MACHINE LEARNING-INSPIRED RESOURCE MANAGEMENT IN M3D-ENABLED MANYCORE ARCHITECTURES

ANWESHA CHATTERJEE

doi:10.7273/000004397

Monolithic 3D (M3D) integration has emerged as an enabling technology to design high performance and energy-efficient circuits and systems. The smaller dimension of vertical monolithic inter-tier vias (MIVs) lowers effective wirelength and allows high integration density. To design an energy-efficient many-core architecture, necessitates efficient resource management of the full SOC system, in terms of power and performance of the system. Voltage/frequency island (VFI)-based power management is a popular methodology for designing energy-efficient manycore architectures without incurring significant performance overhead. In an M3D chip, the vertical layers introduce inter-tier process variations that affect the performance of transistors and interconnects in different layers. Therefore, VFI-based power management in M3D manycore systems requires the consideration of inter-tier process variation effects. In this dissertation, we undertake the problem of resource management in M3D many-core architectures degraded due to inter-tier process variation effects inherent in M3D chips. Firstly, we present the design of an imitation learning (IL)-enabled VFI-based power management strategy that considers the inter-tier process-variation effects in M3D manycore chips. We demonstrate that the IL-based power management strategy can be fine-tuned based on the M3D characteristics. Our policy generates suitable V/F levels based on the computation and communication characteristics of the system for both process-oblivious and process-aware configurations. Subsequently, we propose a machine learning-based online update strategy of IL-based DVFI policies for process degraded M3D architectures. We demonstrate that with no prior knowledge of process-variation parameters, our online strategy captures the inter-tier process variations in the M3D system improving the power-performance trade-off than a process-oblivious offline DVFI policy for the degraded M3D many-core architecture. Furthermore, we show that online update strategy improves the overall energy-efficiency for unseen workloads that are not considered during offline DVFI policy creation.

MACHINE LEARNING-INSPIRED RESOURCE MANAGEMENT IN M3D-ENABLED MANYCORE ARCHITECTURES

(1)