Chinese semiconductor thread II
- Thread starter vincent
- Start date
Xinde Semiconductor achieves breakthrough in TGV technology
Xinde Semiconductor has achieved a significant breakthrough in wafer-level glass interposer 2.5D fan-out packaging technology using TGV (Through Glass Via) structures. As semiconductor processes approach the physical limits of Moore’s Law, advanced packaging solutions like 2.5D and 3D integration have become essential for sustaining performance improvements. The interposer serves as a critical bridge between chips, directly influencing data transmission speed and system scalability.
Traditional silicon-based interposers offer excellent electrical performance and high interconnect density but come with manufacturing costs exceeding 30% of total packaging expenses and complex production processes that limit widespread adoption. Organic interposers are cheaper but suffer from high signal loss at high frequencies, poor thermal expansion matching with chips, and susceptibility to warping—making them unsuitable for demanding high-density applications.
Glass-based TGV interposers provide a balanced solution with superior performance and cost-effectiveness. They offer extremely low high-frequency insertion loss, excellent mechanical strength, near-matched thermal expansion coefficients to silicon chips, and significantly improved manufacturability. This makes TGV ideal for next-generation AI computing, HPC, 5G millimeter wave, and optical communication systems.
Xinde Semiconductor, in collaboration with Stallone’s team at Southeast University, has successfully developed a wafer-level glass interposer technology that meets precise packaging design requirements. The first sample is a 2.5D module for AI acceleration featuring a domestically developed 7nm GPU core, four HBM2E memory chips delivering over 460GB/s bandwidth, and a 400μm thick TGV-based interconnect layer—providing strong hardware support for high-performance computing.
Performance simulations show that the TGV structure has lower insertion loss compared to traditional through-silicon via (TSV) technology. Even under high bit-rate conditions where eye diagrams degrade, TGV maintains better signal integrity than TSV. The technology achieves key advancements in four areas: micro-vias of 62.5μm diameter with a 7:1 aspect ratio and three times higher interconnect density; ultra-fine RDL lines at ≤2μm width/spacing supporting up to 10⁴ I/O/mm² for HBM integration; low warpage control below 50μm on 300mm wafers with a yield exceeding 97%; and exceptional RF performance with glass dielectric loss (Df) as low as 0.004 at 10GHz.
As a leading player in mid-to-high-end semiconductor packaging, Xinde Semiconductor offers comprehensive services across Bumping, WLCSP, Flip Chip PKG, QFN, BGA, SIP, SIP-LGA, and 2.5D packaging. This TGV breakthrough highlights the company’s strong technical capability and innovation in advanced packaging solutions. It not only strengthens China's domestic high-end semiconductor supply chain but also enhances its global competitiveness in next-generation chip technologies.
Traditional silicon-based interposers offer excellent electrical performance and high interconnect density but come with manufacturing costs exceeding 30% of total packaging expenses and complex production processes that limit widespread adoption. Organic interposers are cheaper but suffer from high signal loss at high frequencies, poor thermal expansion matching with chips, and susceptibility to warping—making them unsuitable for demanding high-density applications.
Glass-based TGV interposers provide a balanced solution with superior performance and cost-effectiveness. They offer extremely low high-frequency insertion loss, excellent mechanical strength, near-matched thermal expansion coefficients to silicon chips, and significantly improved manufacturability. This makes TGV ideal for next-generation AI computing, HPC, 5G millimeter wave, and optical communication systems.
Xinde Semiconductor, in collaboration with Stallone’s team at Southeast University, has successfully developed a wafer-level glass interposer technology that meets precise packaging design requirements. The first sample is a 2.5D module for AI acceleration featuring a domestically developed 7nm GPU core, four HBM2E memory chips delivering over 460GB/s bandwidth, and a 400μm thick TGV-based interconnect layer—providing strong hardware support for high-performance computing.
Performance simulations show that the TGV structure has lower insertion loss compared to traditional through-silicon via (TSV) technology. Even under high bit-rate conditions where eye diagrams degrade, TGV maintains better signal integrity than TSV. The technology achieves key advancements in four areas: micro-vias of 62.5μm diameter with a 7:1 aspect ratio and three times higher interconnect density; ultra-fine RDL lines at ≤2μm width/spacing supporting up to 10⁴ I/O/mm² for HBM integration; low warpage control below 50μm on 300mm wafers with a yield exceeding 97%; and exceptional RF performance with glass dielectric loss (Df) as low as 0.004 at 10GHz.
As a leading player in mid-to-high-end semiconductor packaging, Xinde Semiconductor offers comprehensive services across Bumping, WLCSP, Flip Chip PKG, QFN, BGA, SIP, SIP-LGA, and 2.5D packaging. This TGV breakthrough highlights the company’s strong technical capability and innovation in advanced packaging solutions. It not only strengthens China's domestic high-end semiconductor supply chain but also enhances its global competitiveness in next-generation chip technologies.
Way closer than most Western pundits thinks.
Supposedly, SMIC — China’s only advanced-node foundry — is using next-gen lithography machines from the Chinese Academy of Sciences (CAS) and SMEE for production.
IXBT reports that the Moore Threads cards with the latest driver software are much better at gaming than at launch. While previously they only supported DX9 properly they now also support DX10 and DX11. Game performance is also like 70% better in terms of FPS. Sometimes more.
But DX12 and Vulkan support is still lackluster. And the actual performance is still way below what the hardware specs should be able to achieve.
Still they think the drivers are headed in the right direction. Even if slowly. It took two years to get to this point.
Note the test isn't the latest driver, the newest is 310.120.0, released last month, they didn't specify the one they tested, only 300.110, it's not clear since there are 300.110.0, 300.110.1, and 300.110.2, if they tested 300.110.0 there's already been 3 months and 3 driver updates
This will be the pain point for Chinese hardware makers getting their driver stuff optimised. Kind of surprised their DX12 and Vulkan support are still lackluster given those API need a lot less state management from the drivers.
broadsword
Brigadier
This was so outdated already.
Saimai Testing and Control completed nearly 100 million yuan in Series A financing, and is a semiconductor RF test equipment manufacturer
Recently, Suzhou Saimai Measurement & Control Technology Co., Ltd. (hereinafter referred to as "Saimai Measurement & Control") completed a nearly 100 million yuan Series A financing, jointly invested by October Capital, old shareholder Yida Capital, Yuanhe Houwang and others.
According to public information, Saimai Testing & Control was established in 2021 and is headquartered in Suzhou International Science and Technology Park (SISPARK). It is a national high-tech enterprise engaged in the research, development, production and sales of modular test and measurement instruments and dedicated RF test systems. It is committed to providing customers with a complete modular test and measurement platform and system solutions. Its products and solutions are widely used in semiconductors, consumer electronics, scientific research and other fields.
The SGR RF wideband transceiver, independently developed by Saimai Measurement & Control, is based on the open PXI architecture and combines compact size, high test efficiency, and the exceptional performance and reliability of a benchtop instrument. It is widely used in R&D and production testing in fields such as semiconductors, T/R components, satellite communications, and advanced communications prototype verification. Its newly launched SNA vector network analyzer, launched in 2025, features high precision, multi-port functionality, and flexible scalability. It excels in applications such as 5G communications, semiconductors, and aerospace, significantly improving multi-port test efficiency and consistency, helping customers achieve high-quality, large-scale product testing and verification
