ReadReal vs Hype18 Jul 20262:40Semiconductors: chips, fabs & yield

This $400M Machine Prints Chips in a Focus Thinner Than a Virus (Intel Went First)

The Hardest Machine Humans Ever Built Just Made Its First Real Chips

The massive mirror optics module of a lithography machine in dim blue, one mirror catching a warm gold glint
The object this week · generated illustration, no people, no brands
Video

This one has not been published to the channel yet. Subscribe on YouTube and it will turn up there.

The 60-second version
  • It's called High-NA EUV: a $400-million lithography system, twice the price of a normal chip printer, and Intel just became the first company on the planet to ship commercial chips made on it.
  • The real story is what this machine does to the factory: it draws the tiniest features in ONE exposure instead of two (fewer masks, fewer steps, less time each wafer spends collecting defects), while shrinking the focus window to about a third of normal and printing only half a chip at a time — which means stitching two halves together and lining up layers within a couple of nanometers, every single time.
  • The actual breakthrough isn't the optics (Zeiss perfected those years ago) — it's that Intel held ~85% yield on this brutal new process, up from ~65% a quarter earlier, and that's how it clawed back into a foundry race a lot of people had written off.
  • We break down how it works, why it matters for the chips in everything you own, and the honest catch (2.5x cost per shot, select layers only, full payoff around 2030).

Why this matters

This week the most expensive machine humans have ever built made its first real products — and almost nobody noticed. But the headline everyone repeats — "smaller transistors" — is the boring part.

What to do Monday

GitHub just opened its Copilot DESKTOP app to everyone (free, all tiers) — and it runs coding agents in PARALLEL on one repo, each in its own isolated git worktree (own branch + files, zero conflicts). Kick off 3 sessions at once — one writes the feature, one the tests, one the docs — then review and merge the branches that pass.

In the video
  1. 0:00The most expensive machine ever built
Over to you

Genuine question for the people who actually run process: on a 1/3 focus window, would you trust AI to auto-tune focus + overlay lot-to-lot, or keep an engineer signing off every critical layer?

Argue with me on LinkedIn
Sources

References for this piece are in the pinned comment on the video. Nothing is cited here that cannot be linked.

Full transcript, 365 spoken words
Hold two fingers a hair apart. Now shrink that gap until the ONLY thing in focus is a slice thinner than a virus — and paint a flawless circuit inside it, twenty MILLION times, no blur. That's the machine Intel just switched on. High-NA EUV — four hundred million dollars, twice a normal chip printer — and this week Intel became the FIRST on Earth to ship products made on it. But the headline — "smaller transistors" — isn't the hard part. Here's the hard part. A normal machine needs TWO passes to print one fine layer. High-NA does it in ONE. But a sharper lens focuses in a SHALLOWER slice — about a third of normal. That razor-thin depth IS the whole problem. Worse — it can only print HALF a chip at a time. A full chip is two exposures STITCHED together, aligned within a couple of nanometers. Miss by a hair, and the circuit doesn't connect. Here's what matters. The lens was solved years ago. The real feat: Intel held EIGHTY-FIVE percent yield on something this unforgiving — up from sixty-five last quarter. On a focus that thin, that number is the miracle. THAT'S the comeback. Magic yet? No — one shot still costs 2.5 times a normal one, on a few layers of one chip. But the hardest lithography humans have ever pulled off just worked in a real fab. If you want the mechanics behind the machines that decide which chips exist — follow. Now, your FabSpeak Tip of the Week. If you touch code at work, this is a real edge. GitHub just opened its Copilot DESKTOP app to everyone, free — and here's the power move most people missed: it runs multiple coding agents in PARALLEL on the same repo. Each agent gets its own isolated git worktree — its own branch and files — so they never collide. The move: open your repo, kick off three sessions at once — one writes the feature, one writes the tests, one updates the docs — let them run, then review and merge the branches that pass. Three tasks, one codebase, zero conflicts. Save that one. Follow FabSpeak — every week I unlock something new in AI and the factory. See you at the next drop.