Chapter 7: A shift in how chips are made ― drawing without carving a stamp
In Chapter 5, we saw that "the language of design (RISC-V) was opened, and anyone could take on design." But however good a design you draw, if you can't turn it into an actual chip, it ends as a dream. In fact, a quiet shift is happening in that "making," too. The manufacturing-side upheaval that pairs with the opening of the design door. The lead of this chapter is a slightly unexpected word — the "stamp."
7.1 How is a chip "drawn"?
The center of chip-making is the process of "printing" an extremely fine circuit pattern onto a silicon board (a wafer). The method resembles developing a photograph a little: shine light, transfer the pattern. The question is "how do you transfer the 'source data' of that pattern onto the wafer?" And the conventional method has used a "stamp."
7.2 The conventional way starts from a "stamp (mask)"
In conventional manufacturing, you first prepare an original plate called a "mask," which carries the circuit pattern. This is, in effect, a precise stamp (a printing master). You shine light through this mask and transfer the pattern onto the wafer all at once. The strength of the stamp method is being able to press the same pattern in large quantities, at high speed, repeatedly. So it is very strong at mass production, making millions of the same chip.
To put it another way, a New Year's card stamp. Carve the stamp once, and after that you can print the same picture hundreds of times just by pressing. For making things in bulk, there's no more efficient method. Chip mass production has run on exactly this "press the stamp many times" idea.
7.3 But that "stamp" becomes a gate
However, this stamp method has a large wall for the small challenger. Carving the stamp (mask) itself is very expensive and takes time. A set of leading-edge masks is very costly, and can take several weeks to complete. And if you want to fix the design a little, you must re-carve the stamp. Cost and time again.
What does this mean? That "just make one to try" and "fix it a bit and try again" become terribly expensive. A large company with ample funds and mass production in view can bear it, but for an individual or small team, this "stamp cost and wait time" is a heavy gate. Even with the design door opened by RISC-V, this manufacturing gate stood, at the last, in the way.
7.4 Drawing directly, without carving a stamp ― maskless
So the research advancing is "maskless" — literally, a way of making that uses no stamp (mask). The idea is simple. Stop carving the original plate, and "draw the pattern directly" onto the wafer with a fine electron beam. Close to the image of drawing each one by hand.
What's good about this method? The cost of carving the stamp, and the wait for it to be finished, both become unnecessary. Want to fix the design? Just change the data you draw. You can try the next one right away. Further, you can even do experiments unthinkable with the stamp method, like trying a different design, one by one, on a single wafer. In fact, such maskless direct-writing equipment has begun to be introduced in some fabs (equipment from the US company Multibeam was introduced by the manufacturer SkyWater in 2024, and a related company raised new funding in 2025), and the move from research to practical use is advancing, little by little.
※ But it hasn't "replaced" the stamp method (an honest reservation). Maskless is, for now, unsuited to mass production in terms of speed, and shows its strength mainly in prototyping, small volumes, and special uses — "another road." It won't take over all of leading-edge mass production right away. In this corner, we take it not as "the mask is gone," but as "another route that removes the gate of the stamp has begun to grow in earnest." For printing in bulk, the stamp; for trying small volumes quickly, direct drawing — heading toward an age of using each according to the purpose.
7.5 Another way to lower the barrier ― making it cheap by "ride-sharing"
There's one more device for lowering the manufacturing hurdle: "ride-sharing (shuttle)." A single wafer is wide. On it, you have not just one company's design but the small designs of challengers from around the world "ride together," dozens at a time, and make them all at once. Then the manufacturing cost can be split among everyone, and even an individual can have a real chip made for a reachable sum.
To put it another way, sharing a chartered bus. A bus that's expensive to charter alone becomes much cheaper when strangers share the seats. Chip-manufacturing ride-sharing runs on the same idea, and made it real for "an individual to have a chip they designed baked, from a few hundred dollars." The design side (RISC-V) and this manufacturing side (maskless + ride-sharing) have begun to turn as two wheels — this is the true nature of the current "shift in how chips are made."
7.6 This chapter's summary, and the bridge to the next
- Conventionally, the "stamp (mask)" method. Good at printing in bulk. But stamp cost and wait time are a heavy gate for the small challenger.
- Maskless = drawing directly without carving a stamp. "Another route" that removes the gate of cost and wait time. Shows its strength in prototyping and small volumes (not a replacement for mass production).
- Ride-sharing (shuttle) splits the cost. Toward an age when even an individual can make a real chip for a reachable sum.
And so, both design and manufacturing are having their doors opened. Then, where is this upheaval heading? The next chapter, Chapter 8, is the story of the frontier at last. AI evolves from a being that merely answers, into one that judges for itself, and finally moves a "body" in the real world — "Physical AI" — toward robots, cars, and space. Let's go and see that the small chip at your hand is continuous with the world's frontier.