Intel. From Copy Exactly To Copy TSMC?

there are conflicting views on Intel's decision to break with copy exactly...

In a recent post, we described how Intel is quietly signalling its intent to walk away from its long-practised Copy Exactly (CE) philosophy:

There, we highlighted this as a significant move for the company, and not in a good way. In a nutshell, our view is that this amounts to a tacit admission that Intel’s Technology Development (TD) team is no longer able to ramp new processes to the yield levels that would have traditionally been seen as high enough to warrant the process transfer to a high volume manufacturing (HVM) fab, in a timely fashion.

This move is being presented as the “democratization of innovation” between TD and HVM. In other words, the HVM fab will now have to share the responsibility of ramping yields to mature levels with their TD counterparts.

After publishing my thoughts on this move by Intel, I came across another publication, whose author,

Austin Lyons

had also spotted this change, but, in contrast to me, had hailed it as a highly positive move, details here.

Now, I respect the work that

Austin Lyons

does and that’s why I paid attention to what he had to say. I would strongly encourage you to read the full article, in particular as it relates to CE. Here’s some of what he had to say:

Unfortunately, by design, Intel’s local manufacturing teams never develop this continuous improvement muscle. These folks closest to the problem identify it but have to toss it over the wall to TD for an approved fix. Worse, TD’s focus is split between R&D, qualifying the next process, and fixing HVM issues, so this local fab’s problem may not even be the highest priority for TD.

And that’s why I was so encouraged when Naga explicitly stated that Intel Foundry is doing just that. This requires change. Cultural change.

The need to shift away from copy exact isn’t rocket science, but if you’ve only known Intel manufacturing, it can be hard to see it and make it happen.

Naga has been at Intel less than one year and has 20+ years of experience at Micron (which has its own memory fabs).

Sometimes, a fresh perspective is all you need.

Naga is a welcome addition.

Without getting too much into the detail, at the highest level, he’s saying that TSMC has long adopted a different approach where responsibility for process improvement is shared between their R&D and HVM fabs. While he notes that CE may be or have been appropriate for and IDM, it’s not suited for a foundry model with many different customers and process variations. Therefore, Intel should drop CE and copy TSMC’s approach.

Is he correct? Before we get to that, why is this so important? The reason is simple. The single biggest barrier holding Intel back right now is whether or not they will succeed in ramping 18A to high volume, high yield, and thereafter, doing the same with 14A. How you view the likelihood of this happening is paramount to how you view Intel’s future success. In this equation, pretty much nothing else that Intel does in the coming quarters matters. That’s why it’s so important and that’s why the timing of such a seismic shift away from CE at this critical juncture matters a lot. Let’s dig in…