The first chip ever built in volume with extreme ultraviolet lithography is sitting inside a Huawei phone, made by a company the United States added to its export blacklist five months ago. That is TSMC’s N7+ node in one sentence: a twenty-year, multi-billion-dollar industry bet on a new kind of light finally paying off, and the first marquee customer for it is the most geopolitically radioactive name in the business. EUV took ASML two decades and billions to make real, and the node it finally became real on is the one carrying a customer the United States is actively trying to strangle.
EUV is the thing the whole industry has been promising and missing since the early 2000s. Instead of 193nm light bent through water and split across four masks to print one layer, you use 13.5nm light, more than ten times shorter, and print that layer in a single exposure. The wavelength is so short that air absorbs it and glass absorbs it, so the entire optical path runs in vacuum off mirrors instead of lenses, and you make the light by hitting droplets of molten tin with a laser fifty thousand times a second. It sounds insane because it is, and it took ASML two decades and a fortune to get it into a fab. N7+ is where it finally does.
N7+ does not bet the whole node on it. EUV replaces the most painful handful of layers, the ones that needed quadruple patterning on the DUV version of 7nm, while the rest of the process stays on the immersion tools TSMC already knows cold. That buys a modest density bump and, more to the point, fewer masks and fewer steps on the layers that were the worst yield and cost offenders. It is a deliberately cautious first deployment, a way to learn EUV in production on a node that already works rather than gambling a brand-new node on a brand-new light source at the same time.
This is the first real answer to the problem that has shadowed every node since 20nm. The cost per transistor stopped falling once multipatterning took over, because each new node needed more masks and more steps to print, and the density you gained no longer paid for the complexity you added. EUV does not fix that for free, the tools cost more than a hundred million dollars apiece and the throughput is still tight. But replacing four DUV masks with one EUV exposure on the hardest layers is the first time in years the patterning complexity on a new node went down instead of up. Whether it nets out cheaper depends on tool throughput, and that fight is only starting.
Underneath the EUV milestone, the competitive picture only sharpened. AMD’s 7nm server and desktop parts shipped this year and they are beating Intel on performance-per-watt, exactly the outcome the leadership flip at 10nm set up. Intel is still on 14nm for anything that ships in real volume. The gap that opened when Intel’s 10nm stalled is now wide enough that TSMC’s customers are out-competing Intel’s own products using TSMC’s process, and the EUV experience Intel does not have yet only widens it.
Samsung got an EUV-built chip out the door first. Its 7nm EUV line shipped the Exynos 9825 inside the Note 10 back in August, a couple of months ahead of this, which makes the bragging right for first EUV chip technically Samsung’s. The qualifier that matters is volume. Samsung’s EUV output is thin and aimed mostly at its own parts, while N7+ is going into a flagship phone SoC at the scale TSMC runs everything, which is a different order of proof. The foundry race this year is lopsided regardless, TSMC sits north of half the market and Samsung is stuck in the high teens, and EUV is exactly the kind of capital and yield problem that rewards whoever is already running the most wafers.
Full EUV arrives next year at 5nm, where it stops being a few layers and becomes the backbone of the node, and N7+ exists largely so TSMC has run the tools in anger before it bets a whole generation on them. The shadow over all of it is the customer in the opening paragraph. Huawei is one of TSMC’s largest and most advanced clients, and it is now operating under US export restrictions that could, depending on how the rules end up written and enforced, reach the Taiwanese foundry that builds its chips. TSMC is sitting on the best manufacturing on earth and a customer-concentration problem that is no longer only commercial. It is political, and that is a variable no process roadmap knows how to price.
N7+ is a quiet node with a loud meaning. The density gain is small and most people will never know which 7nm their chip used. What matters is that the light source the entire future of the industry was staked on finally works in a real fab, on real wafers, in real volume, after twenty years of being the thing that was always almost ready. TSMC got there first at the scale that counts. It also got there carrying a customer Washington wants cut off and a manufacturing position so concentrated it makes the company indispensable and exposed at once. Everyone spent twenty years treating the light source as the hard part of 7nm. The tin droplets turned out to be more tractable than the export-control lawyers.