Self-adaptative approximate system-on-a-chip: development of the hardware platform.

Abstract

Approximate Computing is a very recent issue that has been attracting the attention of the researchers, mainly due to the increase of RMS software and IoT applications, which process data with intrinsic error, thus allowing computational processing of approximate form. However, the research in this area still needs a lot of progress, one of them, is to build a computational platform that allows to test ideas of AxC with enough productivity, in the entire spectrum of Computer Engineering, from the transistor to Multiprocessed Operating Systems. The main objective of this work is to establish a platform for researchers in the field to test their ideas, train people and advance research. To this end, several free computing platforms have been evaluated, the most promising being the University of Berkeley’s the Rocket Chip SoC Generator. This project, although quite complete for what the research demands, does not have documentation that allows to understand and use the platform and all its capacity in a productive way. This work creates documentation by gradually addressing almost everything that this platform offers. In the end, all knowledge acquired is used to implement and test a Coprocessor on the platform with Approximate Computing technique.