Chinese semiconductor thread II

tokenanalyst

Lieutenant General
Registered Member

Kezhicheng Third Generation Semiconductor Achieves Mass Delivery of Functionalized Diamond Heat Dissipation Substrate for Advanced AI Chips


Henan, China – Kezhicheng Third Generation Semiconductor Carbon-Based Chip Co., Ltd. has announced a significant milestone in the domestic semiconductor supply chain: its independently developed functionalized diamond heat dissipation substrate has achieved mass delivery at the research and development (R&D) level and has been officially adopted by a top domestic microelectronics packaging research institute.

This new material solution is specifically designed to support the advanced packaging R&D of AI computing chips, particularly those utilizing Chiplet architectures and 2.5D/3D stacking technologies. As China accelerates its development in high-end semiconductors, this achievement marks a critical step toward solving the thermal management bottlenecks associated with domestically produced advanced packaging.

The surge in multi-chip integration and high-density packaging has led to a dramatic increase in local heat flux density within single chips. Consequently, hotspot temperatures and thermal gradients have intensified significantly. Traditional cooling materials, such as copper-based substrates and aluminum nitride ceramics, are approaching their physical limits in terms of thermal conductivity. This limitation often results in:​
  • Localized overheating.​
  • Operational frequency reductions.​
  • Reliability issues including electromigration and thermal fatigue within interconnect structures.​
To address these challenges, Kezhicheng leverages its proprietary capabilities in large-scale Microwave Plasma Chemical Vapor Deposition (MPCVD) diamond deposition, ultra-thin precision grinding, and integrated metallization processes to create a functionalized diamond heat dissipation substrate.​
  • Superior Thermal Conductivity: Stable thermal conductivity exceeding 1300 W/(m·K), facilitating rapid heat diffusion from localized hotspots and effectively reducing peak temperatures (intrinsic diamond conductivity can exceed 2000 W/(m·K)).​
  • Thermal Expansion Matching: The substrate's coefficient of thermal expansion is closely matched to silicon, minimizing thermal stress during packaging. This ensures superior flatness, roughness control, and bonding precision.​
  • Process Compatibility: It is fully compatible with complex advanced packaging processes such as flip-chip packaging and interposer integration.​
Joint testing conducted with research institutions confirmed that the diamond substrate significantly reduces chip hotspot temperatures and improves overall thermal gradients. This optimization helps mitigate risks of interface delamination and solder joint failure during thermal cycling, thereby enhancing the stability and long-term reliability of chips under full-load conditions.

The product is now applicable to various R&D verification scenarios, including AI accelerator chips and High-Performance Computing (HPC) chips, helping to shorten development cycles for new packaging architectures.

This batch delivery represents a new milestone in the industrialization of diamond thermal management materials by Kezhicheng. The company has established an independent R&D ecosystem covering deposition, growth, machining, and packaging adaptation, allowing for customized solutions tailored to specific advanced packaging needs.​

Please, Log in or Register to view URLs content!
 

tokenanalyst

Lieutenant General
Registered Member
RadHard SSDs?​

Aikosa Technology Showcases at the 5th Remote Sensing & Communication Innovation Conference​


Founder and CEO of Aikosa Technology, outlined three critical pillars of next-generation storage architecture:​
  • Architecture Shift: Moving away from simple payload attachments to a software-defined model where a single platform manages remote sensing writing, CPU access, and network sharing simultaneously while separating reliability for remote sensing vs. computing tasks.​
  • Engineering Innovation: Utilizing industrial-grade Commercial Off-The-Shelf (COTS) clusters to leverage SSD controller power. This allows the offloading of Network File System (NFS), distributed file systems, and Key-Value (KV) caches, achieving aerospace-grade reliability through natural redundancy.​
  • Software & Philosophy: Emphasized that true reliability relies on reconstructing foundational software (file systems, RAID, drivers, etc.). The team advocates a "crash-only philosophy": crashes are acceptable if lossless recovery is guaranteed; the focus is not just on hardware but on the software platform it runs on.​
Aikosa Technology's presentation highlighted that spaceborne storage has evolved from a peripheral support role to a foundational function. By transitioning from stacking components to reconstructing reliability via COTS clusters and advanced software, Aikosa provided the industry with a systematic roadmap for future intelligent computing in space.

1783358047092.png

Please, Log in or Register to view URLs content!
 

tokenanalyst

Lieutenant General
Registered Member

Ziguang Tongchuang PG2L50M Successfully Passes AEC-Q100 Certification​


Unisplendour Corporation has added a new member to its automotive-grade product line. The PG2L50M-5AMBG256 chip has successfully passed the AEC-Q100 Grade2 certification, a globally recognized authoritative standard for automotive electronics reliability , further solidifying the company's foundation for product applications in the automotive field.

The PG2L50M-5AMBG256 chip is a new high-performance, cost-effective solution launched by Unisoc for high-speed video interface bridging applications. It utilizes a mature 28nm CMOS process, 45K logic scale, and a hard-core MIPI D-PHY with 2.5 Gbps per lane, supporting Tx and Rx, and can directly drive up to 4 MP 60 fps high-definition screens. It can be freely configured with soft-core MIPI and LVDS 1.25G interfaces, providing users with greater flexibility and adaptability to more application scenarios. It supports a hard-core DDR4 interface, providing a faster storage solution for complex image processing applications. Features SEU error detection and correction, AES-256 data stream encryption combined with RSA-4096 verification, greatly ensuring security and reliability. With its ultra-low latency, the PG2L50M-5AMBG256 chip offers more competitive solutions for niche applications such as CMS electronic rearview mirrors, automotive LiDAR, and high-end cockpit displays.

Following the PGL25G-6AFBG256, the PG2L50M-5AMBG256 chip has once again obtained AEC-Q100 automotive-grade certification. This is not only an authoritative recognition of Unisoc's product R&D capabilities and quality control system, but also an important step for the company in its strategic layout in the intelligent vehicle market and in perfecting its automotive-grade product ecosystem.

Please, Log in or Register to view URLs content!
 

tokenanalyst

Lieutenant General
Registered Member

Tongfu Microelectronics is advancing the construction of advanced packaging capacity and increasing its packaging and testing capabilities for memory chips and high-performance computing.​


Tongfu Microelectronics is aggressively expanding its advanced packaging capacity through a significant refinancing plan aimed at raising approximately RMB 4.22 billion via share issuance. These funds will primarily finance the construction of high-end facilities for memory chips, high-performance computing (HPC), and automotive electronics, including projects like the Tongtu Tongda Advanced Packaging Base with an investment of roughly RMB 7.5 billion. This strategic push is designed to bolster working capital while building critical capabilities in technologies such as multilayer stacked packaging, flip chip assembly, wafer-level packaging (WLP), and panel-level processing (FOPLP) to meet growing industry demands.

The company's expansion focuses heavily on the intersection of artificial intelligence and storage solutions, specifically targeting markets for AI computing power, GPUs, CPUs, servers, and solid-state drives. By developing its Nantong Memory Phase II project and enhancing testing infrastructure at its Nantong Advanced Packaging Base, Tongfu aims to solidify its position as a leading global provider for high-end chip packaging. The firm has confirmed it possesses relevant technological reserves for cutting-edge innovations like Chiplets and High Bandwidth Memory (HBM), positioning itself to leverage the rapid growth in AI server infrastructure and HPC applications.

Industry analysts view Tongfu's moves as well-positioned to capitalize on the surging demand for AI-driven semiconductor components, but they caution that actual revenue growth in these advanced sectors will depend heavily on future customer orders and the successful ramp-up of production lines. Ultimately, this expansion is expected to significantly enhance Tongfu Microelectronics' overall competitiveness in the evolving landscape of high-performance computing and memory packaging.


Please, Log in or Register to view URLs content!
 

tokenanalyst

Lieutenant General
Registered Member

The Material Selection of High Numerical Aperture EUV Mask Absorbers Based on Waveguide Effects and Aerial Image Metrics.​

Abstract​

The imaging quality of extreme ultraviolet lithography (EUVL) depends on the optical properties, the thickness of the mask absorber material, and the illumination shape of the source. For high numerical aperture (NA) mask absorbers, we can analyze how to mitigate their mask three-dimensional (M3D) effects based on waveguide effects using test light sources. Then, we can improve key metrics such as the normalized image log slope (NILS), the (BFV), the throughput criterion (TpT), the threshold to size (TtS), and the critical dimension (CD) error of the aerial images. This study effectively screens high-NA EUV mask absorber materials based on the waveguide effect in the absorber. It provides a reasonable explanation for the imaging trend, presents a method to improve the screening efficiency of absorber materials, and determines the final absorber material parameters suitable for high-NA lithography systems by considering the limitations of actual materials and the results of lithography simulation. The above research has potential to experience for the selection of absorber materials in future hyper NA lithography systems.​

1783368838663.png

Please, Log in or Register to view URLs content!
 

tokenanalyst

Lieutenant General
Registered Member

Implicit Wafer Plane Inspection of Masks Using a Lithography Aware End-to-End Neural Network​

Tencent​

Abstract:​

Due to optical proximity corrections and other resolution enhancements necessary for sub-45nm nodes, mask patterns are dramatically different from those in the wafer plane, which makes mask inspection very challenging, as many defects are nonprintable (i.e., false positive, FP). Wafer plane inspection (WPI) has been introduced to reduce such FPs. However, because of the need for rigorous computational lithography (ComLitho) in WPI, it is computationally expensive. Deep neural networks (DNNs) have been utilized for mask inspection and have achieved impressive results. In this study, we propose a new mask inspection paradigm, termed implicit WPI (iWPI), using deep learning that is aware of the impact of lithography on defect printability but does not require ComLitho at run-time. A DNN for end-to-end semantic segmentation is trained using a large number of samples generated by ComLitho. The reference patterns and the corresponding generated defective ones are fed into the ComLitho engine to screen out printable defects in the wafer plane, and the data are used to train the DNN for pixel-level defect detection. Experimental results in die-to-die mode show that iWPI can effectively reduce FPs, while improving the inspection speed by 13 times over WPI and also reducing memory usage. It has great potential for in-line deployment in semiconductor production as it can be run on existing inspection equipment.

Please, Log in or Register to view URLs content!
 
Top