arXiv subjects

Shiyang Wu

Publications and source records attributed to Shiyang Wu.

3 recordsLinked to original sources

arXiv2025

Nonmagnetic ground state of marcasite FeTe$_{2}$: The competition between crystal field splitting and on-site Coulomb repulsion

The magnetic ground states in crystalline systems are significant for both fundamental condensed matter physics and practical materials engineering. Marcasite FeTe$_{2}$, characterized as a small-gap semiconductor, exhibits anomalous magnetic behaviors in low-temperature experiments. In this study, first-principles density functional theory calculations combined with scanning tunneling microscopy/spectroscopy are employed to investigate the magnetic ground state of marcasite FeTe$_{2}$. It is revealed that the competition between crystal field splitting and on-site Coulomb repulsion plays the key role in the formation of localized magnetic moments in FeTe$_{2}$. The ground state of FeTe$_{2}$ bulk is confirmed to be nonmagnetic, while the magnetic responses of FeTe$_{2}$ observed at low temperature are suggested to be related to the magnetic Fe atoms on the crystal surfaces. Our work proposes a straightforward competing mechanism for determining ground-state magnetism of various localized-moment crystalline systems.

cond-mat.str-el↗

arXiv2024

ChipExpert: The Open-Source Integrated-Circuit-Design-Specific Large Language Model

The field of integrated circuit (IC) design is highly specialized, presenting significant barriers to entry and research and development challenges. Although large language models (LLMs) have achieved remarkable success in various domains, existing LLMs often fail to meet the specific needs of students, engineers, and researchers. Consequently, the potential of LLMs in the IC design domain remains largely unexplored. To address these issues, we introduce ChipExpert, the first open-source, instructional LLM specifically tailored for the IC design field. ChipExpert is trained on one of the current best open-source base model (Llama-3 8B). The entire training process encompasses several key stages, including data preparation, continue pre-training, instruction-guided supervised fine-tuning, preference alignment, and evaluation. In the data preparation stage, we construct multiple high-quality custom datasets through manual selection and data synthesis techniques. In the subsequent two stages, ChipExpert acquires a vast amount of IC design knowledge and learns how to respond to user queries professionally. ChipExpert also undergoes an alignment phase, using Direct Preference Optimization, to achieve a high standard of ethical performance. Finally, to mitigate the hallucinations of ChipExpert, we have developed a Retrieval-Augmented Generation (RAG) system, based on the IC design knowledge base. We also released the first IC design benchmark ChipICD-Bench, to evaluate the capabilities of LLMs across multiple IC design sub-domains. Through comprehensive experiments conducted on this benchmark, ChipExpert demonstrated a high level of expertise in IC design knowledge Question-and-Answer tasks.

cs.AR↗

arXiv2024

Piecewise Analysis of Probabilistic Programs via $k$-Induction

In probabilistic program analysis, quantitative analysis aims at deriving tight numerical bounds for probabilistic properties such as expectation and assertion probability. Most previous works consider numerical bounds over the whole program state space monolithically and do not consider piecewise bounds. Not surprisingly, monolithic bounds are either conservative, or not expressive and succinct enough in general. To derive better bounds, we propose a novel approach for synthesizing piecewise bounds over probabilistic programs. First, we show how to extract useful piecewise information from latticed $k$-induction operators, and combine the piecewise information with Optional Stopping Theorem to obtain a general approach to derive piecewise bounds over probabilistic programs. Second, we develop algorithms to synthesize piecewise polynomial bounds, and show that the synthesis can be reduced to bilinear programming in the linear case, and soundly relaxed to semidefinite programming in the polynomial case. Experimental results show that our approach generates tight piecewise bounds for a wide range of benchmarks when compared with the state of the art.

cs.PL↗