This paper details the assembly and fiber attachment process for the Lightmatter Passage™ M Series 3D Silicon Photonics Engine and performance, which uses Chip on Wafer (CoW) packaging to integrate 15 ASIC top die chiplets onto a 7200 mm2 organic substrate with a multi-reticle 3D photonic interposer. The paper will also present performance and bit error rate (BER) test results, demonstrating the system's reliability against temperature cycling and polarization changes, as well as its thermal capabilities supporting a total thermal design power (TDP) exceeding 900 W.
Heating, Ventilation, and Air Conditioning (HVAC) systems generate large-scale multivariate time-series (MTS) data that capture rich information about user behaviors and system dynamics. However, analyzing such data remains challenging due to its unlabeled nature, mixed-type features, noise, missing values, and strong temporal dependencies. In this manuscript, we propose MAPLE, a structured multi-stage framework for unlabeled MTS analysis in large-scale HVAC environments. MAPLE integrates label encoding for mixed-type (numerical and categorical) features, fine-grained point-level clustering, and time-segmented device-wise sequence labeling, enabling robust label-based downstream analysis without manual annotation. By leveraging Change Point Detection, the framework explicitly captures time-of-day and seasonal variations in device operation. Our analysis reveals distinct time-zone-specific behavioral patterns, with routine and consistent usage in the early morning, greater variability during daytime hours, and increasingly diverse, personalized control behaviors in the evening, underscoring the importance of temporal segmentation for modeling nuanced operational dynamics. An expanded ablation study further demonstrates that point-level clustering and temporal segmentation contribute uniquely and synergistically to clustering quality, interpretability, and classification performance. Experiments on a real-world HVAC dataset from LG Electronics, comprising over 8 million records from more than 42,000 devices, show that MAPLE achieves a Silhouette score of 0.6324 and an average classification accuracy of 0.9230, validating its effectiveness for large-scale HVAC behavior analysis.
Automotive optical sensors are essential in enabling advanced driver-assistance systems (ADAS), which significantly enhance vehicle safety and push the boundaries of autonomous driving technology. By accurately detecting and interpreting the environment, these sensors allow ADAS to respond to various driving conditions, reducing the risk of accidents. As the demand for safer and more efficient vehicles grows, these sensors will continue to play a key role in advancing the capabilities of autonomous driving systems. Additionally, automotive optical sensors are migrating to advanced wafer nodes and integrating control unit elements to support the sophisticated functionalities and applications demanded by the increasingly complex automotive market. The new trend requires more advance packaging technologies, Heterogenous Integration (HI) and standard platforms. This evolution poses significant challenges in assembly, testing, and meeting stringent reliability standards. Amkor Technology has proactively addressed these challenges through innovative approaches in automotive optical packaging. Leveraging extensive experience in automotive applications and diverse packaging solutions, a robust package platform has been developed explicitly tailored for optical sensors. This platform uses an adaptable assembly toolbox and integrates advancements in extending existing technology to streamline development cycles while optimizing packaging characteristics. This presentation will delve into Amkor’s standardized package platform, showcasing its versatile design solutions for optical sensors. It will explore how these solutions effectively navigate technical obstacles, adhere to Automotive Electronics Council (AEC-Q100) reliability standards, and drive forward the evolution of ADAS and autonomous driving technologies. These advancements in automotive optical sensor packaging will continue ensure enhanced performance and safety across the automotive landscape.
The continued adaptation of microelectromechanical systems (MEMS) technology for various applications in consumer, automotive, industrial, communications and even health & lifestyle markets has accelerated innovations in design and assembly for MEMS packages. These innovations address various challenges in integration, reliability, scalability and cost. The common driver that has a significant impact on the identified challenges is the form factor. Aside from enabling sensor functionality, the commercial scalability and ease-of-integration to end-customer devices are some of the keys drivers for MEMS and sensor packages’ form factor. Form factor influences the integration potential of a sensor module and determines how many components can be packaged altogether thereby reducing the overall footprint. It also establishes the path for manufacturing scalability which helps in accelerating the development timeline and shortening time-to-market. Form factor also defines board mount compatibility and reliability requirements and has become a common line-item on most package technology roadmaps. Amkor Technology understands these challenges and continues to provide solutions by leveraging its extensive experiences in diverse package platforms. As a result, it has developed a toolbox for advanced packaging technologies supporting Heterogenous Integration (HI) and standard packages for consumer, industrial and automotive applications. The extensive experience, supported by a capable design and the assembly toolbox becomes the development framework for small form factor MEMS and sensor packages. This paper focuses on the design aspects and package structures as well as assembly techniques to enable form factor reduction for MEMS and sensor packages. It describes the evolution of MEMS package platforms and the strategies to align with the industry’s demand for higher integration potential, better reliability, efficient scalability and commercialization.
Strip level warpage control is critical for the manufacturing and assembly of advanced packages. Unfortunately, it is very challenging to simulate strip level warpage accurately. One of the challenges is the strip gravity weight effect. In this paper, nonlinear simulations were performed using Ansys Workbench to observe the effect of gravity on strip warpage. The model consisted of the strip on top of a supporting surface, or table, to more closely resemble how the warpage was actually measured. To account for the gravity effect in simulations with various warpage shapes, contact conditions were included to capture the interactions between the strip and the supporting surface. Simulation results show the weight effect greatly reduced warpage on pre-mold strips, resulting in more accurate warpage simulation results when correlating to actual strip warpage measurement data. There was a significant difference in the weight effect at the chip-attach stage compared to the post-mold stage, with the post-mold stage showing less of a reduction in warpage. Other strip configurations and geometry effects were also studied in this paper, such as: strip size, number of blocks, a slot placed between blocks, and a cored vs. coreless substrate.
