Ambiq Brings Low-Power AI to the Edge
Author: Dylan McGrath
The execution of AI continues to move as quickly as possible from cloud-connected data centers to the edge. But cramming more compute performance into battery-powered IoT devices is challenging from a space and power perspective. Ambiq’s Apollo510 microcontroller (MCU) pairs the company’s subthreshold energy-efficient design technology with vector acceleration performance to bring more AI capability to edge devices.
Ambiq has long been known for its energy efficient devices based on its proprietary subthreshold technology. The company was founded in 2010 by CTO Scott Hanson based on technology developed during his research at the University of Michigan.
Apollo510 is the first entry in Ambiq’s fifth-generation MCU family, Apollo5. It integrates a Cortex-M55 CPU running at up to 250 MHz with 4 MB of on-chip nonvolatile memory (NVM) and 3.75 MB of on-chip SRAM as well as a healthy collection of peripheral interfaces. Compared to Apollo4, it offers a substantial improvement of 3–11 times in AI performance and four times in energy efficiency for AI workloads. It also beats rivals in energy efficiency; Apollo510’s active power consumption is about one third that of its closest competitor.
The Apollo510 is manufactured in TSMC’s ultra-low leakage 22nm process. It is sampling now, with volume production scheduled for the fourth quarter. The company withheld pricing.
