Kai Chang

Scientific and engineering verticals often suffer from data scarcity and strict executability requirements: models must generate not only fluent text but also syntactically valid, tool‐compilable scripts. We present a schema‐first alignment framework for building compact, executable domain‐specific LLMs in low‐resource settings. The framework integrates three core components: (i) large‐scale synthetic QA data generation from expert documentation to instill foundational domain knowledge; (ii) a code‐centric IR DPO workflow that converts verified tool decks into interpretable intermediate representations (IR), performs equivalence‐preserving diversification and constructs preference pairs to directly optimize instruction compliance and code executability; and (iii) a controlled evaluation of retrieval‐augmented generation (RAG), showing that while RAG benefits general LLMs, it can marginally degrade the performance of already domain‐aligned models. We demonstrate the framework by instantiating TcadGPT for semiconductor technology computer‐aided design (TCAD). Using 1.5 M synthetic QA pairs and an IR‐driven DPO dataset, TcadGPT attains 85.6% semantic accuracy and an 85.0% syntax pass rate on SDE executability tests, substantially outperforming state‐of‐the‐art general LLMs such as GPT‐4o. To probe portability beyond TCAD, we apply the same recipe to the open‐source FEM solver Elmer , observing consistent improvements in script‐level success rates over general‐purpose baselines. All datasets, benchmarks, and code (including P1, P2, and IR DPO) are released for reproducibility. Together, these results suggest that the proposed framework provides a robust and reproducible path toward executable LLMs in specialized, data‐scarce professional domains.

Polar skyrmions, as noncollinear polarization textures, have attracted a considerable amount of attention so far because of their novel physical properties. Multistate switching of polar skyrmions is paramount for brain-inspired computing applications, but experimental realization remains challenging because of the inaccessibility of feasible boundary conditions. Here, we explore in situ electrical switching of CuInP2S6-based polar skyrmion bubbles with respect to resistance states on silicon substrates. Because of Cu+ mediation and the large size of CuInP2S6 skyrmions, we can electrically manipulate the boundary conditions and create a series of intermediate skyrmion states. Consequently, we acquire eight nonvolatile resistance states from a single skyrmion bubble, in which information storage density far exceeds that of the current hard drives. On the basis of these devices, we build skyrmion-based solid neurons and simulate spiking neural networks with a recognition accuracy of 98%. Our study shows the uncharted application prospects of polar skyrmions in silicon-based, high-density memory hardware and neuromorphic computing.

We demonstrate that spatially engineered magnetic anisotropy can stabilize skyrmion and meron spin textures in magnetic nanodisks even in the absence of Dzyaloshinskii–Moriya interaction (DMI). Using a constrained analytical model and micromagnetic simulations, we show that competing perpendicular and in-plane anisotropies can generate noncollinear topological textures in nonchiral magnetic systems. We further show that DMI and dipolar interactions lift the helicity degeneracy and select preferred chiral configurations; micromagnetic simulations were used to identify physically stable states. These results establish anisotropy-patterned nanodisks as a platform for studying DMI-free topological spin textures and their controllable magnetic response. We also show that arrays of anisotropy-engineered skyrmions can control spin-wave transmission by manipulating their vorticity arrangement, pointing to reconfigurable magnonic elements based on nonchiral topological textures.

We systematically analyze both the intrinsic and extrinsic spin Hall conductivity in Janus monolayers using an effective minimal Hamiltonian and the Kubo-Streda formalism with Feynman diagrams. Our analytical and numerical results reveal that the vertex cancellation distinguishes Janus monolayers from the pristine counterparts, i.e., the transition metal dichalcogenide, because of the existence of Rashba- and Zeeman-type spin-orbit couplings. The interplay between intravalley and intervalley scattering processes significantly modifies the extrinsic spin Hall conductivity in the conduction bands for different occupations. These intervalley scattering processes play a crucial role and provide us an experimentally accessible way to tune the extrinsic spin Hall effect in Janus.