Handmade CPUs
January 18, 2025 2:46 AM Subscribe
Long before ASML created "most important machine in the world", stencils for VLSI chips were handmade with Rubylith film and then photoreduced to create the actual photomasks used in IC fabrication.
I'm a (retired) chip designer, came to this in the early 90s back when "taping out" meant an actual mag tape (rather than mylar tap - I'm not sure if they used the same term when it was mylar tape :-)
As a teenager (in the 70s) I did make double sided PCBs by laying out double sized red/blue mylar tape, black donuts for vias and letraset for text - essentially the same technique described above but with fewer layers
posted by mbo at 3:01 AM on January 18 [9 favorites]
As a teenager (in the 70s) I did make double sided PCBs by laying out double sized red/blue mylar tape, black donuts for vias and letraset for text - essentially the same technique described above but with fewer layers
posted by mbo at 3:01 AM on January 18 [9 favorites]
Here's someone who made a batch of 10× 100 transistor chips in his garage in the late 2018 (text link).
posted by ambrosen at 4:24 AM on January 18 [4 favorites]
posted by ambrosen at 4:24 AM on January 18 [4 favorites]
My Dad had stories of layout drawings the size of basketball court floors, photographed and then run through an enlarger the wrong way to make masks for early chips
posted by Aardvark Cheeselog at 10:40 AM on January 18 [1 favorite]
posted by Aardvark Cheeselog at 10:40 AM on January 18 [1 favorite]
I've talked about this before, but my late dad's t-shirt screen printing company shared a lot of tech with early analog/optical chip and PCB production.
Notably we used Rubylith film by the truckload for masking off color separations for multicolor prints, and even a surplus but vintage chip/PCB industry graphics camera.
We'd start with some "keyline" image drawn on actual paper or vellum, photostat that to transparent film with black keyline graphics, and then using registration tabs or marks we'd tape down a sheet of Rubylith over that transparent keyline on a light table and manually cut and peel out the parts of Rubylith for each color plate we wanted to expose, and on some designs that could be as much as 18 plates and discrete colors.
The line and cut work we did with Rubylith was as fine and precise as any early analog chip masks, and if anything some of it may have been finer details because we weren't working with uniform lines and shapes, so our "traps" on any given stack of plates could have details well below 1 mm.
One of our specialties was large format and even "all over" printing where the printing area on the shirt could exceed about 40x60 inches, so we also needed a very large graphic arts camera.
Typically these graphics arts cameras were arranged vertically with the copy board (source image) on a horizontal board near the ground, and the vacuum plate film holder at the top with the bellows and lens between all of that.
The pro versions of these were typically about the size of a large washing machine unit or chest freezer, and a maximum film size in the 24-32 inch range.
Somehow my dad found a surplus chip fab camera all the way across the country that had something like a 50-70 inch vacuum film bed and an even larger copy/source image stand, and a hunk of prime Nikkor glass in the middle about the size of a football, which is absolutely huge as these things go. He had to rent like a 40+ foot box truck to drive it back across the country.
Illumination on the copy bed was provided by four quartz lamps, and the copy bed itself was a huge metal frame with glass on the business end and a flexible baffled plate backer so the source images could be vacuumed right up against the glass for high precision and low errors and distortion.
Those lamps were hot enough to light paper on fire at about 5-10 centimeters without touching the bulbs. They may or may not have been used sometimes to light doobies.
The film end was equally massive and had a vacuum bed big enough to play a little air hockey on if you reversed the pump. It also tilted from horizontal for loading film to vertical for shooting, and locked into place like a massive naval cannon breech.
And the whole arrangement was horizontal. The base of the camera was something like 20-25 feet long and basically a high precision box girder with roller bearing tracks that the copy bed and camera end aligned to, with the optics and camera body about 2/3rds of the way towards the film end, and the whole thing stood like 5-6 feet high.
I can't remember the actual enlargement/reduction values but it was something insane like 10% reduction to 500% enlargement, which was a huuuuge range in one camera.
For the camera room we actually had to build a whole room around that camera with a hot/light end in one room and the cold/dark room on the other, where there was also a contact frame and huge developing trays for the massive amounts of film we were processing.
So if we had a ton of film to process and shoot we'd have someone in the light end and someone (often me) in dark end batch processing shots. The person in the light end would be setting up the objective copy, they'd yell that it was ready, I'd load the film sheet, expose it, then pull the film and put it right into the developer tray. And by about the time I was putting that sheet into the fix the next shot was ready to shoot and I'd load the next sheet of film, and so on.
Sometimes I was also processing contact frame shots where we were taking keyline or plate images to make the transparencies we actually used to expose the emulsion on the screen printing screens and frames themselves.
I still think about that massive camera and wonder what it processed before being demoted to printing t-shirts.
And I also wonder what my life would be like if we had transitioned into making PCBs or something instead.
We definitely talked about it because the tech for processing the optical art parts of it were basically identical, and PCB handling and printing is way, way easier to automate than textiles or t-shirts because the substrates (fiberglass composite boards) are rigid and can be handled by simple machines, where textiles and t-shirts (still!) need human hands to move and process.
I know we didn't like the chemicals and acids involved with PCBs, so that was part of the reason why we didn't. The regulation of those chemicals and metal waste byproducts was a lot more restrictive and expensive to deal with.
Today the way they make masks for modern chips is totally insane wizard magic where the masks barely even resemble the finished etch or product, because they have to account for and design around optical diffraction and diffusion around features that small in the 10 nanometer range.
They use advanced computer aided photonic simulation and modeling to distort these masks so that when the light reflects off of them it lands on the substrate die in a way that ends up making those high precision useful features. No humans touch the "art" part of the mask. They design the chip features and layouts in software, and that gets passed to the modelling software that actually forms the shapes used on the masks which are *not* simple lines, squares or boxes, but more like wavy, fuzzy shapes that trick photons into doing what they want them to do so they expose the wafers right.
If you took these current masks (enlarged to human visible scales) back to early chip designers doing manual optical masking the old fashion way with Rubylith and actual tape and stuff they'd probably think you were insane or a wizard or both.
posted by loquacious at 12:29 PM on January 18 [26 favorites]
Notably we used Rubylith film by the truckload for masking off color separations for multicolor prints, and even a surplus but vintage chip/PCB industry graphics camera.
We'd start with some "keyline" image drawn on actual paper or vellum, photostat that to transparent film with black keyline graphics, and then using registration tabs or marks we'd tape down a sheet of Rubylith over that transparent keyline on a light table and manually cut and peel out the parts of Rubylith for each color plate we wanted to expose, and on some designs that could be as much as 18 plates and discrete colors.
The line and cut work we did with Rubylith was as fine and precise as any early analog chip masks, and if anything some of it may have been finer details because we weren't working with uniform lines and shapes, so our "traps" on any given stack of plates could have details well below 1 mm.
One of our specialties was large format and even "all over" printing where the printing area on the shirt could exceed about 40x60 inches, so we also needed a very large graphic arts camera.
Typically these graphics arts cameras were arranged vertically with the copy board (source image) on a horizontal board near the ground, and the vacuum plate film holder at the top with the bellows and lens between all of that.
The pro versions of these were typically about the size of a large washing machine unit or chest freezer, and a maximum film size in the 24-32 inch range.
Somehow my dad found a surplus chip fab camera all the way across the country that had something like a 50-70 inch vacuum film bed and an even larger copy/source image stand, and a hunk of prime Nikkor glass in the middle about the size of a football, which is absolutely huge as these things go. He had to rent like a 40+ foot box truck to drive it back across the country.
Illumination on the copy bed was provided by four quartz lamps, and the copy bed itself was a huge metal frame with glass on the business end and a flexible baffled plate backer so the source images could be vacuumed right up against the glass for high precision and low errors and distortion.
Those lamps were hot enough to light paper on fire at about 5-10 centimeters without touching the bulbs. They may or may not have been used sometimes to light doobies.
The film end was equally massive and had a vacuum bed big enough to play a little air hockey on if you reversed the pump. It also tilted from horizontal for loading film to vertical for shooting, and locked into place like a massive naval cannon breech.
And the whole arrangement was horizontal. The base of the camera was something like 20-25 feet long and basically a high precision box girder with roller bearing tracks that the copy bed and camera end aligned to, with the optics and camera body about 2/3rds of the way towards the film end, and the whole thing stood like 5-6 feet high.
I can't remember the actual enlargement/reduction values but it was something insane like 10% reduction to 500% enlargement, which was a huuuuge range in one camera.
For the camera room we actually had to build a whole room around that camera with a hot/light end in one room and the cold/dark room on the other, where there was also a contact frame and huge developing trays for the massive amounts of film we were processing.
So if we had a ton of film to process and shoot we'd have someone in the light end and someone (often me) in dark end batch processing shots. The person in the light end would be setting up the objective copy, they'd yell that it was ready, I'd load the film sheet, expose it, then pull the film and put it right into the developer tray. And by about the time I was putting that sheet into the fix the next shot was ready to shoot and I'd load the next sheet of film, and so on.
Sometimes I was also processing contact frame shots where we were taking keyline or plate images to make the transparencies we actually used to expose the emulsion on the screen printing screens and frames themselves.
I still think about that massive camera and wonder what it processed before being demoted to printing t-shirts.
And I also wonder what my life would be like if we had transitioned into making PCBs or something instead.
We definitely talked about it because the tech for processing the optical art parts of it were basically identical, and PCB handling and printing is way, way easier to automate than textiles or t-shirts because the substrates (fiberglass composite boards) are rigid and can be handled by simple machines, where textiles and t-shirts (still!) need human hands to move and process.
I know we didn't like the chemicals and acids involved with PCBs, so that was part of the reason why we didn't. The regulation of those chemicals and metal waste byproducts was a lot more restrictive and expensive to deal with.
Today the way they make masks for modern chips is totally insane wizard magic where the masks barely even resemble the finished etch or product, because they have to account for and design around optical diffraction and diffusion around features that small in the 10 nanometer range.
They use advanced computer aided photonic simulation and modeling to distort these masks so that when the light reflects off of them it lands on the substrate die in a way that ends up making those high precision useful features. No humans touch the "art" part of the mask. They design the chip features and layouts in software, and that gets passed to the modelling software that actually forms the shapes used on the masks which are *not* simple lines, squares or boxes, but more like wavy, fuzzy shapes that trick photons into doing what they want them to do so they expose the wafers right.
If you took these current masks (enlarged to human visible scales) back to early chip designers doing manual optical masking the old fashion way with Rubylith and actual tape and stuff they'd probably think you were insane or a wizard or both.
posted by loquacious at 12:29 PM on January 18 [26 favorites]
As I pointed out in the ASML thread, we'd supply disruption issues for larger chips recently, in part because nobody maked the machines to make those chips anymore. See the MCH2022 talk: Where did all the parts go - the 202x component availability trashfire
As an aside, Bunnie Hung's last two CCC talks rocked:
IRIS: Non-Destructive Inspection of Silicon
Open Source is Insufficient to Solve Trust Problems in Hardware
posted by jeffburdges at 3:46 PM on January 18 [3 favorites]
As an aside, Bunnie Hung's last two CCC talks rocked:
IRIS: Non-Destructive Inspection of Silicon
Open Source is Insufficient to Solve Trust Problems in Hardware
posted by jeffburdges at 3:46 PM on January 18 [3 favorites]
Other handmade chip designs......
posted by lalochezia
Here's an entry on Ken Shirrif's fantastic blog giving more background about this chip and this weaving, including the early history of Intel and the story of their plant at Shiprock, New Mexico where they optimistically hired 366 Navajo workers, but the plant closed a decade later after an armed takeover. It's an amazing story and gives more context to this incredible weaving.
posted by crazy_yeti at 7:07 AM on January 19
posted by lalochezia
Here's an entry on Ken Shirrif's fantastic blog giving more background about this chip and this weaving, including the early history of Intel and the story of their plant at Shiprock, New Mexico where they optimistically hired 366 Navajo workers, but the plant closed a decade later after an armed takeover. It's an amazing story and gives more context to this incredible weaving.
posted by crazy_yeti at 7:07 AM on January 19
« Older A very weird book | Shadowdark Newer »
This thread has been archived and is closed to new comments
posted by autopilot at 2:47 AM on January 18 [4 favorites]