The battle for the bottom -

The battle for the bottom

For some time, system power consumption has been one of those “oh, yeah” characteristics (as in “oh, yeah, about your servers, you're going to need a bigger cooling system”). Server farms aside, concerns about system power consumption have certainly been on the mind of anyone building a coin-battery-operated system. Now, with the industry's laser focus on the IoT, wearables, or anything expected to run on rechargeable batteries, we're going to hear a lot more about MCU power consumption as a differentiator as vendors search for the bottom in MCU power consumption.

Texas Instruments' recent announcement of its 32-bit upgrade to its MSP430 line is a case in point. The new MSP432 line packs a generous complement of memory and peripheral blocks around an ARM Cortex-M4F, but for those on a tight power budget, one of the more interesting aspects of this new series its its power characteristics. Based on ARM's low-power Cortex-M4F core (M4 with floating point), the new series features 95 µA/MHz power consumption in active mode.

Power optimization in MCU on-chip peripherals is nothing new, but TI appears to have taken special care to pare down power requirements even more with a low power 14-bit ADC (375 µA at 1 Msamples/s); an integrated LDO and DC/DC converter (TI cites a 40% power improvement); on-chip ROM containing its DriverLib (35% less power than a Flash-based library according to TI); and selectable RAM retention (30nA savings per disabled RAM bank, says TI).

We're likely to see more from MCU makers as they find ways to achieve lower power ratings with a variety of techniques including power-optimizing technologies such as sub-threshold voltage in Ambiq Micro's Apollo MCUs, which are also based on the ARM Cortex-M4F core. Ambiq quotes active-mode power at 30µA/MHz executing from Flash in Apollo MCUs, which integrate RAM, Flash and a 10-bit 1 Ms/s ADC.

It's been difficult to gauge claims about power consumption given its complex dependencies not only on system hardware configuration but also software mix and runtime operating parameters. In an attempt to offset this confusion, EEMBC, which is behind the well-known CoreMark benchmark suite, added ULPBench (as in ultra-low-power benchmark) for comparing power consumption in MCUs.

As Jack Ganssle described recently, EEMBC offers a hardware energy monitor as a tool for measuring energy accurately. As shown below, TI (currently) wins out among M4F cores, but a member of Atmel's SAML21 ARM Cortex-M4-based family (seems to) outshine all in the (current) ULPBench report reproduced below. Unlike TI's posted scores, Atmel's are not (currently) certified by EEMBC, so the jury is still out. You'll want to keep checking back in to see the latest in this battle for the bottom.

EEMBC ULPBench (as of 25 Mar 2015, 2:22pm Eastern) (Courtesy of EEMBC)

2 thoughts on “The battle for the bottom

  1. “SOMNIUM are members of the EEMBC Automotive subcommitee and see ULPBench, and the EnergyMonitor hardware as key technologies.nnWe've been using them to measure the benefits of our new SOMNIUM DRT software development tools product, which automatically a

    Log in to Reply

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.