Embedded ML - TinyML

§ Intelligence is moving down to the ‘edge’ to keep

processing as close as possible to the source

of data

§ Edge AI refers to any device with the capabilities to

process AI workload

§ Tiny Edge – sensor class devices or small application

processors, typically Cortex M class

(M33, M4 or equivalent/below) run TinyML applications

§ TinyML Characteristics:

§ Ultra-low power applications and systems (<mW range)

§ Decent inference speed but more focused on efficiency

§ Algorithms, networks, and models down to < hundreds of kB

§ Efficient ML accelerators

§ Low cost, optimized, well integrated systems