Physics
Deep microscopy physics
Scattering theory, simulation, tomography, diffraction, and quantitative microscopy — the physical foundations, not just image processing.
About NeuralSoftX
End-to-end deep-learning workflows for industrial electron microscopy — from physics-based synthetic data to deployed inference.
NeuralSoftX delivers the complete pipeline: multislice-simulated training data with a full detector forward model; neural networks trained under user-specified physical and operational constraints (dose budget, acquisition regime, detector type, sample class, noise envelope); and deployable inference — ONNX-packaged or pipeline-integrated — that runs on the client's instrument or review station. The company takes on problems where commercial instrument software, off-the-shelf AI models, and classical image-processing routines stop performing reliably: low-dose and fast-scan acquisitions, drifting or beam-sensitive specimens, scarce labelled data, CBED and 4D-STEM preprocessing, patterned-wafer defect inspection, and instrument-specific acquisition regimes. The methods combine electron–matter physics, multislice simulation, classical image processing, and deep learning.
Founder
Dr Iván Lobato
Physicist, computational microscopist, and developer of the MULTEM simulation platform.
Credentials
Credentials
PhD in Physics from the University of Antwerp. 30+ peer-reviewed publications in Science, Nano Letters, PNAS, Physical Review Letters, Ultramicroscopy, and Acta Crystallographica. 1,800+ citations, h-index 21. Cosslett Award for Best Invited Paper at Microscopy & Microanalysis 2018.
Experience
Ten years in quantitative EM
Senior Researcher at the University of Antwerp (2014–2022) and Associate Investigator at the Rosalind Franklin Institute (2022–2025) — positions focused on quantitative electron microscopy, tomography, ptychography, deep learning for EM, and GPU computing.
Commercial present
Commercial present
Dr Lobato runs NeuralSoftX alongside a commercial SEM role at Semplor. The combination keeps NeuralSoftX engagements focused on a small number of industrial projects where the work requires deep microscopy physics — not generic ML consultancy. Engagements are scoped milestone-by-milestone.
Why it exists
Industrial labs have difficult microscopy data — but often no path from raw acquisition to a robust computational workflow.
Commercial instrument software covers the straightforward cases. Academic research delivers breakthrough methods but rarely packages them for industrial use. The space in between — methods that match the instrument, the sample, and the business constraint, and that actually ship as deployable software — is where NeuralSoftX operates.
Methodology
Hybrid methods
Classical algorithms where they are reliable. Machine learning where it creates genuine advantage. Combinations when the real workflow needs both.
Delivery
Deployable software
Methods packaged as ONNX inference tools, on-prem workflows, or pipeline integrations — not as notebooks the client has to rebuild.
Synthetic data
Simulation as leverage
MULTEM and related simulation work make model development possible even when industrial labelled data is scarce.
Track record
Research depth that supports commercial delivery.
2025 → Present
NeuralSoftX CommV · Semplor
Commercial microscopy work focused on deployable computational methods and industrial SEM-related workflows.
2022 → 2025
Associate Investigator — Rosalind Franklin Institute
Computational methods for quantitative electron microscopy, ptychography, diffraction workflows, and advanced reconstruction problems.
2014 → 2022
Senior Researcher — University of Antwerp
Deep learning for EM, tomography, simulation, GPU computing, and development of MULTEM as an open-source platform.
2010 → 2014
PhD in Physics — University of Antwerp
Scattering theory, parameterisation work, and the foundations of the MULTEM simulation software stack.
Broader industrial delivery
Production ML track record outside electron microscopy.
In parallel with microscopy work, NeuralSoftX collaborates with Liquisens on industrial ML workflows — water-quality prediction (chemical oxygen demand, legionella detection) and process-monitoring regression and embeddings for production-line data. Client names remain under NDA. This experience grounds NeuralSoftX's commercial execution credibility without diluting its microscopy-first positioning.
Contact