Your browser is not supported. Please update it.

20 September 2022

Google to fund open source US chips to boost R&D and start-ups

The USA and China are engaged in a battle to gain self-sufficiency from one another in strategic technologies, and also to establish global leadership. Much of this activity has focused on semiconductors. China has a huge decade-long programme to build up key chip capabilities and IPR, and recently, the USA’s CHIPS Act set aside funds worth $280bn to increase national capabilities in chip foundries and innovation.

One way to increase a base of innovation and technology is to adopt open approaches, and despite the risks this entails in terms of IPR ownership (as highlighted by O-RAN), a degree of open source is increasingly popular with governments looking to accelerate hi-tech progress. The open source processor architecture, RISC-V, is gaining some ground against the incumbent Intel and ARM platforms, though much of that development is being led by Chinese firms.

And now, the US government’s National Institute of Standards and Technology (NIST) is collaborating with Google to co-develop licence-free, domestically produced chips to support technology and product R&D. The semiconductor industry may have been built on proprietary architectures and giant commercial companies like Intel, but there is a more complex set of dynamics at work in strategic areas of technology such as 5G/6G, advanced AI and cloud processing.

The hyperscalers have all developed some of their own semiconductor designs, working with various partners. At a time when computing and telecoms equipment makers were moving away from inhouse designs to merchant processors, Google, Microsoft, Meta and AWS were pulling in the other direction, and had the scale to do it. Their objective was to drive down cost and power consumption, and rather than defending their proprietary architectures for reasons of competitive edge, they were more likely to seek even greater economies of scale by opening up the technology, as seen in Meta’s Open Compute Project or Google’s AI-oriented tensor processing unit (TPU).

This latest initiative is not about commercial platforms, but a way to provide lower cost advanced chips to researchers and start-ups, to encourage their innovations. The chips will be manufactured by SkyWater Technology’s foundry in Bloomington, Minnesota. Skywater will produce 200-millimeter wafers, which customers can then “dice into thousands of individual chips at their own processing facilities”, NIST said.

NIST, working with university researchers, will design open source circuitry for chips that will be available “without restriction or licensing fees”, according to NIST. Google will pay for the upfront production costs and subsidize the first production run.

NIST aims to design as many as 40 different chips for various applications under this open source scheme, saying that “researchers will be able to pursue new ideas without restriction and share data and device designs freely”. Universities that will contribute to the designs include Brown, Carnegie Mellon, George Washington, the University of Maryland and the University of Michigan.

“Large companies that design and manufacture semiconductors often have ready access to these types of chips,” NIST said in a release. “But the cost can run into the hundreds of thousands of dollars, presenting a major hurdle to innovation by university and start-up researchers. By increasing production to achieve economies of scale and by implementing a legal framework that eliminates licensing fees, the collaboration is expected to bring the cost of these chips down dramatically.”

NIST is hosting a virtual workshop this week on the use of chips in measurement science and prototyping, which will include a public working group meeting on its collaboration with Google.

Of course, such initiatives do not apply to the most advanced semiconductor designs and geometries, which are at the heart of the rivalry with China and the desire to place more foundries on US home soil (the European Union has similar, if somewhat less-well funded plans). Currently, more than 60% of the world’s chips are made in Taiwan, many by its leading foundry, TSMC. The US government has taken a carrot-and-stick approach to TSMC, luring it with large grants and subsidies to built out more foundry capacity in the USA, while also barring it from supplying its former second-largest customer, Huawei, at the same time as supplying US customers, which include its largest, Apple, and others such as Qualcomm.

The CHIPS Act provides substantial extra funding for chipmakers to expand US-based capacity, whether local players like Intel Foundries, or TSMC and Samsung, the world’s leading players. Some of the new or expanded facilities that are planned in the USA in the next 5-7 years include the following (list compiled by FierceElectronics):

  • GlobalFoundries and Qualcomm:  An expanded GF footprint in New York has been boosted by Qualcomm’s commitment to buy a guaranteed $7bn in chips between now and 2028, for various applications in 5G, automotive, WiFi and IoT.
  • Intel: $20bn for two fabs in central Ohio, under construction in summer 2022, to make advanced logic chips. Intel also broke ground in 2021 on two new fabs in Chandler, Arizona, a $20bn investment, and an advanced packaging facility is being constructed in New Mexico.
  • Micron:  A $15bn fab for its Boise, Idaho HQ to make memory chips. Active prospects for billions of dollars more in fabs, including near Austin, Texas.
  • Samsung: A $17bn fab in Taylor, Texas, was announced in late 2021. The South Korean company has set out preliminary long-term plans for 11 fabs in Texas to be built over the coming years, in a total investment plan worth nearly $191bn.
  • Texas Instruments: A potential $30bn investment plan includes four fabs in Sherman, Texas. Ground was broken in May for new 300mm wafer fabs.
  • TSMC:  TSMC has several US fabs, including one in Camas, Washington, and a new $12bn facility is under construction in Phoenix, Arizona to make chips using the 5nm process, though its work on the 3nm process is still centered mainly on Taiwan.
  • Wolfspeed: $5bn for a silicon carbide wafer fab in central North Carolina.