Experts at the table, part 1: What’s missing from power standards and who will benefit once they’re created.
Semiconductor Engineering sat down to discuss power, standards and the IoT with Jerry Frenkil, director of open standards at Si2; Frank Schirrmeister, group director of product marketing of the System Development Suite at Cadence; Randy Smith, vice president of marketing at Sonics; and Vojin Zivojnovic, co-founder and CEO of Aggios. What follows are excerpts of that discussion, which was held in front of a live audience at the IEEE SA Symposium on EDA and IP Interoperability in San Jose.
SE: Do we have enough standards for power as the Internet of Things takes root?
Frenkil: We need more than we have, which is why there are a number of efforts underway right now. IEEE 1801 was conceived as a power-intent standard, not as a power modeling standard—and in this case I’m using modeling to refer to the estimation of power consumption. So 1801 has been extended to a certain extent because of that. IEEE P2415 and IEEE P2416 are working to fill the gaps. When we get those worked out and in production, we’ll be in good shape. But until then, we have work to do.
Smith: We need to go a lot further. Ten years ago some people had the right idea that power was going to become a first-class problem. That’s where we are now. Ten years ago we were focused on functionality and making sure everything is going to work—still trying to figure out how to deploy new verification methods. Power and security are becoming first-class problems now. You can’t get your product to market if you don’t solve these things. In terms of standards, we started with intent and moved to modeling and analysis. We still haven’t focused on design, automation and creation aspects. We’re not there yet. And part of the reason is that we’re still rooted to software-controlled power. When we’re using software, we’re burning tremendous amounts of energy to try to save energy. That’s ridiculous. We have standards for intent today. We need better analysis tools. We have to come up with more for top-level modeling to enable design.
Zivojnovic: IP is really about these small devices that make our lives easier. Imagine a small startup that has an idea to change the world by combining sensors, logics, Bluetooth or similar communication. Opening 1801 is not software-centric enough. It’s too detailed. It deals with time defaults, isolation and level shifters. It’s the wrong stuff to open the book. Of course, once someone starts putting transistors together and routing car supply lines and distribution, that’s a different game. On top of that, 1801 doesn’t have any clock information. It doesn’t tell you when clocks turn on and off. 1801 absolutely serves its purpose. But we shouldn’t overstretch it. What we should be thinking about is what these clever startups are looking for. They’re going to make these IoT devices and this is where the innovation is. It has to have a software face.
Schirrmeister: Standards are not yet in the way of getting things done for these smart startups. If you look at the system level, the new standards are great, but we have done these things by hand for the better part of a decade. We used to annotate data into virtual platforms. The question is whether standardization will accelerate all of this. When it comes to figuring out the different modes and how software switches through them and how you can present that in a better way than the device trees of the past, it takes the fear away that anything you do manually might not be useful next week. Companies are doing this by throwing manpower at it. But putting a positive spin on this, standardization could accelerate things.
Smith: We have to recognize that when it comes to power, we have two very diverse classes of customers. There are developers of apps processors, where big companies may put 80 or more engineers on the project and they can afford to do so based on the products they’re building. And then the vast majority of other designs have one, two or three people. They can’t do it by brute force. They don’t have the number of engineers available. We have to come up with standards that allow IP and EDA vendors to build the automation needed for the adoption of new methods.
Schirrmeister: Yes, standards can help accelerate things, especially for companies that don’t have the brute force. But then again, none of this has been in the way of managing power. It’s a matter of where power sits in the hierarchy of needs. Where does it rank? If it’s in the top list, it will be dealt with.
SE: With the IoT it’s not just what’s designed, it’s also how it’s used, which can vary greatly from one person to the next. How do we deal with these use cases more effectively?
Zivojnovic: When a standards working group comes together it makes a prediction about how things should be done in the future. We would not have such extraordinary brute force in software without languages, which have very clearly standardized how you write code. You can communicate and exchange and test in a better way. This is great functionality. But we don’t have that for power. We don’t know how good or bad we are until we start communicating. We make a product, we sell it, and we say it’s the best in the market. But if there’s a small company that has a great idea, they’re going to have trouble with that project. So while standards are not a roadblock, they should be part of our investment in the future. The problem is that when computer science was started, no one thought about power. They created languages, architectures, products, but power wasn’t part of that. Through standards, we’re now looking at whether we should have done that a little differently. Power and energy are having a revival now. We are revisiting what we did well and poorly in the beginning.
Schirrmeister: I’m in favor of power being a first-class citizen in the hierarchy of needs. I was at a conference where they were talking about power being one of two of the top needs for smart phones. The first was WiFi. But people are actually concerned about power. We have shifted from PPA (power/performance/area) to PPT (power/performance/thermal). That’s an exciting change.
Frenkil: Today, if you want to design some new IoT device and you’re concerned about power—and you should be—there are so many things you can do, so many methods you can deploy, to determine what’s the best architecture and the best protocol. That’s better than a few years ago. We’ve learned since then. To accelerate that move forward, we need these models. It’s difficult to compare this communications protocol with that communications protocol. To really optimize this, we need better data and we need better models. We’re not there yet. The modeling efforts we’re involved in will lead to more optimization opportunities and better-optimized designs. It’s not that people aren’t trying today, but with limitations on resources or data or time, they can’t optimize as much as they would like.
Schirrmeister: Models are important, but what are you actually modeling? You can model your device. There are articles that say the cell network is the data hog, not the device. Having the protocols optimized for all of this becomes very important. Modeling the scenario of what you’re trying to do becomes very important. There’s parallel optimization going on. There are models for software, and on the verification we’re doing portable stimulus where you model the scenario. That becomes a combination of standards. You have the modeling of the scenario defining what you’re doing and how that translates into how hardware and software will behave and contribute to power consumption. Then we can optimize across the whole spectrum.
Smith: The pain of power is everywhere in the IoT. If you’re at an edge node, every bit of power you can squeeze out of design means your battery will last that much longer. With a hub, whether it’s a phone or plug-in devices, a lot of these devices are facing Energy Star ratings. Companies have come to us for help because their modem ran at 5.5 watts and the limit was 5 watts. This happens. And if you look at data centers, you can’t put a data center in Manhattan right now for any amount of money because they can’t get you the power.