Embedded Intelligence is the result of technological advancements that create an environment that allows the analysis of data on the level of cognitive reasoning with AI on the edge. Likewise, control and automation tasks traditionally carried out on a centralised server are shifting to distributed computing devices, making use of decentralised control algorithms.

The drivers of embedded intelligence?

1. Cyber-Physical Systems (CPS): Computing is ubiquitous and is embedded in an increasing number of objects in our life. These systems, often intelligent, are referred to as Cyber-Physical Systems (CPS) if they display integration of computation with physical processes. Their role is becoming increasingly dominant (autonomous cars, robots, home automation, etc) because of the functional benefits they provide (safety, security, autonomy). Because of their close integration with the physical world, the systems have to provide high-performance and low-power consumption as well for efficient processing of deep-learning algorithms. Many CPS have the capability to adapt functionality at run-time either through updates or by learning.

2. IoT and System-of-Systems (SoS): Modern highly-connected digital systems rarely operate independently. Value is added through cooperation between the systems to solve complex problems by exploiting functionalities generated through cooperation. Hence System-of-Systems emerge from the composition of embedded and cyber-physical systems (CPS) to reach an objective that none of the constituent systems can perform or reach on their own. Correct collective and autonomous behaviour of the composed systems is a challenge and a degree of intelligence is usualy required for such systems to work together.

3. Edge Computing: Edge computing allows near real time processing of data at the edge to solve tasks at the source. The shift to the edge supports the increased autonomy of CPS (autonomous cars, robots, home automation) which demands complex decision-making based on AI combined with strict timing constraints. Solution at the extreme edge (small sensors etc.) will require even more efficient computing systems, because of their low cost and ultra-low power requirements.

4. Embedded High Performance Computing (EHPC): In the IoT/SoS infrastructure some CPS have to provide supercomputing performances in a reduced space, weight and power consumption, ensuring at the same time, an adequate level of robustness and reliability. Similar computing requirements will be increasingly relevant also for embedded applications in new demanding areas such as autonomous driving cars and more generally in autonomous systems (such as automotive, train, aircraft) that are based on embedded vision, complex decision making and sensor data processing (Radar, Lidar, Positioning). EHPC is usually provided by a heterogenous multicore (CPU,GPU,DSP,FPGA) System-on-Chip.

The benefits of embedded intelligence?

Considering the above we can see how Embedded Intelligence can generate a better business compared to cloud-based solutions:

1. Latency: Round-trip data transfer response times for a cloud backend are typically of the order of 100 milliseconds. For many applications this is not a problem but sometimes latency is (safety) critical (e.g. autonomous cars) or improves the perception of products usability (e.g. mobile app).

2. Real-time: Analytics performed one the edge can approach real-time performance if carefully implemented. This is an important consideration in time critical applications. For example a time discrepency in motion control for pick-and-place robots on a production line can lead to a defective product being shipped or cause infrastructure damage.

3. Scalability: As data volumes grow in centralized cloud based solutions data transfer becomes a bottleneck. Edge-based solutions do most processing locally and therefore suffer less from volume problematics. As an example consider the relative scalability of analysing video streams from thousands of devices in the cloud or individualy on each device.

4. Privacy: Sending data to the cloud creates a vuneralability to online attack. Edge solutions often process on device or operate in a closed network. The risk of attack is therefore minimized.

5. Automation: Edge Intelligence is a great automation enabler. Consider process control in pharmaceutical production. Hundreds of sensors constantly monitor the process state and equipment diagnostics. AI can use that data to iteratively optimize processes “on-the-fly”.

6. Cost: We have seen how Edge Intelligence can reduce latency in decision making and improve scalability both of which reduce costs. It also increases energy efficiency by greatly reducing data transfer bandwidth and minimizes the need for costly cloud computing. This must however be offset by the cost of supplying hardware with sufficient local computing power.