AI-Augmented Runtime Reconfigurable Hardware Architectures for Energy-Efficient Edge Intelligence
Year: 2026
Type: article
Abstract: It is edge intelligence that has enabled the execution of learning and inference operations near the source of data to achieve low-latency, privacy-preservation data processing capabilities. Embedded edge platforms are however run under strict energy, latency as well as resource constraints, which greatly restrict the efficiency of traditional fixed hardware accelerators. Even though reconfigurable hardware architecture lets out architectural flexibility; it has become a significant challenge to establish the most desirable runtime configurations in the context of dynamic and heterogeneous workloads. The paper is a proposal of an AI-enhanced runtime reconfigurable hardware architecture, developed to facilitate an energy-efficienting edge intelligence using malleable architecture control. The offered system combines a lean learning-based control plane and embedded reconfigurable data plane, enabling the dynamic virtualization of computational assets, dataflow frameworks, and calculating precision. A decision engine that uses AI continually analyzes workload properties and system-level performance indicators, such as latency and energy use, to inform reconfiguration easy choices such as dynamic partial reconfiguration, accelerator choice and precision scaling. To achieve a significant degree of tight coupling between learning-based control logic and reconfigurable hardware modules without incurring a high overhead to be used in embedded implementation, a hardware-software co-design approach is used. The suggested architecture is deployed and tested on an embedded programmable equipment platform with realistic workloads of edge intelligence. The experimental performance is shown to have a high energy efficiency and latency values, by up to 3045 percent of energy savings over the fixed and heuristic-based reconfigurable systems, at minimum runtime reconfigurability overhead. These findings affirm that AI runtime reconfiguration offers a successful and scalable solution to adaptive energy-constrained edge intelligence ability in embedded systems of the next generation.
Author: Prerna Dusi
Institution: Kalinga University
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7220133510
