Experts At The Table: Which Comes First?
By Ed Sperling
System-Level Design sat down to discuss hardware and software priorities with Neil Hand, group director for marketing for Cadence’s SoC realization; Johannes Stahl, director of product marketing for system-level solutions at Synopsys; Prasad Subramaniam, vice president of design technology at eSilicon; and Bernard Murphy, CTO at Atrenta. What follows are excerpts of that conversation.
SLD: Is it critical to move to the next process node with designs?
Hand: The ability to analyze is critical. Can you meet performance at a different node with a mix of IP. Is it available? Is the software available?
Stahl: If the market leader goes to 20nm for a tablet, then the competition has to follow. There is no choice.
SLD: But isn’t the iPhone back a node?
Hand: When the iPhone 3 won out, it was a couple of technology nodes behind Android. If you look at every generation of that phone, when everyone else had 3G they didn’t have 3G. When everyone else had GPS they didn’t. They focused on what was their application and their differentiation.
Stahl: They were focused on how to get the maximum out of their market.
Hand: They invested in the user experience. Some of that was software, some of it was other areas.
Murphy: And the other smart phone makers have to figure out how much more functionality they can stuff in there to differentiate themselves. That means they have to go to 28nm.
Hand: And if you have more real estate, you can have more cores and distributed functionality rather than complex interactive software that’s constantly switching context. You can shut it down and use your area in different ways. But at the same time you need be sure you will have the market volume to pay for it.
SLD: What does the Apple cloud do for the design?
Murphy: One of the competitive solutions is very focused on 3D stacking, particularly to put a lot of DRAM on a processor so you can have a lot of processors in an array with DRAM on top of them. That gives you a lot of compute bandwidth, so you can have tons of applications running extremely fast.
Hand: If you look at the iCloud, it’s about moving storage off the device. The next step, if there is one, may be a micro-console for gaming, which is basically a smart decoder with all the processing is done in the cloud. It’s all virtualized. Longer term, if you start seeing that then it will affect architectures. It’s more about how much memory you need on a device.
Subramaniam: It’s moving the storage off the device, but it’s also enabling multiple devices to synchronize on the same storage. So if you have a smart phone, a tablet and a compute, all of them can share the same photos or data. Anytime you open up one of these devices it has current information. You don’t have to worry about data being out of date. But I’m still skeptical about processing moving away from the local device. Until the network is reliable enough and you can have instant bandwidth, then I’m not sure having processing done away from your device is acceptable to anyone.
Stahl: There is a lot of activity going on with the FCC and companies like AT&T. They are really trying to gobble up every spectrum they can find to deliver the bandwidth. We will run out of bandwidth in three or four years if they don’t do something. That will enable more stuff to sit in the cloud. Otherwise you can’t get that information to your device.
SLD: This all comes back to what will future architectures look like. Will it be on the device, on the cloud server or a mix?
Stahl: Connectivity will be a huge part of this, too.
Subramaniam: But the end device we know today isn’t going away.
Hand: You’re right. What do you do if it’s all in the cloud?
Murphy: You have to consider why Apple came up with the iCloud. On the iPhone you have Mobile Me. You don’t really need much on the handset. But one thing you can’t do in Mobile Me is iTunes because of digital rights. iCloud has solved that problem. You can have the iTunes stuff remotely.
Hand: But all the iCloud stuff is DRM-traced. It comes down to how you can maximize that cost. People pay several hundred dollars for a device, and if you don’t have to put all of that memory in the device and you can get more capabilities and more features with less memory, their margins get higher. You don’t have to pay for the chips in the device.
Murphy: What applications are going to drive that?
Hand: Entertainment.
Stahl: Video. If you look at the average user, with 16GB of memory you go over capacity very quickly with 20 videos. If you have infinite space that’s okay.
Subramaniam: Having infinite space is no good if you do not have infinite connectivity. You will still need a cache on the device.
Hand: But that’s how the iPhone is doing it. You do have a certain amount on the device. That local buffer may be 32GB, but you create a much large space to play with.
SLD: What drives the designs in the future and how does it change?
Hand: It still goes back to the class of applications.
SLD: But that’s applications in terms of the device, not just the software applications, right?
Hand: Correct. It’s what is the usage of the technology. In some cases, there will be a specific application like a GPS that requires real hardware to drive it. If it’s a 3D game it requires very specific hardware. But generally it will be the cost of applications.
Murphy: There are a lot of new applications emerging that are going to use a lot of the same technology, though.
SLD: To differentiate yourself, though, the market may be outside of the device itself and into more vertical solutions.
Murphy: I’m hearing a lot about security with remote cameras and sensing. All these things have been going into smarter and smarter and less and less reliable cars, too.
Hand: That verticalization is being driven by opposite forces. While verticalization makes it easier to have a well-defined set of requirements and differentiation, it also narrows the scope of the SoC. That goes against the move toward 28nm and 20nm. It’s attractive from a differentiation standpoint, but what does it mean for node migration? You go vertical, but you go with an older generation because you don’t have the volume to support it.
