Shifting silicon

MORRIS CHANG is preparing for retirement. After 30 years in the role, the founder of Taiwan Semiconductor Manufacturing Company (TSMC), the island’s largest firm, will step down as chairman in June. He will hand the reins over to the current co-CEOs, C.C. Wei and Mark Liu, the former becoming sole CEO and the latter chairman. Later that month the company will ship new semiconductors manufactured with its latest technology. For the first time the world’s most powerful chips will be made by TSMC, not by Intel, its American rival.

Intel and TSMC are different sorts of company. Intel is an integrated device manufacturer (IDM). It both designs and manufactures chips. TSMC is a “foundry”, making chips for designers without factories, or “fabs”, which cost a fortune. TSMC’s latest fab will cost $20bn. The Taiwanese company pioneered this model and is its dominant exponent. In 2017 it had 56% of the foundry market, according to Trendforce.

Intel led the pack in squeezing more computing power onto chips. The company turned Moore’s law—which states that computing power doubles every two years at the same cost—into a self-fulfilling prophecy. To do so they shrunk “nodes”, the width of the channel etched into silicon chips. The narrower the channel, the more computing power can be squeezed in. Intel currently makes chips using a ten-nanometre (billionth of a metre) node. TSMC’s new ones are made with a seven-nanometre node. TSMC’s rise to technological leadership is reflected in its valuation. In 2017, for the first time, its market capitalisation exceeded Intel’s.

How the company surpassed the king of chipmaking is hotly debated. It is hard and expensive to shrink nodes. Smaller firms have stopped trying. One reason may be that by 2017 TSMC was investing close to $3bn (8% of revenues) on research and development. Mr Liu claims TSMC spends more on node technology than Intel and Samsung, another IDM, combined.

The answer may also lie in the strength of the foundry model itself. Intel is renowned for making computer processors and Samsung for smartphone chips. TSMC serves both customers. It is ready to provide chips for new technologies as they arise. In 2017 crypto-currency miners brought in revenues of $1bn. Their rise was one “we truly did not anticipate,” says Mr Liu. As he puts it, the firm’s top five customers always account for roughly half of revenues, but the names change. This variety helps TSMC to innovate.

Chipmaking also now requires a close partnership between manufacturers and designers. Mr Liu describes this in painterly terms. “A decade ago a customer would design a simple pattern and the factory would make it for them. But current designs have many shades and colours.” That may benefit large entrenched players. Whereas switching from one foundry to another was once trivial, now companies work within the TSMC “ecosystem”’ for years before chips are manufactured. Crypto-currency firms like Bitmain, a Chinese hardware manufacturer, which has been collaborating with TSMC for three years, are among hundreds of companies it works with. To switch fabs requires companies to duplicate R&D invested in TSMC’s technology.

High switching costs may not be a product of technological complexity alone. GlobalFoundries, a smaller American competitor, argues that TSMC is deliberately increasing these costs, using loyalty rebates, exclusivity clauses and penalties. It has asked the EU to investigate. TSMC says the claims have no merit.

For now, TSMC is in a sweet spot. It uses steady revenues from firms like Apple, which are unwilling to switch to IDM firms like Samsung that are also competitors, to fund R&D that other foundry firms cannot match. This sharpens its technological edge, which in turn attracts new customers. Whether this can continue is unclear. Moore’s law is running out of steam. Beyond the next cycle of shrinking nodes the future is less certain. As Mr Chang prepares to leave, investors will hope Messrs Liu and Wei are chips off the old block.