Postdoc in Machine Learned Semiconductor Material Properties Quantum Transport Simulations

Organisation/Company ETH Zürich Research Field Computer science » Other Physics » Computational physics Physics » Other Researcher Profile Recognised Researcher (R2) Final date to receive applications 21 Apr 2026 - 21:59 (UTC) Country Switzerland Type of Contract Temporary Job Status Full-time Is the job funded through the EU Research Framework Programme? Not funded by a EU programme Is the Job related to staff position within a Research Infrastructure?

No

Postdoc in Machine Learned Semiconductor Material Properties for Quantum Transport Simulations

The simulation of electronic devices has a long and successful history of accompanying experimental developments, be it for transistors or memory cells. Nowadays, to be of practical relevance, such technology computer aided design (TCAD) tools should operate at the ab‑initio and quantum mechanical level. Moreover, they should capture the interplay between electrical (voltage‑induced currents), thermal (excitation of crystal vibrations), and structural (migration of atoms) effects with an atomistic resolution.

This can be achieved by self‑consistently coupling molecular dynamics (MD), density‑functional theory (DFT), and quantum transport (QT) simulations of both electrons and phonons.

The Computational Nanoelectronics Group of ETH Zurich recently started implementing a novel, state‑of‑the‑art TCAD tool called Qua Tr Ex  that can perform ab‑initio QT calculations at unprecedented scale. As Qua Tr Ex  aims to solve for the transport and interactions of various quanta (electrons, phonons, etc) directly at atomic resolution, it requires ab‑initio material inputs corresponding to the simulated device components, such as the Hamiltonian and Dynamical matrices, electron‑phonon coupling elements, forces and energies, etc.

Computing these inputs for device‑scale structures, with methods such as DFT, currently poses a bottleneck in the application's capabilities.

The Computational Nanoelectronics Group was recently awarded a grant from the Swiss National Science Foundation entitled Machine Learning for Optimised Ab‑initio Quantum Transport Simulations (MALOQ). It officially started on January 1st 2026 and will conclude on December 31st 2029. The goal of this research effort is to apply machine learning (ML) techniques, in particular (equivariant) graph neural networks to accelerate the creation of all physical quantities that enter ab‑initio QT simulations of nanoelectronic devices.

In this context, we are seeking a post‑doctoral fellow who will be part of a team that also comprises two PhD students and will closely collaborate with the Qua Tr Ex  developers.

As part of the MALOQ project, you will train state‑of‑the‑art ML models to learn atomic, electronic, and vibrational properties of large‑scale atomic systems representing the building blocks of semiconductor devices. The aim is to predict these properties for arbitrarily large structures, at a DFT‑level of accuracy.

As a starting point, you will extend the large‑scale equivariant GNNs we develop for Hamiltonian matrix prediction to treat dynamical matrices. This ML framework will then also allow us to produce the derivatives of both quantities, which correspond to the electron‑phonon and an harmonic phonon‑phonon coupling elements. With these, dedicated scattering rates can be computed and then used in quantum transport simulations.

Down the line, we aim to pre‑train a common GNN backbone model capable of predicting electronic, structural, and thermal quantities while leveraging underlying symmetries for computational efficiency. There will be a significant computational component in deploying multi‑GPU codes to efficiently train on the large, densely‑connected and graph‑structured data encountered in our systems of interest.

Your contributions would be across the spectrum from methodological development, implementation, and application to realistic semiconductor device systems made of thousands of atoms. All codes will be made freely available to the scientific community through Git Hub.

• A track record in building and deploying ML models for applications in…