SemiAccurate talked about the process side of Global Foundries a few days ago, but being a foundry is much more than printing designs. Another major portion of their talk was about six new IP related items customers can now choose from.
One of the more important services a foundry offers is helping customers make a product. This involves both the physical fabbing and providing a menu of compatible IP usually in the form of battle-tested blocks of logic. These pre-tested chunks can range from a small circuit or I/O design to a full ARM core as a hard macro. The customer just needs to pick what they want, pay the license fee, and hopefully ends up with a ready to roll bit of functionality they no longer have to make on their own. On this front, GloFo had six announcements at Common Platform 2013.
The new IP being brought to bear is from Cyclos, Adapteva, Synopsis, Rambus, and ARM, plus a new process from ST-Micro. They range from I/O blocks to full RISC cores, and can be picked from a larger menu of offerings by foundry customers. Lets take a look at each of them.
The new items on the menu at Global Foundries
First up is Cyclos, a startup you might have heard about around the time of AMD’s Piledriver launch. Cyclos has a technology called resonant clock mesh which is said to radically drop power consumed by a device’s clock tree. Since clocks are usually the number two power consumer in modern CPUs, this could be a big deal. Piledriver was the first production part to use Cyclos’ resonant clock mesh, and that 4.4GHz part is said to save about 10% of the total chip power as a result. As of now, anyone using GloFo’s 28SLP process can add resonant clock mesh IP to their design.
Adapteva is very different from Cyclos, they offer a complete core called Epiphany IV to 28SLP customers. Adapteva’s cores are small, very small, and have a very wide mesh between them that is reminiscent of Tilera’s designs.
The company claims that a 64-core mesh of CPUs takes less than 10mm of die area, pulls sub-2W, and delivers 100GF of performance. If it seems really GPU-like to you, the fact that there is an OpenCL API will make a lot of sense. The oddest part about Adapteva is not the architecture, the offering, or the API, it is the fact that it is a crowd-source funded design, Kickstarter projects go 28nm.
Skipping one bullet brings us to Synopsis, maker of semiconductor tools and designs for just about everyone. The big news at Common Platform is that nearly the entire set of their tools and libraries will be available to customers making pretty scribblings on the 14XM process. That should save a lot of time and headaches. Those same customers also have access to Rambus interfaces on the same node, so high speed I/O should be quite doable with FinFETs.
On the more complex end of the scale, that same 14XM process now has licensable ARM A9 cores available. It is not just one A9 core though, there are three different A9s optimized for power use, performance, or area used. On a 14/20nm hybrid process, you have to wonder how much area really matters on such a small core. In any case, if you want the latest and greatest cell phone CPU from 2011 on 14XM, you can pick from three variants. The unsaid part about A15 and ARM64 CPUs is probably more interesting, but until it is officially announced, we will feign ignorance.
Stepping back to the previously skipped point, we have ST-Micro and their 28nm FD-SOI process. We covered it at length in our writeup of the resultant NovaThor SoC at 2.5GHz, a device that is quite an impressive advertisement. If you are building something on 28nm at Global Foundries, this process is now an option for your project. If you get anything close to the gains ST-Ericksson got for roughly the same low effort, there doesn’t seem to be any reason to avoid FD-SOI on your 28nm device.
With these offerings, Global Foundries takes another large step toward being a complete one stop shop for foundry customers. Considering they were an in-house fab with only one customer a few short years ago, this is quite an impressive set of choices. That said, there is a long way to go before they can please all comers, but the steps taken so far are solid enough.S|A
Latest posts by Charlie Demerjian (see all)
- Thing go bump(gate) in the night for Nvidia’s GP100 Pascal GPU - May 3, 2016
- Cavium’s Octeon TX blends compute and packet moving - May 2, 2016
- Qualcomm releases Zeroth API to developers - May 2, 2016
- Another detail about Qualcomm server SoCs revealed - Apr 27, 2016
- AMD finally really honestly launches the dual Fury - Apr 26, 2016