[sdiy] Blimey

[brianw]

I went looking for a definition of FD-SOI, and found the following links:

https://www.st.com/content/st_com/en/about/innovation-and-technology/fd-soi.html

https://semiconductor.samsung.com/news-events/tech-blog/fd-soi-the-disruptive-innovation-samsung-foundry-is-leading-to-overcome-the-limits/

Looks promising for analog, but it remains to be seen whether designs like the chips from THAT Corp will be updated (redesigned) to use it.

Brian


On Sep 4, 2026, at 11:45 AM, Mike Bryant wrote:
> The only major general purpose DI process I know of nowadays is the GaN process at Imec.  There probably are some in universities for specialist work, but no volume capability.
> 
> Most DI designs moved to FD-SOI which uses more expensive wafers but gives a much higher performance solution for r.f. and high speed analogue as the intrinsic capacitances are much lower.
> 
> But I've never seen a high performance audio device built on FD-SOI.
> 
> From: brian
> Sent: 04 September 2026 19:11
> To: SDIY <synth-diy at synth-diy.org>
>  
> Yes, very unique.
> 
> The story was told at an AES meeting (maybe more than once) about THAT Corp buying this 4" fab from the military or some other company that was retiring it. Even at the time, it was basically a dead process that nobody wanted, when almost all US fab had moved overseas already, but THAT Corp could use the fab for Dielectric Isolation capabilities - effectively like discrete transistors. It looks like THAT Corp kept this process alive as long as possible, but there's just no way to sustain it.
> 
> I think it was Bill Whitlock who gave the AES talk ... but it might have been someone in the audience from THAT Corp who told the history of the fab. Bill designed the InGenius chip that works almost like a perfect transformer, so he is tied to the company despite being president of Jensen Transformers.
> 
> Thanks for the notes about power transistors. I was thinking that might be the closest technology that might be able to use this process. I'm not surprised that power transistors can use a different process, because they're already isolated by design. i.e. I assume that power transistors don't need the DI technology THAT Corp uses in their long list of to-be-discontinued chips.
> 
> My question is whether dielectric isolation is possible with a smaller scale process on a larger wafer.
> 
> Brian
> 
> 
> On Sep 4, 2026, at 9:22 AM, Mike Bryant wrote:
> > It's moreorless a unique process.  And the volumes are far too low for most foundaries to be interested.
> >
> > Ain't going to happen.
> >
> > From: Brother Theo
> > Sent: 04 September 2026 17:07
> > To: Mike Bryant
> > Cc: brian; SDIY <synth-diy at synth-diy.org>
> > 
> > Maybe one of the overseas fabs will pick up the product line.
> > --Timster
> >
> > On Fri, Sep 4, 2026 at 5:08 AM Mike Bryant wrote:
> > Just to add one point - a processed 4" wafer costs about $300 to $1000 before test and packaging.  I suspect their process is near the top end as it is quite specialist.  So I expect we're talking about $25k-ish for each device you want to do a lifetime buy on.
> >
> > Hopefully the distributors will do this for some of the more popular devices.
> >
> >
> > From: Mike Bryant
> > Sent: 04 September 2026 09:53
> > To: brian; SDIY <synth-diy at synth-diy.org>
> > Subject: Re: [sdiy] Blimey
> > 
> > In my day job as a semiconductor consultant we see obsolete fabs come up for sale all the time.  Over half of them have become supermarkets or warehouses.  There's about a thousand times difference in the silicon area output of a modern 12 inch fan and an early 1980s 4 inch fab, and another ten times in the value per area.
> >
> > A 4 inch fab cannot make the devices people want - memory and AI chips.  It cannot even make most of the supporting chips around them which are almost all on 8 inch now.  Even power transistors are moving to 12 inch, and about three quarters of all discrete transistors in the world, no matter what the brand, are made in a single huge 8 inch fab plant in Guangdong.
> >
> > So no government is even going to notice this one closing.  And as the press release says, they couldn't keep the fab full with their orders.  Fabs usually need to be at 85-90% capacity just to break even - the profit is in the final 10-15 percent as the basic materials used are dirt cheap - it's almost all value add.
> >
> > It used to be that the equipment from old fabs was bought up by Russia to make its military devices, but that door is now closed of course, so the only value of this fab is the shell.
> >
> > Their press release says they will take lifetime orders by end of this month, then run them all until they have finished them before then shutting the fab.  I expect MOQ (minimum order quantity) will probably be 25 wafers worth so about 50,000 devices.  One possibility to reduce upfront cost would be to have them process the wafers and deliver them to you in sealed containers, then only have them cut, tested and packaged a few at a time.  There are numerous specialist small companies still doing DIP and other obsolete packaging.