AMD and Global Foundries agreement not what it seems

A good deal, but not about yields

AMD Fusion LogoThe AMD (AMD) and Global Foundries Wafer Purchase Agreement (WPA) that was released yesterday made little to no sense. On a conference call today, AMD’s Interim CEO Thomas Seifert filled in the missing pieces, it all makes sense now.

Few things are more beloved by journalists than a 5:30am PST financial conference call, but this one was worth it, especially in light of the questions left hanging by yesterday’s announcement. We stated that on the surface, it sure sounded like AMD was tearing Global Foundries a new reticle for use in debugging their 32nm process. That however contradicted the facts we had heard on the ground, as of late last year, there simply were not 32nm yield problems. So why was the press release written the way it was, and is really going on?

There are several key bits to the puzzle, most stated in the Q&A session following the conference call itself. The first was that the known good die payment model is for 2011 only, and then things move back to the ‘cost plus’ model in 2012 and beyond. The first clue is in the AMD PDF for the call, on page 7 it says, “In addition, AMD agreed to make an additional quarterly payment to GF during 2012 if GF meets specified conditions related to continued availability of 32nm capacity in 2012.”

The second clue was that Mr Seifert said that original discussions were started last year “when ramps were more challenging.” He also intoned that it took a long time to renegotiate, something we suspected too. This is the long way of saying that there were some long chats between AMD and Global Foundries when thing were looking ugly on the 32nm process last summer. The result was pricing per known good die, something that, contrary to what was intoned in the call, is simply not done in the industry.

Given the relationship between AMD and Global Foundries, plus the number of other foundries that can supply 32nm SOI chips in volume, there is no potential second source. Known good die pricing was one of two realistic solutions, the other was to simply not make any wafers until 32nm yields improved to the point where the chips could be produced at an economically viable cost. If Global Foundaries’ one 32nm SOI volume customer held off on running wafers, Glofo would have an interesting job ramping volume yields without any volume.

Luckily, as SemiAccurate exclusively reported, the yields did improve dramatically late last year, so the known good pricing model was kind of a moot point. It was in force by then though, and another clue was that Mr Seifert also said that the agreement covers 2011, including the recently closed Q1. This means it has been in effect for 3+ months, and was likely mostly finalized before the yield ‘breakthrough’ was known about.

Tying it all together was an easy to overlook comment in the Q&A session about crossover points and margins for the year. That was the key to it all, it isn’t about hard numbers, yields, or payments, it is all about smoothing out lumps now.

Basically AMD was extremely worried about 32nm yields last year, enough so that they twisted Global Foundries arm to give them an insurance policy. Fair enough, especially in light of the closer ties between the two companies in the future. Yield issues were solved, so the impetus for this agreement was removed about the day the first production wafer ran.

The same goes for the upside on capacity, or the flexibility that was talked about in the call. It is about ensuring AMD has enough chips to meet the demand for their customers, not being able to order less as some suggest. That is the polite way of saying, “If you promised us 75% yields and 5000 32nm wafers a quarter, and you only deliver 25% yields, you will find a way to deliver us 15000 32nm wafers at the same time for the same cost.” Can you feel the love?

In 2012, there are quarterly payments from AMD to Global Foundries, a curious bit that doesn’t seem to make sense on the surface. In light of the rest though, it looks like AMD is going to take whatever they can get in 2011 at whatever yields GloFo can deliver. If those yields are low, AMD wins, but currently SemiAccurate’s sources say that they are above any trigger points that would cause GloFo pain.

These payments seem to be a way for AMD pay for yields when and if they go above the required point, or if AMD needs more capacity. If yields were as bad as feared when this agreement was signed, there might be no payments at all in 2012. If they are above the required points, then you will see the $400 million payments described in the call. AMD seems to think there will be payments to GloFo, so that should be another big clue about the yield progress. It also nicely explains the money AMD is paying GloFo, $1.5-1.9 billion minus $1.1-$1.5 billion is what again?

So, if things are going swimmingly, and the need for this agreement was obviated before the first wafers ran, why not just tear it up and pretend it never existed? Well, there is one other benefit for AMD, but not necessarily to Global Foundries.

Normally, a new process does ramp yields, and there is a curve. This means that chips made on a new process are expensive and can have both functional (speed/power use) problems and supply shortages. Both impact what a chip company can sell and for how much. For AMD, the worst case is that they initially get bad bin splits and low volume at a very high cost.

Even if the normal production ramps fix this in short order, they might have a quarter or two of lower than hoped for revenue at a higher than hoped for cost. On top of this, supply shortages for a hot product tend to peeve OEMs, burning bridges just as things are getting better. AMD would end up with much lower revenues, much higher costs, and the metric that Wall Street seems to care about most, margins, would tank.

That is where the payments to Global Foundries come in. It smooths out AMDs margins for 2011 when they need it, and allows them to avoid some parts of the initial financial hit that come with ramping chips on a new process. It also frees up money for AMD to use in banging the proverbial drum about the chips, spending more on sales, advertising, and generally using their cash to a more productive end.

Normally, two companies, even two very close companies like AMD and Global Foundries, would never do something like this, it goes against several widely held business practices. AMD is partially owned by the owners of GloFo, and AMD still owns a chunk of Global Foundries too. One thing that those owners, ATIC, do understand is money, especially long term strategic investments. What ATIC appears to be doing here is taking a small hit to current Global Foundries income in 2011 to make AMD, another of their investments, and the key factor in GloFo’s success, stronger. For them, it looks like a long term play, and is a win/win.

So there you have it, in a convoluted way, this announcement and agreement is not really a big deal. What was started in a panic about yields ended up not being about that at all, it just moves some payments around to allow for better resource use. The net result is that AMD gets more even numbers in 2011, cost per good die effectively equals set margins, and any differences in cost are deferred a year. As of now, there doesn’t seem to be any net win or lose in cash, potential financial hiccups are avoided, and things look steady for the investment community. They like that. Quite tidy and happy, except for the ‘incentivize’ headline.S|A

The following two tabs change content below.

Charlie Demerjian

Roving engine of chaos and snide remarks at SemiAccurate
Charlie Demerjian is the founder of Stone Arch Networking Services and is a technology news site; addressing hardware design, software selection, customization, securing and maintenance, with over one million views per month. He is a technologist and analyst specializing in semiconductors, system and network architecture. As head writer of, he regularly advises writers, analysts, and industry executives on technical matters and long lead industry trends. Charlie is also available through Guidepoint and Mosaic. FullyAccurate