Alternatives Implementations

By: Maynard Handley (, July 14, 2020 10:10 am
Room: Moderated Discussions
Doug S ( on July 13, 2020 11:25 pm wrote:
> Maynard Handley ( on July 13, 2020 10:37 pm wrote:
> > I used to think this (mask costs) was a big deal. I’m no longer convinced.
> > A 5nm mask set supposedly costs $15M. For comparison Apple sells ~20M Macs per year.
> Do you have a source for that? I was under the impression a 7nm mask set was in the ballpark of
> $30 to $60 million, though since TSMC's 5nm is using EUV there are fewer masks required (due to
> the lack of double/quad patterning steps) but I'd be surprised if it was half as many since the
> DP/QP stuff is only done for a few layers at the finest pitch. Not that I dispute your overall point
> that mask costs aren't a big deal when you have Apple's volumes and the Mac's cost structure.

I'm not sure where I got that number, but a search right now turned up:
(do a search for mask set)

Obviously that's for 7nm, but it's for EUV, so I don't think it's wildly off.

BTW I mentioned mac volumes not as rah rah, but as a calibration of how many chips intel sells in, for example, the "high end laptop/desktop" bucket. Obviously the PC vendors have much higher volumes, but I don't know them, and even the Apple volumes justify a certain level of die segmentation via different masks.

BTW, as an interesting fact, while trying to (re)find this number I also found which prices the A12 at $72. This is interesting both insofar as it's higher than the usual estimate people throw out (~$30..$40) but also insofar as an iPhone SE2 has an A13 and costs $400. Which makes one wonder how reliable these cost estimates are -- $72 seems a substantial fraction of $400, and a lot more also has to go into an SE2.

> What I wonder about though - and maybe those in the industry can enlighten me - is that you don't really
> just need ONE mask set, right? No design is perfect out of the gate, you do the original design, prepare
> a mask set, run a hot lot and get samples back and test them and you find problems. Then you update
> the design to fix problems and you need new masks. You won't change every layer so the entire mask
> set doesn't need to be replaced, but maybe 1/3 or 1/2 of it does since the layers at the finest pitch
> will be the ones most likely to change. And you may do this through several revs.
> The other thing I'm unclear on is if you want to produce something at high volume using multiple
> lines at the fab, do you need separate mask sets for each line? Maybe that's kind of irrelevant
> since if you are producing at such high volume you are running on more than one line all that
> cost is divided out by enough chips that the mask cost per chip is still pretty small?
> > But I think saying "oh well it is only 75 cents per chip so who cares" is certainly too low, though
> > certainly correct within an order of magnitude. Even then if we WERE talking $7.50 per chip it will
> > still an "oh well who cares" because that's still small compared to the current Intel share of the Mac's
> > BOM. Apple is probably paying Intel $100 for every Mac (even if the i3 CPUs in the lowest end ones cost
> > under $100 I'll bet it is over that mark once you include the southbridge) so they have a lot of slop
> > to play with and still come in cheaper. Or at least no more expensive. I'm making the assumption that
> > cutting cost is NOT Apple's primary reason for switching to ARM, so they aren't going to worry about
> > spending a few extra dollars on masks if that's what it takes to meet their actual goals.

Well you can look at the number and slice it two ways.
One way is to say that at a certain (not that high) volume it makes sense to create a new mask set for a design that's cheaper because it better matches the alternative requirements.

Another way (this is what I was trying to drive at) is that it still doesn't make sense to slice your number of designs extremely fine because there are other NRE costs (like design and validation) that swamp mask costs.
Hence if you're Intel you sell cheaper chips not by as many designs as SKU's but mainly via fusing (you know my opinions about the [mostly non]importance of die scavenging).
The question then is where's the economics cutover point for it to make sense to go from one mask set (plus fuses) to a smaller die and different mask set. I fully accept that NRE costs limit the number of "baseline" models the company, and any company, is working with to surprisingly few; my only contention is that I don't think mask sets is the primary reason for this, IMHO it's design/validation (and even more so for Intel than for say the ARM and GPU vendors).

Of course in the particular context of Apple, they come out looking great because (at least so far) they see no need to artificially cripple any model, so no fuses, even the lowest-end offered is awesome.
For someone like QC that sits halfway between Apple and Intel in terms of "offered" diversity, how many "baseline models" are they working with? Any guesses?
Three? And the rest is fusing?
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Skylake-SP area breakdownDavid Kanter2020/07/12 06:13 PM
  Skylake-SP area breakdownanon22020/07/12 07:01 PM
    Skylake-SP area breakdownTravis Downs2020/07/12 08:02 PM
      Skylake-SP area breakdownanon2020/07/12 08:44 PM
  Skylake-SP area breakdownTravis Downs2020/07/12 08:03 PM
    Skylake-SP area breakdownDavid Kanter2020/07/12 08:20 PM
      To elaborateDavid Kanter2020/07/12 08:22 PM
        To elaborateTravis Downs2020/07/13 07:03 AM
          To elaborateAnon2020/07/13 07:36 AM
            To elaborateAdrian2020/07/13 01:45 PM
              To elaborateAnon2020/07/13 02:06 PM
                To elaborateChester2020/07/13 08:30 PM
  Alternatives ImplementationsKyle Siefring2020/07/13 06:02 PM
    Alternatives ImplementationsTravis Downs2020/07/13 08:41 PM
    Alternatives ImplementationsMaynard Handley2020/07/13 10:37 PM
      Alternatives ImplementationsDoug S2020/07/13 11:25 PM
        Mask costsDavid Kanter2020/07/14 08:13 AM
        Alternatives Implementationstarlinian2020/07/14 08:22 AM
          Alternatives ImplementationsDoug S2020/07/14 10:03 AM
          Alternatives ImplementationsMaynard Handley2020/07/14 10:12 AM
        Alternatives ImplementationsMaynard Handley2020/07/14 10:10 AM
          Alternatives ImplementationsDoug S2020/07/14 10:47 AM
            Alternatives ImplementationsBrett2020/07/14 01:38 PM
            Alternatives Implementationstarlinian2020/07/14 02:30 PM
Reply to this Topic
Body: No Text
How do you spell avocado?