Electron Channeling Contrast Imaging for Rapid III-V Heteroepitaxial Characterization

An accelerating voltage of 25 kV was used for this study. The accelerating voltage will determine the electron beam penetration depth; with higher accelerating voltage, there will be BSE signal coming from greater depths in the sample. The high accelerating voltage was chosen for this system because it allows for visibility of dislocations that are far from the surface of the sample, buried at the interface. Other types of defects/features may be more or less visible at different accelerating voltages depending on the type of sample.

As previously discussed, the invisibility criteria will determine what features have strong contrast at the specific diffraction condition in use and the resultant imaging contrast of those features. Just as in TEM, this can be used to provide guidance to the operator as to what imaging conditions will be necessary to observe the particular defects of interest, or in the case of some unknown defect, a range of different diffraction conditions can be used to provide further information to help elucidate the nature of that defect. For example, to clearly image an array of misfit dislocations (MDs) that are orthogonal to each other, a number of different diffraction conditions can be used, depending upon the goal of the operator. This was previously demonstrated by the authors for ECCI characterization of MDs in GaP/Si,¹⁴ and is shown here in Figure 4 where four images of the same MD network, taken from a 50 nm thick GaP/Si sample, were captured using different diffraction conditions.

Figure 4A presents an ECP map indicating the diffraction condition, g, used in each of the images displayed in Figure 4B-E. Figure 4B is an image of the MD network as imaged under the g = [̅220] condition. As previously discussed, dislocation contrast is determined by the invisibility criteria, g · b = 0 and g · (b x u) = 0. In (001)-oriented zinc blende crystals, compressive strain is relieved by dislocations with u = [̅110] and [̅1̅10] line directions – vertical and horizontal, respectively, in the coordinates of Figure 4 – with four distinct Burgers vectors possible for each. For the g = [ ̅220] diffraction condition all four possible Burgers vectors associated with the horizontal u = [1̅1̅0] line direction give non-zero values for both invisibility criteria and thus provide strong contrast. Those on the vertical u = [̅110] direction yield g · (b × u) = 0, but also g · b ≠ 0, and should thus provide only weak contrast, as can be seen in Figure 4B. Note that the off-axis tilt of the dislocations in the horizontal direction is a result of the use of an intentionally misoriented Si(001) substrate (i.e., offcut 6° toward [110]).²²   The opposite contrast levels displayed by the horizontal MDs (i.e., dark and bright) are related to the sign of g · (b × u), thereby providing an additional level of distinction between the different dislocations. Previous work by the authors comparing experimental and simulated offcut GaP/Si ECCI data indicated that of the four possible Burgers vectors for the u = [1̅1̅0] (horizontal) line direction, only two are actually observed, potentially due to a preferential dislocation nucleation and glide mechanism resulting from the offcut substrate;²³ whether the same occurs in the u = [̅110] (vertical) direction is difficult to ascertain due to the lack of offcut induced dislocation skew.

Figure 4C shows the same MD network with the diffraction condition antiparallel to that of Figure 4B, g = [2̅20]. Because the dislocations that are perpendicular to g = [̅220] are also perpendicular to g = [2̅20], they still possess high contrast, but with opposite polarity due to the change in sign of the diffraction condition. This means that contrast reversal can be used in combination with the standard invisibility criteria using a set of known g vectors to determine the sign of the Burgers vector of a given defect.  Indeed, the Figure 4B and 4C images were taken using the same Kikuchi band, but on opposite edges. In Fig. 4(d), the vertical oriented MDs, which are orthogonal to those highlighted in Figure 4B-C now exhibit strong contrast due to the use of an orthogonal diffraction vector, g = [220], while the horizontal dislocations exhibit very weak contrast. Finally, in Figure 4E, both sets of MDs are visible when using the diffraction condition g = [400], which is non-parallel to either set and thus yields non-zero invisibility criteria values for all the possible Burgers vectors and line directions.

In addition to providing TEM-like data within an SEM, a particular strength of ECCI is the ability to perform some such analyses in a rapid manner, significantly faster and simpler than would typically be possible via TEM. One example of this is presented in Figure 5, where ECCI was used to perform a multi-sample analysis of misfit dislocation evolution over a range of GaP-on-Si film thicknesses (30 nm to 250 nm), with the goal of accurately determining the critical thickness (the thickness necessary to induced strain relaxation via the formation of dislocations) for dislocation nucleation, h_c, as well as developing a better understanding of dislocation glide dynamics. Figure 5A shows an ECCI image of a 30 nm thick sample, which exhibits no observable MD features. This thickness is thus most likely sufficiently below h_c such that no nucleation events have yet occurred. This is consistent with previous TEM-based studies suggesting that the GaP-on-Si h_c is somewhere in the range of 45nm – 90nm.^24,25 However, it is possible that some nucleation events have actually occurred but have not yet produced any observable misfit length. In this case, the just-nucleated dislocations should still be observable – indeed, there are a number of contrast features in the image that could be related to this, or to a slight surface roughness – but may be difficult to adequately resolve due to a lack of strain-driven loop expansion.

As the film thickness increases, presented in Figure 5B (50 nm) and Figure 5C (100 nm), interfacial misfit segments are seen to appear and extend, relieving excess misfit strain via glide; the thicker the epilayer the longer the resultant misfit lengths and the greater the number of MDs visible. The appearance of observable misfit dislocations in the 50 nm sample, Figure 5B, indicates that the critical thickness has been reached (at least at the growth temperature), giving a critical thickness estimate of somewhere in the range of about 30 nm – 50 nm, which represents a significant narrowing, and perhaps a slight shift, of the previously reported range. Additional high-temperature (725°C) annealing experiments (not shown here) were found to yield observable, though short, misfit lengths in the 30 nm the nucleation,¹⁴ suggesting that the critical thickness value may actually closer to the lower limit or mid-range. At significantly higher GaP thickness, such as the 250 nm sample shown in Figure 5D, the MDs themselves are no longer directly observable due to the previously mentioned depth-dependent broadening/damping of the ingoing electron wave front. Instead, the associated near-surface threading segments are visible, as well as broad contrast features likely related to misfit dislocation induced heterogeneous strain fields. This ability to non-destructively observe and count threading dislocations in such films at TEM-like spatial resolutions, which typically requires time-consuming plan-view TEM foil preparation and yields comparatively small areas of analysis, is another important strength of the ECCI technique.

While the main focus in this paper is the use of ECCI to characterize misfit dislocations in GaP/Si, it is important to note that it can also be applied to the characterization of other crystalline materials and other types of defects. Figure 6 presents examples of the latter. Figure 5A displays an ECCI micrograph of surface-penetrating threading dislocations in a 250 nm thick GaP-on-Si(001) sample, taken at higher resolution than that of Figure 5D. Of note here is that even the fringed tail of the thread can be seen, a feature regularly observed via plan-view geometry TEM (PV-TEM). Similarly, Figure 6B displays an ECCI micrograph of a stacking fault in the same sample – an important tell-tale sign of non-optimal GaP nucleation for this particular test structure – which also displays observable extinction fringes. This fringing has also been observed via ECCI in metals samples by other researchers.^1,26 These types of micrographs can be obtained via ECCI much more quickly than via TEM since the sample requires no preparation or processing. All the while, the potential resolution achievable with ECCI is comparable to that of conventional PV-TEM, making ECCI an effective tool for rapid characterization of density and distribution of extended defects, such as dislocations and stacking faults, as demonstrated above.

In this work the procedure for ECCI was described. Because the ECCI signal is diffraction-based, it can be performed under different, specific diffraction conditions much in the same way the TEM operates, making it possible to image various types of defects. This makes ECCI an excellent alternative to TEM for detailed microstructural characterization in cases where rapid turn-around and/or large numbers of samples are needed, or where non-destructive, wide-area characterization is desired. Here, ECCI was demonstrated through the characterization of misfit dislocations at the lattice-mismatched interface of heteroepitaxial GaP-on-Si samples, but it has a large range of applicability and can be used for other types of defects and crystalline structures.