SMIC 14nm went into risk production this year, and with it China climbed the cliff. SMIC 14nm is the country’s first homegrown FinFET, the three-dimensional transistor that loomed over the company like a wall when it was stuck at 28nm, and reaching it is the most important thing SMIC has done since it was founded. The gap to the leaders is still wide, five years or so. But the kind of gap changed. SMIC is no longer a mature-node foundry locked out of the modern transistor. It is a FinFET foundry now, late but real, and that is a different company.
The breakthrough has a name attached to it, and the name explains a lot. Liang Mong Song joined SMIC as co-chief executive in 2017, and Liang is one of the most consequential process engineers alive: he spent years at TSMC, then went to Samsung and architected the 14nm FinFET that let it leap ahead of TSMC in 2015, and now he is in Shanghai doing it a third time, for a third company, for China. Wherever Liang goes, FinFET follows. He is the closest thing the industry has to a one-man technology transfer, legal this time, carried in a single brilliant head, and SMIC’s 14nm is substantially his.
14nm does not put SMIC near the front. TSMC is shipping 7nm and moving to EUV while SMIC reaches the node TSMC ran four years ago. What it does is prove SMIC can do leading-edge engineering at all, can develop a hard modern process and put it into production, can climb the kind of cliff that mature-node competence never prepares you for. That capability, more than the node itself, is the asset. Once you can build one FinFET node, the next ones are hard but no longer a different species of hard. SMIC just joined the small club of companies that can play this game.
The club is smaller than it was a year ago. UMC and GlobalFoundries both walked away from the leading edge in 2018, deciding the FinFET economics past 14nm were not worth chasing, which leaves three companies pushing forward at the front: TSMC, Samsung, and Intel. SMIC arriving at 14nm just as two old foundry names gave up advancing is a strange piece of timing. It is still the world’s number five foundry by revenue, low single digits of market share, a rounding error next to TSMC’s half. But it is climbing while others are quitting, and in 2019 the direction matters more than the absolute position.
The first big customer is Huawei, which matters in 2019 in a way it would not have a year ago. Huawei’s chip-design arm needs somewhere to build, and as a Chinese champion it has every reason to build at the Chinese foundry, so SMIC 14nm gets a marquee customer and Huawei gets a domestic option. But Huawei is also, as of this spring, on a United States blacklist, cut off from American technology in a trade fight that is getting sharper, and its troubles hang over this partnership. For now SMIC itself is untouched, a Chinese foundry buying tools on the open market like always. The storm is over Huawei, not yet over the fab that builds its chips.
The United States has decided that Chinese technology champions are a strategic problem, Huawei is the first big target, and the comfortable assumption that chips are a borderless global business is starting to wobble. SMIC has spent the better part of two decades buying its tools and talent from abroad on the premise that the market is open to anyone who pays. In late 2019 that premise is still mostly true. It is also, for the first time, something you can imagine becoming untrue, if the trade fight keeps escalating and starts reaching past the chip designers to the foundries that build for them.
14nm is a genuine triumph and SMIC has earned the moment. Perso, I expect SMIC keeps moving now, 7nm in a few years, the gap to TSMC slowly closing the way it always seemed it eventually would once the FinFET wall was behind it. The path looks open. The one shadow on it is political, not technical, and even that feels manageable from here; Huawei is in trouble but SMIC is a foundry, a step removed, and surely the tools keep flowing to a company that just buys machines and builds chips. That is the assumption underneath the optimism. In late 2019 it is still safe to make. The interesting question, the one nobody in Shanghai wants to dwell on, is what SMIC’s roadmap looks like if it ever stops being safe.