A new approach to efficient heterogeneous computing

Abstract With global scale applications, we have stepped into an era of heterogeneous computing. The three big challenges of this time are energy efficienxy, system complexity, and dependability. With this work, I answer the question on how to quickly design systems that have just the right flexibility, energy efficiency, and accuracy. My main contribution is the agile tradeoff-centric design methodology. With this approach design teams can utilize their full creativity and concentrate on the tradeoffs with the most leverage to create efficient and adaptable platforms. I apply this methodology to three applications domains: industry 4.0, big data, and finance. There I uncover more than twenty novel tradeoffs and create four of those platforms. I show how those systems bring one to five orders of magnitude improvements in energy efficiency to each of these domais over the state-of-the-art.