallel to [I io] of the Hg,-,Cd,Te The corresponding reciprocal vectors g , and gx of the milling. Hence some or all of these dislocation loops may be due to Ar ion bombardment. The electron beam used for HREM images may possibly damage the sample, but no apparent damage occurred within the first few tens of seconds, which is enough time to yield good quality HREM images.  The number of satellites in a diffraction experiment is a good criterion far the structural quality of a superlattice electron beam used to make these images are indicated only if the width of the ,interfaces is small compared in the dark-field diffraction contrast images shown in with the period: As shown in figures 3(a) and 4, the This implies that be used to explain the presence of only two orders of they are dislocation loops rather than being caused satellites in the corresponding electron diffraction pattern by precipitates. Furthermore, the loops indicated by in figure 3(b); the number of observed satellites is a arrowheads in figures 8(a) and (b) have different shapes rough qualitative measure of the interfacial abruptness of to those in figures 8(c) and ( d ) (note the variation of the heterojunctions. The corresponding (004) x-ray rocking central region for each loop indicated). This excludes the curve also has only first-and second-order satellites, but possibility that precipitates with a spherically symmetric computer simulations of the satellite intensities are in strain field have caused these loops because these good agreement with the experimental x-ray intensities precipitates will, in principle, have similar shapes when monolayers wide. If the interface is assumed to be abrupt then the satellite intensities are one or two orders of magnitude too large. Furthermore, if an interface with twice the above width is used then the second-order satellites disappear. Superlattices with larger periods that have been grown under the same experimental conditions have up to seventh-order satellites in their (004) rocking curves 1271.

Te precipitates
As is normally the case, these HgTe-Hg,,Cd,Te superlattices were grown with a constant Hg flux. Bm'er material grown under the prevailing conditions has an x value of 0.70. An upper limit on the average Cd concentration of the barriers themselves was established by annealing one of the superlattices in Hg vapour, pressure 80 Torr, at 250 "C for 24 hours 1271. The Cd concentration of the resulting Hgl-,Cd,Te alloy was determined from the bandgap to be 23.0 at.%. This means that the average Cd concentration, E,, in the barriers is 36 at.% if no Cd is present in the wells. For these narrow wells this does' not seem to be a good assumption. Therefore .ib = 0.36 i r a n upper limit for this superlattice. For example. if X, = 0.05 then i b = 0.33. These results are to a good approximation consistent with the maximum value in the barrier of 0.4, which is necessary to calculate the correct optical absorption coefficient [34]. It is somewhat surprising that interdiffusion has not totally obscured these interfaces. Kim et a1 1191 found that interdiffusion is a sensitive function of the distance from the sample surface, i.e. interdiffusion is two orders of magnitude less at an interface depth of 700 nm than at 10 nm. Their published values for the Hg diffusion constant at 180°C are approximately 1 x I x cm*s-' for depths of IO, 350 and 700 nm respectively. The interfaces observed in our investigation are all close to the buffer and thus far from the surface of the superlattice, which is 2.84 p m thick. The time spent during growth at a distance of 350 nm or less from the surface was 30 minutes. Therefore, in spite of the growth time of nearly 4 hours at I80 "C the values for the diffusion constant according to Kim et a1 are consistent with the fact that the HgTe-Hgo,6Cdo.4Te interfaces have not been completely obliterated, i.e. Large variations in the'Hg,-,Cd,Te layer thickness are apparent in the region near the %-buffer interface as indicated by white arrows in figure 2. These layers have a thickness of about 2.5 nm rather than the average thickness of 1.9 nm. Since this variation occurred only within 15 nm of the buffer, it can be regarded as due to a deviation of the initial fluxes from their later steady-state values.
and I x

HgTe-Hg,,Cd,Te and SGbuffer interfaces
In general, a periodic arrangement of misfit dislocations occurs at the interface of heterostructures to accommodate the strain relaxtion caused by the lattice mismatch. However, in our samples, the lattice mismatches between HgTe and Hgo.6Cdo.aTe and between the SF and 2222 the Cd,,.B6Zno,wTe substrate are only 0.1 and 0.02% respectively. In addition, the formation of misfit dislocations at the SLbuffer interface is also hindered by the low substrate temperature (1 80 "C) used during the MBE growth. The thermal expansion coefficients of HgTe and Hgl-,Cd,Te differ only slightly, i.e. 4.0 x 10-6/"C for HgTe, 4.9 x 10-6/"C for CdTe and an intermediate value for Cd,Te [35].
Therefore, the resulting small strain fields and the very thin layers can be expected to induce only a small number of misfit dislocations. According to our HREM observations, no misfit dislocations were identified in either the HgTe-Hgl-,Cd,Te interfaces or the S!-buffer interface, as shown for example in figure 2. It should be pointed out that this statement implies only an upper limit for the misfit dislocation density of approximately IO9 to 10" cmWz due to the very nature of H E M investigations, i.e. the relatively low number of observations. Dislocations can be more easily observed in bright-field and dark-field images, as for example in figure 1, but the dislocation type cannot be determined without elaborate analysis. Chami et a1 [25] have shown that channelling is a more appropriate method to determine the low misfit dislocation densities that occur in nearly lattice-matched heterostructures.
Consequently, the relatively good quality of the superlattices investigated can be partly attributed to the generation of only a few misfit dislocations [32]. It should be pointed out that the CdTe buffer may play an important role in improving the quality of the films. This can be concluded from figure 1. in which a rather disordered buffer-substrate is followed by a much smoother and sharper %-buffer interface.

Conclusion
An extremely short-period MBE-grown (001) HgTe-Hgo.6Cd,,.4Te superlattice on a Cdo.&10,04Te substrate was studied by TEM. The superlattice was found to have a constant period after the initial five or so periods. The average layer thicknesses for the HgTe and Hg,-,Cd,Te layers, which were determined to be four and six monolayers wide respectively, were corroborated by x-ray diffraction measurements. Interdiffusion of the HgTe and the Hgl-,Cd,Te layers is consistent with the published diffusion constant for Hg [19]. Some nanometre-size defects, such as Te precipitates with a nearly spherically symmetric strain field, as well as dislocations and dislocation loops were observed. These precipitates have been shown to be the monoclinic, highpressure phase of Te.