Redefining Systems Around Power

By Ed Sperling
Engineers have been talking about system-level power budgets since Moore’s Law reached 65nm, but as power becomes a critical element of any design with or without a plug the definition of what constitutes a system is changing.

While most SoC engineers think of the system as an IC, power increasingly is playing a significant role in the subsystem, and even in the larger systems that contain thousands of subsystems. That includes everything from implantable medical devices to airplanes and automobiles, in which case the full device is now considered the system.

This has led to some fundamental changes in who calls the shots in a design, what parts get included and what makes one part more competitive than another. It also puts a different emphasis on the design flow, making power and the ability to test for power critical elements of the flow, while opening new opportunities for EDA and ESL tools that can take physical effects into account.

Consider, for example, what’s happened in pacemaker engineering. One of the great advances in the 1960s was a version of a lithium battery to power these devices, but the early versions were relatively crude. They did little more than provide a consistent pulse. Much has changed since then. New designs call for pacemakers that can run 10 to 12 years without a battery change. Most times the devices are in deep sleep mode, but they can quickly wake up to respond to any irregularities, and some new models have built-in defibrillators that charge quickly using a capacitor. Moreover, they include an analog version similar to the original models that kicks in if there is a malfunction in the digital portion while beeping an alert.

“We’re seeing some of these devices where they have an ARM processor that they slow to 5MHz,” said Serge Leef, general manager of the System-Level Engineering Division at Mentor Graphics. “The leads are connected into tissues and they can listen to the electrical activity of the heart at that speed. The device can respond with a pace as needed, whether it’s a short burst or a longer pace.”

These devices, which were once considered simple technology, are now as high-tech as smart phones. And like smart phones, they have an overall power budget that must be met.

Redefining the system
While most engineers can grasp the design implications of a pacemaker, which is basically another smart mobile device, try applying the same concept to a jet, a missile or a car. Drive by wire has given way to fly by wire as aircraft builders attempt to reduce the overall weight of the vessel and improve fuel economy.

The twist is that all of these electronics—including more electronics in each seat, more sophisticated controls in the cockpit and throughout the plane—are much more densely packed together than in the past and they produce heat. That used to be less of a problem when jets were constructed of metal, but the new composite materials in the new Boeing 787 and the future Airbus engines that were added to reduce weight are far less conductive. So rather than just figuring out ways to remove the heat caused by current leakage, airplane manufacturers are keeping a close eye on overall power budgets.

Heat, power and airflow in aerospace design. Source: Ansys

“The challenge here is that aerospace is so highly regulated that change is difficult,” said Robert Harwood, aerospace and defense industry director at Ansys. “A big part of the 787 delay was caused by materials. The same is true of the components.”

That means heavy re-use of existing components, when possible, and improvements in power management.

“Managing heat is a huge problem,” Harwood said. “One airplane company executive said that in commercial aircraft, thermal management has become more important than aerodynamics. More sophisticated electronics generate more power and heat. So to decrease the weight and improve the passenger experience you need more powerful electronics. That means these systems get smaller and hotter.”

Models and standards needed
It also means that systems have to be modeled on a much higher level to be able to make the kinds of power tradeoffs that SoC architects now make.

“We’re going to have to go through some effort to build higher-level models,” said Mike Meyer, a system-level design fellow at Cadence. “We’re going to have to take them up a level of abstraction to the larger system level, not just the electronic system.”

That’s easier said than done inside of these kinds of systems companies, however. Processes have developed in some of these companies over decades. Changing the flow means changing the way they do business.

“This is very much something that is their internal process,” Meyer said. “We’re seeing this in the verification space. At the system level, their verification is tailored to their needs. As time goes on there’s going to need to be some standardization. And then when you extend it up to the software and the physical device, that’s not just going to include EDA. It’s going to have to include software and packaging. We don’t know how that’s going to look, but there’s going to have to be a broader support structure. One of the problems is that this is very high value, but there aren’t that many people that are doing it. Right now they’re getting by with Excel.”

One of the keys to making inroads in this process may be from the software and IP side. Software now needs to be developed for all systems and subsystems, and being able to improve that process would be invaluable to systems developers in multiple markets. IP re-use, meanwhile, is viewed as a way of reducing risk in complex systems—particularly if the IP can be reused for multiple generations of systems.

“People are composing systems by putting those subsystems together,” said Grant Martin, chief scientist at Tensilica. “In some cases they’re evolving those bigger systems by taking RTL subsystems that deal with past standards and linking them up with new subsystems to deal with new or current standards such as in the baseband area. We need to understand beyond the individual processor core what the right kind of partitioning is onto heterogeneous multiprocessors and then how those subsystems work together into the overall system being developed. So we’re having to develop our own system knowledge. People are willing to look to IP for individual functions and subsystem functions because they’re learning how to compose them together and to reuse IP to reduce risk in complex systems.”

Military cooldown
That’s as true in a smart phone as in an unmanned vehicle designed by the military. Ansys’ Harwood said that the U.S. Department of Defense is moving toward consolidated platforms for the three branches of the armed forces to improve efficiency because each was developing its own. One of the consistent themes, he noted, is managing thermal issues and power, which are critical in weaponry.

It’s particularly critical in missiles, where batteries are known to detonate a device before it hits its target. Instead, these weapons use turbines to charge capacitors when they are plummeting vertically.

Mentor’s Leef said many defense contractors are concerned about how much ground these missiles can cover because when they are moving horizontally they consume energy, while they actually generate energy as they drop. This requires much more accurate battery readings than the kind that appear on a cell phone saying the battery is about 20% charged. In missiles, an extra percentage point or even a fraction of a point can be the difference between hitting a target and falling short—and understanding the power implications that can affect that number are equally as critical.