A bill releasing nearly $50 billion to boost U.S. chip production has sparked euphoria among chipmakers. But that joy was short-lived when the United States banned the shipment of advanced chips to China this month.
Export controls aim to stifle China’s computing advancements by denying access to US-sourced chip design software, silicon and semiconductor manufacturing equipment. Nvidia, which has a strong presence in China, said export controls would affect its GPU business as it would not be able to ship its A100 and H100 GPUs to market.
Export controls complement U.S. government efforts to bolster the domestic semiconductor ecosystem through the U.S. CHIPS and Science Act of 2022. The bill opens up $39 billion in funding to companies like Intel, TSMC and Texas Instruments to open fabs on American soil. About $11 billion will be used to boost R&D, education and workforce development programs in the semiconductor industry and academia.
Intel executives at last month’s innovation show acknowledged that Chinese companies are important customers and are building a manufacturing strategy to be geographically diverse. This plan will help it continue to serve its customers in China.
The global expansion will help the company quickly adapt to geopolitical changes and continue to supply chips without breaching local regulations and restrictions. Regional supply chains will also act as a fail-safe to avoid chip shortages and ensure multiple sources of critical chipmaking materials and tools.
The US, EU and China view chips as an important pillar for advancing economies in an increasingly digital world. The shortages exposed weaknesses in the global semiconductor supply chain, with manufacturing largely concentrated in China, Taiwan and Korea. Chip prices have risen, and low-cost chips such as power management circuitry have led US and European automakers to shut down their production lines.
The chip shortage has prompted the US and EU to strengthen regional chip supply chains. The United States has passed the CHIPS and Science Act, and the European Commission is proposing its own CHIPS Act to stabilize the local semiconductor ecosystem. China and the EU have invested heavily in the development of sovereign chips based on the open-source RISC-V architecture, which will reduce reliance on the proprietary US-based Arm and x86 architecture.
Intel has restructured its manufacturing around geographic hubs in strategic areas. In September, Intel opened $20 billion factories in Ohio, which CEO Pat Gelsinger called “Silicon Heartland.” The company is investing nearly 17 billion euros in new factories in Magdeburg, Germany, dubbed “Silicon Junction”. The company is also spending $7 billion on a new test and assembly facility in Malaysia and upgrading its manufacturing facilities in Israel. India courted Intel to open a factory, but the company rejected the offers.
The breadth of Intel’s geographically distributed manufacturing was exemplified by a representative at the Intel Foundry Services booth at the Innovation Show last month. Researchers or companies will be able to use Intel’s services to have chips manufactured at its regional facilities while circumventing issues such as export restrictions. Intel will adapt within the regional framework governing chip design and deployment.
For example, the European Processor Initiative – which is funded by the European Commission and designs the RISC-V architecture – will be able to have a chip manufactured at Intel’s factory in Europe. PPE participants, like SiPearl, have expressed interest in using advanced chip manufacturing facilities on European soil because chips can be delivered quickly.
Intel is helping to create a chip made in Europe, which increases the company’s chances of receiving factory subsidies from the European Commission’s EU Chip Act. Intel is already helping the Barcelona Supercomputing Center – part of the European Processor Initiative – design chips based on RISC-V.
China’s export restrictions, which were announced shortly after Intel’s innovation show, are far-reaching, and chip company executives said HPCwire that there is a lot of confusion around the policies and they are still wondering how it will apply to their operations.
But the manufacturing-specific limitations are clear: companies cannot export more advanced chips to China than 16nm or 14nm chips with 3D structures, 18nm DRAM chips, and NAND flash chips with more than 128 layers.
Beyond the limitations, Intel could work with Chinese low-end chip organizations from their facilities in Asia. But Intel may not be a manufacturing option for Chinese companies, which make RISC-V processor designs that require factories using American technologies. The Chinese Academy of Sciences is developing a RISC-V server chip named XiangShan. Chinese cloud provider Alibaba is also supporting the development of RISC-V chips.
Gelsinger is betting that its long-term foundry strategy, called Integrated Device Manufacturing (IDM) 2.0, will be judged on the technological merits. Export restrictions could change with a new ruling party after the 2024 U.S. election, and Gelsinger doesn’t see any near-term distractions.
“We have to manage our cash carefully, but we are investing for the long term. It’s our strategy for building process technologies – for undisputed leadership, five nodes and four years, as we’ve called it. We are building the capacity to accelerate these technologies,” Gelsinger said at a press conference during Innovation.
Signals of Intel’s geographic manufacturing separation can be seen with a new consortium on the production and design of chips for US national security applications.
The United States Military, Aerospace and Government (USMAG) alliance, which was announced Tuesday and is led by Intel Foundry Services, is designed to create processes that “meet stringent design and production requirements” for military customers, U.S. aerospace and government agencies, the chipmaker said in a statement.
The alliance indicates a greater effort to create political inroads to secure more foundry business from US customers. The program could be a pathway for similar programs he is creating in Europe and China.
Header image: A photo from November 2021 shows employees working at Intel’s D1X factory in Hillsboro, Oregon. (Credit: Walden Kirsch/Intel Corporation)