2.1. Experimental setup

The cRED experiments were performed on a Themis Z microscope equipped with a Gatan OneView IS camera (4096 × 4096 pixels, pixel size 15 µm) and a JEM 2100F TEM equipped with a Gatan Orius SC200D camera (2048 × 2048 pixels, pixel size 7.4 µm). The OneView camera is well suited for cRED data acquisition, because it has essentially no readout dead time when in movie mode. The in situ data capture mode with 1024 × 1024 pixel resolution (binning × 4) was employed. cRED data were collected using a single-tilt TFS holder (±40°) without applying a beam stopper. We found that the Themis Z is very stable both electrically and mechanically, and the crystal tracking procedure described by Cichocka et al. (2018) is not a prerequisite for keeping the crystal centred in the electron beam during data collection. Before data acquisition, a standard TEM alignment routine was performed. All experiments were performed in the parallel illumination mode using a 50 µm C2 condenser aperture. A suitable magnification (typically ×13 000) in the image mode at the SA magnification range is chosen to search for a suitable crystal. The crystal is then moved to the centre of the screen. In order to ensure the crystal stays in the area selected by the aperture or electron beam during crystal rotation, it is important to adjust the crystal height to the mechanical eucentric position of the goniometer. This is achieved by either enabling an α-wobbler (±15°) or manually tilting the goniometer and minimizing the crystal drift by changing the Z height of the crystal. Diffraction patterns were focused to obtain sharp spots in the diffraction mode. The rotation speed was 1.44° s⁻¹ and the exposure time was 0.30 s per frame, leading to 0.432° per frame. A cRED data set with a total rotation range of ∼80° and 185 ED frames was collected in approximately 55 s.

Two different beam settings available on the Themis Z were tested, namely selected-area electron diffraction (SAED) and nanoprobe electron diffraction (NED) modes. In the SAED mode, a 40 µm SA aperture was inserted to limit the area used for diffraction, whereas in the NED mode the field of view was restricted by the beam size. Spot size 5 or 6 was usually used in the SAED mode, and spot size 11 in the NED mode. The electron dose on the specimen was controlled by varying the monochromator focus.

For the JEM 2100F equipped with a Gatan Orius SC200D detector (2048 × 2048 pixels, pixel size 7.4 µm), the exposure time and rotation speed were set up to be 0.5 s per frame and 0.444° s⁻¹, leading to 0.222° per frame and resulting in 209 frames within the total rotation range of 46.42° collected in 104.5 s. The relatively small tilt range was due to the limit of the single-tilt holder for the microscope.

2.2. Data processing and structure determination

Diffraction images were collected as TIFF files (.tif) and converted to SMV format (.img) using the process_DM Python script (Smeets, 2019). The collected frames were processed with the XDS software (Kabsch, 2010) for spot-finding, unit-cell determination, indexing, space-group assignment, data integration, scaling and refinement. The previously determined lattice parameters and space group (Olson et al., 1981) were used as input, and the REFLECTING_RANGE_E.S.D. parameter in the XDS.INP file was set to be 0.7 to include very sharp diffraction spots in the indexing procedure. Data statistics indicators provided in the output CORRECT.Lp file were used for further data quality comparison. The reflection file for structure solution and refinement was obtained by merging several individual data sets from different crystals using the XSCALE sub-program. The structure was solved by SHELXT and refined by SHELXL (Sheldrick, 2008, 2015b) using atomic structure factors for electrons (Doyle & Turner, 1968) with the help of the OLEX2 software (Dolomanov et al., 2009).

4.1. Tests of InsteaDMatic on Themis Z and JEM 2100F microscopes

First, we tested InsteaDMatic on the Themis Z with a Gatan OneView CCD camera, collecting cRED data from different crystals. A typical experiment was recorded in order to illustrate the procedure of cRED data acquisition (see Movie S1). The best Themis Z data set demonstrated a completeness of 77.7% in the resolution shells ranging from 2.36 to 0.80 Å (see Table S1), enabling ab initio crystal structure solution from this one individual data set. Unfortunately, the completeness of most individual data sets does not exceed 50% for the orthorhombic structure, and often only merged data can provide the correct structure (see below).

We found that the OneView camera is well suited for experiments that require continuous read-out of the sensor. To check if the script would work on other cameras, we tested it on an Orius SC200D detector installed on a JEM 2100F. A `single-crystal' data set collected over a rotation range of 46.42° reached a completeness of 34.5% in the resolution shells from 2.36 to 0.80 Å. Due to the limited tilting capability of the microscope, the data completeness is low, prohibiting a correct crystal structure solution by direct methods, e.g. SIR2014 (Burla et al., 2015) or SHELXT (Sheldrick, 2008, 2015a).

4.2. cRED in SAED versus NED mode

Traditionally, collection of electron diffraction data has been performed via diffraction area selection of a region of interest (ROI). However, the ROI selection can also be accomplished by adjusting the illumination settings. Almost parallel illumination with a sub-micrometre beam diameter can be obtained either by Köhler illumination (Wu et al., 2004; Meyer et al., 2006; Benner et al., 2011) or by inserting a small C2 condenser aperture (Kolb et al., 2007; Dwyer et al., 2007). NED provides full control of the beam diameter and in principle allows collection of data on a smaller area than SAED (Gemmi et al., 2019). However, in the literature there is a lack of direct comparison of data quality collected on the same sample by cRED in SAED and NED modes. Here, an attempt has been made to reveal the difference between these modes using the same area of the sample for collecting diffraction data.

In SAED mode, a diffraction field of about 750 nm was selected by inserting an SA. In NED mode, the beam was condensed to illuminate the 750 nm area, and the electron dose rate was kept equal to that in SAED mode by adjusting the monochromator focus. The two resulting data sets registered on the same isolated crystal are presented in Table 1.

Based on the previous crystallographic reports on the ZSM-5 single-crystal X-ray diffraction (SCXRD) structure (Olson et al., 1981; van Koningsveld et al., 1987), the lattice parameters a = 20.022 Å, b = 19.899 Å, c = 13.383 Å and the space group Pnma (No. 62) were used as input for XDS. Both SAED and NED data sets fit well with the expected orthorhombic structure and the refined unit-cell parameters are close to the published values within the accuracy of the 3DED method. Fig. 3 shows the reconstructed reciprocal lattice of ZSM-5 based on the cRED data collected in SAED mode from Table 1.

Figure 3 (a)–(c) Typical 3D reciprocal lattices of ZSM-5 reconstructed and visualized by REDp (Wan et al., 2013). The corresponding crystal image is shown as an inset in (b). (d) A wire-and-stick ZSM-5 crystal structure representation.

Among factors affecting the cRED data quality, electron dose has the utmost importance. Our experiments have shown that the optimal electron dose rate range for ZSM-5 data acquisition is approximately between 0.03 and 0.10 e Å⁻² s⁻¹ (Fig. 4). In the optimal range with no saturation, the higher the dose the better the I/σ. Excessive electron dose (>0.20 e Å⁻² s⁻¹) causes read-out biases of the OneView camera, whereas a low electron dose rate (<0.03 e Å⁻² s⁻¹) leads to significant deterioration of the signal-to-noise ratio and, as a consequence, to poor data statistics. Examples of the raw SAED/NED diffraction patterns collected at different electron doses are shown in Fig. S1. Since XDS relies upon the lowest measured intensities to guide subtraction of the background, the scaling of the Bragg intensities as a function of resolution shells unavoidably leads to significant deterioration of weak but still useful high-resolution signals, and consequently to higher R values in the high resolution shells (1.00–0.80 Å). For X-ray diffraction a common practice would be to truncate the data at the resolution at which the data still show correlations (indicated by * on the CC_1/2 value) (Karplus & Diederichs, 2015). However, for electron diffraction, we found that including data out to a CC_1/2 value of ∼70% leads to an improvement in the refined model.

Figure 4 Effect of the electron dose rate on CC1/2 in different resolution shells. cRED data were collected with ROI selection either by selected-area aperture (SAED) or by nanoprobe illumination (NED). The ROI diameter was 750 nm. All SAED data were collected from the same ZSM-5 crystal sequentially, in ascending order of the electron dose rate. All NED data were collected from a second crystal following the same procedure. The lines in the figure are guides for the eye.

Another important factor for data collection is the stability of the CompuStage, since currently InsteaDMatic does not provide an opportunity to track the continuously rotating crystal during data collection. We have shown that the specimen movement controlled by the CompuStage controller is smooth and the crystal does not move out of the beam, even without its position being realigned during cRED data acquisition. In a typical experiment, we observed a total drift of only a few nanometres for a 100 × 100 nm crystal rotated from −40 to 40°, accompanied by a jump of ∼50 nm at the beginning of the rotation (see Movie S2). It should be noted that crystal drift becomes more severe at high tilt angles and ED intensities are also commonly systematically disrupted by uncompensated Z height changes. It is therefore a better strategy to merge a number of data sets collected from different crystals in the range from −40 to 40°, instead of collecting cRED data from one or two crystals with a large tilt range, e.g. from −70 to 70°.

For the structure solution, five individual cRED data sets collected from different crystals in SAED mode were merged, chosen by performing hierarchical cluster analysis based on 15 data sets using an in-house-developed program called edtools (https://github.com/stefsmeets/edtools). For the NED data, six out of ten input cRED data sets were chosen for merging. Hierarchical cluster analysis helps to find structurally similar data with high correlation coefficients between scaled diffraction intensities and to reach high completeness by merging only few data sets (Wang et al., 2019). However, it is worth noting that simple averaging of unit-cell parameters obtained from individual 3DED data sets may result in irrelevant interatomic distances in the final structure. Hence the unit-cell parameters of standard ZSM-5 (as-made ZSM-5, determined by SCXRD; van Koningsveld et al., 1987) were used for the structure solution and refinement: see the International Zeolite Association Database of Zeolite Structures (https://www.iza-structure.org/databases/).

The structure of ZSM-5 can be solved using either SIR2014 direct-space (Burla et al., 2015) or SHELXT dual-space methods (Sheldrick, 2008). We note that a minimal I/σ signal-to-noise ratio of ca 2 (at 1.0 Å resolution limit) is required for revealing the framework of ZSM-5 by means of direct methods, whereas dual-space methods are not so sensitive to the I/σ ratio. There are 38 symmetry-independent atoms in the ZSM-5 structure, of which 12 are Si atoms and 26 are O atoms. There are four O atoms located at special positions. The atomic positions of all 12 Si and 26 O atoms were found successfully using both NED and SAED data, and used as an initial structural model. The details of the structure refinement are provided in Table 2. Anisotropic refinement of the NED model leads to R₁ = 0.1758, goodness of fit (GoF) = 1.609, Si—O bond lengths in the range 1.555–1.635 Å and O—Si—O angles in the range 105.5–112.7°, with no additional restraints applied. For the SAED data, the refinement converged with R₁ = 0.1992, GoF = 1.584, Si—O bond lengths in the range 1.551–1.635 Å and O—Si—O angles in the range 105.6–116.0°. Two restraints were applied to keep the Si—O bond lengths reasonable. In full agreement with the SCXRD model (Olson et al., 1981; van Koningsveld et al., 1987), the framework structure of ZSM-5 obtained from cRED data has a three-dimensional channel system with ten-ring straight channels of 5.4 × 5.6 Å in diameter running parallel to [010] and ten-ring sinusoidal channels of 5.1 × 5.4 Å in diameter running parallel to [100], as shown in Fig. 5.

Figure 5 The framework structure of ZSM-5 viewed along the b axis as refined using (a) NED and (b) SAED data, showing anisotropic atomic displacement parameters for Si (yellow) and O (red) atoms. Displacement ellipsoids are drawn at the 50% probability level.

A comparison with the reference model obtained from SCXRD (van Koningsveld et al., 1987) was carried out using the COMPSTRU program (de la Flor et al., 2016). All deviations of atomic positions between the reference ZSM-5 structure and those determined from cRED data are listed in Table S6. The deviations for the model obtained from the merged NED data set are on average 0.03 (1) Å for Si and 0.05 (2) Å for O, while those for the model obtained from the merged SAED data set are 0.05 (1) Å for Si and 0.07 (3) Å for O. This shows that cRED data collected using both NED and SAED provide reliable structural models. The NED data have a higher number of reflections with I > 2σ(I) (3903) than the SAED data (2854) (Table 2), which gives a slightly better structural model. The accuracy of the models is comparable to that obtained from our previous studies using single cRED data sets collected in SAED mode on a JEM-2100 LaB₆ microscope equipped with a Timepix quad hybrid pixel detector (Wang, Yang et al., 2018).

In contrast with the SAED mode, where the ROI to be used for data collection is pre-defined by the selected-area aperture size, the NED mode provides higher flexibility in adjusting the size of the area to be illuminated, and hence in fitting the size of each individual crystal so that the background in the ED frames is largely eliminated. This may be highly beneficial for studies of beam-sensitive materials since it paves the way for tailoring of the electron dose received by a specimen in a controllable manner.