Paul van Gerven

4 March

Building a leading-edge foundry from scratch is a fool’s errand, but that’s not to say no good can come of it.

“You’ll be pleasantly surprised by the yield you’ll achieve in Japan,” legendary Sony co-founder Akio Morita told the equally formidable father of TSMC Morris Chang when the latter – still employed at Texas Instrument at the time – was looking into opportunities to build fabs in Japan. This was in the late 1960s, and Japan delivered. TI ended up building several well-yielding fabs, and many Japanese electronics companies would follow suit. By the late 1980s, Japan had become the world’s semiconductor powerhouse, sporting six of the top ten global chipmakers.

The lead eroded quickly. By the year 2000, only three Japanese companies were left among the semiconductor elite. Toshiba held on longest but had dropped out of the top ten by 2020, too.

Now, much like Europe and the US, Japan wants to restore at least some of its former semiconductor glory. In a two-pronged approach, it has convinced TSMC to set up shop to manufacture trailing-edge logic (6-12nm). For the leading edge, however, it will take matters into its own hands. Leapfrogging a handful of nodes – Japan’s current most advanced logic process is 40nm – the eight-company Rapidus consortium intends to start contract-manufacturing 2nm chips by 2027.

It’s a high-risk venture, to say the least, even with generous support from the Japanese government. Imec and IBM are on board to get the process technology up to speed, but neither organization has (recent) experience running a fab or the supply chain that supports it. No number of hired heavy hitters can conjure up a leading-edge process in a couple of years on what’s a desert island in terms of advanced semiconductor technology. If an industry stalwart like Samsung struggles with yield, what hope is there for Japan? History is no help here.

Bits&Chips event sponsor registration-early bird

Or is it? The Rapidus plan features striking similarities with the Mega project, which saw laggards Philips and Siemens joining hands in the mid-1980s. Supported by ample government money, the electronics giants set out to skip a node to be first to 0.7-micron technology, only to pull the plug a few years later. Siemens found a gracious way out by allying with Toshiba, but Philips suffered an embarrassing defeat that was all over Dutch media in the early 90s.

In hindsight, however, Mega wasn’t the failure it was portrayed to be. The CMOS technology developed for Mega proved fertile ground to create new or update existing products, which were manufactured alongside Philips’ highly profitable analog and mixed-signal devices in the brand-new ‘Mega’ fab in Nijmegen. Already by the mid-1990s, the chip plant was expanded and the semiconductor division went from zero to hero within the battered conglomerate.

Rapidus will befall a similar fate. The venture will miss deadlines, face huge cost overruns or simply won’t prove competitive. But after the dust of the implosion has settled, the investments can be steered toward more realistic goals. In a semiconductor market that’s projected to reach 1+ trillion dollars by 2030, investing in manufacturing capacity will never be completely in vain.

In a particularly beneficial scenario – for Rapidus, not necessarily the rest of the world – the shrink might slow down so much that sweet spots form behind the leading edge. If performance or energy efficiency gains come at too high a cost, moving to the next node is no longer a given for a number of applications. This offers a chance for chip manufacturers to compete right behind the industry leaders without having to keep up.

One can’t help but admire Japan’s bravery. Even if Rapidus’ goals are unattainable, it’s comforting to know that the venture won’t necessarily fail either.

Main picture credit: Rapidus