美国芯片企业重金押注研发

AI 资金为半导体研究实验室注入强劲动力 - IEEE Spectrum
IEEE Spectrum 11 月刊已上线!
美国芯片企业重金押注研发
致技术内行
IEEE Spectrum 是 IEEE 的旗舰出版物,而 IEEE 是全球最大的致力于工程与应用科学的专业组织。我们的文章、视频和信息图表为读者提供技术、工程与科学领域的最新动态。
创建账户,畅享更多免费内容与权益
保存文章以便稍后阅读需要 IEEE Spectrum 账户
The Institute 的内容仅对会员开放
下载完整 PDF 期刊仅限 IEEE 会员
下载本电子书仅限 IEEE 会员
访问 Spectrum 数字版仅限 IEEE 会员
关注话题是 IEEE 会员专享功能
加入全球最大的致力于工程与应用科学的专业组织,即可访问 Spectrum 的全部文章、档案、PDF 下载及其他权益。了解更多关于 IEEE 的信息 →
加入全球最大的致力于工程与应用科学的专业组织,即可获得本电子书,以及 IEEE Spectrum 的全部文章、档案、PDF 下载及其他权益。了解更多关于 IEEE 的信息 →
免费访问数千篇文章
创建账户即可获得专属内容与功能:保存文章、下载合集、发表评论——全部免费!如需完整访问权限与更多权益,请订阅 Spectrum。
美国芯片企业重金押注研发
AI 热潮正在推动更快的反馈循环
Lucas Laursen 是 IEEE Spectrum 的技术政策编辑。
在 Applied Materials 的一间洁净室里,一名技术员正在操作前开式晶圆传送盒(FOUP),这种容器用于在加工与制造工序之间运送晶圆。
10 月 12 日,Applied Materials 在加州桑尼维尔——硅谷的核心地带——揭幕了一座耗资数十亿美元的研究设施。该公司曾在 2023 年首次宣布,计划向其设备与工艺创新及商业化中心(EPIC Center)投资 40 亿美元,但现在表示投资额还将更高:约 50 亿美元。这一加码是私人投资更广泛增长的一部分,这种增长既在扩大美国的半导体研发板块,也使其变得更加复杂。
EPIC 的这一宣布,紧随两家半导体公司 8 月的类似动作之后。存储芯片制造商 Micron 披露,将在 10 年内向其位于爱达荷州博伊西的存储研究实验室投资 100 亿美元;SK Hynix 则表示将斥资 40 亿美元,在印第安纳州西拉法叶建设一座先进封装设施。(这两家公司均未回应 IEEE Spectrum 关于这些资金中有多少将用于研发的提问。)
美国联邦政府一直鼓励总部设在友好国家的芯片巨头——例如 TSMC 和 Samsung——在美国投资半导体研究。它还对境外零部件加征关税,并把部分自有资金投入到 Intel 等陷入困境的美国本土芯片企业。但一些分析师和高管认为,当今这波私人投资浪潮背后的真正动机,是企业预期能从当前这轮 AI 热潮中获取巨额利润。塔夫茨大学的 Chris Miller 是《芯片战争》一书的作者,他说:“最好的解释大概与政策或政治无关,而更多是 AI 需求推动的营收增长大幅跃升。”
AI 资金与工具加速芯片制造
Applied Materials 首席战略官 Tristan Holtam 也持相同看法:“这真的全靠 AI。”他说,该公司估计,得益于 AI 需求,前十大半导体公司的营收自 2021 年以来增长了约 150%,而这些公司反过来将研发投入提高了 67%。Holtam 补充说,这些头部半导体公司今年合计投入 670 亿美元用于研发,令联邦资金相形见绌。联邦资金大部分来自 2022 年的《芯片与科学法案》,该法案在数年间拨款 110 亿美元用于研发。
面向 AI 的芯片需求为更多芯片研发提供了资金和动力,而作为研究工具的 AI 也在帮助真正从事研发的人。总部位于马萨诸塞州比勒里卡的半导体材料科学公司 Entegris 正将 AI 应用于其早期研究与设计,该公司首席创新官 Olivier Blachier 表示。在他所在的公司,员工正在为下一代产品定义性能目标,并让 AI 工具推荐有望匹配这些目标的新材料。他说:“如果你是 AI 工程师、化学家或材料科学家,你可以把科学理解与 AI 驱动的建模和仿真结合起来,从目标性能要求反向推导。”该公司还在试点用 AI 对潜在新材料进行初步筛选,这类似于制药行业在实验室尝试合成和评估物质之前对新分子所做的筛选。
在 Applied Materials,Holtam 表示公司正在利用 AI 加速新材料的发现、设备的生成式设计以及算法和软件开发。他宣称:“对 EPIC 来说,时机再好不过了。”EPIC 是 Applied Materials 新设立的实验室,专注于设备与工艺。
一幅艺术家渲染图描绘了 Applied Materials 耗资 50 亿美元新建的 EPIC 研究中心,该中心于 2026 年 10 月 12 日启用。Applied Materials
重塑芯片研发,而不仅仅是调整其规模
物理上的邻近性正成为芯片制造商的一大关键优势,这正在重塑该行业在地理布局和设施投资上的取向。例如,日本半导体封装公司 Resonac 于 4 月在美国开设了其首个研发设施。Miller 表示:“这些公司有非常充分的理由希望靠近自己的客户,并靠近客户的客户,从而为未来 5 到 10 年做出正确的投资。”
关于原型芯片制造设备如何运作的知识,并不像软件那样容易在供应链中上下传递。“把全新技术交到我们客户——也就是芯片制造商——手中的传统做法,是把一台原型设备从我们的实验室运到他们的试产线。这是一种让人提心吊胆的局面,”Holtam 说。
“这一直是一个巨大的争议点,”他补充道。“当芯片公司拼命催促时,我们愿意在什么样的成熟度水平上允许一台原型设备离开我们的大楼。”应用材料(Applied Materials)分析了自身 15 年的内部研发,以找出哪些环节可能因减少这类惊心动魄的运输、增加并肩协作的实验室工作而受益。随后,它试点了“客户进实验室”的做法,并取得了 Holtam 所称的“非常、非常令人鼓舞的结果”。
这在一定程度上促使应用材料公司邀请顶级客户入驻新的 EPIC 实验室。截至目前,TSMC、Samsung、Micron、Broadcom 等都已表示同意。即便是互为竞争对手的客户,也将在 EPIC 内拥有各自的设施,并配备数字防火墙、物理隔离屏障及其他安全措施。“我们正在消除交接环节,”Holtam 说,同时让自家的研究人员更容易向客户预告即将推出的技术。
应用材料公司明确的目标是,让其 EPIC 中心把从早期发现到芯片商业化所需的时间缩短数年。这一过程通常需要 10 到 15 年。
美国商务部曾根据《芯片法案》为广纳各方的半导体研发设施宣布了类似目标,但这些项目如今要么进度落后,要么被无限期搁置。在一个以高度集中著称的行业里,它们本可能帮助初创企业获得相对芯片巨头的先发优势。事实上,EPIC 迄今宣布的合作对象都是其他行业领军企业,因此初创公司或许只能排在队尾。Blachier 说,总体而言,目前“中小型企业必须与大公司合作,才能对接这些巨头”。
该行业寻求加速的另一条路径,是与大学建立更紧密的联系。大学也是通往美国国家半导体研究战略的桥梁,因为大学与联邦政府有着长期的研究资助合作关系。
“拥有更长期的视野以及获得支持的投入,是有好处的,”Blachier 说。他指的是欧盟和一些亚洲国家的国家芯片战略,这些战略催生了像 Imec 这样成功的竞争前研究设施。他说,美国若能拥有更多这类设施,“肯定会有用,而且往往能让整个生态受益。”
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(U.S. Chip Firms Bet Big on R&D)
AI Money Supercharges Semiconductor Research Labs - IEEE Spectrum
The November issue of IEEE Spectrum is here!
U.S. Chip Firms Bet Big on R&D
FOR THE TECHNOLOGY INSIDER
IEEE Spectrum is the flagship publication of the IEEE — the world’s largest professional organization devoted to engineering and applied sciences. Our articles, videos, and infographics inform our readers about developments in technology, engineering, and science.
Enjoy more free content and benefits by creating an account
Saving articles to read later requires an IEEE Spectrum account
The Institute content is only available for members
Downloading full PDF issues is exclusive for IEEE Members
Downloading this e-book is exclusive for IEEE Members
Access to Spectrum 's Digital Edition is exclusive for IEEE Members
Following topics is a feature exclusive for IEEE Members
Join the world’s largest professional organization devoted to engineering and applied sciences and get access to all of Spectrum’s articles, archives, PDF downloads, and other benefits. Learn more about IEEE →
Join the world’s largest professional organization devoted to engineering and applied sciences and get access to this e-book plus all of IEEE Spectrum’s articles, archives, PDF downloads, and other benefits. Learn more about IEEE →
Access Thousands of Articles — Completely Free
Create an account and get exclusive content and features: Save articles, download collections, and post comments — all free! For full access and benefits, subscribe to Spectrum .
U.S. Chip Firms Bet Big on R&D
The AI boom is driving faster feedback cycles
Lucas Laursen is the technology policy editor at IEEE Spectrum.
![]()
In an Applied Materials clean room, a technician handles a front-opening unified pod, which is used to transport wafers between processing and manufacturing steps.
On 12 October, Applied Materials unveiled a multibillion-dollar research facility in Sunnyvale, Calif., in the heart of Silicon Valley. The company first announced in 2023 plans to invest US $4 billion in its Equipment and Process Innovation and Commercialization (EPIC) Center, but now says it will invest even more: about $5 billion. The boost is part of a wider growth in private investment that is both enlarging the U.S. semiconductor R&D sector and making it more complex.
The EPIC announcement follows others, in August, by two semiconductor firms. Memory chipmaker Micron disclosed that it would invest $10 billion over 10 years at its Boise, Idaho, memory research lab, and SK Hynix said it will spend $4 billion on an advanced packaging facility in West Lafayette, Ind. (Neither of those companies responded to IEEE Spectrum’s questions about how much of those funds would be for R&D.)
The U.S. federal government has encouraged investment in U.S. semiconductor research by chip giants based in friendly nations, such as TSMC and Samsung. It has also enacted tariffs on foreign components and put some of its own money into faltering domestic chip companies including Intel. But the real motive behind today’s wave of private investment, according to some analysts and executives, is that companies anticipate enormous profits from feeding the current AI frenzy. “The best explanation is probably not about policy or politics and probably more about a big jump in revenue growth driven by AI demand,” says Chris Miller of Tufts University, and author of the book Chip War.
AI Money, Tools Accelerate Chipmaking
Applied Materials’ chief strategy officer Tristan Holtam agrees: “It’s really all about AI.” The firm estimates that the top 10 semiconductor companies’ revenue has grown about 150 percent since 2021 thanks to AI demand, and they have in turn raised their investment in R&D by 67 percent, he says. Those top semiconductor companies are collectively investing $67 billion this year in R&D, Holtam adds, dwarfing federal funding. Most of that federal money comes from the 2022 CHIPS and Science Act, which allocated $11 billion for R&D over several years.
While demand for AI-ready chips provides funding and motivation for more chip R&D, AI in the form of research tools is helping the people actually doing the R&D. Entegris, a semiconductor materials science company headquartered in Billerica, Mass., is applying AI in its early-stage research and design, says Olivier Blachier, the company’s chief innovation officer. At his company, workers are defining performance goals for next-generation products and asking AI tools to suggest promising new materials to match. “If you’re an AI engineer, a chemist, a materials scientist, you can combine scientific understanding with AI-driven modeling and simulation to work backward from a target performance requirement,” he says. The company is also piloting AI for preliminary screening of potential new materials, similar to what the pharmaceutical industry does with new molecules before trying to synthesize and evaluate materials in the lab.
At Applied Materials, Holtam reports that the company is making use of AI-enabled acceleration in new material discovery, generative design of equipment, and algorithm and software development. “The timing is perfect for EPIC,” Applied Materials’ new lab devoted to equipment and processes, he declares.
An artist’s rendering depicts Applied Materials’ new $5 billion EPIC research center, which was opened on 12 October 2026.Applied Materials
Reshaping Chip R&D, Not Just Resizing It
Physical proximity is emerging as a key advantage for chipmakers, and that is reshaping the industry’s geographic and facilities investments. The Japanese semiconductor packaging company Resonac opened its first R&D facility in the U.S. in April, for example. “These companies have a very strong reason to want to be close to their customers and close to their customers’ customers so they can make the right investments for the next five to 10 years,” according to Miller.
The knowledge of how prototype chipmaking equipment works does not move up and down the supply chain as easily as software. “The traditional way of working to get brand new technology into the hands of our customers, the chipmakers, was to ship a prototype piece of equipment from our lab into their pilot line. That’s a kind of heart-in-your-mouth type of situation,” Holtam says.
“That was always a huge point of contention,” he adds. “At what level of maturity we’re willing to allow a prototype outside of our building when the chip company is pulling on that as hard as possible.” Applied Materials analyzed 15 years of its own internal R&D to look for the places likely to benefit from fewer such heart-stopping shipments and from more shoulder-to-shoulder lab work. Then it conducted pilot programs of its customer-in-the-lab approach and got what Holtam deemed “very, very encouraging results.”
That helped prompt Applied Materials to invite top customers to set up shop inside the new EPIC lab. So far, TSMC, Samsung, Micron, Broadcom, and others have said yes. Even competing customers will get their own facilities under EPIC’s roof, with digital firewalls, physical barriers, and other security features. “We’re eliminating handoffs,” Holtam says, and making it easier for their own research staff to give customers a preview of what’s coming.
Applied Materials’ stated goal is for its EPIC center to reduce, by several years, the time it takes to go from early-stage discoveries to chip commercialization. That’s typically a 10- to 15-year process.
The U.S. Department of Commerce announced similar goals for big-tent semiconductor R&D facilities under the CHIPS Act, but those are now behind schedule or indefinitely paused. They might have helped startups get a leg up on the chip giants in a notoriously concentrated industry. Indeed, EPIC’s partnership announcements so far have been with other industry leaders, so startups may be at the back of the queue. Right now, generally, Blachier says, “small and medium companies have to cooperate with bigger companies to feed into the giants.”
Another way the industry is seeking acceleration is by closer ties with universities. Universities also serve as a bridge to the U.S. national semiconductor research strategy since universities have long-running research funding partnerships with the federal government.
“There is a benefit in having longer-term horizons and supported investments,” Blachier says, referring to EU and some Asian countries’ national chip strategies, which have created successful precompetitive research facilities such as Imec. More facilities like that in the U.S., he says, would be “definitely useful, and tend to benefit the ecosystem.”
Related Articles Around the Web
原文为英文,正文由自动抓取并保留原文。
原文发布于 IEEE Spectrum Semiconductors(U.S. Chip Firms Bet Big on R&D),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。