STM Journals Peer-reviewed research from Consortium e-Learning Network

Recent Trends in Parallel Computing Review Article

Energy-Efficient Parallel Computing for Edge Devices: Techniques, Challenges, and Emerging Research Directions

  1. Nidhi Chahal Department of Electronics and Communication Engineering, Chandigarh Engineering College-CGC
  2. Samar Singh Department of Electronics and Communication Engineering, Chandigarh Engineering College-CGC
  3. Ritik Department of Electronics and Communication Engineering, Chandigarh Engineering College-CGC

Abstract

Edge computing brings computation closer to data sources, which supports low-latency applications but places substantial demands on energy-constrained and heterogeneous devices. Parallel execution across multicore CPUs, GPUs, FPGAs, and other accelerators can improve responsiveness, but it also complicates scheduling, power management, and resource allocation. This review examines six closely related technique families: energy-aware task scheduling, dynamic voltage and frequency scaling (DVFS), task offloading and resource allocation, artificial-intelligence-based optimization, hardware acceleration, and mathematical optimization. Representative studies show that energy efficiency is obtained through different mechanisms: battery-aware placement and scheduling, coordinated control of voltage and frequency, selective migration of computation, learned policies for dynamic environments, specialized data paths, and formal constrained optimization. Reported improvements are promising but are tied to particular devices, workloads, baselines, and measurement methods; therefore, they cannot be treated as directly comparable. The review also identifies recurring limitations, including limited hardware-level validation, weak cross-platform generalization of learning-based methods, inconsistent benchmarks, and fragmented optimization across system layers. Potential research directions include joint scheduling-DVFS-offloading control, hardware-software co-design, transfer learning, sparse-aware execution, federated optimization, energy-aware accelerators, and renewable-aware edge infrastructure. The central conclusion is that practical edge efficiency requires integrated policies that balance energy, latency, reliability, privacy, and implementation overhead rather than optimizing any single metric in isolation.

Keywords

References (12)

  1. Llorens-Carrodeguas A, Sagkriotis SG, Cervelló-Pastor C, Pezaros DP. An energy-friendly scheduler for edge computing systems. Sensors (Basel). 2021 Oct 28;21(21):7151.
  2. Liu Y, Qu H, Chen S, Feng X. Energy efficient task scheduling for heterogeneous multicore processors in edge computing. Sci Rep. 2025 Apr 7;15(1):11819.
  3. Zhang Z, Zhao Y, Li H, Lin C, Liu J. DVFO: learning-based DVFS for energy-efficient edge-cloud collaborative inference. IEEE Trans Mob Comput. 2024;23(10):9042–9059.
  4. Zhang Z, Wu Z, Liu J, Mottola L. SparseDVFS: sparse-aware DVFS for energy-efficient edge inference [Preprint]. 2026. arXiv:2603.21908.
  5. Yan J, He F, Sang Q, Tong B, Sun P, Gong Y, et al. Metadata-guided adaptable frequency scaling across heterogeneous applications and devices [Preprint]. 2025. arXiv:2509.22707.
  6. Hou L, Li S, Shen Z, Fu J, Wu J, Jin J. Energy efficient offloading policies in multi-access edge computing systems with task handover. IEEE Trans Mob Comput. 2026 Jan 19:9419–9436. doi:10.1109/TMC.2026.3655433.
  7. Mohapatra H. Task offloading and edge computing in IoT—gaps, challenges and future directions. Comput Mater Contin. 2026;87(3).
  8. Nishad DK, Verma VR, Rajput P, Gupta S, Dwivedi A, Shah DR. Adaptive AI-enhanced computation offloading with machine learning for QoE optimization and energy-efficient mobile edge systems. Sci Rep. 2025 May 1;15(1):15263.
  9. Pérez I, Figueroa M. A heterogeneous hardware accelerator for image classification in embedded systems. Sensors (Basel). 2021 Apr 9;21(8):2637.
  10. Huang M, Shen A, Li K, Peng H, Li B, Su Y, et al. EdgeLLM: a highly efficient CPU-FPGA heterogeneous edge accelerator for large language models. IEEE Trans Circuits Syst I Regul Pap. 2025;72(7):3352–3365.
  11. Gelenbe E. Minimizing delay and power consumption at the edge. Sensors (Basel). 2025 Jan 16;25(2):502.
  12. Zhou T, Lin M. CPU frequency scheduling of real-time applications on embedded devices with temporal encoding-based deep reinforcement learning. J Syst Archit. 2023 Sep;142:102955.
Support