Compound-specific memory demand determines when a research group and a computing centre benefit from sharing information in two-dimensional semiconductor screening
- 1Department of Materials Science and Engineering
- 2University of Utah
Abstract
Screening two-dimensional semiconductors with density functional theory (DFT), the standard electronic-structure method, involves two parties who know each other only qualitatively: a research group holding the candidate list and a computing centre that knows its nodes. Sharing information helps either party only if a calculation’s demand on the node depends on the compound. Thirty transition-metal dichalcogenide layers (MX2; M = Ti, Zr, Hf, Mo, W; X = S, Se, Te) in 1T and 2H structures were screened with Quantum ESPRESSO at the PBE level, recording each calculation’s duration and memory, then supercells up to 3328 electrons and preregistered paired comparisons of no sharing against a job-placing broker. Nine layers have PBE gaps in the 1.0–2.0 eV window of transistor channels and photodetectors, seven after spin–orbit coupling; single-layer 2H cells keep the same nine, and chalcogen-vacancy hop barriers follow the gap, also in Janus layers. In primitive cells and 3×3 supercells, demand followed calculation stage and cell size, which the centre already knows, and no arrangement raised throughput. At 8×8 supercells, compounds of equal electron count differed in memory by 23–32% near node capacity; one disclosed memory-class bit per job raised the centre’s throughput 1.133-fold under memory-reservation charging, but not on single-layer pools, whose memory classes nearly coincide. The group gained no resolvable advantage and job timings revealed the compound, so the hypothesis that the broker benefits both parties was rejected. A gap screen needs batching rather than disclosure; defect and heterostructure supercells justify disclosing one memory class per job.
Related
- Pressing Mode Determines Densification Uniformity and Dimensional Stability in Sulfide Solid-State Batteries
- Metasurface information processor based on 4D bionic hardware and software for ultralow energy consumption computing
- Adsorption regime changes the information needed for data-driven metal–organic framework screening
- Self-Powered Photodetection and Optical Information Recognition in a Mixed-Dimensional WSe₂/CdSe Heterojunction.
- Deciphering the Material Basis of an Herbal Formula via Integrated Two-dimensional Chromatography, In Silico Screening, and Comparative Pharmacokinetics: A Case Study of Kaiyu Jiangzhuo Formula
BibTeX
@misc{ackolf2026compoundspecific,
title = {{Compound-specific memory demand determines when a research group and a computing centre benefit from sharing information in two-dimensional semiconductor screening}},
author = {Kinston Ackölf and Taylor D. Sparks},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010664/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010664/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Compound-specific memory demand determines when a research group and a computing centre benefit from sharing information in two-dimensional semiconductor screening
AU - Kinston Ackölf
AU - Taylor D. Sparks
PY - 2026
DA - 2026/10/19
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010664/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010664/v1
AB - Screening two-dimensional semiconductors with density functional theory (DFT), the standard electronic-structure method, involves two parties who know each other only qualitatively: a research group holding the candidate list and a computing centre that knows its nodes. Sharing information helps either party only if a calculation’s demand on the node depends on the compound. Thirty transition-metal dichalcogenide layers (MX2; M = Ti, Zr, Hf, Mo, W; X = S, Se, Te) in 1T and 2H structures were screened with Quantum ESPRESSO at the PBE level, recording each calculation’s duration and memory, then supercells up to 3328 electrons and preregistered paired comparisons of no sharing against a job-placing broker. Nine layers have PBE gaps in the 1.0–2.0 eV window of transistor channels and photodetectors, seven after spin–orbit coupling; single-layer 2H cells keep the same nine, and chalcogen-vacancy hop barriers follow the gap, also in Janus layers. In primitive cells and 3×3 supercells, demand followed calculation stage and cell size, which the centre already knows, and no arrangement raised throughput. At 8×8 supercells, compounds of equal electron count differed in memory by 23–32% near node capacity; one disclosed memory-class bit per job raised the centre’s throughput 1.133-fold under memory-reservation charging, but not on single-layer pools, whose memory classes nearly coincide. The group gained no resolvable advantage and job timings revealed the compound, so the hypothesis that the broker benefits both parties was rejected. A gap screen needs batching rather than disclosure; defect and heterostructure supercells justify disclosing one memory class per job.
ER -