Intelligence at the Edge
32
Intelligent Sensor with Embedded
Processing Unit
Intelligent sensors with MLC offer the features to run
predefined AI algorithms such as decision-tree-based
models. An Intelligent Sensor Processing Unit (ISPU), with
an embedded ultra-low power microcontroller, offers the
flexibility to run a broad range of algorithms at very low
power and high efficiency. For example, algorithms based
on decision trees, binary neural networks, and anomaly
detection can efficiently run in an ISPU sensor module.
Figure 6 shows a high-level block diagram of an ISPU
connected to an external host. The ISPU block consists of
a set of sensors, a dedicated RAM and a custom-designed
microcontroller to run AI algorithms. While powering up
the ISPU sensor module an external host uploads the
software/algorithms into the RAM of the ISPU module. The
module is equipped with I2C and SPI serial interfaces to
facilitate this communication between the module and the
external host. During the operation, the module can act
autonomously at ultra-low power, firing interrupts to wake
up the external host only when the application requires it.
Applications of Intelligent Sensors
With the fast expansion of the Internet of Things
(IoT), the number of sensors deployed is increasing
significantly and the amount of data collected is growing
exponentially. These increases require complex multi-
layer machine learning algorithms to create predictive
models. Sensor data can be processed and analyzed
at the edge depending on the target applications and
their requirements. Alternatively, the sensor data can be
transmitted to a gateway or to the cloud, although this
data transmission to the gateway and cloud increases
the latency, raises security issues, and increases power
consumption as large amounts of data need to be
transmitted in real-time.
The edge usually consists of one or multiple sensor
nodes with processing components and a limited power
budget because most nodes are battery-operated. The
latest machine-learning capabilities and AI embedded in
sensors are designed to address the issues mentioned
above. The trend of innovative AI inside the sensors
enables a wide range of applications. Figure 7 shows
some of these applications with dedicated algorithms that
can run within the sensor. They all release the external
processing units for more complex and critical functions.
ISM330DHCX INEMO
INERTIAL SIP
MODULE
Learn More
LSM6DSRX INEMO
INERTIAL
MODULE
Learn More
LSM6DSOX INEMO
INERTIAL
MODULE
Learn More
ISPU:
Intelligent Sensor
+Processing Unit
External
Microcontroller/
Microprocessor
SPI/I2C/13C
Interrupts
Figure 6: Intelligent Sensor Processing Unit (ISPU) and an external host. (Source: STMicroelectronics)
