3 ms·

Ultraprecise method of aligning 3D semiconductor chips invented

170 points by thebeardisred 2y ago

heisenbit 2y ago
Isn't this very similar to the way optical position encoders work?
- scheme271 2y ago
  I think the difference is that you don't need to change the focus of the optical sensors in order to verify alignment. So you don't have to worry about movement while the focus is changed.
lawlessone 2y ago
Could this also be used for manufacturing lots of other microscopic things? layer by layer? "Instead, their method finds errors up to 0.017 nm along side-to-side measures (x and y axes) and 0.134 nm when assessing the distance between the two chips (z-axis)." Could you make some very very sensitive and tiny seismic sensors with this? edit: " Arbabi also points out that this method can be used to make displacement sensors that can be used for measuring displacements and other quantities. "Many physical quantities that you want to detect can be translated to displacements, and the only thing you need is a simple laser and a camera," he says. For instance, "if you want a pressure sensor, you could measure the movement of a membrane." Anything that involves movement—vibration, heat, acceleration—can in theory be tracked by this method. "
- g5pw 2y ago
  I made a home-brew seismic sensor using something similar, a hard disk head arm assembly, a cd-rom laser (which has an anisotropic lens and four photodiodes) and a Red Pitaya used as PID, so I guess it can be done!
  - pfdietz 2y ago
    Oh man, this is giving me The Amateur Scientist vibes. What would C. L. Stong be doing with today's technology?
Etheryte 2y ago
Maybe I'm missing something here, doesn't this simply move the precision problem to a different part of manufacturing? Previously you had to be precise with aligning the chips, now you have to be precise with how you put those alignment marks on the chips you want to align. Am I missing something here? Or is it considerably easier to put the marks on the chips with sufficient precision?
- dooglius 2y ago
  I would think the alignment marks would be included in the photomasks, so they would be part of the chips themselves
- lambda 2y ago
  Putting marks on the chip with high precision is much easier; that's done by the same kind of lithographic process that's used for building up all the other layers of the chip, which is generally via exposing a photosensitive layer of material with light through a mask, and they already have ways of keeping those mask layers in alignment. But aligning multiple chips together is a different process, and while it sounds like they previously had ways to do this via simple optical inspection of those alignment marks, that's less accurate than a holographic alignment using a laser.
  - hinkley 2y ago
    Gates on chips are essentially precise marks. In metal machine shops there are apparently certain things you do all at the same time before you move the parts around because once you move them you can’t ever get them in exactly the same spot again - due to tolerances in the tools. If for instance you want to grind a precision bolt out of solid stock, you slap an oversized piece of bar into the chuck. If it’s slightly off-center it doesn’t matter. If it’s slightly skewed it doesn’t matter, because you’re going to grind a (nearly) perfectly straight section out of the middle of the bar, and at exactly a 90° angle to the chuck. Then cut the threads. And then, you’re going to take the bolt out of the chuck, and use a different tool to cut the head of the bolt. But it doesn’t matter if the head is four thousandths of an inch off of dead center of the bolt because it’s just a surface for the tool and a surface to spread the force of the tension in the bolt. Which is very similar to the idea of building a bunch of chiplets and putting targeting marks on them at the same time, then putting those chips into a device that slices them up and prepares the surfaces for assembly.
iandanforth 2y ago
The last bit about using this same technique for sensors is pretty cool. Ultra-sensitive microphones or touch sensors would be pretty awesome.
- sandworm101 2y ago
  Ultrasensitive microphones are easy and have existed for decades. From a shotgun mic at a rock concert to a laser interferometer detecting gravity waves, the problem is the same. The issue is not detecting vibration but shedding off those sounds you dont want to hear. This new device will certainly be able to detect vibrations, but is likely useless outside of ultraquiet environments.
jrh3 2y ago
Nothing is invented, only discovered.
- mhb 2y ago
  Maybe that's very clever or just dumb semantic nitpicking, but unless you say more no one will know which you think it is.
  - jrh3 2y ago
    I think only discovered. It was always there, but someone(s) just had to do the work to find it.
    - mitthrowaway2 2y ago
      There's a word for this concept, discovering useful techniques that were always there but which nobody else had discovered before; we call it "invention".
      - lmpdev 2y ago
        That’s just Plato’s Theory of Forms
- perching_aix 2y ago
  > Nothing is invented, only discovered. https://en.wikipedia.org/wiki/Invention https://en.wikipedia.org/wiki/Invention https://dictionary.cambridge.org/dictionary/english/invention https://dictionary.cambridge.org/dictionary/english/inventio... https://www.merriam-webster.com/dictionary/invention https://www.merriam-webster.com/dictionary/invention https://www.britannica.com/technology/invention-technology https://www.britannica.com/technology/invention-technology my opinion alarm is jumping off the desk over here, sheesh
  - jrh3 2y ago
    It's a great mental exercise. Reading it, my opinion alarm went off. (Thank you for the source material!)
RachelF 2y ago
This is not novel in general - the same technique has been used in lens alignment for decades.
kabouseng 2y ago
You also need to take into account your manufacturing precision you can achieve with fabricating the lenses, how accurately you can position the lenses relative to the work piece etc. From my cursory reading of the paper it assumes perfect lenses and positioning, and only simulates the alignment procedure. Still, a worthy paper, but as others have mentioned, not much different than the methods used for optics alignment.

170 points by thebeardisred 2y ago

heisenbit 2y ago

Isn't this very similar to the way optical position encoders work?

scheme271 2y ago
I think the difference is that you don't need to change the focus of the optical sensors in order to verify alignment. So you don't have to worry about movement while the focus is changed.

lawlessone 2y ago

Could this also be used for manufacturing lots of other microscopic things? layer by layer? "Instead, their method finds errors up to 0.017 nm along side-to-side measures (x and y axes) and 0.134 nm when assessing the distance between the two chips (z-axis)." Could you make some very very sensitive and tiny seismic sensors with this? edit: " Arbabi also points out that this method can be used to make displacement sensors that can be used for measuring displacements and other quantities. "Many physical quantities that you want to detect can be translated to displacements, and the only thing you need is a simple laser and a camera," he says. For instance, "if you want a pressure sensor, you could measure the movement of a membrane." Anything that involves movement—vibration, heat, acceleration—can in theory be tracked by this method. "

g5pw 2y ago
I made a home-brew seismic sensor using something similar, a hard disk head arm assembly, a cd-rom laser (which has an anisotropic lens and four photodiodes) and a Red Pitaya used as PID, so I guess it can be done!
- pfdietz 2y ago
  Oh man, this is giving me The Amateur Scientist vibes. What would C. L. Stong be doing with today's technology?

Etheryte 2y ago

Maybe I'm missing something here, doesn't this simply move the precision problem to a different part of manufacturing? Previously you had to be precise with aligning the chips, now you have to be precise with how you put those alignment marks on the chips you want to align. Am I missing something here? Or is it considerably easier to put the marks on the chips with sufficient precision?

dooglius 2y ago
I would think the alignment marks would be included in the photomasks, so they would be part of the chips themselves
lambda 2y ago
Putting marks on the chip with high precision is much easier; that's done by the same kind of lithographic process that's used for building up all the other layers of the chip, which is generally via exposing a photosensitive layer of material with light through a mask, and they already have ways of keeping those mask layers in alignment. But aligning multiple chips together is a different process, and while it sounds like they previously had ways to do this via simple optical inspection of those alignment marks, that's less accurate than a holographic alignment using a laser.
- hinkley 2y ago
  Gates on chips are essentially precise marks. In metal machine shops there are apparently certain things you do all at the same time before you move the parts around because once you move them you can’t ever get them in exactly the same spot again - due to tolerances in the tools. If for instance you want to grind a precision bolt out of solid stock, you slap an oversized piece of bar into the chuck. If it’s slightly off-center it doesn’t matter. If it’s slightly skewed it doesn’t matter, because you’re going to grind a (nearly) perfectly straight section out of the middle of the bar, and at exactly a 90° angle to the chuck. Then cut the threads. And then, you’re going to take the bolt out of the chuck, and use a different tool to cut the head of the bolt. But it doesn’t matter if the head is four thousandths of an inch off of dead center of the bolt because it’s just a surface for the tool and a surface to spread the force of the tension in the bolt. Which is very similar to the idea of building a bunch of chiplets and putting targeting marks on them at the same time, then putting those chips into a device that slices them up and prepares the surfaces for assembly.

iandanforth 2y ago

The last bit about using this same technique for sensors is pretty cool. Ultra-sensitive microphones or touch sensors would be pretty awesome.

sandworm101 2y ago
Ultrasensitive microphones are easy and have existed for decades. From a shotgun mic at a rock concert to a laser interferometer detecting gravity waves, the problem is the same. The issue is not detecting vibration but shedding off those sounds you dont want to hear. This new device will certainly be able to detect vibrations, but is likely useless outside of ultraquiet environments.

jrh3 2y ago

Nothing is invented, only discovered.

mhb 2y ago
Maybe that's very clever or just dumb semantic nitpicking, but unless you say more no one will know which you think it is.
- jrh3 2y ago
  I think only discovered. It was always there, but someone(s) just had to do the work to find it.
  - mitthrowaway2 2y ago
    There's a word for this concept, discovering useful techniques that were always there but which nobody else had discovered before; we call it "invention".
    - lmpdev 2y ago
      That’s just Plato’s Theory of Forms
perching_aix 2y ago
> Nothing is invented, only discovered. https://en.wikipedia.org/wiki/Invention https://en.wikipedia.org/wiki/Invention https://dictionary.cambridge.org/dictionary/english/invention https://dictionary.cambridge.org/dictionary/english/inventio... https://www.merriam-webster.com/dictionary/invention https://www.merriam-webster.com/dictionary/invention https://www.britannica.com/technology/invention-technology https://www.britannica.com/technology/invention-technology my opinion alarm is jumping off the desk over here, sheesh
- jrh3 2y ago
  It's a great mental exercise. Reading it, my opinion alarm went off. (Thank you for the source material!)

RachelF 2y ago

This is not novel in general - the same technique has been used in lens alignment for decades.

kabouseng 2y ago

You also need to take into account your manufacturing precision you can achieve with fabricating the lenses, how accurately you can position the lenses relative to the work piece etc. From my cursory reading of the paper it assumes perfect lenses and positioning, and only simulates the alignment procedure. Still, a worthy paper, but as others have mentioned, not much different than the methods used for optics alignment.