Productivity And The IoT

Experts at the table, part two: What needs to be improved in the design process to make it better suited to address IoT apps? Will IoT drive a new way to design chips?

popularity

The market for devices that connect almost everything to the Internet of Things is projected to explode, creating opportunities for companies that haven’t been traditional chip developers to decide to start developing devices. Semiconductor Engineering sat down to discuss this topic with Jack Guedj, corporate VP of Tensilica products at Cadence; Bill Neifert, CTO at Carbon; Marc Evans, director of customer solutions at CEVA; Drew Wingard, CTO at Sonics; and John Koeter, VP of marketing and AEs for IP and Systems at Synopsys. What follows are excerpts from that discussion. For part one, click here.

SE: What are some more of the interesting IoT apps that you are hearing about?

Neifert: We see the starts of the attempts of smart watches and the fit bands and these things happening — I wear a watch all the time — but there would definitely be something for me at least if I had to take that watch off and power it like I do my cell phone. If I could put that on, and leave it for a week, then I’d probably be pretty happy especially if it brought me some added functionality whether it be displaying my messages or whatever it is. I think that’s an attractive consumer case but we’re a little far from the power profile in the silicon that’s available today to get there given the standard APEs and just running an app on it.

Koeter: Another interesting example is connected garbage cans. If you think about that for a second, it actually makes a tremendous amount of sense. These are actually be deployed in places like Paris, Nice and other places. The reason is if you have a sensor that’s connected to the Internet on your garbage can and it’s empty, then the garbage truck doesn’t have to stop at your location. Garbage trucks are big and heavy and slow — starting and stopping them takes both a lot of time and a lot of energy.

SE: What do we need to do to improve this productivity?

Wingard: Maybe just taking the application processor and bringing it down may not be the right model, so what do we need to do to help people design these things quickly?

Koeter: First of all it’s got to be about cost and it’s got to be about energy harvesting and efficiency. These stripped down apps processors, who wants to charge their watch every day?

Guedj: Unless you’ve got microwave power as you walk in.

Neifert: Power harvesting is an interesting one. I’m walking around with my cell phone all day long, and as my Fit Bit tells me, I’m generating a lot of energy and it would be nice if I could somehow get my phone to harvest some of that.

Wingard: Why isn’t your Fit Bit harvesting energy? It’s measuring it.

Neifert: And I still have to charge this thing once a week. It’s one thing to translate it to mechanical, it’s another thing to actually store it in electrical, and how much can that be miniaturized? When I look at IoT and the things that need to be done to enable all of this, personally, I don’t see limitations in what’s available from today because in some cases, you don’t want to take the application processor itself and scale it down — the methodologies that are making the application processor successful — because you have to worry about power there and all of these various things — you should be able to take those methodologies and apply them for these more specialized applications. Finding a market for them is part of it. Definitely, you need to be more power efficient because no one wants to plug it in all the time, so there will be additional applications here but I think it’s a refinement of what’s being done today. I don’t see any revolution that needs to be done in the processes in order to be successful in the IoT.

Wingard: I kind of expected my processor brethren to be commenting on — I think, to get to the power levels required, we’re going to see more decentralized computing as a general theme. There’s been a lot of conversation about the always-on aspects of some of the IoT things, which tend to need to get to much lower power dissipation levels while still doing a modest amount of sensing and computing which is something an application processor hasn’t had to do. Application processors still today get away with principally one processor complex–I think all these IoT devices won’t get away with that; I think they’ll have distributed processing complexes for different things. I know Synopsys has described some sensor hub kind of technologies, as a sensor subsystem. We’ve talked about how the connectivity is an important part, so there may well be a connectivity subsystem with it’s own processing, and then eventually the thing turns on for real and there’s some kind of compute subsystem as well. I think across all of the IoT things that we’ve looked at, there’s always a sensor part, there’s always a wireless communications part, there’s always a compute part, sometimes there’s a display part, sometimes there’s not but there’s a set of these subsystem and I think there’s a lot of architectural innovation available at how do you do that at the right power levels. Then, how do you put those things together at the next level up because this whole concept of getting the productivity and agility involves some fundamental mix and match, and again we’re going after such aggressive form factor requirements, it means the sensor subsystem can’t be fixed. It has to be a pallet of building sensor subsystems so you have to have the exact combination of sensors that you need on this.

Koeter: In our particular case, it’s more than that because the way our sensor subsystem is built is you can implement a box in hardware or software — it’s that kind of flexibility you need. You’ll have your different tradeoffs between if you’re focused on cheap in which case that’s smaller area, in which case you’ll want to do the functions in software or do you want absolute lowest power, in which case you’ll want to do a lot of those math processing functions in hardware.

Evans: I think one thing that’s interesting and opens up quite a bit of opportunities is that unlike apps processors, where there is Qualcomm, and then a couple of others, nobody knows what this IoT is and there’s a lot of companies trying to figure out where their niche is in it, and how they’re going to assemble their secret sauce for differentiation. From the perspective of an IP provider, we see a lot of traction in the always-on that’s everywhere right now; there’s a lot of traction in some amount of face detection to see if somebody is in the room or recognize people; obviously the connectivity and sensor fusion — and those seem to be coming together in this one, always-on functionality where proven IP in that domain becomes one baseline ingredient. It will also be interesting to see if there becomes some kind of application style market on IoT a la iPhone apps, Android apps and what that really means for the rest of the system

Guedj: There are some tradeoffs here — apps processors are decentralized and decentralizing. You’ve got millions of gates so you can afford to do that. If you take the statement that you don’t really know what’s going to work, if you don’t know what’s going to work, you take a general purpose device. You take a processor, a standard CPU like an ARM or any of the processors we have, and you can run anything, but it’s not efficient so the tradeoff is do you know exactly what you want to build and how much of the specialization you want to build in, or do you want to start general purpose and then migrate to something that’s function specific?

Koeter: That’s a good point. We haven’t seen it yet but we have a line of business we call ASIPs (application specific integrated processor) — it’s going to be interesting whether maybe that might be a play here because it has 10x the power/area benefit compared to a general purpose processor but you have to develop yourself. The IoT market is so new right now we’re seeing nothing but standard DSPs. Maybe it might go that way over time.

Neifert: But I would imagine the need …. because power seems to be the biggest driving force in all of this. And as you go to that, you need to specialize more — is that going to raise a whole new level of the processors? Is ASIPs the way to go? It was interesting: Intel got all of its money based on PCs, ARM rose on mobile, is IoT going to bring rise to a new way to do the processors for this in order to meet the power constraints or are they going to adapt the other stuff. I certainly don’t think there is enough of a trend out there yet to say what’s going to win.

SE: Architecturally then, what are some guesses on how that looks for IoT?

Koeter: For example, a medium edge/mainstream kind of wearables — typically the sweet spot process node is 55nm — typically an embedded Flash process. The basics of a mainstream Fit Bit-like device is pretty straightforward. It has a very low power processor, a bunch of standard peripherals, typically on-chip memory, Flash a couple megabytes maybe, and typically a USB port for charging it, and some sort of sensor subsystem on it to process the senses. That’s the basic architecture on one of those things.