We have explained previously how Ensemble and Balletto microcontrollers pair a neural processing unit (NPU) with a CPU to produce very fast AI inferences in edge applications, while also performing system control functions as a conventional MCU does.
At the edge, a small battery is the most common power source. Is this compatible with AI? Many worry about the power consumption of AI operations, because we hear about cloud service providers buying nuclear power stations just to power their AI data centers.
Many edge devices have to offer a long run-time from a small battery
In fact, the CPU+NPU architecture of Alif products makes it completely viable to run AI at the edge: because the NPU performs inferencing operations very fast, the time during which it is drawing power from the battery is very short.
But there’s another reason why AI operations running on Alif products are compatible with the battery even in as small a device as an earbud: the aiPM™ (autonomous intelligent Power Management) technology built into Ensemble and the wireless Balletto MCUs provides ways to save power that were previously unknown in mainstream MCUs.
The aiPM circuitry is incredibly sophisticated, and provides for granular control of the power supply to functional blocks within the chip. Essentially, aiPM ensures that parts of the chip are only powered on when they are required by the application – the aiPM system can manipulate in real time up to 12 individual power domains in the chip.
How does it work in practice?
aiPM technology uses power saving features such as frequency scaling, clock gating, powering off unused regions, and sleeping for extended periods. The aiPM circuitry supports five modes: RUN, READY, IDLE, STANDBY, and STOP.
Tested when operating at a supply voltage of 3.3V, in STOP mode an Alif MCU draws current of less than 1.6μA with the real-time clock running and wake sources active.
In RUN mode, the device draws 29μA/MHz with the Arm® Cortex®-M55 CPU running code from SRAM.
Simplifying the bill-of-materials and system development
The power savings achieved through the use of the aiPM technology are an incredibly valuable benefit, helping to make sophisticated AI possible in small, power-constrained embedded devices operating at the edge.
But there is another advantage of the fine power control provided by aiPM: in many cases, the integration of power management inside the MCU eliminates the need for an external power management IC (PMIC). This is valuable, because it not only saves board space and lowers bill-of-materials cost. It also means that developers avoid the need to devote time and energy to learning how to implement an external PMIC’s functions, to qualifying this additional component, and to updating or replacing it in the event of obsolescence.
So the aiPM feature of Ensemble and Balletto MCUs is not only good for power savings and battery life – it also streamlines the entire system design.