Energy-Efficient Computation and Communication

Energy-Efficient Computation and Communication

The wearable systems we envision, as well as existing mobile computing systems, rely on limited battery power and have limited thermal budget.  Our Energy-Efficient Computing and Communication project aims at developing modeling, optimization and run-time management techniques for heterogeneous systems.

eLab is also part of the DARPA Domain-Specific SoC (DSSoC) program, which is part of the DARPA Electronics Resurgence Initiative (ERI) initiative. More information on our 17M DSSoC ProjectRelated stories: Science MagazineIEEE Spectrum

Our work combines mathematical modeling and control theory with the  state-of-the-art mobile platforms, such as Samsung Exynos, Intel Bay Trail and Qualcomm Snapdragon. We develop power, performance and temperature models and dynamic power/thermal management algorithms. Then, we implement these models inside the Linux kernel and replace the default operating systems that comes with the mobile device. This enables us to make accurate measurements and comparisons with the latest commercial solutions. The video below illustrates the power and temperature modeling procedure applied to Odroid XU3 powered by Samsun Exynos SoC.

Recent publications under this project:

Ganapati Bhat, et al. “Algorithmic Optimization of Thermal and Power Management for Heterogeneous Mobile Platforms,” in IEEE Trans. Very Large Integr. (VLSI) Syst., November 2017. [link]

Ujjwal Gupta, et al. “DyPO: Dynamic Pareto Optimal Configuration Selection for Heterogeneous MpSoCs,” in ACM Tran. on Embedded Comp. Sys. (ESWEEK Special Issue), October 2017. [link][poster]

Ganapati Bhat, Suat Gumussoy, and Umit Y. Ogras. “Power-Temperature Stability and Safety Analysis for Multiprocessor Systems,” in ACM Tran. on Embedded Comp. Sys. (ESWEEK Special Issue), October 2017. [link][poster]

Ujjwal Gupta, et al. “Dynamic Power Budgeting for Mobile Systems Running Graphics Workloads,” in IEEE Trans. on Multi-Scale Computing Systems, February 2017. [link]

Research overview:

Mandal, S. K., “Network-on-Chip (NoC) Performance Analysis and Optimization for Deep Learning Applications,” Technical Report,  June, 2021. [slides][report]

Basaklar, Toygun, “Multi-Objective and Energy Efficient Reinforcement Learning for Edge AI Applications,” Ph.D. Oral Defense. [slides]

Protocol Development Kit for Flexible protocol design:

This tool developed at ASU aims to enable the design of fluid protocols. Users can enter environmental parameters along with the required performance as inputs. The tool uses these to suggest an appropriate protocol scheme. Simulations are performed within the tool and then a hardware verification is done using Intel T2200 development boards.

Power and Temperature Modeling for Heterogeneous MpSoCs:

Empirically validated power and thermal models for Odroid XU3 will be released soon. They are currently being tested in EEE598 System-level Design for Multicore Architectures course.

The suite of experimental platforms currently used at eLab