Special Issue on the 7th International Sino <scp>MOS</scp> ‐ <scp>AK</scp> Workshop

As device structures become increasingly complex, with the continuous emergence of novel materials, unconventional architectures, and new physical phenomena, the coupling of multiple physical domains, including thermal, electrical, and optical effects, is becoming ever more prevalent. At the same time, rising development and manufacturing costs place additional demands on modelers to deliver representations that are both accurate and computationally efficient across the entire chain from device physics to circuit behavior. Modeling serves two complementary purposes: Theoretical models provide insight into the operating principles of devices while also guiding design optimization and enabling engineers to fully exploit intertwined physical effects. Analytical modeling, however, often requires careful trade-offs among accuracy, generality, and simplicity. Models must be predictive enough to inform design while offering meaningful physical insight. In modern semiconductor devices, which often feature three-dimensional geometries, solving the coupled semiconductor physics equations analytically is extremely challenging or even impossible. Closed-form solutions are typically unattainable, so judicious simplifications are necessary to ensure that models remain tractable and practically useful. The papers in this Special Issue address these challenges by balancing physical fidelity with computational efficiency. They deepen our understanding of device physics while providing models that are both insightful and practical, with applications spanning cryogenic electronics, wide-bandgap devices, and radiation-hardened systems. Su et al. [1] present a charge-based analytical model for bulk MOSFETs, that is, valid down to 10 mK. Their work clarifies the interface-trap-dominated mechanisms that lead to threshold voltage divergence between NMOS and PMOS devices and quantifies significant analog parameter enhancements, including a 73% increase in PMOS cutoff frequency at 4 K. These findings are essential for quantum-control electronics. Complementing this, Mao et al. [2] provide a comprehensive review of four physics-based compact models for GaN HEMTs, namely MVSG, ASM HEMT, EPFL, and QPZD. They analyze how each model addresses challenges such as trapping effects, self-heating, and process variability, and highlight emerging opportunities for combining physical models with machine learning to accelerate parameter extraction and quantify uncertainties. In the area of radiation-tolerant electronics, Xu et al. [3] introduce a machine-learning approach using an ant-colony-optimized neural network. By adaptively sampling critical waveform regions, their method achieves an RMS error of only 0.82% in predicting single-event transient currents, surpassing the fidelity limits of traditional double-exponential pulse models and enabling high-precision radiation effect simulation for aerospace applications. Meanwhile, Deng et al. [4] demonstrate a practical strategy for AI-assisted SPICE integration. They employ geometry-parameterized scaling laws for spiral inductors and machine-augmented Power MOS trans-conductance models to accelerate parameter extraction by an order of magnitude while preserving full SPICE compatibility. This approach significantly streamlines industrial design workflows. Collectively, these contributions point to a trend toward physics-informed, data-driven co-design methodologies. By combining rigorous physical insight with computationally efficient, machine learning–aware workflows, they enable robust optimization of devices and circuits across a wide range of applications, from quantum interfaces to aerospace systems. Future research should prioritize the development of standardized interfaces between AI tools and physical models, the extension of models to three-dimensional integrated wide-bandgap architectures, and the establishment of co-design frameworks for emerging ultra wide–bandgap materials capable of operating in environments ranging from near-zero Kelvin to orbital radiation conditions. We sincerely thank all authors for their outstanding contributions, which have advanced the frontier of semiconductor modeling science. Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.