Fri, Apr 14, 2023 page9

• Intel’s foundry bet might split the market in three

  • By Tim Culpan / Bloomberg Opinion
  • • 
    • 
    • 

  Intel Corp’s bold plan to enter the custom chipmaking industry is aimed at countering the dominance of Taiwan Semiconductor Manufacturing Co (TSMC) and boosting supply of leading-edge manufacturing. Instead, the move will likely split the market in three, leaving the US giant stuck in the middle.

  Chief executive officer Pat Gelsinger announced his intention to open up Intel’s fabrication plants (fabs) to external customers two years ago, accompanied by a US$20 billion investment into two new facilities in Arizona.

  Early last year, it added a further US$20 billion plan for a site in Ohio.

  “Our ambition is to be the No. 2 foundry in the world by the end of the decade,” Randhir Thakur, president of Intel Foundry Services, told Nikkei in November last year.

  That means overtaking either TSMC or close rival Samsung Electronics Co. In truth, this is not too hard if you fiddle with the definition of a semiconductor foundry. Samsung, for example, is considered to be the second-largest because market researchers include all the sales the South Korean giant gets from making chips that it designs and uses in its own devices.

  There is a logic to this calculation: If Samsung did not manufacture those chips, someone else would have. So by this standard, Samsung’s foundry revenue last year was US$30 billion, while Intel’s total sales was US$63 billion and TSMC’s US$67 billion.

  Being second by revenue share is not really the point, though. What Intel is also aiming for is to become a technology leader, which means developing advanced manufacturing processes that can compete with TSMC and Samsung. Right now, it is trailing by a few years.

  Once the global leader, the US company surrendered ground over the past decade as its rivals barreled ahead on research and development, and capacity expansion.

  For Intel to catch up it needs to not only advance quicker than TSMC in the race to the next process node, but then turn the ability to research the latest chip technology into producing high-volume output with a low rate of defects — known as yield.

  The pace of development, outlined by Moore’s Law, means that to just tread water with rivals requires moving to the next step every two years or faster.

  This complicated and increasingly expensive battle has left rivals in the dust. Two decades ago, fellow Taiwanese foundry United Microelectronics Corp (UMC) was considered neck-and-neck with TSMC in manufacturing technology. Today, its most-advanced node — at 28 nanometers — is a product TSMC first released 12 years ago.(1) UMC is the world’s third-largest foundry, ahead of Shanghai’s Semiconductor Manufacturing International Corp (SMIC) and Malta, New York-based GlobalFoundries Inc, both of which trail even further behind on technology.

  This leader-laggard paradigm has resulted in a stable bifurcation of the market. TSMC and Samsung dominate the top end, churning out chips used in smartphones, artificial-intelligence servers and cryptominers. Everyone else handles the low end, including components used in vehicles, smart speakers and industrial robots. About 50 percent of the foundry market consists of products made at 16 nanometers and smaller, an area almost exclusively occupied by those two leaders.

  An important aspect of this dynamic is that frontrunners consistently advance to the next node, while progress by laggards remains slow. Last year, for example, revenue for TSMC’s best technology (5 nanometers) jumped 79 percent and that of its prior best (7 nanometers) climbed 15 percent. That 5-nanometer product was not even available three years ago and the company already announced manufacturing of its 3 nanometers service in December last year.

  By contrast, UMC’s best offering, launched a decade ago, achieved just 14 percent revenue growth last year.

  Recent US curbs on access to manufacturing technology aims to halt any further advancement by Chinese companies such as SMIC and memorychip maker Yangtze Memory Technologies Corp. Yet it would not make much difference to UMC and GlobalFoundries, and they are still likely to fall further behind over the coming decade.

  Intel is likely to be caught in the middle. Even if Gelsinger manages to close the gap on TSMC and Samsung in manufacturing technology, as his own aggressive development road map predicts, he needs to convince the world’s most important chip customers that Intel can be trusted not only with their designs, but also to deliver products in volume, on time and with minimal defects.

  Luring leading-edge customers is an important part of the development process, because the supplier and customer work closely together, learning from each other in a relentless battle against the clock.

  Today, the largest buyers of advanced semiconductor foundry capacity include Apple Inc, Nvidia Corp and Advanced Micro Devices Inc (AMD). The maker of iPhones dumped Intel as a supplier of Mac processors two years ago, while Nvidia and AMD are fierce competitors in the market for high-performance computing such as servers.

  Instead of catching up to and maintaining pace with rivals year after year, there is a possibility Intel will drop back by a few development cycles within half a decade. That is when we would see a trifurcation of the market, with Intel at neither the front nor the back. This mushy-middle could be challenging to navigate: Intel would not be able to charge the same as the leaders and would not enjoy the low-cost advantages of the laggards.

  There could still be a profitable market beyond supplying to itself if Intel can convince strategic customers such as the administration of US President Joe Biden and security-sensitive businesses in defense, aerospace and data management that only a US company with US-made chips can be trusted.

  This strategy relies on Intel persuading such customers that bleeding-edge technology is not necessary — and quite often it is not — with almost-leading being sufficient for the task.

  That means Intel might need to rely not only on leaps of faith and technological breakthroughs, but savvy salesmanship and the good graces of the US government.

  (1) UMC has a 22-nanometer node, but this is considered only a minor increment from 28 nanometers.

  Tim Culpan is a Bloomberg Opinion columnist covering technology in Asia. Previously, he was a technology reporter for Bloomberg News. This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
• Most Popular

  • 1

    Taiwan thanks Japan for family registry change

  • 2

    MAC warns of PRC arrests, detentions

  • 3

    Taiwan blacklists two Chinese schools

  • 4

    Inter denied top spot in Serie A by rivals Juve

  • 5

    US exempts security funds for Taiwan