Samsung started mass production of 3nm chips this week, the first company in the world to ship a gate-all-around transistor, beating TSMC to the most important change in chip design since FinFET. On paper it is a triumph, the kind of genuine first that justifies the whole leapfrog strategy. Samsung 3nm exists, in production, before anyone else’s. And almost nobody is talking about it, because it arrived two weeks after the 8+ Gen 1 proved a Samsung chip runs better when TSMC builds it.
Gate-all-around is the successor to the fin. FinFET wraps the gate around three sides of the channel; gate-all-around wraps it around all four, stacking the channel into nanosheets the gate completely surrounds, which gives back the control even fins start to lose at these sizes. Samsung calls its version MBCFET, and shipping it in volume before TSMC is a real engineering achievement, not a slide. For once the leapfrog is not a press release. The transistor is new, it works, and it is Samsung’s first.
And it does not matter nearly as much as it should, because being first to a transistor is worthless if you cannot build it well or sell it. The early reports on Samsung 3nm yields are not good, and the customer list is thin, reportedly leaning on small orders and crypto-mining chips rather than the marquee mobile and HPC names that make a node a business. Contrast that with TSMC’s 5nm, a year and a half old, packed with Apple and AMD and Nvidia, printing money on a more mature process. Samsung has the newer transistor and a thin order book. The customers are the only scoreboard that pays in this industry, and they are sitting at TSMC. TSMC’s own 3nm is still months from volume and stays on FinFET rather than gambling on a new transistor yet, and it already has Apple booked. Samsung got to 3nm and to gate-all-around first, and the biggest names are waiting for the slower FinFET version from the foundry they trust.
This is the Samsung foundry story compressed into a single node. First to FinFET, first to EUV, now first to gate-all-around, a decade of genuine technical firsts, and at the same time a steady bleed of trust over chips that ran hot and yields that came in low and customers who did the math and left. Samsung keeps winning the engineering race and the business keeps going somewhere else. 3nm is both halves at once, the most advanced transistor in the world built by the foundry the industry has the most reason to doubt. The firsts were never what Samsung lacked.
There is a real argument that this time first pays off. Gate-all-around is where the whole industry has to go, TSMC included, and Samsung will have years of production experience with it before the competition ships its own. If GAA is hard, and it is, then a head start learning to yield it could be the thing that finally turns a Samsung first into a Samsung lead. The company is betting the customers come once GAA is the only game in town and Samsung is the one who has been building it longest. It is not a crazy bet. It is the same bet Samsung always makes, that the future arrives on its schedule.
Maybe it works this time. Maybe the GAA head start is the one that compounds, the customers come back when they have nowhere else to go, and 2022 is remembered as the year Samsung quietly got ahead for real. Perso, I would not put money on it, not because the transistor is wrong but because the thing Samsung needs to fix is not in the transistor. It is the yield and the trust, and those do not improve just because you got somewhere first. Samsung shipped the future this week, and the industry mostly shrugged, glanced at the 8+ Gen 1 benchmarks, and kept its orders in Taiwan. The most advanced transistor in the world is sitting in a fab the biggest customers no longer trust to build their chips well. You do not fix a trust problem by shipping the next transistor first. That is the only move Samsung has ever known how to make.