>>>That makes no sense. SOITEC talks up their technology >because they need to convince
>>>companies to use it. Frankly, they have a poor history of >doing so. There was a
>>>huge hyping of PD-SOI (and ZRAM). Neither of which panned >out really. PD-SOI is a niche technology, and
>
>
>Actually the problem might not be SOITEC at all, and or PD->SOI being a niche technology.
>The problem is that PD-SOI might have *required*(PD-SOI >will be gone soon to) some
>peculiarities to be the best at some particular node >size... and the BEST most developed
>SOI process is IBM, which licenses its fab tech to others.

The problem is the technology (PD-SOI) that SOITEC was pushing. It simply wasn't appropriate for most products.

It was very expensive and had two main benefits:

1. Higher speeds (although at modern nodes, it's only ~5-7%)
2. Lower soft error rate in SRAM

Additionally, it's better for optical stuff, but that's mostly irrelevant, since nobody was doing optical on the same die as high performance CMOS in that time frame.

On the other hand:

1. Designing I/O and PHYs is a royal pain in the ass.
2. No ESD, you need to dig through the insulator to create a 'bulk-like' region of the die.
3. Floating body effect
4. It's much more expensive
5. Not compatible with standard tools and 3rd party IP*

*This wasn't a problem for IBM, they have their own IP and tools; AMD also had their own fab. For TSMC though, it was a huge issue.

>It may take a considerable amount of time even for a >company like Intel, to develop
>SOI to be as effective as IBM, but by going around IBM >patents. It could simply destroy their time-to-market >strategies.

That's an irrelevant point. Most of these companies have (or can obtain easily) cross-licensing agreements. Intel already runs SOI in their research fab. Just not for high volume production.

>And i'm not seeing a company like Intel, licensing from >others techs so crucial
>to make their chips, least of all under the current CEO...

Intel, IBM, AMD, etc. all have cross license agreements with each other. That's not an issue.

Not invented here may be an issue, but I don't think it's really the dominant one.

Remember: TSMC also investigated PD-SOI and decided to stick with bulk.

So when I hear about SOITEC pushing new technology, I look at their track record, and it's not very good. Perhaps this time it will be different. But SOITEC has a history of pursuing interesting technology that isn't appropriate for the broader market.

The two largest logic manufacturers, Intel and TSMC, both chose to pursue FinFETs, so it is clearly an economically viable technology. Intel's results also show that it is technically quite solid as well. While you might argue that they are just being 'cheap', the reality is that most companies use TSMC and are obviously happen with their fabs.

The FD-SOI supporters are much smaller companies, with rather different concerns. ST seems to have been motivated by time to market, and perceives FD-SOI as a way to get a competitive advantage over TSMC customers. That's crucial, since ST needs to justify why they are still an IDM. Just as importantly, FD-SOI has low fixed costs, which are important for a smaller company with less business on cutting edge nodes.

IBM is quite small and focused on proprietary designs.

We'll see how many GF customers end up adopting FD-SOI.

But I'm relatively skeptical.

David