While inexpensive, harvested power is highly variable in both space and time. When we minimize their storage capacity, systems become cheaper, smaller and can still function efficiently when computation is performed as a function of the harvested energy. There are many new research topics in the hardware/software codesign of such systems, which is my central PhD topic.
Energy Management Unit (EMU):
In the absence of large energy storage devices, one key design parameter is how to exchange energy between a transducer and an electronic load. If the I-V properties of the transducer are not aligned with those of the load, an intermediary is unavoidable. The Energy Management Unit (EMU) is one such device, which enables quantized energy transfers between a transducer and a load. With an EMU, designers can minimize the system's total size and cost and still supply generic loads in a scalable and efficient manner.
|Figure. EMU Block Diagram.|
In the real-time domain, computing systems must not only be functionally correct but also respect precise time constraints. Of the many factors that can affect the response time of these systems, the most influential is the arbitration scheme of limited resources: the scheduling algorithm. For my master thesis, I developed SF3P, an application-level framework for prototyping and composing different real-time schedulers.
Scheduling Framework For Fast Prototyping (SF3P):
SF3P introduces a new layer at the application level that implements different scheduling algorithms to be used by SF3P tasks running on top of it. This would not be possible in a standard linux system because the linux scheduler is priority-based, and has limited scheduling options. Naturally, since SF3P itself runs on top of the standard pthread library, all algorithms must be implemented using a priority-based mechanism.
|Figure. User-space Scheduling of Real-Time Tasks.|
Andrés Gómez 2017©