Chasing After Phantom Power

A lot of effort is being invested in power reduction techniques for mobile devices, where battery life is an important buying decision and power can translate into heat that can make a device uncomfortable to use. But are people willing to pay more for a device that consumes less power if it’s plugged into a wall?

And even if they are concerned about the power drawn during operation, what about when a device is in standby? This is the realm of phantom or vampire power—power that is consumed while a device performs no useful function except to wake up with a remote control or some other external stimulus—and it’s a growing concern.

“Around 2005, some countries decided that the problem was serious enough and found that the carbon dioxide (CO²) emissions are around 1% of the total,” says Abhishek Ranjan, senior engineering director for the Calypto system division of Mentor Graphics. “Compare that to the pollution created by the airline industry, which is 3% of total CO2 emissions. This is a significant number, and for households it amounts to 6% to 8% of total electricity consumption.”

This is not yet a global concern, but it is gaining attention. “Europe is way ahead in terms of thinking about the impacts of power and wastage in general,” says Krishna Balachandran, low power product management director at Cadence. “They have a different attitude to these things and I do not see any of that in this country, even today. We do not have this culture, and neither do I see it in Asia.”

While many have worked hard at reducing operating power, that can be a small piece of the total pie. “Idle power can account for as much as 20% to 40% of overall energy,” says , low power architect in the verification group of Synopsys. “For example, a 160W active power TV can consume 5W in idle power. In a 24-hour span, if there are 2 hours of active TV usage, 320Wh is consumed, whereas idle power consumes about 110Wh – almost a quarter of the total.”

As consumers, we are not yet aware of the total cost of ownership for many of the products that we buy and thus the economic incentives are not there, which in turn means they are not there for businesses to make improvements.

The is likely to change that because we are about to see another 10X growth in leaf node volume. “This is hugely worrisome for energy delivery,” points out Jadcherla. “Billions of connected devices draining the power grid. That’s why governments notice them sooner.”

Ranjan believes that there are two types of markets, and in one of those you compete on power. “For the Apple, Samsung and Microsoft’s of the world that are into consumer electronics, every milliwatt of power savings provides them with an advantage. They will do power optimization with multiple power islands, complex wake-up sequences, and dynamic voltage and frequency scaling. There are other types of customers that just need to be under the power budget that is either set by regulation or some market threshold. It is not significant enough that they will lose market share if they do not meet the power budget. This could be networking and routing companies that are plugged in and don’t care as much about power. They are not doing enough because it is not hurting them in a financial way.”

In addition, we are expecting more from our connected devices. “Think about some of the expected use models,” says Drew Wingard, chief technology officer at Sonics. “They will use the most power when they are doing their primary function and that will not change, but the amount that they perform in standby is increasing. Video game consoles often have a dedicated chip whose primary use case is in the off-state because the consoles want to be continually connected to the network. This could be for over-the-air updates. A Smart TV has a number of apps, and many of these involve IP-based communications. It is the equivalent of getting a Twitter feed where these devices want to gather information while the TV is off so that it can be instantly displayed when the user requires it.”

But it is not just entertainment devices facing this trend. White box items want to have an Internet connection as well. “You value the network connection in the off state,” adds Wingard. “I may want to tell the furnace to wake up because I am on my way home and there are things that I may want to have prepared in my home for my arrival.”

Phantom power in smart homes is particularly worrisome. “This problem is going to get much worse,” says Balachandran. “Consider a toaster that is talking to a refrigerator that is talking to something else. All of them are pinging each other, and that will take increasing amounts of power. They have to operate standalone where they consume almost nothing and wake up only on an event, when something changes and has to be communicated to something else. It is the same for the IoT or office automation, otherwise they will waste large amounts of power.”

Approaches for reduction
The most significant factor in phantom power is leakage. Balachandran says there are four buckets to consider for leakage reduction. First are design techniques. Second is process innovation. Third is IP that has been optimized for leakage. And fourth are EDA tools that minimize leakage. Another couple that can be added to that list, including efficient architecturesand preventing actions that cause no outcome.

For design techniques, “most companies are using the traditional techniques for reducing leakage, shutting off what you don’t need, low VDD standby, or using high Vt transistors in the areas that need to stay on but have low leakage,” explains Balachandran. High Vt is a tradeoff between performance and leakage in that the transistor performance degrades as the threshold voltage is increased, but leakage is reduced.

Jadcherla adds that “we are seeing a migration to an ‘off by default’ architecture, and a functional block of an IC should be at the lowest possible voltage at all times. Idle power reduction requires significant changes to the architecture, OS and the underlying hardware needs to support it.”

Another way to reduce standby power is by migrating to newer process nodes. “Devices that were not 1-watt compliant at 45nm or even 28nm may be compliant at 14nm,” says Ranjan. There is also interest in processes such as those using FD-SOI, which offers lower leakage, as well as low-power processes using bulk CMOS at older nodes.

There are higher-level issues as well. “Many designs are doing computations that don’t actually need to be done,” points out Ranjan. “These can be difficult to identify. You can see activity, but you may not be able to see if that activity leads to any meaningful outcome. A lot more thinking has to go on in this area.”

Balachandran adds some architectural considerations as well. “Consider the flow of information for the IoT from the edge nodes to the hub. There are no standard protocols that cover that last mile. There are all kinds of low-power protocols, but there is no standard for the IoT. This makes it hard for hub designs to be able to handle disparate data and protocols and be able to aggregate them and communicate that back and forth between these devices.”

But the lack of standards in the architecture is somewhat mitigated by standards in EDA. “We believe that the end of the power format wars and the adoption of IEEE 1801 and its variants was a huge step,” says Wingard. “This created broad support in various software platforms and helps people who are not used to doing aggressive power-managed designs do things such as power gating. Power reduction involves a lot of unfamiliar techniques, so is this something new that we are about to pile onto the designers? Will they have to learn about isolation cells and level shifters? What style of power gating architecture will they need to use to limit the inrush current? Is the market ripe for an IP-based subsystem solution?”

Education is an important piece of this. “Many customers are now beginning to understand the need for power optimization,” says Ranjan. “It was not the same five years ago when they saw no economic advantage. They could not see why they should put in any amount of investment in reducing power, but today, with the markets becoming more aware, they now know that if they do not take power into account, they could have problems.”

And it is important that they do understand the costs as well. Jadcherla says that “at the semiconductor level, there is a 15% to 20% overhead for supporting power management features such as power gating, standby voltages and retention.”

There are hidden issues, as well. “If you have never thought of the supply terminal of your logic as being anything other than on, you have never thought about power gating or the things that you need to do to make it safe,” adds Wingard. “How do you make sure that when you are turning things on, you don’t accidentally ask for so much current that you affect the circuitry that was left on in the standby state because of an IR drop problem? How do you avoid power latch up?”

Each of the issues on its own may be simple but “you are managing a mind numbingly large number of simple state machines,” explains Wingard. “The individual steps of sequencing power gates or triggering the isolation cells or stopping or restarting clocks are all very simple state machines, but there are a lot of steps and they have to be done in a specific order—and there are dependencies between them. The complexity comes from the interaction.”

And other things that may have been fairly simple in the past can take on extra complexity. “Recently, Cadence hosted a front-end design summit,” says Balachandran. “Within that we had a panel session and one panelist was Leah Clark of Broadcom. She said that the more you try and control power in a design in terms of fine-grained power control, the more you have to spend time designing the power regulators. These things don’t come for free. Can they be placed on-chip or do they have to remain off-chip, how much will they consume and leak. How much you end up saving is questionable.”

Wingard agrees. “The power regulator can create a large waste. What many people do is to create an on-chip regulator for the always-on state and it may not have to be the same level of quality or provide as much current. You may be able to run at a much lower frequency, such that you can afford more margins. Instead of +/- 5% you may be able to get away with +/- 15%. This is much less expensive in terms of power and area and will be much easier to integrate. There could still be an external power management IC that is providing the real business voltage and shut this down in the standby state.”

As more designers face the need to minimize phantom power, they will be looking at the EDA, IP and semiconductor industries for help. Luckily, they now have quite a bit of experience thanks to work in the mobile space.