Yesterday evening, All About Circuits took Ambiq up on its invitation to a sneak preview of its Apollo microcontroller running under the new ULPBench benchmark before the big announcement today. As expected, the MCU came out on top, but the surprise was by how much it beat the nearest competitors from STMicroelectronics (188) and Texas Instruments (124): it was almost double at 366.
Ambiq’s CEO, Mike Noonan, and EEMBC president, Markus Levy, were quick to point out that in the pre-preview demos the MCU--a 32-bit Cortex M4-based device with a floating-point unit–actually scored 377. Check out the video below for the demo in action:
Now, normally, when you’re 2x the competition in anything, you don’t care too much about a few percentage points difference in scores between demos. But with MCUs, particularly for the Internet of Things (IoT), every microwatt counts, so Noonan can be forgiven for being a tad competitive (and Levy can be forgiven for being excited about his new benchmark). In theory, twice the efficiency means twice the battery life. Or you can even possibly dispense with the battery and use some form of energy harvesting. But be careful. Look at the ULPBench benchmark itself, which you need to understand before coming to any decision based on Ambiq’s enormous win.
The benchmark came about as result of every MCU supplier saying that its MCU consumed the least power and therefore was better, particularly for IoT applications. They may be right, of course, but only for their particular set of test parameters. But those parameters may not be your application’s parameters, so ULPBench standardizes a known set of typical parameters and conditions so we can at least get a baseline reference point.
In summary, ULPBench standardizes a typical low-power design workload and measures the actual energy required to complete that workload. This approach normalizes the many different behaviors of MCU operation, such as active current, sleep current, wake-up time, core efficiency, and cache efficiency. It then synthesizes this data into a single value developers truly care about: the amount of energy required to complete their specific application.
Ambiq's Apollo MCU in a graphic nutshell.
The key to its low power consumption resides in its Subthreshold Power Optimized Technology (SPOT), which lets it operate at 0.5 V versus the more typical 1.8 V, taking advantage of the square relationship between power consumed and operating voltage (P is proportional to V2). The microcontroller consumes 34μA/MHz executing instructions from Flash, and sleep mode currents can be as low as 140nA. The Apollo MCU is at the heart of the Misfit Shine 2 wearable fitness and sleep monitor, and that energy consumption was a key criterion for its selection by Misfit designer engineers.
"The Misfit has an unprecedented number of sensors and can go up to six months without needing a recharge," says Noonan.
Ambiq won the design seat for the Misfit Shine 2, a sensor hub that can go six months without a recharge, helping remove one of the biggest obstacless to persistent use of fit or heatlh monitoring devices.