Transmission electron microscopy

Whereas a narrow beam is rastered across the sample in SEM, transmission electron microscopy (TEM) uses a static electron beam that can be a few micrometres in diameter. By accelerating the electrons to hundreds of keV and preparing samples that are about 100 nm thick, the transmitted electrons image the sample in a

similar way that a transmission optical microscope can. Unlike SEM, the lateral resolution is not diminished by scattering events because the electrons are much more energetic and the samples are relatively thin. The instrument resolution in TEM is usually limited by aberrations in the electromagnetic lenses, and is typically a few Å. The effective resolution is also degraded as the electrons lose 5–50 eV of energy when they pass through a sample. This leads to a spreading of the focal length, which degrades the point-spread function and hence limits the lateral resolution [47]. Nevertheless, the resolving power of TEM is approximately an order of magnitude better than in SEM [30]. Whereas image contrast is often influenced by surface topography in SEM and by absorption in optical microscopy, contrast in TEM images is governed by density and thickness variations, phase contrast and diffraction of the transmitted electrons.

The electron gun in a TEM is similar to that in an SEM. In TEM, however, the sample is placed in front of the objective lens that produces the first image of the sample [47], as indicated in figure 2.14(a). The other post-specimen lenses are configured according to whether an imaging mode or diffraction analysis is used. Bright-field imaging is the most intuitive mode and uses directly-transmitted electrons to create the image. A centred aperture in the back focal plane of the objective lens is used to exclude the diffracted electrons and only image the central beam. As a result, strongly diffracting areas will appear dark in a bright-field image, and bright areas indicate where more electrons are directly transmitted. On the other hand, the objective aperture in dark-field imaging selects one of the diffracted beams and blocks out all of the other beams. In this case, areas that diffract strongly in the selected direction will appear bright. In practice, the aperture is still placed in the centre of the back focal plane and the incident beam is tilted, as shown in figure 2.14(c). This is known as centred dark-field imaging, and avoids astigmatism and spherical aberration that can be introduced with displacement dark-field imaging [47]. The third imaging mode, which was not used in this thesis, is known as high-resolution TEM. By recombining the central beam with selected diffracted beams, phase and amplitude information are retained and this mode can image crystal lattices with atomic resolution.

The analytical modes in TEM include diffraction analysis, EDX and electron energy loss spectroscopy (EELS) [30]. The former provides crystallographic information about the sample and the latter two techniques, which will not be discussed here, provide compositional information. The Bragg law governs the diffraction by atomic planes in crystalline samples, and the sample can be tilted to align the incident beam with different crystallographic planes. As shown in figure 2.14(b),

Figure 2.14: The principal components in a TEM column are shown in (a), the location of the back focal plane and image plane are illustrated in (b) and centred dark-field imaging is performed by tilting the incident beam, as shown in (c). These schematics were reproduced from [47]. A cross-section of the samples prepared for plan-view TEM is also shown in (d).

electrons transmitted through the specimen form an image, which is rotated 180◦_.

In addition, the objective lens brings the beam to a focus in the back focal plane, which lies between the objective and the image planes. This is manifested as a diffraction pattern, which is not rotated with respect to the sample [47]. For imaging modes, the intermediate lens is focussed on the rear image, but it is focussed on the back focal plane for diffraction analysis. In selective-area

diffraction, a small area of the sample, such as a single grain, can be investigated. The specimen is tilted so that the electron beam impinges parallel to a major crystallographic direction of the crystal, called a zone axis. The diffraction pattern can be analysed to identify the scattering planes, and hence the crystal phase and orientation. Diffraction analysis can also be used to characterise structural defects, such as stacking faults [42].

Features in selective-area diffraction patterns (SADPs) are characterised by the diffracting vector g, which points from the directly-transmitted spot to the

diffracted feature. Intuitively, g is the normal to the diffracting planes and its

magnitude is inversely proportional to d_(hkl), or the d-spacing. The magnification factor of a SADP is conventionally expressed in terms of the camera length (L). This is effectively the length of a simple diffraction camera that would create the same magnification, in the absence of any lenses. For small Bragg scattering angles θ, such that 2 sinθ≈2θ, it can be shown thatd_(hkl)|g|=λL, whereλis the electron wavelength [47]. A small camera length can be used to view high-order diffraction features, whereas large values of L may be chosen to investigate fine structure in the SADP. For crystalline samples, the d_(hkl) values and angles between different

g vectors can be compared with data in the literature and theoretical values for

a specific crystal structure. For example, the annealed TiO2 films investigated

in chapter 8 show clear similarities to the anatase phase. This titania structure belongs to the I41/amd space group, with well-known dimensions, a and c. In this

tetragonal space group, the interplanar spacing for a plane (h k l) is theoretically related by 1/d2_(hkl) = (h2+k2)/a2+l2/c2 and the angles between different planes can also be calculated [24]. In this way, measurements from SADPs can be compared with likely candidates in order to identify the crystal phase. The zone axis of the crystal can be calculated from the cross-product of two independent planes in a SADP. For a thorough analysis, a single crystal can be tilted from one zone axis to another. After extracting each zone axis from the two SADPs, the angle between these directions can be compared with the specimen tilt angle. This technique was used successfully in chapter8. All TEM images and diffraction patterns presented in this thesis were captured by Dr. Jenny Wong-Leung using the Philips CM300 system in the Research School of Earth Sciences, and 300 keV electrons were used for all data.

The titania films examined by TEM for this thesis were all less than 200 nm thick and deposited on Si. To prepare each sample for plan-view TEM, a 3 mm disc was ultrasonically cut and then the substrate was polished to about 100 µm. The middle of the substrate was further thinned with a Gatan 656 dimple grinder,

leaving about 40 µm thick Si in the centre. The disc was then mounted on a teflon stick, with wax covering all but the dimpled region. The HF/CH3COOH/HNO3

solution given in table 2.1 was used to etch the remaining Si in this opening, and optical transparency was frequently checked to monitor the thickness. When a pinhole was visible through the Si, the sample was cleaned in solvents and gently dried. As shown in figure 2.14(d), an electron-transparent TiO2 border remained

around each pinhole that was suitable for TEM analysis.

When samples are prepared for TEM, or indeed any characterisation technique, it is important to be mindful of potential preparation artefacts [48]. Depending on the sample structure and TEM modality, different preparation methods may be chosen. The potential for the sample to be damaged or contaminated should be considered with this choice. With care, however, one can be confident that the results are truly properties of the sample material. For the plan-view TEM discussed in chapter 8, the titania films were deposited on Si, which provided mechanical stability during the mechanical thinning steps. During thinning, the disc was mounted face-down on a Pyrex stub and the film surface was protected by CrystalbondTM _{adhesive. By only dimpling the middle of the sample backside, the}

rigidity was maintained by the 100 µm Si around the edge. During the subsequent etching, the titania surface and the undimpled Si were protected by wax. In fact, the chosen acidic solution showed no evidence of damaging TiO2 whereas Si was

selectively etched at about 10 µm/min. A pinhole formed through the Si but in some samples, the thin titania film initially remained intact. When the wax was dissolved during the final cleaning step, some of the suspended film was always broken (the intrinsic stress in the thin film may have also contributed to this). Given that these films were less than 200 nm thick but they still remained after etching suggests that the acids did not attack the titania through the hole in the Si backside. Hence, the titania that ultimately remained around the edge of the pinhole was not damaged during sample preparation. Therefore the film properties deduced from TEM were determined by the original deposition and annealing processes. Furthermore, trichloroethylene and then isopropanol were used in the final cleaning. These solvents dissolve adhesives, wax and organic matter, so no contamination would be expected in the films studied by TEM.

Figure 2.15: The standard processing steps used to fabricate quantum dot infrared photodetectors. The particular methods investigated to tune the devices are detailed in the relevant experimental chapters.