MEP Reconstruction Software

High-throughput, GPU-accelerated multislice electron ptychography for atomic-resolution imaging beyond the conventional depth and thickness limits.

3datomic — MEP Workstation v2.0
Phase Retrieval
Phase Map
MEP Output
Convergence
Atom Positions
Lattice Map
GPU Reconstruction Active — 3.2 min elapsed

GPU-Accelerated Engine

CUDA-optimized multislice propagation achieving 100× speedup over CPU baselines. Handle large-scale 4D-STEM datasets (512×512×64×64) in minutes, not hours.

A100 / H100 / RTX Pro 6000 Ready
🔬

Break Thickness Limits

Proprietary multi-slice algorithms reconstruct thick specimens (>100 nm) with atomic precision—unreachable by conventional SSB/Wigner distribution methods. Ideal for bulk interfaces and 3D-IC TSV structures.

3D-IC / GAA / CFET Compatible
🎯

Geometric Calibration

Built-in probe aberration correction (C1, C3, C5), scan distortion rectification, and tilt-axis refinement—ensuring quantitative accuracy for every reconstruction, with traceable calibration metadata in every output.

Quantitative & Traceable
🔧

Fab-Ready Integration

Seamless data pipeline from JEOL ARM / Thermo Fisher Spectra / Nion HERMES. Automated batch processing, metadata preservation, and LIMS-compatible JSON/XML reporting for FA labs.

LIMS-Compatible Reporting

Compatible TEMs & Detectors

MEP reconstruction requires pixelated detectors and aberration-corrected STEM platforms. Our software natively supports the leading 4D-STEM hardware ecosystem.

Je
JEOL
Japan

ARM Series

Cold-FEG double Cs-corrected STEM with sub-Å probe and highest stability for atomic-resolution 4D-STEM ptychography.

  • ARM200F / ARM300F / JEM-ARM200CF
  • Double Cs corrector (probe + image)
  • Cold FEG, info limit ~0.6 Å
  • High-voltage stability < 0.01 ppm/min
Fully Verified
TF
Thermo Fisher
USA / Netherlands

Spectra / Themis Series

X-FEG / X-CFEG monochromated platforms with X lens corrector. Excellent for low-dose ptychography and EELS-correlated 4D-STEM.

  • Spectra 300 / Themis Z / Themis G2
  • X-FEG or X-CFEG + monochromator
  • X lens probe corrector
  • EELS integration ready
Fully Verified
Ni
Nion
USA

HERMES / UltraSTEM

Ultra-high vacuum cold-field emission STEM with Nion-designed quadrupole-octupole correctors. Best-in-class probe coherence for ptychography.

  • HERMES-100 / UltraSTEM 100MC
  • Cold-field emission, 25 meV energy spread
  • Nion QO corrector (C5-capable)
  • UHV environment, minimal contamination
Fully Verified
Hi
Hitachi
Japan

HF-5000 / HD-2700 Series

Dedicated STEM platforms with probe aberration correction and high-tilt capability. Cost-effective entry point for 4D-STEM ptychography.

  • HF-5000 (200 kV, Cs-corrected)
  • HD-2700 (200 kV, dedicated STEM)
  • High-tilt tomography holder compatible
  • Robust automation for batch acquisition
Beta Support
Fe
FEI / Thermo
Legacy

Titan Themis / Krios G3i

Legacy but widely deployed platforms. With pixelated detector retrofit and probe corrector upgrade, capable of high-quality 4D-STEM ptychography.

  • Titan Themis (S)TEM with DCOR
  • Krios G3i (Cryo-EM, 4D-STEM capable)
  • X-FEG or Schottky emitter
  • Detector retrofit recommended
Retrofit Required
Cu
Custom
Open Hardware

Open-Source / Custom 4D-STEM

Custom-built 4D-STEM setups using open-source scan controllers (e.g., TEM Extensibility Interface) and home-built pixelated detectors. We provide SDK integration support.

  • TEM Extensibility Interface (TEI)
  • Custom scan coil drivers
  • Home-built Medipix/Timepix arrays
  • SDK & API integration available
SDK Integration
📷

EMPAD

Cornell / TechnoSoft

128×128 pixel array, 1M fps, 24-bit dynamic range. The gold standard for 4D-STEM ptychography.

128×128 px 1M fps 24-bit
🔲

Medipix3 / Timepix3

CERN / ASI / Quantum Detectors

Event-driven readout, zero noise, 55 µm pixel pitch. Timepix3 adds TOF capability for energy-resolved 4D-STEM.

256×256 px Event-driven Zero noise
📹

Merlin

Quantum Detectors

Medipix3 ASIC-based, 1.2M fps burst mode, radiation hard. Widely used in materials science 4D-STEM.

256×256 px 1.2M fps Rad-hard
🎬

4D Camera

Berkeley Lab / NCEM

512×512 direct electron detector, 4,000 fps, optimized for 4D-STEM diffraction pattern capture at high speed.

512×512 px 4K fps Direct
🔋

Gatan K3 IS / K2 IS

Gatan / Ametek

Direct detection electron counting. K3 IS supports in-situ 4D-STEM with high DQE and large field of view.

~4K×3K px Counting In-situ
🔭

Direct Electron DE-64

Direct Electron

Large-format direct detection, 64 MP, optimized for low-dose diffraction and large momentum transfer capture.

64 MP Low-dose Large FOV

pnCCD (S65)

PNDetector / HZDR

Split pn-junction CCD, 1,000 fps, radiation tolerant. Excellent for high-energy electron diffraction (MeV range).

264×264 px 1K fps MeV-ready
🌟

Stellaris

ASI / Amsterdam Scientific

Hybrid pixel detector with adaptive gain. Optimized for simultaneous EELS and 4D-STEM acquisition.

Hybrid Adaptive EELS-sync
🔷

ARINA

DECTRIS

Hybrid-pixel electron-counting detector purpose-built for 4D-STEM. Up to 120 kfps with noise-free readout and high dynamic range — ideal for ptychography with dwell times below 10 µs.

192×192 px 120 kfps 30–300 keV Retractable

Built for Atomic Precision

Our technology stack combines cutting-edge algorithms with high-performance computing to deliver results you can trust.

100×
GPU Speedup
vs CPU baseline
<1 pm
Position Precision
Atomic column tracking
>100 nm
Thickness Range
Multislice propagation
4D-STEM
Native Format
Ptychography ready

🧅 Multislice Reconstruction Engine

Our forward model treats the specimen as a stack of thin slices and explicitly propagates the probe through it, capturing multiple (dynamical) scattering instead of assuming it away. Mixed-state probe modes absorb partial coherence; joint position refinement corrects scan errors during iteration.

Why it matters: SSB / WDD / single-slice methods assume single scattering — they break down beyond 20–30 nm. Multislice keeps atomic precision past 100 nm and enables depth sectioning.

⚡ GPU-Native Compute Pipeline

Reconstruction is built GPU-first: batched FFTs, fused CUDA kernels, and mixed-precision arithmetic keep the propagation loop on-device end to end. Multi-GPU domain decomposition scales near-linearly for large fields of view.

Why it matters: a typical 256×256-scan dataset reconstructs in minutes on a single A100 — ~100× faster than CPU codes. Iterate on parameters the same day, not next week.

🎯 Quantitative Calibration Chain

Probe aberrations (C1 / C3 / C5), scan distortion, detector response, and specimen tilt are refined jointly with the reconstruction rather than assumed from nominal values — every output carries traceable calibration metadata.

Why it matters: this is what turns a pretty picture into measurement — sub-picometer column statistics, strain tensors, and polarization maps you can defend in a paper or an FA report.