Online information source for semiconductor professionals

To say that 2007 has been a complete mess for AMD is an understatement. Though at a recent financial analyst meeting in early December, AMD executives painted the picture that the AMD glass was half-full rather than half-empty! However, no one bought it, even after numerous referrals back to the point of optimism from the executives in attendance. 

AMD has lost $1.6 billion-plus in 2007 and, based on the ‘half-full’ analysis, plans to be profitable again in 3Q08!

One of the ‘half-full’ hopes put forward by analysts was that the muted ‘asset-lite’ plan that Hector Ruiz had said would be investigated earlier in the year (as losses topped $600 million in a given quarter) would be announced or at least that some flesh would be put on the bones of the plan at the New York meeting.

So it seemed strange that throughout the meeting not one executive mentioned ‘asset-lite.’ Only when asked in the Q&A session did Ruiz say that it was ‘not prudent and silly’ to discuss those plans in public. For competitive reasons, we can only assume?!

Rumours have been flying around that TSMC will be a major foundry partner for AMD in the future and that Fab-38 or Fab-36 and/or the whole complex in Dresden would be sold off, thus saving AMD billions in capital expenditure as it simply hasn’t got that amount in the piggy bank anymore to keep pace with Intel Corporation.

Since the acquisition of ATI, AMD has gained much knowledge around working with major foundries. ATI has been heavily dependent over the years on TSMC while AMD had just started working with Chartered Semiconductor. Going more aggressively down the asset-lite route would therefore seem attractive and one that would be relatively risk-free!

In late July 2007, I wrote a blog highlighting the easy steps AMD could take to an asset-lite business model, as AMD themselves had not seemed to provide any guidance on the issue. My plan comprised 10 steps, many of which have since come to fruition!

Firstly, AMD stopped tool installing at Fab 38 to save cash. It is doubtful that it will be equipped in 2008.

Chartered was kept at minimum monthly wafer contract levels and set about migrating to the 65nm node for dual-core processors. Though not officially announced, strong reports from Taiwan have emerged that suggest TSMC will fab Fusion and some new/old low-cost bulk CMOS microprocessors.

In many respects, AMD has implemented an asset-lite plan, especially when you consider that all of ATI’s production is outsourced. Older dual-core MPUs can be fabricated at Chartered to a greater degree, leaving Fab 36 to be ramped to full capacity and run a mix of quad-core and dual-core MPUs for maximum efficiency.

Fab 38 is idle now and no fab loading in 2008 means that foundries can pick up the demand requirements should there be any!

But many analysts are expecting much more than this to be seen as an asset-lite strategy that delivers improved margins and profitability. What is being asked of AMD is that they feed foundries with a higher proportion of production, close completely Fab 38 and cancel plans to build another 300mm fab in Luther Forrest, New York State that the company is currently taking into consideration.

Ultimately, if Fab-36 cannot be migrated past a given technology node, the company would become fabless and the asset-lite path complete.

Unless I am very much mistaken after listening to the December conference call, these last details are as far apart from the current management’s thinking that I believe this is the key reason why Ruiz hasn’t gone further in announcing more asset-lite plans.

And why? Because there aren’t any!

We are expecting a major announcement with TSMC for its Fusion (SoC-MPU/CPU) next year and this we believe will be produced on TSMC’s low-power Bulk CMOS 45nm process. This is a new platform for AMD and not one that ultimately requires SOI wafers.

Not only that, but with the temporary shuttering of Fab-38, AMD would find it tough fabricating more chips than already loaded at Fab-36. Fab 36 is also ramping native quad-cores and needs to move to 45nm production in 2008. This is as much as it could handle.

We also get the strong impression now that the New York fab will go ahead in 2009 and be the centerpiece of AMD’s internal manufacturing strategy at the 22nm node and beyond when we could see the true integration of multiple MPU & GPU cores on a single die. The NY fab would be the first and only AMD facility built to do that as the Dresden fabs would be trailing-edge by then.

We also see little talk from AMD about not migrating to the 32nm node at Fab 36 as it continues to spend R&D dollars within the IBM Alliance. If the 45nm node was to be AMD’s last internal process node, then like the announcement from Texas Instruments on stopping internal development, it would have done this by now so that it would not need to keep spending R&D resources with IBM.

Looking forward, AMD has a plan to keep internal manufacturing of leading-edge devices for many years to come. The cash-strapped 2007 was tough but not catastrophic for the company. 2008, according to AMD executives, will get better and lookout in 2009 may see the glass half-full once again as the company’s manufacturing side returns to business as usual!

Ironically, AMD may even have the cash or share price recovery in 2009 to re-open Fab 38, just as it starts construction of the NY fab! That would take AMD’s fab portfolio to three 300mm fabs!

Now that’s not very asset-lite, is it?

But when one takes into consideration the much broader range of semiconductors AMD would be fabricating by then, and a likely good percentage of revenues coming from multiple foundries, then surely that set-up is still classified as an asset-lite strategy?

It seems to me that AMD has already described its asset-lite plans in some detail over the last 18 months or so with only the increased TSMC links yet to be confirmed. But having said that, there may be a further link between AMD and Chinese foundry SMIC after last week’s 45nm process deal between IBM and SMIC, especially as it relates to graphics chips!

The problem is that many financial analysts don’t think that this is an asset-lite strategy and I tend to agree. However, what the financial community wants as far as asset-lite is, in reality, for AMD to become fabless. I doubt they would be satisfied with any degree of asset-lite plan unless it meant going fabless.

So for Hector, the asset-lite mumblings are a ruse to keep the financial community at bay. As far as it relates to actually cutting the number of fabs it operates in the coming years, AMD could end up with three in production in the 2010-11 timeframe.