5

I just read the answer to this question asking how much a custom ASIC costs. It says that

When it comes to making an ASIC, the cost of the masks is HUGE. It is not uncommon at all for a set of masks (8 layers, 35 to 50 nm) to run US$1 Million!

I'm curious why the cost is so huge. Where exactly does all of that money go? Is it the material that's expensive, or does it take a lot of man-hours to make the mask, or something else?

  • I can't speak to the exact numbers. I worked in the "back room" so to speak. But there are lots of reasons I'm familiar with. The cost of an electron beam emitter gun, by itself alone, is in the vicinity of $50k to $100k each. They use Wehnelt emitters which wear out quickly, when the owner doesn't properly maintain its temperature -- requiring replacements on 2-month cycles instead of the 2-yr cycles that are possible with good management (hard to achieve, but doable.) The brighter they operate the guns, the faster they can make masks. But the faster the gun wears out, too. – jonk Jun 15 '20 at 02:37
  • Also, masks require fancy software. That's because, as the feature sizes have shrunk, the optical system is no longer able to produce the expected image at the beam focal point. Instead, there is some very complex mathematics that is used to predict the resulting mask and to then control the electron beam. In a sense, this is like knowing that to get a (+)-sign shape in the mask, one needs to instead generate an (X)-sign trajectory. If that makes any sense. So the software is complex and requires continual work to keep it up to date. Not to mention the NRE for all this. – jonk Jun 15 '20 at 02:37
  • Welcome. Mask are made to be used in lithography, so ultra-precision is needed. The diamond lens used to focus the image may cost $10 million USD, not to mention the cost of a extreme UV light source. –  Jun 15 '20 at 02:39
  • Sorry, I didn't realize that. Is there a stack exchange site where questions on ASIC manufacturing would be on topic? In any case, thank you for your answers, I'll read some more on this very interesting topic. –  Jun 15 '20 at 02:46
  • @user359787 Try https://meta.stackexchange.com/questions. There are over 170 sites, so possibly one of them handles more abstract questions about electronics. –  Jun 15 '20 at 02:56
  • 8
    I disagree with Sparky and consider this on topic. No need to look for another forum unless 4 more people with vote-to-close power agree with him. Or if you don't get a satisfactory answer (which is possible because I don't think there are many users here who actually work in semiconductor manufacturing). – The Photon Jun 15 '20 at 03:12
  • The only thing I can add is that I once designed a board to be used in a lithography machine and the mark-up from our BOM cost to our sales price was staggering (possibly justified by the low sales volume). And our customer would mark it up again before selling it to an actual manufacturer like Intel or TSMC. – The Photon Jun 15 '20 at 03:13
  • 1
    Masks for obsolete processes (Small wafer size and low resolution and fewer masks in a set) still used in some specialty products are at least two orders of magnitude cheaper than state-of-the-art, based on some very niche work we did a few years ago. – Spehro Pefhany Jun 15 '20 at 05:11
  • 6
    @ThePhoton There may not be many semiconductor industry people here because every time a question that leans more to the semiconductor side of things is asked they are told it is off topic. – Matt Jun 15 '20 at 10:44

1 Answers1

1

I do not have the precise data, but I've worked in ASIC development on products that easily exceeded tens of millions of units in production.

Machines and technology are expensive in part because of the R&D that goes into it. But also in the precision required to make them.

One challenge is that the physical end precision required is smaller than the useable wavelength used during photolithography.
That requires special techniques relying on interferences to get the desired results. Those interferences are "planned" in the masks. Also, at these precisions the masks must be so perfect that every artefact is effectively represented. A single interrupted line implies a missing connection, a missing transistor, etc. Precision impacts transistor properties, etc. Any single "error" is enough to not be able to produce the ASIC and you'ld find out after minimum one to two months if the error is undetected.

Making sure that every mask produced is "perfect" and so precise requires a bunch of R&D in many fields and high precision in building the machines that we need relatively few off.

You can put that into perspective with the fact that it is currently reported that a semiconductor fabrication facility costs up to 20 billion USD and 3 to 5 years to build (cutting edge technology).

Besides the masks, that also drives other costs up, in particular ASIC verification costs. Given the cost and delay in making the first prototype, design errors must also be avoided and that in itself also requires a lot of technology and engineering efforts.

So all that makes cutting edge technology only interesting for mass production. So that decreases the number of masks that are actually required in that technology which also drives the individual mask price cost up.

le_top
  • 3,111
  • 11
  • 22