Source author record

Jan Seidel

Jan Seidel appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

cond-mat.mtrl-sci Artificial Intelligence cond-mat.mes-hall cond-mat.str-el Distributed, Parallel, and Cluster Computing Networking and Internet Architecture physics.app-ph

Catalog footprint

What is connected

4works

7topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

Collective Communication for 100k+ GPUs

The increasing scale of large language models (LLMs) necessitates highly efficient collective communication frameworks, particularly as training workloads extend to hundreds of thousands of GPUs. Traditional communication methods face significant throughput and latency limitations at this scale, hindering both the development and deployment of state-of-the-art models. This paper presents the NCCLX collective communication framework, developed at Meta, engineered to optimize performance across the full LLM lifecycle, from the synchronous demands of large-scale training to the low-latency requirements of inference. The framework is designed to support complex workloads on clusters exceeding 100,000 GPUs, ensuring reliable, high-throughput, and low-latency data exchange. Empirical evaluation on the Llama4 model demonstrates substantial improvements in communication efficiency. This research contributes a robust solution for enabling the next generation of LLMs to operate at unprecedented scales.

preprint2021arXiv

Optical Tuning of Resistance Switching in Polycrystalline Gallium Phosphide Thin Films

The nanoscale resistive switching characteristics of gallium phosphide (GaP) thin films directly grown on Si are investigated as a function of incident light. Firstly, as-grown GaP films show a high RON/ROFF (~10^4), shown to arise from the formation of conductive channels along the grain boundaries. It is proposed that point defects (most likely Ga interstitials) and structural disorder at the grain boundaries provide the ideal environment to enable the filamentary switching process. Next, we explored if such defects can give rise to mid-gap states, and if so could they be activated by photonic excitation. Both first-principles calculations as well as UV-vis and photoluminescence spectroscopy strongly point to the possibility of mid-gap electronic states in the polycrystalline GaP film. Photoconductive atomic force microscopy (phAFM), a scanning probe technique, is used to image photocurrents generated as a function of incident photon energy (ranging from sub band-gap to above band-gap) on the GaP film surface. We observe photocurrents even for incident photon energies lower than the band-gap, consistent with the presence of mid-gap electronic states. Moreover the photocurrent magnitude is found to be directly proportional to the incident photon energy with a concomitant decrease in the filament resistance. This demonstrates GaP directly integrated on Si can be a promising photonic resistive switching materials system.

preprint2013arXiv

A Nanoscale Shape Memory Oxide

Stimulus-responsive shape memory materials have attracted tremendous research interests recently, with much effort focused on improving their mechanical actuation. Driven by the needs of nanoelectromechnical devices, materials with large mechanical strain particularly at nanoscale are therefore desired. Here we report on the discovery of a large shape memory effect in BiFeO3 at the nanoscale. A maximum strain of up to ~14% and a large volumetric work density can be achieved in association with a martensitic-like phase transformation. With a single step, control of the phase transformation by thermal activation or electric field has been reversibly achieved without the assistance of external recovery stress. Although aspects such as hysteresis, micro-cracking etc. have to be taken into consideration for real devices, the large shape memory effect in this oxide surpasses most alloys and therefore demonstrates itself as an extraordinary material for potential use in state-of-art nano-systems.

preprint2011arXiv

Anisotropic conductance at improper ferroelectric domain walls

Transition metal oxides hold great potential for the development of new device paradigms because of the field-tunable functionalities driven by their strong electronic correlations, combined with their earth abundance and environmental friendliness. Recently, the interfaces between transition-metal oxides have revealed striking phenomena such as insulator-metal transitions, magnetism, magnetoresistance, and superconductivity. Such oxide interfaces are usually produced by sophisticated layer-by-layer growth techniques, which can yield high quality, epitaxial interfaces with almost monolayer control of atomic positions. The resulting interfaces, however, are fixed in space by the arrangement of the atoms. Here we demonstrate a route to overcoming this geometric limitation. We show that the electrical conductance at the interfacial ferroelectric domain walls in hexagonal ErMnO3 is a continuous function of the domain wall orientation, with a range of an order of magnitude. We explain the observed behaviour using first-principles density functional and phenomenological theories, and relate it to the unexpected stability of head-to-head and tail-to-tail domain walls in ErMnO3 and related hexagonal manganites. Since the domain wall orientation in ferroelectrics is tunable using modest external electric fields, our finding opens a degree of freedom that is not accessible to spatially fixed interfaces.