择天记

SK海力士将采用英特尔EMIB封装技术：打造下一代HBM_我的网站

一 |     

    Illustration: Liu Xiangya/GT
Recent media reports have questioned whether a natural gas plant built to power an Amazon data center project in Texas could become the largest climate polluter in the US. The controversy, whatever the eventual outcome, offers a reality check for America's artificial intelligence (AI) drive. 
It exposes a growing contradiction: The US is racing to expand its AI capabilities, yet its protectionist trade policies are making it harder and more costly to access some of the clean-energy technologies needed to sustain that expansion. This raises a broader question: Can an energy-intensive AI race afford the costs of renewable energy protectionism?
The US is entering a new era of rising electricity demand. Data centers, the backbone of the AI economy, are emerging as one of the fastest-growing sources of power consumption. Much of that demand is still being met by fossil fuels: The International Energy Agency reports that natural gas supplies more than 40 percent of the electricity used by data centers in the US, making it their largest source of power. 
So, it's not surprising that the expansion of data centers has raised concerns over their environmental impact and the pressure they could place on local power systems and electricity bills. A Gallup survey conducted in March found that seven in 10 Americans opposed the construction of AI data centers in their local area, including 48 percent who strongly opposed such projects.
The findings point to a broader challenge for the US: The race to develop AI is increasingly becoming a race to meet growing energy needs. Addressing this challenge will require more than advances in computing technology; it will also depend on an energy system capable of delivering large amounts of reliable, affordable and cleaner power. That, in turn, will require faster development and broader deployment of clean-energy technologies, from solar power to energy storage.
Yet in the clean-energy sector, the US has increasingly relied on protectionist trade measures that limit access to cost-competitive products from global markets. The country has placed greater emphasis on expanding domestic manufacturing capacity, but rebuilding entire clean-energy supply chains at home is a costly and time-consuming process. Even if expanded domestic production is achieved, it is likely to come at a higher cost, making the deployment of renewable technologies more expensive and potentially slower.
The solar industry offers a clear illustration of this policy direction. The US has continued to expand trade barriers in the sector. Reuters reported that the US government announced on Thursday a series of price floors and a 15 percent tariff on products made from polysilicon, a raw material used in solar panels.
The challenge lies in the limited scale of the US polysilicon industry. Reuters reported that the country has two polysilicon factories. Against this backdrop, relying on domestic polysilicon production while restricting access to imports runs counter to the goal of expanding solar power in the US. The country risks creating barriers that ultimately constrain its own access to the global supply chains needed for growth.
The pressing issue for the US is the speed at which new power demand is emerging. The expansion of data centers is creating electricity needs that cannot wait for domestic clean-energy capacity to develop gradually. Global supply chains can provide the scale and speed required in the near term. By narrowing access to these sources, the US risks turning clean-energy policy into a drag on the infrastructure needed for its AI race.
The US has placed AI high on its economic and technological agenda. The outcome of this race will matter greatly, as financial markets are also watching whether America can turn its AI efforts into commercial success.
This leaves the US with a difficult choice: Can it afford the cost of clean-energy protectionism while racing to build AI infrastructure? The answer may be no. Trade barriers that limit access to competitive renewable technologies could ultimately become a constraint on the AI expansion that Washington is seeking to accelerate.
The author is a reporter with the Global Times. [email protected]

。    8月24日消息，据Wccftech报道，SK海力士近日详细阐述了其HBM技术路线图的加速路径，核心举措包括引入EMIB封装方案，并明确将3D集成作为长期目标。     当前，HBM通过TSV（硅通孔）技术将多层DRAM芯片堆叠于基础裸片之上，最高可支持16层堆叠。

二 | HBM与计算模块（XPU，涵盖GPU、TPU等）相互独立，通过2.5D封装共同安装于同一中介层。    

    每个HBM模块配备1024位I/O接口（共16通道），通过物理接口层（PHY）与XPU相连。而下一代HBM4将首次将I/O接口扩展至2048位，这将是自2015年HBM问世以来的最大规格变革。     在封装工艺上，目前主要存在两大技术路径：热压键合配合非导电膜（TC+NCF）和整体回流焊配合模塑底部填充（MR+MUF）。前者在应对芯片翘曲方面更具优势，但热阻较高且生产效率偏低；后者生产效率更高、热阻更低，却更易出现翘曲及间隙填充缺陷。

三 |     

    面向未来，SK海力士计划引入混合键合技术，以消除芯片堆叠中的凸块并缩小间距。同时，公司正在开发名为“I-HBM”的局部散热技术（类似三星的HPB方案），旨在优化传热路径，更高效地分散热量，为更高层数的堆叠铺路。    在2.5D封装解决方案方面，SK海力士目前已在传统CoWoS-L、CoWoS-R和CoWoS-S之外，新增了英特尔的EMIB技术，形成更丰富的异构集成选项。     此外，市场传言SK海力士与英特尔有望在存储领域组建合资公司，进一步深化合作。同时，与三星类似，SK海力士也在积极推进3D垂直堆叠方案，将HBM直接置于计算模块之上，以突破现有封装架构的带宽和能效瓶颈。

四 |     

         

。

Current article:http://f3v.michengyuraochixun.buzz/d0c3c4/1vio.html

Published on:00:00:00