Conference Agenda

Silicon nitride (SiN) passivation layers play a crucial role

in ensuring the long-term reliability of power semiconductor devices by

protecting sensitive regions from environmental and electrical stresses.

However, thermo-mechanical loads induced during fabrication and as

sembly processes can generate significant stresses within the passivation

stack, potentially leading to crack initiation and reliability degradation.

In this work, a finite element-based thermo-mechanical analysis is per

formed to qualitatively assess the stress state induced in a SiN passi

vation scheme by three key manufacturing steps: SiN deposition, die

sintering, and package-level (STPAK) sintering. Simplified geometrical

models and material descriptions are adopted to isolate the contribu

tion of each process and to rank their relative impact. Results indicate

that SiN deposition, dominated by intrinsic and thermal stresses due

to cooling from deposition temperature, is by far the most critical pro

cess, generating stresses exceeding 500 MPa at room temperature. In

contrast, die and package sintering processes produce comparatively mi

nor stress variations. The study provides useful design-oriented insights

for process optimization and reliability-aware passivation design, while

clearly identifying the limits of quantitative interpretation.

This paper proposes a statistical CAD modeling framework to generate par-ametric geometries representative of manufacturing variability based on measurement data. The method combines distribution fitting and percentile-based modeling of design variables with correlation information to reproduce realistic relationships among geometric parameters. In contrast with Principal Component Analysis approaches, the proposed framework preserves the original CAD design variables, preserving the physical interpretability of the original CAD design variables and enabling direct interpretation. The meth-odology is implemented through an automated workflow that generates sta-tistically consistent CAD instances by propagating deviations according to the extracted correlation matrix. A case study on a machined component is presented to demonstrate the capability of the approach to capture process-dependent variability and correlated behavior among parameters. The results confirm that the proposed method can generate statistically representative CAD models suitable for simulation-driven analyzes.

This review examines 3D Gaussian Splatting as an emerging explicit representation for 3D reconstruction and real-time novel-view synthesis, with emphasis on industrial engineering, biomedical imaging, and cultural heritage. The core principles of Gaussian-based scene modeling are summarized, and recent methods are organized into static reconstruction, dynamic modeling, efficiency-oriented approaches, and learning-based extensions. The 3DGS role is assessed against photogrammetry, mesh-based workflows, and neural implicit fields in terms of rendering efficiency, geometric fidelity, metric reliability, scalability, and interoperability. Application-specific evidence is then analyzed in medical visualization and surgical scene reconstruction, cultural heritage digitization and restoration, and industrial inspection and digital twins, highlighting domain-specific opportunities, maturity levels, and adoption barriers. Open challenges remain in geometric accuracy, standardization, data exchange, and downstream integration. 3DGS emerges as a promising but still maturing explicit representation for application-specific 3D data visualization, documentation, inspection, exchange, and analysis.

The Geo3DTouch project addresses equitable access to STEAM education for blind and visually impaired people (BVIP) through inclusive design. This article presents the project’s multi-disciplinary frameworks across three innovative tracks: computational tactile reconstruction using AI-driven and deterministic models to convert 2D visual data into highly intelligible 3D tactile bas-reliefs; ac-cessible 3D modeling utilizing a custom Braille plugin for Tinkercad and Blender to empower pre-service teachers to design tactile geometric aids; and inclusive immersive environments through Neotrie VR, which utilizes high-contrast inter-faces and Mixed Reality to foster embodied spatial reasoning for low-vision learners. User testing and pilot interventions validate that integrating physical re-lief, structured CAD workflows, and immersive digital adaptations significantly enhances shape recognition, spatial orientation, and classroom engagement. Ulti-mately, these findings offer a scalable methodology for future multisensory and accessible STEAM learning ecosystems.

Data transfer between non-conforming meshes arises in multiple engineering and scientific domains, yet existing methods often decouple interpolation from error assessment and lack a unified treatment of visualization and quality control. This paper presents MeshMapX (MMX), a methodological framework that integrates conservative interpolation, error metrics, interpolation analysis, and interactive visualization in a single environment. MMX is capable of diagnosing and exposing critical aspects of the transfer process that would otherwise go unnoticed. Built-in error metrics and mesh-map visualization provide direct feedback on conservation quality, spatial distribution of errors, and unmapped regions. An optimization module is also envisioned as part of the framework, though it remains under development. As a case study, MMX is applied to the transfer of electromagnetic load data onto the structural mesh of the DTT (Divertor Tokamak Test facility) divertor cassette, a representative multiphysics problem in nuclear fusion research. The analysis demonstrates how MMX successfully identifies the root cause of force balance discrepancies — namely, geometric differences between the source and target models — which would have been difficult to detect without integrated diagnostic tools. MMX offers a transparent and extensible solution for error-aware, conservative data transfer, independent of the specific physical domain.

This work introduces RE-CAD, a pipeline that converts multiview images into editable Computer-Aided Design (CAD) command sequences. Building on ARE-Net and DeepCAD, it addresses the lack of recoverable CAD design procedures and the scarcity of paired image–CAD sequence data. A synthetic multi-view dataset was therefore generated from DeepCAD models using an automated Blender pipeline, with a black background and a High-Dynamic-Range imaging (HDRI) variant; the latter adds ray tracing, reflections, and depth blur to approximate real-world imaging conditions.

The architecture uses a pretrained ResNet-18 encoder to process 12 views per object, a Gated Recurrent Unit (GRU) pooler to aggregate their features, and the pretrained DeepCAD decoder. The encoder–pooler was trained by Mean Squared Error (MSE) against 256-dimensional target vectors produced by the pretrained DeepCAD encoder. Hyperparameters were optimized through two Optuna studies, followed by multiple training sessions and 5-fold cross-validation. The reported losses evaluate latent-space alignment, not command validity or geometric reconstruction accuracy. Validation loss plateaued while training loss decreased; an exploratory comparison also showed a lower, more stable trend in validation loss for the 8% subset than for the 5% subset.