STMicroelectronics 2022
27
application that embeds the generated NN libraries and
the COM agent to communicate with the host system,
so the validation result can be checked via the terminal
emulators. To simplify this mode, you can enable the
"automatic compilation and download" option, so that the
tool automatically generates, compiles, programs, and
runs a temporary project corresponding to the current
network. The validation inputs can use either random
numbers or customized inputs.
Multiple networks can also be added, as supported by the
RAM and Flash memory of a selected STM32 MCU device
for validation at one time. Furthermore, X-CUBE-AI has
advanced settings that allow the engineer to manipulate the
memory scheme to further reduce the memory footprint,
such as configuring the network to use external Flash or
RAM, using activation buffers for input and output buffering,
and configuring custom layers. If one desires to use a
command line, all the core features are available through a
complete and unified Command Line Interface (CLI).
Regarding the STM32 NN library, only the specialized C
files are generated for each imported model. For example,
network.c and network.h files for the topology, and
network_data.c and network_data.h files for the weights
and bias, which you can find it in the generated project
folder "..\Src" and "..\Inc". The network_runtime.a library
or NN computing kernel library is provided as a
static library, which you can find in the project
folder "..\Middlewares\ST\AI".
B-CAMS-OMV
CAMERA MODULE
BUNDLE
Learn More
STM32H747I-DISCO
DISCOVERY KIT
Learn More
Embedded ML is rapidly evolving, and
constrained resources present the biggest
challenge. Achieving the optimal trade-off
between performance, memory footprint,
and power consumption is always the
primary objective. With the advent of
new developer tools, highly optimized
algorithms, and specialized AI hardware for
MCUs, such as neural network processing
cores and dedicated ML acceleration
hardware, the future of embedded ML
applications is very promising.
This article is a very brief introduction to the
embedded ML. To know more about ST's AI
ecosystem for STM32, please go to www.st.com/
stm32ai. To learn more about the X-CUBE-AI tool,
please refer to UM2526 – Getting started with
X-CUBE-AI Expansion Package for AI. ■
