A heterogeneous AI server package integrating XPU–HBM chiplets and co‐packaged optics (CPO) on a glass core substrate is presented. This work addresses system‐level mechanical, electrical, and power co‐optimization for multi‐module integration on glass, rather than substrate process comparison, by evaluating CTE‐induced warpage, TGV‐based power delivery, electromigration robustness of inorganic‐coated RDL layers, and high‐speed signal transmission between XPU and CPO modules. The results demonstrate improved signal integrity, power efficiency, and reliability for large‐scale, energy‐efficient AI server packages.
Aluminum laminates, widely used in food packaging, are difficult to recycle because of their complex multilayer structures, leading to disposal primarily through landfilling or incineration. In this study, we propose a method for separating aluminum foil and plastic layers from aluminum laminates using electrical pulsed discharge (EPD). Since EPD can selectively heat the aluminum foil, its energy consumption is expected to be lower than that of conventional thermal processes. However, the applicability of EPD to various aluminum laminates and its separation mechanism have not yet been fully elucidated; thus, this study aimed to investigate these aspects. EPD was applied to four aluminum laminates with varying layer thicknesses to investigate their separation behaviors. Successful separation was achieved for all samples except the thinnest one, with high aluminum recovery rates (93.6-98.7 wt%) and purity (96.6-99.7 wt%). Also, the chemical analysis of the recovered resins showed almost no change before and after the pulsed discharge. To elucidate the separation mechanism, high-speed camera imaging, heat transfer and thermal stress simulations were conducted. The results revealed that successful delamination requires both sufficient energy to decompose the interfacial resin and adequate thermal stress to propagate the delamination. To evaluate the environmental impact of the EPD method, its energy consumption and greenhouse gas (GHG) emission were compared with those of other separation methods. EPD-based recycling process demonstrated superior energy efficiency and GHG emission reduction compared to conventional processes.
The demand for high-speed signal transmission technologies is increasing, and co-packaged optics, which significantly reduces electrical transmission distances, is attracting growing attention. This report presents the performance evaluation results of a glass substrate with integrated polymer optical waveguides that we have been developing. The optical waveguides were successfully fabricated with a minimum pitch of 10 micrometers using imprint lithography. For the 10-micrometer-pitch optical interconnects, it was confirmed that crosstalk can be effectively suppressed by increasing the relative refractive index difference. In addition, the reliability of the optical waveguide material was also evaluated. Within the evaluated period, the waveguides were confirmed to exhibit sufficient reliability in terms of single-mode propagation and bending loss.
The use of intelligent reflecting surfaces (IRSs) to enhance coverage in millimeter-wave bands has been gathering attention. To utilize IRS in mobile networks, it is necessary to determine the reflected power of the IRS. As the electrical size of the IRS must be large to enhance the reflected power, its application in the near-field region (Fresnel region) should increase. In addition, an actual IRS is affected by unaccounted errors such as reflection phase errors, reflection amplitude errors, and mutual coupling. A practical evaluation method for electrically large IRS is required for highly accurate reflected power estimation. To address these problems, we propose dividing electrically large IRS into sub-IRSs to synthesize the reflected E-field from each sub-IRS. The reflected E-field was estimated using the measured radar cross-section (RCS) of the sub-IRSs. To compensate for the differences in the incident and reflection angles of each sub-IRS, we proposed estimating the difference using a geometrical approach. The power reflected from each sub-IRS is estimated by shifting the reflection angle in the RCS pattern. The numerical results verified that the reflection power of the electrically large IRS was estimated with high accuracy while the location of each transmitting point and observation point met the far-field condition of the sub-IRS. The measurement results in an anechoic chamber also verified that the proposed method could estimate the reflected power with high accuracy compared with the conventional physical optics (PO)-based approach. The median power difference of the proposed method and measured results was approximately 3 to 4 dB, and the correlation between them exceeded 0.77, which are superior to those of the conventional PO-based approach.
