SMIC 28nm reached real volume this year, roughly four years after TSMC shipped its own 28nm, and the size of that delay is the whole story. SMIC 28nm took years longer than it should have, stuck on yield problems the leaders solved a process generation ago, and arriving in 2015 it lands into a world that has already moved past planar transistors entirely. The node is an achievement and a warning at the same time. SMIC can still reach the nodes the leaders reached. It just keeps arriving years after they have left.
28nm was supposed to be a milestone and it turned into a slog. SMIC spent years fighting yields, struggling with the harder version of the process, the one with the better transistors, and leaning for a long time on an older, cheaper variant customers did not want as much. The trouble was instructive. It showed that without the shortcut the lawsuit closed, the original engineering SMIC now has to do itself is genuinely hard and genuinely slow, that catching up by your own effort means hitting every wall the leaders hit, with none of their accumulated experience to climb it faster. 28nm is where that reality bit down.
While SMIC fought 28nm, the leading edge left planar transistors behind. TSMC and Samsung have moved to FinFET, the three-dimensional transistors that 16 and 14-nanometer nodes are built on, and they are building the most advanced phone chips in the world on them. SMIC at 28nm is now not one or two nodes behind but standing at the bottom of a cliff: the jump from planar 28nm to FinFET is a different kind of climbing, new physics and new tools and a wall of difficulty SMIC has never faced and the leaders cleared years ago. The leaders did not just get smaller, they switched to a transistor SMIC has never built, on tools it already has trouble buying, and that turns a delay of a few years into something closer to a moat.
None of this threatens the business, which is the strange thing. 28nm is, for many years to come, a sweet spot for an enormous range of chips, and a deep, cheap, high-volume 28nm line inside China is a genuinely valuable asset that will pay for itself for a decade. SMIC’s mature-node floor is solid and getting deeper. It holds only a few percent of the global foundry market, a distant fifth behind TSMC’s better-than-half, Samsung, GlobalFoundries and UMC, but inside China it is the anchor, and that is the position that pays. The business is fine: profitable, fabs full, growing every year. China did not pour state money into SMIC to build a comfortable mature-node shop, though, and at 28nm in 2015 the leading-edge ambition is stalled exactly where the climb turns vertical.
Getting to FinFET would take something SMIC has not yet shown it can summon: a genuine leading-edge engineering effort, the kind that develops a hard new transistor mostly from scratch, and probably an injection of senior talent who have actually built FinFET before. The companies that have done it are few and guarded, and the knowledge does not transfer easily or legally, as SMIC learned the expensive way. Whether SMIC can assemble that on its own, or poach and buy its way to it, is the question the next few years answer.
SMIC at 28nm is a solid foundry stuck at a hard wall, and which of those two facts ends up mattering more is still anyone’s guess. Perso, I lean toward the optimistic read: China wants leading-edge chips too badly to accept 28nm as a ceiling, the money will keep coming, and sooner or later the right engineers arrive and FinFET gets cracked, late but real. The pessimistic read is that the cliff is simply too steep, that original leading-edge development is a different game from mature-node catch-up and SMIC may never quite play it. What neither read takes seriously, because in 2015 it sounds far-fetched, is the possibility that the question gets taken out of SMIC’s hands entirely, that the tools and talent it would need to climb the cliff stop being things money can buy. That is not a 2015 worry. It should be.