Page 257 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

3. Energy Harvesting and Management 249

FIGURE 12.1
(A) The characteristic curves for a photovoltaic cell: the acquired power depends on the
applied controlling voltage v p . (B) The control power transfer module identifies the optimal
voltage according to (Eq. 12.1) and applies it to the cell; other architectures can be
considered, e.g., see Ref. [4].

where g is a small constant accounting for the step taken along the gradient descent
direction. Such a solution requires acquisition of current i p and v p over time through
suitable sensors.

3.2 ENERGY MANAGEMENT AND RESEARCH CHALLENGES
Energy management is a very important issue in any CBS and IoT system given the
fact units are mostly battery powered and need to be kept as simple as possible to
reduce their cost.
Energy management can be carried out both at hardware and software/applica-
tion level by leveraging on
• Voltage/frequency scaling. By scaling power voltage and clock frequency, the
power consumption of the device reduces. In fact, for a CMOS technology the
power consumption scales quadratically with the voltage and linearly with
the working frequency [4]. Machine learning and fuzzy logic techniques
can be adopted, for example, to profile the application at compile time and
identify both at compile and run time when and how the control variables