AMD (AMD) just amended it’s wafer purchase agreements with Global Foundries in a way that appears to take a poke at the foundry’s yields. This does not appear to match what is happening technically, so that brings up the question of what is going on. Update soon. Will be posted as part II.
The short story, you can read it in this press release, is that the Wafer Purchase Agreement between AMD and Global Foundries has changed. AMD used to pay per wafer, as is the industry norm for dealing with foundries, and they still do for 45nm wafers. On 32nm wafers, AMD is now paying per good chip, something that is never done, even when things are horrifically bad. On top of that, AMD also takes a $492 million non-cash gain in Q1 relating to Global Foundries, the Chartered Semiconductor acquisition, and how that deal is structured.
Reading the title of the release, AMD is ‘incentivizing’ Global Foundries to “continue improving 32nm yields.” This one isn’t hard to parse, and if you hear the voice of a particularly strict grade school teacher when you read this, you will be forgiven, it has that not so subtle tone of a veiled threat. What they are implying is that yields on the 32nm process are bad enough that it isn’t worth making chips on it unless there are some financial tweaks to the contract. Those tweaks are the per good die pricing.
One problem with this theory is that since last November when SemiAccurate exclusively reported about a yield breakthrough on that very process, there have been no credible reports of problems on the process. As late as last week, several sources told SemiAccurate that things were going swimmingly. Llano has been in volume production since February or so, and it is still set to launch in just over a month. If there were problems serious enough to cause a change in the Wafer Purchase Agreement during the last week, there hasn’t been enough time to try and debug it, much less renegotiate a contract like this.
Something is off here, and it is not technological. First, if any foundry is having severe problems ramping up a process, the last thing that they need to be told is that they should improve. They know this. Any multi-billion dollar company tends to be aware of large problems in its core business, and fixing that is usually a top priority. If there are problems, customers pointing it out in a public release doesn’t make a foundry work faster either.
Similarly, such statements don’t do much if there aren’t yield problems in the first place, and in this case, we can find nothing that point to such problems. We looked, hard. All we found were that things were slightly better than last time we checked.
Another curious thing is that if AMD is so worried about Global Foundries 32nm and its half-shrink to 28nm, why are they upping their wafer starts in 2012 by about 1/3? In the release, AMD is set to purchase between $1.1 and $1.5 billion from Global Foundries in 2011, and that rises to between $1.5 and $1.9 billion in 2012. That sure doesn’t sound like a customer that is unsure of where things stand to us.
Add in that the owner of Global Foundries has a large stake in AMD itself, and it is obvious that things are nothing like the release is intoning. In fact, all the release says is that there is a changing pricing model and that AMD is going to buy many more wafers from GloFo in the near future.
AMD is going to split production of the recently taped out Southern Islands GPU family between TSMC and Global Foundries, so that accounts for a good number of wafers. On top of that, during their Financial Analyst Day last November, AMD’s Chekib Akrout stated that the successor to Ontario and Zacate, Krishna, will be fabbed at Global Foundries. Between these two and the CPU already at GloFo, almost all of AMDs line will be made on the ‘incentivized’ processes within a year.
That brings us back to the original question, what is the point of this press release? It isn’t yields on the 32/28nm process, those seem to be OK, and it isn’t a rift between the two sides, if anything this brings them closer together. Whatever the real reasons are, they will likely be answered during a call AMD has for the financial community Monday morning. It is going to be quite interesting hear what is said, and more interesting to hear how it is said.S|A
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018