Exclusive: Intel's new CEO explores big shift in chip manufacturing business

[osnium]

I'm speaking primarily about foundry here, though my comments do apply to the design side as well. Neither Apple nor Nvidia have great motivation to poach from Intel's foundry staff. They design chips, but don't build them.

And yes, Intel did lose a lot of experience with the voluntary retirement packages, but a lot of talent still remains. And TBH the most experienced talent may have been the best talent to lose on the manufacturing side, because they are going to have to learn a completely new way to run the fabs as external foundries. Your most experienced talent has the most to unlearn.

On the process design side this is less true. The changes to how Intel operates are going to be less extreme, though I do believe the mentality in process design has to change to focus on looking for larger process windows up front rather than the "we'll fix this in the next node" mentality.

The fact that other semiconductor manufacturers are struggling (Wolfspeed declaring bankruptcy, Microchip doing furloughs, TI Lehi doing layoffs) means that Intel foundry employees who are not touched in the latest layoffs will also stay because there are not too many other options out there.

 

[hist78]

Would could policy #1 entail if Intel does not successfully execute on 14A? TSMC Arizona all the way with some commitment to do some process development in Arizona? Could it involve some sort of merger with GlobalFoundries along with a government golden share akin to Nippon Steel and US Steel? Given all this uncertainty, would you expect any and all competent Intel foundry employees to jump ship?

I'm not sure if U.S. government policy can go into such detailed levels, like 20A, 18A, or 14A, just for one company, Intel. From Trump Administration 1 to the Biden Administration, and now Trump Administration 2, Intel itself has gone through several changes. If government policy dives too deep into specific nodes, it risks being hopelessly dragged in every direction.

For example, Intel’s “5 Nodes in 4 Years” plan includes Intel 7, Intel 3, Intel 4, Intel 20A, and Intel 18A. Intel 7 reached high volume production, but it wasn't technically or cost competitive. Intel 3 and Intel 4 are good, but Intel moved their production from the US to Ireland. Intel 20A was canceled after the company declared that Intel 18A's development was going so well there was no longer a need for 20A. Now, recent rumors suggest Intel 18A may not be competitive enough to attract large external orders.

How can an effective government policy keep up with the twists and turns of private companies, especially when even Intel may not have full control of its own direction?

I haven’t even mentioned that Intel missed the critical and lucrative AI wave. Can we really expect the US government to know the semiconductor industry better than Intel?

The U.S. government can incentivize and create a favorable environment to encourage domestic semiconductor manufacturing. But the rest will ultimately be determined by the market itself.

If Intel succeeds, that’s great. But what if Intel fails? That’s why the TSMC Arizona fab project is so important to the overall strategy.

 

[KevinK]

The Review contains many very nice photos of all kinds of CPO chips/boards/hardware, for a non-semi-expert like me very digestible and simple....Here an Intel CPO-chip from 2021:

This recent Lightmatter rollout a couple months ago is heavy on the marketing but explains the current AI interconnect limitations and challenges quite well, along with sample applications of CPO to solve those problems. Very digestible as well. Seems like race is on for silicon photonics and CPO for the AI data center sweet spot.