Alternate routes to high-repetition-rate probing

For future experiments, a FIB-based transfer process as shown in Fig.6.2may be used as an alternative to the manual film placement. In this case, a small 1T-TaS₂flake had been transferred onto a standard TEM copper grid after ultramicrotome preparation. A micromanipulator was then used to extract one mesh cell, transfer it onto a silicon nitride membrane and mount it using FIB-deposited platinum. In analogy to the preparation of the DF specimen structure, the membrane had a front-side gold coating and a circular through-hole prepared in advance.

However, static TEM diffraction after preparation of the FIB-based structure has only revealed the hexagonal lattice of the material, and no trace of CDW superstructure reflections. It has to be assumed that implantation of scattered gallium ions during the preparation process severely damaged the electronic structure of the material (342), effectively suppressing CDW formation.

This indicates that even more care has to be taken in future FIB-based approaches not to contaminate the 1T-TaS₂film. The same problem has also been reported to occur when using reactive ion etching processes for the preparation of 1T-TaS₂thin films. While energy-dispersive x-ray spectroscopy reveals almost perfect stoichiometry for ultramicrotomed samples, reactive-ion-etched samples exhibit large fractions of oxygen and carbon (302).

Generally, the combination of pulsed laser excitation and FIB-deposited platinum is prob-lematic. The resulting amorphous platinum compound contains a large fraction of organic byproducts of the gaseous precursor. For platinum deposition, the organometallic precursor trimethyl(methylcyclopentadienyl)platinum (MeCpPtMe₃) is a common choice. Some of the

6.1 Preparation and characterization of thin films for UTEM experiments

A B C

50 µm 50 µm 15 µm

Figure 6.2: Transfer of a 1T-TaS₂film using FIB manipulation. (A) A micromanipulator is secured to the copper grid using ion-beam-assisted platinum deposition. The grid bars are cut by FIB milling before lifting the extracted portion of the grid. (B) Subsequently, the grid portion is positioned on a gold-coated silicon nitride membrane and fixed using platinum deposition. (C) In a last step, the micromanipulator is detached using FIB milling. A through-hole in the membrane (below the 1T-TaS₂flake) prepared in advance defines the region of interest for TEM.

A B C

500 nm 500 nm

VO₂ Sapphire

GoldPlatinum

10 µm Silicon nitride

membrane

Lamella

Figure 6.3: Crystallization of FIB-deposited platinum.(A) Scanning electron micrograph of a cross-sectional VO₂lamella after placement on a pristine silicon nitride membrane (dark area in the bottom left corner). Some platinum residue is visible on the membrane (white quarter circle). (B) TEM image of the lamella with the different layers indicated. (C) After exposure to ultrashort laser pulses, platinum crystallization is visible on the silicon nitride membrane.

precursor is also deposited in undesired regions of the specimen. Under laser illumination, these leftovers crystallize and modify the surface, changing for example its optical properties. This is shown in Fig.6.3using the example of a VO₂lamella. The effect of this kind of laser annealing is probably very similar to thermal annealing as a post-deposition treatment (361).

As an alternative, stroboscopic repetition rates could be increased by optimizing the thermal conductivity in the experimental region of interest. While the specimen structure used in the nanoimaging experiment features a free-standing film of 1T-TaS₂, the use of a sufficiently electron-transparent substrate with a low scattering cross-section and high thermal conductivity may help to reach that goal. Standard choices for TEM are amorphous carbon and silicon nitride membranes, but there are also more uncommon options such as single-crystalline silicon and even diamond membranes.

Material Thermal conductivity References

Silicon nitride membrane (amorphous) 4.9 W/(m⋅K) (341)

Silicon (bulk) 148 W/(m⋅K) (326)

Silicon membrane (single-crystalline) 22 W/(m⋅K) (362) Carbon membrane (amorphous) 0.2 to 2.2 W/(m⋅K) (363)

IIa diamond (bulk) 2300 W/(m⋅K) (326)

CVD diamond (bulk, polycrystalline) >1200 W/(m⋅K) (364) Diamond membrane (polycrystalline) Reduced by a factor

of 5 to 8 w.r.t. IIa diamond (365)

Table 6.2: Thermal conductivities of TEM membranes (in-plane) and bulk materials.All values are given at or close to room temperature.

As shown in Table6.2, these materials differ widely in their thermal conductivities. The obvious choice from this point of view is diamond due to its extremely high thermal conductivity.

Polycrystalline diamond membranes with thicknesses of a few tens of nanometers can be produced by chemical vapor deposition (CVD) and a following etching process. While bulk CVD diamond reaches at least half of the thermal conductivity of IIa diamond (natural diamond that contains the lowest density of impurities), the value is found to be reduced in polycrystalline membranes by a factor of 5 to 8 with respect to IIa diamond (365). With that, the thermal conductivity would still be a factor of two higher than that of bulk silicon (326), and more than one order of magnitude higher than that of single-crystalline silicon membranes (362).

To test the usefulness of diamond substrates in UTEM experiments, we obtained membranes from two companies, “Diamond Materials” and “Applied Diamond”. The respective membrane thicknesses are 50 and 110 nm, and 40 nm. Figure6.4A shows a TEM image of one of the diamond membranes. In the out-of-focus image, one can see the diamond grains whose different

6.1 Preparation and characterization of thin films for UTEM experiments

A B C

500 nm 2 µm 10 nm^-1

Figure 6.4: TEM images and diffraction patterns of polycrystalline diamond membranes. (A) TEM image of a 50 nm diamond membrane (Diamond Materials) imaged under in-focus (top half) and out-of-focus (bottom half) conditions. The out-of-focus image visualizes the nanoscale grain boundaries in the film. (B) TEM image of a 40 nm diamond membrane (Applied Diamond) with two overlapping ultramicrotomed 1T-TaS₂ flakes.

(C) Diffraction image taken from the upper flake in B. The polycrystalline diamond membrane gives rise to diffraction rings in addition to the structural and CDW reflections of 1T-TaS₂.

orientations lead to the granular contrast in the in-focus image. Generally, there is no large difference to other membrane materials in terms of handling, and 1T-TaS₂flakes obtained by ultramicrotomy can be deposited on the membranes for TEM observation (Fig.6.4B). In the diffraction images, strong scattering from the diamond grains overlaps with the hexagonal pattern of 1T-TaS₂(Fig.6.4C).

Further inspection of diffraction patterns obtained from empty diamond membranes reveals that a large fraction of the electron beam is elastically scattered by the membrane alone. We estimate that 83 % and 38 % of the electrons remain unscattered while passing through the 50 nm and 110 nm membranes, respectively. The two 40 nm membranes by the second manufacturer left 63 % and 77 % of the incident electrons unscattered. This large scattering cross-section leads to multiple scattering effects that are evident from additional diffraction rings centered around the brightest structural reflections of 1T-TaS₂.

We find that the gain in thermal coupling of the specimen is not high enough to tolerate the pro-nounced elastic scattering background. While heating intensities on the order of 0.15 mW/µm² were necessary to statically heat most of the specimen structure discussed in section6.1.5to temperatures above the NC/IC phase transition temperature (Fig.5.3A), we experimentally found that values about 100 times smaller are necessary to reach the same local temperature in the specimen shown in Fig.6.4B. In conclusion, it is much more efficient to confine the specimen excitation to the region of interest of the UTEM experiment than to improve the thermal coupling by choice of substrate only.