Groo (charlie.delete@this.semiaccurate.com) on August 3, 2022 11:15 am wrote:
> anon3 (Anon3.delete@this.3.com.nonotthat3com.test) on August 1, 2022 10:33 pm wrote:
> > Groo (charlie.delete@this.semiaccurate.com) on August 1, 2022 12:28 pm wrote:
> > > Then there was the fact that the second gen
> > > media that was supposed to go into Barlow never came out. Or the third. And there is a lot more.
> > >
> > > This is the long way of saying your supposition is correct.
> >
> > Yep, classic story: shove too-expensive product into market, get weak uptake (but mostly
> > from your usual bleeding-edge types who clear paths), wait for the cost reduction to actually
> > build the market, cancel the cost reduction, panic, kill the product family. Business as
> > usual! Oh wait, those last few are only for Intel. Deep pockets, and all that.
> >
> > It was very telling that they weren't able to shrink 3DXpoint at all, or at least not by any
> > amount worth the trouble to ship. Normally I'd blame the weirdo PCM stuff, which is at least
> > genuinely weird, but then I remember the other achievements of this particular R&D department.
>
> Agreed. That said one thing no one seems to talk about is that Xpoint suffers from additive errors with layer
> count, flash does too but to a far lower degree. To me what is telling is they never made a 4-layer version of
> what they had, that should have been an obvious and 'easy' next step that for some reason never happened.
>
> -Charlie
>

Is there any activity in selling the relevant patents and IP?
It seems to me plausible that there's still something there that can be productized, it just needs
- a different engineering team with fresh eyes (maybe dice first, use known good dice to form synthetic wafers, and stack those synthetic wafers ala CoWoS or string stacking?)

- a marketing team willing to serve a variety of new customers with new demands (I'm thinking of eg simple wearables where slow RAM is OK'ish, and the power savings from DRAM refresh are appealing, as is a single chip solution replacing both DRAM and flash), as opposed to a marketing team that saw their mandate as "compel a captive market to have to upgrade to Xeon Platinum". In particular the biggest sensible market might very much prefer dice that handle say 8GiB rather than dice that start at 256GiB, and that are willing to offer some flexibility in the exact pinouts to fit a small form factor.