Online information source for semiconductor professionals

The issue of AMD's using TSMC for microprocessor fabrication hasn't gone away and remarks by Rick Tsai, TSMC's CEO and President have further fuelled the fires.

AMD has had two opportunities recently to put some meat on the bones of its asset-lite plans but has declined to do so for ‘competitive reasons' apparently!

So as I have recently become the official ‘unofficial' spokesperson for handling AMD's rumour mill (http://techreport.com/onearticle.x/12840) responses, I thought it best to try and clear up details of the deal being touted between TSMC and AMD. I will also outline some ‘logical' paths that AMD will more than likely be heading down in its asset-lite journey.

But before heading down that road, a basic point about owning fabs needs to be clarified, a point that often gets completely forgotten in the fast-paced hype-o-sphere.

Chip manufacturers go asset-lite or totally fabless when they can't or don't want to go to the next wafer size or invest any further in fabs. If you have recently invested in fabs you don't stop using them as the write-off costs are crazy and investors will say you are mad and need firing!

AMD has migrated to 300mm by building Fab 36 and will migrate Fab 38 to the larger wafer size when it can afford to do so, otherwise it will waste money on a full closure. Fab 38 is not a difficult conversion so it was an easy decision to make and undertake.

However, managing cash flows is required in this undertaking, and so having foundry partners to spread the risk is a good idea. But AMD has more demand for its products than its fabs can actually produce. This comes from the ATI acquisition and when new MPU/GPU platforms start appearing by mid-year 2008, AMD will need even more fab space.

So the use of foundries is based on more rather than less in AMD's case and without cutting the current number of fabs. But as you read on that might all change!

So let's walk through what is happening in 10 easy steps.

Step 1. Conserve desperately-needed cash at this time by slowing/halting the capital spending on the Fab 38 300mm conversion. Slow the tool install at Fab 36 without hitting supply/demand problems; push higher yields and shorter cycle-times until year-end demand requires CapEx spend.

Step 2. Don't use Chartered for anything more than the minimum contract requires as production is on 90nm where profit margins are low. Ramp 65nm asap if Fab 36 can not meet world demand, otherwise push out the 65nm ramp at Chartered by a few quarters.

Step 3. With the intro of Fusion (SoC-MPU/CPU) next year, cater for this by putting in place the foundry partners. Confusion with Fusion is that ATI-type markets for applications will be on 45nm Bulk CMOS while AMD-type markets will still expect SOI.

Step 4. Share out Fusion production to TSMC for the Bulk CMOS 45nm devices, which will require a low-power process that will need high-k at least, with metal gates preferred. This may not be possible until late 2008 from TSMC but first phase low-power would do for the mid-year launch as this is possible from TSMC.

Step 5. Use own fabs for SOI Fusion requirements as they should be able to handle that. Backup of course is Chartered if the price/competitive market allows the fabrication of 65nm MPUs through 2008 and into 2009 with a 45nm shift.

Step 6. Do we complete the Fab 38 conversion in 2008 or 2009? If we haven't done this by end of 2008 forget it and go ahead with the NY fab instead.

Step 7. When the NY fab is built you can pull back outsourcing to foundries to help fill the new fab with new fast-selling CPU/GPU platforms and 88 core MPUs!

Step 8. Don't build NY fab at all. Keep Dresden fabs filled, don't migrate past 22nm using double patterning immersion lithography and end up using these fabs from trailing-edge-legacy products until required and close the place down. The shift at 2012/14 timeframe means all products are now being produced at foundries.

Step 9. By 2014 expect 65 percent margins, lots of frequent flyer miles and membership of the Fabless Semiconductor Association.
 
Step 10. Do none of the above and go bankrupt or sell up and retire early!

Update: 27/07/07
No sooner had this blog been posted than this story appeared: http://money.cnn.com/news/newsfeeds/articles/newstex/

AMD has put into action Step 1 and could easily end up at Step 6!

Update 2: 27/07/07
Strangely I didn't get this press release from AMD yesterday but Chris Tom over at AMDZone posted it yesterday evening US time, long after I was tucked up in bed.

http://www.amdzone.com/modules

However, it confirms a couple more things!

Firstly, that the 65nm migration at foundry partner Chartered will happen in the second half of this year, though Chartered has already highlighted that its 65nm ramp has not met expectations as previously projected. I see this as a clear push-out by AMD, and should actually happen late in 2007.

That means Step 2 has also been implemented.

TSMC will also produce the ATI side of System on Chip (SoC) elements of the hybrid MPU products using Bulk CMOS.

That's Step 3 sorted out.

AMD's Fab 36 gets the credibility it deserves in ramping 65nm with ‘high yields' so early in the ramp phase that corporate decisions have been made easier in preserving cash.  This is a critical time when more money may have been needed (and given) to ensure the migration and allow delivery on previously made promises.

Once again the Dresden team at AMD has come through and allowed AMD to manage the expenses of ramping Fab 36 without hitting demand.

That's the other issue raised within Step 1 rectified.