PRATHAM: A power delivery-aware and thermal-aware mapping framework for parallel embedded applications on 3D MPSoCs

In emerging 3D-ICs, thermal hotspots and high IR-drops in the power delivery network (PDN) can significantly limit overall system performance. The high core counts required to support parallel embedded applications in these 3D-ICs also notably increases communication energy. As inter-core communication patterns, IR-drop distributions, and 3D thermal profiles all influence system performance and power, it is critical that a system-level application mapping framework consider their combined effect. In this paper, for the first time, we propose a system-level parallel embedded application allocation framework (PRATHAM) that integrates a holistic solution evaluation methodology while considering the impact of network-on-chip (NoC) communication, drops in supply voltage, and the on-chip thermal profile on overall system performance, and co-optimizes on-chip IR-drop, thermal, and communication profiles, for improved overall system performance and lower energy consumption. Our experimental results indicate that PRATHAM reduces energy-delay-squared product (ED²P) by up to 43.6% over recently proposed PDN-aware and thermal-aware 3D-mapping frameworks, respectively.