How Long Will Your Battery Last?

The current ways of measuring energy efficiency are useless. It’s time to change them.

popularity

Designing an IP block or memory subsystem to fit within a power budget is essential for building energy efficiency into hardware and software. There’s only one catch—it’s meaningless to the end customer.

Twenty years ago this was a pretty straightforward formula. If you used a device consistently, whether it was a computer or a calculator or an electric motor, then you would burn up X number of watts during that time. If you ran a car at 55MPH you would see an efficiency gain that you didn’t get at 65MPH.

These formulas don’t work anymore. Two people can use the same device in the same place with the same connectivity and they’ll get wildly different results—both in energy usage and in performance.

What’s changed is that the number of functions per device has gone up significantly. A phone is no longer just a phone, and a computer is no longer just used for word processing or spreadsheets. That means the patterns of measuring energy consumption have changed dramatically for most people, and they frequently change from one task to the next. They may focus on search one day using Wi-Fi or while connected to an Ethernet cable, and the next day they might be using a combination of 3G or 4G.

Apple, for one, has received a fair amount of negative press for the projected battery life of its forthcoming watch. While most of the comments focus on the battery life with intense usage—an estimated 2 to 4 hours—it can run for up to 96 hours for some users. Whether those numbers are acceptable to consumers is beside the point. What’s worth noting is just how far apart these numbers can be.

The reason is that some of the processing is being done on the edge device—in this case a wearable electronic device—and some of it is being done in the cloud. And depending on the individual application, the amount done locally and remotely will vary greatly, and so will the battery life. Leave one application on in the background and it will likely continue drawing some power. Leave on an application that requires an antenna to communicate with a server somewhere, and it will draw even more.

That leads to the next part of the metrics. Multi-core and multi-function devices are complex electrical engineering feats. These devices have built-in power management features that are astoundingly complex. They can buffer components from surges, wake up certain regions of an SoC or PCB gently or with full-on energy. They also can detect the proper thermal environment, ambient plus internal heat, and ratchet down the processors or MCUs so they don’t burn up the device. So while they may run at top speed for a few seconds to a few minutes, they certainly won’t be running at top speed after 10 minutes unless they have an internal fan or some liquid cooling mechanism.

There are no published metrics anywhere for this kind of device behavior, and there may never be. As devices become more complex, with more complex interactions and possible user scenarios, the power and performance numbers behind those devices become equally complex to the point of being meaningless for most people. But it is possible to alert users while using those devices that their battery will drain more quickly in one use mode versus another, and that one application is more power-hungry than another and not turning off an application will result in battery loss, a higher energy bill, or loss of performance.

These are system-level measurements, but the information has to be gathered at the chip and software level, which is where the processing is done and the measurements can be made. So far, there has been zero progress in this area other than a meter that levels battery life. That’s a long way from helping end customers understand how their actions will impact energy consumption, and no matter how good the underlying engineering or how tight the power budgets, unless this information reaches the user its value will be limited.