Yoni Aizik

H-EARtH: Heterogeneous Multicore Platform Energy Management

By scheduling each workload according to its most advantageous core and managing voltage and frequency, the heterogeneous energy-aware race to halt (H-EARtH) algorithm optimizes CPU platform energy.

Challenges and methodologies for efficient power budgeting across the die

To provide higher performance by staying within a constant power budget is challenging. Today, operational cost and specifically energy cost constitute a large portion of an organizations Total Cost of Ownership (TCO), so it is crucial that microprocessor consumes the lowest possible power for a given workload. In this paper, we describe the major challenges faced by the microprocessor architects in achieving higher performance within constant power budget or energy-efficiency, with special focus on design technologies and solutions. We present a die level power simulation solution that facilitates exploration of system-level power management algorithms that budget power for various microprocessor die components. We demonstrate usage methodology via real-life use case studies and recommend future areas of research that can enhance the proposed power management simulation methodology.

Power-Aware Design via Micro-architectural Link to Implementation

Micro-architectural power macro-models dictate the power budget of a new chip design. Based on the early feasibility studies, power specification of new features are defined and then verified all through the design cycle. In this paper, we introduce a novel power-aware design paradigm that aligns power macro-models by mapping power-significant events at all levels of design hierarchy. We apply this paradigm on a state-of-the-art 65nm high-performance micro-processor and demonstrate significant benefits in power optimization at RTL implementation.Moreover, this approach facilitates a feedback loop from the design implementation to higher level (micro-architectural) power models and thus has built-in potential for more accurate power models.