Accurate lifetime prediction of power semiconductor devices is critical for the reliability of power electronic systems. Power cycling tests generate only a small number of failure samples due to their high cost and long duration. This data scarcity makes it difficult to train reliable prediction models. At the same time, modern deep learning models require significant AI expertise to configure and tune, which creates a barrier for reliability engineers. To address these two problems, this paper proposes a domain-constrained multi-agent framework for automated remaining useful life (RUL) prediction of power semiconductor devices. The framework integrates three collaborative agents: a Stats Agent for statistical reliability assessment, a Physics Agent for physical plausibility validation, and a Model Agent for algorithm selection and hyperparameter tuning. A Central Controller manages the workflow and coordinates agent communication. The agents interact through a Propose-Critique-Refine (PCR) mechanism, in which the Model Agent proposes configurations and the other two agents critique them from statistical and physical perspectives. This iterative negotiation eliminates the need for manual parameter tuning. The framework is validated on the NASA IGBT accelerated aging dataset under Leave-One-Device-Out Cross-Validation. Results show that the framework achieves an R2 of 0.914 at a 50% observation ratio and remains above 0.79 even at 30%, demonstrating strong generalization under limited data conditions.

Due to the better performance of the Wide Band Gap (WBG) devices, there has been a paradigm shift toward WBG-based power modules for diverse applications like Electric Vehicles (EVs). However, the high parasitic inductance value of power modules hinders these devices from unlocking their full potential. Therefore, this paper comprehensively reviews SiC-based Single Side Cooling (SSC) power modules that benefit from low parasitic inductance. The paper also discusses the need to develop newer power modules using modern packaging methods. The surveyed power modules are categorized into three main groups, namely wire bonding, hybrid, and 3D packaging methods. This classification contains several vital parameters of the studied power modules, such as nominal and Double Pulse Tests (DPT) power ratings, parasitic inductance, size, etc. The main features and characteristics corresponding to the reviewed power modules' packaging methods and techniques are also briefly described. Finally, a thorough discussion about challenges and future trends is highlighted before concluding the paper.

Higher switching speeds and power densities enabled by SiC MOSFETs make accurate, time-resolved junction-Temperature (Tj) estimation under realistic switching conditions increasingly important for performance validation, thermal design, and reliability assessment. This work presents an electro-Thermal co-simulation framework for SiC MOSFET double-pulse testing, combining measured switching waveforms with a compact thermal network (Cauer model) to translate transient switching-loss energy into (Tj) evolution. A Peak-Power-Threshold (PPT) method is introduced to robustly identify the switching interval and compute instantaneous power and switching energies (Eon, Eoff) from experimental data, enabling consistent comparison between measurement and simulation across few operating points. The proposed approach links dynamic loss extraction to junction-Temperature estimation in a unified workflow, supporting more reliable interpretation of double-pulse tests and improving confidence in electro-Thermal estimation for high-performance SiC power converters.

Solder joint reliability related to failures due to thermomechanical loading is a critically important yet physically complex engineering problem. As a result, simulated behavior is oftentimes computationally expensive. In an increasingly data-driven world, it is popular to use efficient data-driven design schemes. Among the family of efficient optimization methods, Bayesian optimization with Gaussian process regression is a key representative. The authors argue that additional computational savings can be obtained from exploiting thorough surrogate modeling and selecting a design candidate based on multiple acquisition functions. This is feasible due to the relatively low computational cost, compared to the expensive simulation objective. This paper presents a novel heuristic framework for performing Bayesian optimization with adaptive hyperparameters across multiple optimization iterations. A comparative study shows the ability of adaptive Bayesian optimization to save on expensive objective evaluations with respect to the worst-performing regular Bayesian optimization scheme. As an engineering use case, the solder joint reliability problem is tackled by minimizing the accumulated non-linear creep strain under a cyclic thermal load. Results show that adaptive Bayesian optimization can at least match the performance of regular Bayesian optimization in terms of raw objective performance, but achieves this with half of the computational expense budget. This practical result underlines the methodological potential of the novel adaptive Bayesian data-driven methodology to achieve more efficient results and significantly cut optimization-related expenses. Lastly, to promote the reproducibility of the results, the data-driven implementations are made available on an open-source basis.

Physics-informed neural networks (PINNs) on complex domains are limited by input representations that encode geometry but do not reflect the physics of the governing PDE. We propose an operator-aware PINN for solid mechanics problems that embeds precomputed eigenmodes of the problem’s own discrete operator as geometry and physics-aware features within a weak-form variational formulation. The displacement field is represented by a neural mapping enriched by operator-aligned features rather than being constrained to a finite element trial space. A hybrid chain-rule formulation propagates spatial derivatives through both coordinate and eigenmode branches, yielding strain fields consistent with operator structure while preserving neural expressivity. The approach maintains the stability of variational formulations while mitigating the spectral bias and geometry-encoding limitations of classical PINNs. For inverse elasticity, the same eigenmodes act as virtual test functions in a weak-form residual, mitigating the stiffness-collapse pathology. The framework provides an instance-based alternative to differentiable finite element method without labeled training data. It achieves competitive accuracy and connects spectral operator theory with physics-informed neural computation.

The paper presents a degradation for remaining useful lifetime framework for smart power stages (SPS) with an integrated temperature sensor for electronic control units (ECU). A reduced order state space model with physically meaningful time constants is derived from finite element simulations. The state space model is the core of the damage estimation of the SPS during operation. A framework for remaining useful lifetime estimation is presented simulating degradation based on an estimated mission profile for cars. Temperature cycles are extracted and used to estimate the time constant drift representing package degradation. An exemplary approach on how to integrate degradation monitoring into a lifetime approximation of the package is presented. The state space model is extended for estimator based predictions to integrate the model in future test benches for real life monitoring and remaining useful life predictions of active devices.

Modeling and monitoring advanced power semi-conductor packages are challenging due to their multiphysics behavior, structural complexity, and the high computational cost associated with detailed models. Conventional high-fidelity approaches, such as finite element models, provide high accuracy but are often unsuitable for applications like health monitoring, while purely data-driven models may lack physical interpretability and robustness under degradation.This paper proposes the concept of a multi-agent digital twin framework to address these challenges by decomposing the overall system into interconnected virtual domains with different roles and fidelity levels. The approach is demonstrated for health monitoring of a SiC MOSFET in a DC-DC buck converter. A three-agent architecture is implemented, consisting of a temporal ANN for system identification, a decision layer for failure detection and localization, and a parametric reduced-order thermal model for failure assessment. Experimental results show that the proposed framework can accurately identify system behavior, detect and localize failures, and quantify parameter variations associated with degradation.

Low-temperature atmospheric plasma (LTP) is widely used in industrial processes, such as disinfection, surface modification and wastewater treatment. The dielectric barrier discharge (DBD) is regarded as one of the most robust and reliable methods for generating LTP in ambient air. Compared to conventional AC excitation, pulsed powering offers several advantages (i.e., lower energy use and heat production). The present trend is to use short and fast pulses (in the nano- and picosecond range). In this review, the key design parameters of a DBD (barrier thickness, relative permittivity and gap distance) are discussed. Material-specific phenomena like surface charging and degradation are analyzed. The complex interactions between the pulse source and DBD are examined. By mapping the interdependencies, this review aims to support the rational design and optimization of pulsed DBD systems, and to facilitate their broader industrial use.

Pressureless sintered Ag pastes are promising die-attach materials for power electronics, yet practical sintering-profile optimization still relies heavily on trial-and-error, and the link from thermal kinetics to fracture-relevant microstructure and strength remains insufficiently established. In this work, a thermal-kinetics-guided workflow was developed to design paste-specific pressureless sintering profiles for two commercial Ag pastes (a spherical-particle paste and a flake-based paste) and to interpret the resulting strength–fracture response using SEM-based, heterogeneity-aware learning. Multi-heating-rate TGA was used to define the conversion fraction (α), while DSC was used to identify the dominant thermal-event window and extract the characteristic peak temperature for Arrhenius pairing. The TGA-defined conversion evolution was then modeled using a JMAK/Avrami form with Arrhenius temperature dependence to predict the isothermal holding time required to reach a target conversion at a selected dwell temperature. Relative to supplier-recommended profiles, the optimized profiles increased die-shear strength by 35% for the spherical-particle paste and by 206% for the flake-based paste, and promoted a fracture-mode transition from interfacial debonding toward mixed/cohesive fracture. Pearson/Ridge baselines and attention-based multiple-instance learning (MIL) linked strength and fracture-mode distribution to porosity/connectivity-related descriptors; paste-wise normalization mitigated paste-specific baselines and enabled MIL to reveal profile-induced microstructure–performance co-variation. Overall, this study establishes a practical workflow that couples thermal-kinetics-guided profile design with mechanical and fractographic validation and interpretable microstructure–property attribution, supporting mechanism-informed optimization of pressureless sintered Ag die-attach pastes.

The rapid growth of generative Artificial Intelligence (AI), automotive electrification and Internet of Things (IoT) is driving an unprecedented demand for high-performance Integrated Circuits (ICs), projected to push the semiconductor industry to a $1 trillion business by 2030 (Burkacky et al., 2022; PWC, 2025). This drove the adoption of advanced process (FinFET, Gate-All-Around, Forksheet, CFET, Backside Power Delivery Network) and 3D package technologies (chiplets, Heterogeneous Integration, Co-packaged Optics). Unfortunately, component stacking in 3D ICs such as Package-on-Package (PoP) products creates an optical barrier during Electrical Fault Isolation (EFI). This paper presents a novel Failure Analysis (FA) hardware and sample preparation solution that enables System-Level FA on mobile System-on-a-Chip (SoC) PoP devices, where the DRAM is stacked atop the logic controller device. The primary challenge in performing EFI on the bottom die is providing direct line-of-sight (LoS) access while preserving the top DRAM functionality through extremely small interconnects called Through Interposer Vias (TIVs). For the first time, this groundbreaking hardware solution overcomes the challenge of connecting a DRAM atop the SoC silicon, utilizing the smallest possible interposer pin pitch (∼210 um) and advanced DRAM card design that incorporates stringent design rules to enable up to 6.3 Gbps using a DDR training test and runs standard Android stress applications. The results of the FA hardware solution for SLT platform will be discussed, together with several FA use cases that were enabled through this innovative solution. Lastly, this paper will discuss the limitations of this hardware, together with opportunities for improvement and further work. This novel solution paves the way for the failure analysis of 3D IC devices on a system level platform, which are utilized not only in mobile applications, but also for cloud, server, and quantum computing ICs assembled in complex packages.

The evolution in modern-day electronics extends beyond miniaturization and significantly increases the number of power cycles operating with low temperature swings (few tens of degrees). This shift has raised the reliability requirements and expectations for electronics systems to last longer. The transition of vehicles and autonomous systems to zonal architectures, driven by electrification, has increased power-on hours of electronic components, leading to higher system complexity and more challenging reliability testing. Currently, standardized stress- based tests and predictive models have constraints, as they are designed for large temperature swings and are inaccurate for evaluating low T conditions. Additionally, the damage caused by Δ millions of low thermal swings is less explored in simulation and qualification, making it important to analyze the influence of low ΔT on solder joints to understand field reliability risks.An Arduino-based testing setup was developed to apply active thermal cycles to WLCSP40 using relay-controlled switching with 6-second cycles. Samples were tested from 50k to 800k cycles to analyze the long-term damage accumulation under such conditions. In addition to experiments, a thermomechanical simulation in COMSOL incorporated ΔT values obtained from thermal camera characterization to establish realistic boundary conditions. It helps to correlate experimental results and identify critical locations within the solder joints where damage is most likely to occur.Cumulative plastic strain analysis and cross-sectional imaging revealed cracks in the solder bulk at 40 °C, demonstrating microstructural fatigue under low ΔT cycling. Furthermore, based on strain per cycle values, the results exhibit how the active and passive thermal cycles differ.

Serving as a significant attachment component, Active Metal Brazed (AMB) substrates are prone to oxidation during storage and processing. The copper surface oxidation of AMB substrates seriously affects the reliability of high-power modules. The formation of Cu2O/CuO oxide layer on the surface degrades interfacial qualities in subsequent packaging processes, including sintering, wire-bonding, and transfer-molding. To investigate the mechanism behind, this paper constructs a physically grounded multiscale oxidation framework that explicitly bridges molecular dynamics with reactive force field (ReaxFF-MD) and mesoscale continuum modeling. For ReaxFF-MD simulations, it aims to resolve oxygen adsorption, dissociation, diffusion behavior, and Cu-O network formation on the Cu(100), Cu(110), and Cu(111) surfaces at temperatures of 300 K and 600 K. Diffusion coefficients and interface reaction kinetic parameters are then quantitatively extracted from MD simulations and transferred as a bridge in reaction-diffusion continuum model with flux damping governed by a characteristic structural thickness. The results demonstrate that the early oxidation process of the copper surface is modulated by structural evolution and orientation anisotropy. The proposed multiscale framework provides a physical mechanism basis for understanding the early oxidation mechanism of copper and its impact on the reliability of AMB packaging.