What is going on is pretty simple, 7nm development is extraordinarily risky, costly, and resource intensive. The company is also separating it’s ASIC business into a wholly owned but stand alone unit so it’s customers can make 7nm designs at other foundries. Officially they are putting the brakes on process development but of the foundries who have taken this step in the past, none seem to have restarted development.
That brings up the question of why GlobalFoundries(GF) is pulling the plug now. They are at capacity, the industry seems flush with customers needing advanced nodes, and anyone offering services is sure to sell out. The problem is money. Development costs at 7nm and beyond are quite frankly insane. Tool costs are spiraling upwards and the advent of EUV is going to add a zero to each very expensive white box the fabs buy. The numbers for any individual step will make your eyes water, in total they are borderline incomprehensible.
If all goes well then you can charge a premium for each wafer but can you charge enough to make the development costs worth it? From a layman’s perspective the answer would be yes you can but there is one big catch, value. There are some things you can only do in an advanced process but most of the designs can be implemented to a reasonable degree on an older process. It may take more area, more design time, and a lot of other tweaks, but it can be done.
The reason you don’t see this done very often is the economics of Moore’s Law made it cheaper to pay a bit more for the new node than to do it on the older node with 2.5x the area as a hypothetical example. There is a distinct crossover point where the new node becomes more expensive than doing it with a hypothetical 2.5x the area on an older process. That leaves the designs that can only work on the newer node as customers and that is a much smaller TAM. This is the long way of saying that the market isn’t as simple as raising the price to cover development costs, it is much more complex than that.
Then there are margins. Right now 14nm wafers are pretty cheap to make, relatively speaking. Those expensive tools have been paid for, additional costs are low, and margins go up. Go back to 28, 45, and even 65nm nodes and you have a proverbial money printing machine. It may not look sexy to technology enthusiasts but twitter and forum posts don’t bring in much revenue to a foundry.
More importantly is the statement that GF is going to be making differentiated variants of their processes to cater to specific processes and clients. This is a smart move which take a relatively low development budget to make a higher margin product. Since the tools are already paid for and the base process are in full production, it is a fairly low risk play too. If the idea is to make a profit, this is the right short to mid-term play.
That said it is sad and potentially the wrong long term play. As we said in the beginning, GF is unlikely to be able to restart development without a vast and likely untenable budget. So what is their future? It is either to keep churning out older nodes for the rest of time or to license another process from a competitor.
The option here, singular, is Samsung and the blueprint is 14nm. GF could wait until Samsung works the bugs out of their 7nm process, pay them truckloads of cash, and be up and running with a 7nm node. Time will tell if the numbers work out here, if they do it will likely be because of a large customer demanding a second source. Nothing like a multi-billion dollar deal to grease the wheels between two ostensible competitors.
Then there is the worst part, the lost jobs. The GF press release did not mention any layoffs but SemiAccurate is hearing that there will be a fair number of redundancies at GF. Since IBM sold their foundry business to GF there are effectively no options in the area for many of these workers, we hope things work out well for them.
So in the end it is about money. The new leadership at GF ran the numbers and came to the conclusion that the development cost wasn’t worth the eventual payouts and that money was better spent elsewhere. We understand the math for the short and mid-term but the long term outlook is not so bright. If GF goes the licensing route and keeps offering new nodes, the future could work out well. If they stay put and take their R&D budget as margin, things are going to peter out fairly rapidly. At the moment SemiAccurate has no clue which way things will go.S|A
Latest posts by Charlie Demerjian (see all)
- Another Epyc Rome TCO data point comes out - Nov 12, 2019
- What is the name of Intel’s Cascade Lake +5 server? - Nov 9, 2019
- AMD launches Threadripper 3, 3950X, and Athlon 3000G - Nov 7, 2019
- Intel messaging hits a new low - Nov 5, 2019
- Intel releases the world’s largest FPGA - Nov 5, 2019