News and New Products

Lower power becomes bigger issue

By Ed Sperling, Editor in Chief -- Electronic News, 4/6/2007

Electronic News/Electronic Business sat down to discuss low power issues with with Vic Kulkarni, president and CEO of Sequence Design; Eric Filseth, VP of marketing at Cadence Design Systems, and Steve Schulz, president and CEO of Si2. What follows are excerpts of that conversation.

Q: How pervasive is the low-power issue?
Filseth: Power has become an issue for lots and lots of people, whereas it used to be an issue for only a few. If you go back about 20 years, the world was bipolar logic. Power is what killed off bipolar logic. As chips got bigger and faster, the power dissipation made it impractical to scale to a very large density. Fortunately there was CMOS, which had a variety of problems. It had latch-up effects and no gain and it was slow, at least in those geometries. But it did have one really good feature, which is that it consumed orders of magnitude less power per gate than bipolar did. It was a material solution. CMOS enabled the whole electronics revolution for the past 20 years. There’s a fantastic chart from IBM that looks at the speed and power of bipolar. It looks identical to CMOS now. But unlike 20 years ago, there’s not an obvious silicon technology. That’s why so many people are looking at a design-based solution.
Schulz: Before we had the level of integration that’s now required, you could make some simple assumptions that bounded it. You could put it between the goalposts, and if it was in that range, you didn’t worry about it. As you get to the kind of complexity of integration now you cannot afford to leave that on the table. You don’t have an economically viable or competitive part. What used to be able to be handled from a process standpoint now requires more sophisticated methodology in design, more sophisticated support in your tool flows and the way data flows through in a consistent manner.
Kulkarni: We have about 150 customers worldwide. Out of those, about 70 to 75 are power-aware and power sensitive. There are four numbers driving this power demand. One is a 3-watt number. If you hold an appliance higher than 3 watts, it’s uncomfortable after about 7 to 10 minutes. SoCs should run at 2.5 to 2.6 watts for applications like cell phones and PDAs. As you go from 3 watts to 6 watts, there is an increase from $1 to $1.10 as you change the package from a simple plastic flat-pack to a plastic pin grid array. The 10 watt number comes in because then you need heat sinks. That’s another $1.50. Then there are the big processors that are 100 watts, but they are going to co-processors on a chip. In the past year, power has become the number one criteria instead of timing and area.

Q: The second factor is cost, beyond just what you’re mentioning. It’s the cost of developing a complex chip, right?
Filseth: There is kind of a perfect storm going on here. There are three factors. One is the growth in complexity of devices. The second is that the move to 90, 65 and 45 nanometers introduces much more concern about parasitic leakage. At 32 nanometers, it’s going to be more of the same. The third is the macro trend of what’s driving technology innovation out of the information technology world and into the consumer and wireless world. Power and cost are inherently much more important. All three things are happening at once.

Q: So what’s the solution?
Filseth: The answer is much more of a platform-based approach. If you’re going to do a 40-million instance chip at 65 nanometers, this is a really expensive design. There aren’t that many markets that support the kind of investment it takes to do that. This may be $100 million of investment cost. Those markets that can support it are trying to re-use as much as possible, spin off derivative designs and variations so they can leverage the core platform in a number of products.
Schulz: If you have to have one platform serve multiple applications, increasingly that’s being done with software. The actual switching of the transistors is tightly coupled with what instructions are being run and how you’re using those instructions. There are things that have a super-linear effect, things that have a direct linear effect on the amount of switching, and so you really need to deal with this as a system problem. It’s not just about one slot, because you’re probably going to lose that benefit unless you treat it in context.
Kulkarni: At Samsung, the cell phone guys have been given a top-down mandate to reduce the power by 100 times on static and 10 times on dynamic. The most important problem they’re trying to solve is how to bring the whole ecosystem together. It’s not one foundry or system. One problem is all teenagers are using the new DMB (digital multimedia broadcasting) standard. It’s continually on, they’re watching movies and chatting with their friends. It’s no longer about leaking power. It’s about talk time and running of displays and the dynamic power.

Q: This is an interesting twist, because what drove the cost up was the price of the fabs and the complex design. This is pressure from the consumers driving it the other way.
Filseth: That’s already happening, especially in some of these extraordinarily complex designs. What you’re starting to see is consolidation in the end-user applications because the development costs—and that includes the software—are so high. The world can’t support 20 different cell phone architectures. The concentration that’s been there on the back end is starting to show up on the front end.
Kulkarni: In the U.K., there’s a new law. When truck drivers back up or change lanes, they’re causing many accidents—especially at night. Now they need 32 mobile cameras around the truck at night. That is creating huge demand for mobile cameras. That’s a huge new application for mobile. There also will be more and more low-power, mobile technology for security issues.

Q: Let’s go back to ecosystems for a second. Who’s going to drive the front-end design? Is it going to be Cadence, the OEM or Si2, or someone else?
Filseth: If someone wants to cut power by 90 percent, you don’t do that in layout. It’s all system architecture stuff.
Schulz: About 80 percent of the opportunity is in the first 20 percent of the flow.
Filseth: You need to do design exploration, but in order to do that you need to really automate the back end. What if I did this, what if I partition my system differently? Is it going to meet the power budget? In order to know that, you need a very automated implementation path behind it so you can try it out. If it takes an army of engineers doing manual work to figure out whether an architectural partitioning will work, you’re not going to be able to do the kind of exploration you need.
Schulz: There’s certainly an overall increase in NRE (non-recurring engineering costs). But even with that, there’s limited resources. What we want to be doing is moving downstream. The only way you can do that is have the automation upstream.

Q: You’ve taken the first step, but now you need other companies coming in?
Filseth: What we announced in January was an automated spec-to-GDS low-power design flow. We took a massive step forward, which is the ability to integrate verification, implementation and design off a single database. It’s not everything.

Q: And IBM has driven the back end with its Common Platform across many other pieces. But what you’re talking about is based on what Cadence does. How do you get to the next step, which is including other parts of the value chain?
Schulz: Si2 has a mission and a charter and a role. We don’t play in the product side. We don’t define the design methodologies. What we do is listen. Because we have 112 member companies now, we have a pretty good idea about the trends and what’s sensitive and urgent and try to help connect dots. Many things that come out of Si2 were things that have been talked about for some time by the user companies, because ultimately they’re the ones buying the tools and who are adding their own internal stuff. These companies are the ones we look to guide us. We help them in their own councils to talk about significant issues, what they have in common, and if they have alignment we test the market a little further. If we see major disconnects, we need to work that. We can’t necessarily solve it, but we do enable the connection of dots.
Kulkarni: Si2 has done their homework in terms of getting the silicon into proof points. It’s not just talk. They have AMD, ARM, NEC now, and they have the same voice to create one voice and one standard. The supplier community has to give the solution to the end customers—to Nokia and Motorola and Samsung and Toshiba—so they can move on to meet the demands of high-end consumer markets with 3D graphics, new displays and cell phones. Cell phones are not nice to have anymore. They are part of survival in many developing countries.

Q: To solve this problem, doesn’t the industry have to start broadening its approach? This can no longer be just the electronics industry. It’s now including consumer devices and software. Who drives this?
Kulkarni: Yes. We just got a booking from XM Radio. Why? Because this is a value-chain equation. XM Radio has set up a group to design an RTL core. They want to create the lowest-power core and give it to the manufacturers of satellite radio so they can compete against MP3 and iPods. It’s the first time ever that a customer’s customer has purchased a product from us.
Filseth: And who would have thought Microsoft would be making video game platforms?
Schulz: We don’t have hardware engineers prescribing how the software engineers should do their job. That hasn’t worked before and it won’t work again. We need to engage those who are involved in the software design—those who are doing the day-to-day job. A system engineer is not a generalist. He’s a specialist in the interfaces. This industry has to do a better job of reaching out to the people who are layering their software on top of our hardware designs and engage them in active and open discussions. That will lead to some of the desired methodology goals for how we work together, and from that will flow some of the decisions about tools and enabling standards. There are standards that have been in use for years, and there are standards that enable the market to advance. Open Access is beginning to be that.