What is 4D STEM and How Does It Work?

Four-dimensional scanning transmission electron microscopy, or 4D-STEM, is a technique that records a full two-dimensional diffraction pattern at every pixel in a two-dimensional scan of a sample, producing a four-dimensional dataset that captures both where the electron beam hits and how the material scatters it.¹ That richness is the whole point. Instead of collapsing scattered electrons into a single brightness value the way a conventional detector does, 4D-STEM preserves the full scattering distribution, and researchers can then mine that information for crystal orientation, strain, electric fields, charge density, and more, all from a single experiment.

How a Conventional STEM Image Compares

In a standard STEM experiment, a finely focused electron beam is scanned across a thin specimen point by point, and a detector underneath collects whatever electrons pass through. Different detector geometries pick up different signals: a bright-field detector gathers electrons that travel nearly straight through, while a high-angle annular dark-field detector catches electrons scattered at steep angles. Each detector integrates all the electrons that land on it into one number per scan position. The result is a two-dimensional map of intensities, essentially a grayscale image of the sample.

The limitation is that once those electrons are summed into a single intensity value, the angular information is gone. You know how bright the signal was at each point, but not exactly which directions the electrons scattered. That angular distribution is where most of the structural detail lives: the angles tell you about crystal lattice spacings, local strain, and internal electric fields. A conventional STEM image throws all that away by design.

Where the Extra Two Dimensions Come From

4D-STEM keeps the angular information. At each of the many scan positions (the first two dimensions, x and y on the specimen), a fast pixelated detector records the entire two-dimensional diffraction pattern (the other two dimensions, kx and ky in reciprocal space). The name “4D-STEM” simply refers to this: 2D real-space scan plus 2D diffraction pattern at each point.² The raw output is a four-dimensional array: for a scan grid of, say, 264 by 264 positions, each with a diffraction image that might be hundreds of pixels on a side, the total dataset grows fast.

This became practical only because of two parallel developments: fast direct electron detectors that can capture thousands of diffraction frames per second, and enough computational horsepower to store and process the resulting data.¹ Early experiments could produce datasets of hundreds of gigabytes in under three minutes.² Without detectors that could keep pace and disks that could swallow the data, the technique would have stayed a curiosity.

What You Can Extract from the Dataset

The beauty of 4D-STEM is that a single acquisition feeds many different analyses, because each diffraction pattern encodes multiple kinds of structural information. After the experiment, you decide what to look for. Researchers routinely extract crystal orientation maps, strain fields, electric-field distributions, and phase-contrast images, all from the same raw dataset. Some of the most common analyses deserve their own explanation.

Crystal Orientation Mapping

One of the most straightforward uses of 4D-STEM is figuring out how the crystal grains in a material are oriented. At each scan position, the diffraction pattern shows bright spots whose arrangement depends on the local lattice orientation. By comparing each experimental pattern against a library of simulated patterns for all possible orientations, the software assigns an orientation to every pixel. The result is a color-coded map showing every grain in the field of view and how it is rotated relative to its neighbors.

Two major open-source toolkits handle this: py4DSTEM, which uses sparse correlation matching to index orientation from diffraction patterns,³ and Pyxem, which performs GPU-accelerated template matching for speed.⁴ Both are freely available and built on the scientific Python ecosystem, which has lowered the barrier for labs that want to try 4D-STEM without proprietary software.

Strain Mapping

Tiny distortions of the crystal lattice, collectively called strain, shift the positions of the diffraction spots. By tracking those shifts with sub-pixel precision across every scan point, researchers build maps of how the lattice is stretched or compressed throughout a material. One powerful approach uses the exit-wave power-cepstrum transform, which has been shown to achieve sub-picometer precision in lattice spacing and sub-nanometer spatial resolution, even when the sample is slightly tilted or randomly oriented.⁵ That level of sensitivity is enough to map ferroelectric domains in thin films or strain profiles inside individual nanoparticles used in fuel-cell catalysts.

For two-dimensional materials like graphene, the problem gets trickier because the sample can ripple and bend, so what looks like strain might actually be a local tilt. A method based on nanobeam electron diffraction at two specimen tilts can disentangle genuine in-plane strain from the sample’s slope, giving honest strain values with nanometer resolution.⁶

Electric Field and Charge Density Mapping

When the electron beam passes through a region with an internal electric field, the whole diffraction pattern shifts sideways. By computing the center of mass of each diffraction pattern, researchers can measure the momentum transfer the beam experienced and work backward to the electric field that caused it.⁷ This center-of-mass technique has been used to map polarization-induced electric fields at interfaces in semiconductor heterostructures, although the measured field values drop as the sample gets thicker because the beam averages over more material.

The same center-of-mass approach feeds charge-density mapping, but it has real limitations. It only recovers the total projected charge density and works best on extremely thin samples that behave as “phase objects,” meaning the beam’s amplitude does not change much as it passes through.⁸ More advanced analysis methods are being developed to push past those constraints.

Phase Imaging and Ptychography

Electrons passing through a thin sample pick up a phase shift, which encodes information about the material’s electrostatic potential. Conventional detectors cannot measure phase directly, but because 4D-STEM records the full diffraction pattern, computational techniques can reconstruct it. The most powerful of these is electron ptychography, an algorithm that stitches together overlapping diffraction patterns from neighboring probe positions to recover both the amplitude and phase of the transmitted wave.

Ptychography has become one of the headline achievements of 4D-STEM. Researchers have demonstrated spatial resolution down to 0.44 angstroms in an uncorrected STEM, beating the conventional resolution of aberration-corrected instruments and matching their best ptychographic results, using nothing more than a standard commercial microscope and a pixelated detector.⁹ That result is striking because aberration-corrected microscopes cost substantially more and are less widely available. The catch is that the best ptychographic resolution has historically required very thin samples. Deep-sub-angstrom resolution through thicker specimens is an active area of research.¹⁰

Simpler phase-contrast methods also live under the 4D-STEM umbrella. Integrated differential phase contrast, or iDPC, uses a segmented detector (typically four quadrants) to approximate the phase gradient. By correcting for image shifts between the quadrant segments, which arise from defocus and other beam aberrations, iDPC can sharpen resolution in a way that resembles contrast-transfer-function correction in conventional TEM, but with a more straightforward implementation.¹¹

Imaging Beam-Sensitive and Soft Materials

Many materials that scientists care most about, including polymers, organic semiconductors, metal-organic frameworks, and biological tissue, are easily destroyed by the electron beam. This makes conventional high-resolution TEM impractical for them. 4D-STEM offers a way around the problem because the total electron dose can be tightly controlled: the beam blanker limits exposure, and the convergent beam spreads the dose over a very small area at each scan point.¹² Instead of flooding the sample with electrons to form a high-resolution image, 4D-STEM collects diffraction data at each point with just enough electrons to see the pattern, then moves on.

This low-dose flexibility has opened up studies of materials that range from highly crystalline to completely amorphous. For amorphous organic thin films, for example, pair distribution function analysis of the diffraction data can reveal short- and medium-range order that would be invisible in a normal image.¹³ You are not seeing atoms individually, but you are learning how far apart they tend to sit and whether there is any local pattern to their arrangement.

Biological Specimens and Cryo-STEM

The dose sensitivity problem is at its most extreme in structural biology. Frozen biological specimens can tolerate only a tiny number of electrons per unit area before their structure is wrecked. Traditional cryo-electron microscopy works around this by averaging images of thousands of identical particles, but 4D-STEM phase-contrast techniques offer a complementary route: they squeeze more information per electron out of each exposure.

An ERC-funded project called 4D-BioSTEM is working to develop methodologies that combine ultrafast detectors, differential phase contrast, and ptychographic reconstruction to push cryo-4D-STEM toward imaging proteins smaller than about 50 kilodaltons, a size range where current methods struggle.¹⁴ Several phase-imaging approaches, including tilt-corrected bright-field STEM, have already been demonstrated on vitrified biological specimens.¹⁵ This is still early-stage work, but the promise is that 4D-STEM could make cryo-EM applicable to a wider range of biological targets.

Watching Processes Happen in Real Time

Because 4D-STEM datasets can be collected quickly enough, the technique pairs naturally with in-situ experiments, where the sample is heated, cooled, strained, or exposed to gases while inside the microscope. One recent study combined in-situ heating with 4D-STEM to watch halide perovskite nanowires transition from a non-perovskite phase to the perovskite phase in real time, tracking the phase front as it propagated along individual wires with nanometer spatial and millisecond temporal resolution.¹⁶

Acquisition speed is the key bottleneck here. An automated “5D-STEM” data acquisition system has reduced the time per full 4D-STEM dataset (264 by 264 scan positions) from about ten seconds to less than one second, making it feasible to collect many snapshots during a single in-situ experiment.¹⁷ Combine that speed with tomographic tilt series and you start approaching time-resolved three-dimensional structural movies of materials as they transform.

Exotic Order in Quantum Materials

4D-STEM has also become a go-to tool for mapping unusual nanoscale textures in functional materials. Polar skyrmions, tiny swirling structures in which the electric polarization direction winds around like a vortex, are one example. By analyzing scanning electron diffraction data at sub-nanometer resolution, researchers have simultaneously mapped the local polarization, strain state, and chirality (handedness) of individual skyrmions as they transitioned to a related structure called a meron, showing that strain is a crucial driving force behind the switch.¹⁸ That kind of multidimensional mapping, getting the polarization, strain, and topology from one dataset, is something no other single technique delivers as cleanly.

The Data Problem

All that richness comes at a cost: 4D-STEM datasets are enormous. A single scan can easily run into multiple gigabytes, and high-throughput or time-resolved experiments push into the hundreds of gigabytes.¹⁹ Storing, transferring, and interactively visualizing this data is a genuine infrastructure challenge, especially for labs that want to compare many experiments or archive results long-term. Active research is exploring lossless compression strategies that reduce file sizes while preserving everything needed for downstream analysis.

On the analysis side, machine learning is increasingly stepping in to handle the scale. Unsupervised methods like non-negative matrix factorization can decompose a dataset of thousands of diffraction patterns into a small number of representative components, each associated with a spatial map showing where that component dominates. Combined with hierarchical clustering, this approach can automatically identify amorphous regions, different crystalline grains, and everything in between, without the researcher having to specify in advance what to look for.²⁰ A separate machine-learning framework uses divisive hierarchical clustering to uncover multi-scale deformations across an entire sample flake, again without prior knowledge of the material’s structure.²¹

Practical Pitfalls and Artifacts

4D-STEM is powerful, but the data it produces is not self-interpreting, and several artifacts can mislead the unwary. The most persistent is dynamical diffraction: electrons bouncing multiple times through a crystalline specimen before exiting. Dynamical diffraction changes the intensity distribution in the diffraction pattern in ways that depend on sample thickness and local tilt. In differential phase contrast imaging, this can create contrast that mimics the signal from a long-range electric field, making it hard to tell a real field from a thickness-driven artifact.²² Precessing the electron beam, tilting it in a cone during acquisition, can suppress some of these artifacts, but adds complexity to the experiment.

Sample thickness matters in other ways too. The center-of-mass method for electric field mapping gives weaker signals as samples get thicker, because the beam averages over more material along its path.⁷ And orientation mapping by template matching becomes less accurate in thick samples because dynamical scattering changes the relative intensities of diffraction spots, making them harder to match against simulated patterns calculated for ideal thin crystals. Adjusting the weighting given to peak intensities during matching can help compensate.³

None of these are showstoppers, but they do mean that 4D-STEM analysis requires careful attention to sample preparation, acquisition parameters, and the assumptions built into whatever analysis algorithm you choose. The technique gives you more data per experiment than any conventional STEM mode, which is both its greatest strength and its greatest demand on the person interpreting the results.

How 4D-STEM Fits Alongside Other Characterization Techniques

4D-STEM does not replace every other microscopy mode. It complements them. For chemical composition, energy-dispersive X-ray spectroscopy and electron energy-loss spectroscopy remain the standard tools, and both can be collected simultaneously with a 4D-STEM scan in modern microscopes. Some instruments now even integrate atom probe tomography capabilities alongside STEM, where the 4D-STEM dataset provides crystallographic and strain context that atom probe’s chemical maps lack on their own.²³

For researchers choosing between techniques, the decision usually comes down to what question they need answered. If you want a quick image of atomic structure at a known zone axis, conventional high-resolution STEM is faster and simpler. If you want to map grain orientations across a large area, electron backscatter diffraction in a scanning electron microscope covers more ground with less sample preparation. But if you need simultaneous access to strain, orientation, phase contrast, and field mapping at nanometer scale, and especially if your sample is beam-sensitive or disordered, 4D-STEM is increasingly the tool of choice.

1
PubMed. Data-efficient 4D-STEM in SEM: Beyond 2D materials to metallic materials
2
Oxford Academic. Four-Dimensional Scanning Transmission Electron Microscopy (4D-STEM): From Scanning Nanodiffraction to Ptychography and Beyond
3
Microscopy and Microanalysis. Automated Crystal Orientation Mapping in py4DSTEM using Sparse Correlation Matching
4
PubMed. Free, flexible and fast: Orientation mapping using the multi-core and GPU-accelerated template matching capabilities in the Python-based open source 4D-STEM analysis toolbox Pyxem
5
PubMed. The exit-wave power-cepstrum transform for scanning nanobeam electron diffraction: robust strain mapping at subnanometer resolution and subpicometer precision
6
PubMed. Quantitative Strain and Topography Mapping of 2D Materials Using Nanobeam Electron Diffraction
7
PubMed. 4D-STEM at interfaces to GaN: Centre-of-mass approach & NBED-disc detection
8
PubMed. Mapping valence electron distributions with multipole density formalism using 4D-STEM
9
PubMed. Achieving sub-0.5-angstrom-resolution ptychography in an uncorrected electron microscope
10
PubMed Central. Imaging thick objects with deep-subangstrom resolution and deep-subpicometer precision
11
Microscopy and Microanalysis. Optimizing Contrast in Automated 4D STEM Cryotomography
12
Accounts of Chemical Research. 4D-STEM of Beam-Sensitive Materials
13
Microscopy. Mapping structure and morphology of amorphous organic thin films by 4D-STEM pair distribution function analysis
14
CORDIS | European Commission. 4D scanning transmission electron microscopy for structural biology
15
Biophysical Journal. Direct Phase-Contrast Imaging in Cryogenic Scanning Transmission Electron Microscopy
16
PubMed. Observing Phase Transition Dynamics in Halide Perovskite Nanowires via in Situ 4D-STEM
17
PubMed Central. In-situ heating-and-electron tomography for materials research: from 3D (in-situ 2D) to 4D (in-situ 3D)
18
Nature Communications. Emergent chirality in a polar meron to skyrmion phase transition
19
arXiv. Inference-Sufficient Representations for High-Throughput Measurement: Lessons from Lossless Compression Benchmarks in 4D-STEM
20
PubMed Central. Unsupervised machine learning combined with 4D scanning transmission electron microscopy for bimodal nanostructural analysis
21
npj Computational Materials. Uncovering material deformations via machine learning combined with four-dimensional scanning transmission electron microscopy
22
PubMed. Suppressing dynamical diffraction artefacts in differential phase contrast scanning transmission electron microscopy of long-range electromagnetic fields via precession
23
Nature Communications. Bringing atom probe tomography to transmission electron microscopes