Page 255 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

3. Energy Harvesting and Management 247

acquired data are complete and correct, and secure operations, assuming that cyber-
attacks are not carried out. By relaxing above assumptions we enter in a new realm
requesting presence of intelligence in all computational architectures; these aspects
are addressed in subsequent sections.

2. SYSTEM ARCHITECTURE
The physical ICT architecture we consider here reflects those considered for CPS
and IoT [6]. As such, it is a very variegated one composed, in its most general
form, of heterogeneous hardware and software platforms. End point units of the
(Internet) connection communicate with servers (possibly also acting as gateways)
in a star, field bus, or general topology depending on the particular application at
hand. Computational complexity and hardware resources (e.g., memory capacity,
energy availability) are application-specific, with units that, not rarely, are operating
system-free. In other cases, units possess a simple operating system (e.g., RTOS), a
more complex one (e.g., Android), or one that is specifically developed for limited-
resource devices, such as Contiki [7], ARMmbeb [8] or, again, specifically targeted
to IoT such as the Google Android Things [9].
An end unit can mount application-specific sensors and/or actuators, with the
interesting case where humans can act both as sensors and actuators. Fog and cloud
computing processing architectures can be elements of the overall architecture: the
final architectural decision depends on the expected and the maximum response time
tolerated by the application.
Where should intelligence be located in the architecture? This complex question
receives a very simple answer: it depends on the energy availability and the
computational and hardware/software resources needed by the application and the
intelligent functionalities to carry out their tasks. Once intelligent functionalities
are taken into account, we should consider hierarchical processing solutions with
intelligencedfor a given functionalityddistributed along the processing architec-
ture. Within this framework, low performing end units provide (very) simple
functionalities but lack a comprehensive, global view of the problem. As such, we
should expect decisions taken in a hierarchical way, with the effectiveness of the
decision increasing with the availability of a larger set of data/features and the
possibility to execute a more complex algorithm. In fact, more processing
demanding algorithms can be run, for example, to identify a better solution to a
specific problem, in systems where larger computational power/energy is available.
Result outcomes from the processing stage are then sent back downward the
communication chain to reach end units.

3. ENERGY HARVESTING AND MANAGEMENT
In CPSs where energy availability is an issue, an accurate and sound management of
energy in addition to energy harvesting represents a major necessity. Due to its