Early on Thursday, a group of US chip designers and manufacturers sent a letter to the White House, asking that the government include “substantial funding for incentives for semiconductor manufacturing” as part of the overall COVID-19 economic recovery plan. The Biden Administration has now pledged to take action to help remedy the situation by “identifying choke points in supply chains.” President Biden will sign an executive order directing a government-wide review of supply chains for critical goods.
Both the request and the Biden Administration’s response are mostly grandstanding. There is no practical way that any action taken by the Biden Administration is going to have a near-term impact on silicon supplies. Increasing foundry capacity takes months to years, not weeks. TSMC has already stated that while it will allocate more space for automotive production, it will be taking that space away from other customers. The leading-edge foundries are shipping every wafer that they can.
According to the 21 CEOs who signed the letter, including those representing Intel, AMD, Micron, and Qualcomm, the reason the United States has not retained a larger share of the global chip manufacturing market is because “the governments of our global competitors offer significant incentives and subsidies to attract new semiconductor manufacturing facilities, while the U.S. does not.” This is the supposed explanation for why the United States manufactured 37 percent of the world’s semiconductors in 1990, but just 12 percent today.
This is ridiculous twaddle. Samsung has literally just proposed building a foundry in Austin. In return for this grand act of corporate benevolence, it wants the county to hand over a 20-year 100 percent tax abatement. It wants the city of Austin to provide a 50 percent abatement over the same period of time, for a total value of over $800 million dollars. Not content with that, Samsung also wants to be excused from the estimated $252 million it would pay in school taxes over the same period. If it succeeds, the state of Texas will be responsible for making up the shortfall to the school district, leaving taxpayers literally on the hook.
The US federal government may not, as a rule, provide enormous incentive packages. That doesn’t change the fact that Samsung feels perfectly fine asking for over a billion dollars in tax relief at a time when it earned more than $34 billion in profit the previous year. Intel set revenue records last year. It may be true that other companies provide aggressive support for silicon manufacturing at the federal level, but silicon manufacturers clearly have no qualms about demanding special treatment.
There are absolutely things the Biden Administration could do to encourage greater semiconductor manufacturing in the United States, but simplistically tying the drop in US semiconductor market share to the presence or absence of government subsidies isn’t persuasive. This chart of leading edge foundries over time is useful:
Back at 90nm, when the cost of advancing to a new node was far smaller and chip designs cost a fraction of what they do today, there were a lot more companies on the leading edge — and most of them weren’t in the US. Of the 18 companies listed in the 90nm column, only Freescale, Texas Instruments, IBM, AMD (GlobalFoundries) and Intel were American companies. Just five, out of 18 firms — and that was nearly twenty years ago.
The reason why the United States accounts for just 12 percent of chip manufacturing today is because partly because the first and most successful pure-play foundry was founded in Taiwan. When mobile chip designers like Qualcomm needed someone to manufacture their chips, they turned to the likes of TSMC and Samsung. The subsequent explosion in mobile SoCs and now AI and edge processors has favored the countries where massive pure-play foundries were established. Intel tried to adopt the foundry model and collect its own stable of customers, but the effort was unsuccessful and the company may have quietly abandoned it.
The reason you see companies leaving the leading edge with every generation is twofold. First, the cost of new foundry upgrades and chip designs rises every generation. Not all types of transistors benefit from new nodes, and not all chips sell in high enough volumes to justify node transitions. Plenty of companies are like GlobalFoundries: Off the leading edge and earning a tidy profit.
The second reason is that the number of customers available at any given node has historically shrunk from one node to the next. This may have changed recently, given the sudden influx of spending from a lot of various AI companies propped up with VC dollars, but for most of the past two decades, fewer and fewer companies have jumped to the leading edge with every generation. With fewer customers available, higher costs, higher design costs, and smaller gains from each passing generation, we’ve seen repeated waves of consolidation in the foundry industry. When GlobalFoundries announced it was leaving the leading edge a few years ago, it didn’t blame a lack of subsidies. It blamed a lack of customers and an inability to make the math work when AMD was its only public big-name 7nm customer.
If the Biden Administration can find supply chain bottlenecks it can address, like boosting the supply of raw materials used to manufacture components, that would be useful, but the chance of a near-term improvement to the situation is probably nil. Building new factories takes time. Bringing new capacity online takes time. There are no quick solutions in semiconductor manufacturing, no matter how much the auto industry squawks to Congress.
Intel Tops Semiconductor Revenue, as TSMC, Nvidia, and AMD Vault Upwards
The 2020 semiconductor revenue forecast is in, and pretty much everyone is a winner.
Intel CEO: Reliance on Asia for Semiconductor Manufacturing Isn’t ‘Palatable’
This push for a more balanced global supply chain is part of Intel's expansion into the client foundry business and the company clearly hopes to capitalize on the perception supply chain issues are the result of too much capacity in the Asia-Pacific region.
Why We Can’t Build Our Way Out of the Semiconductor Shortage
Eight months on, the semiconductor shortage seems likely to stretch into 2022. Increasing chip production is a slow and difficult business.