AI-Driven FPGA Power Prediction and Performance Optimization for Autonomous Drone Systems

Abstract

Power consumption has become one of the most critical design constraints in modern Field Programmable Gate Array (FPGA) based systems. With the rapid growth of complex digital applications such as signal processing, embedded control and artificial intelligence accelerators, designers must accurately estimate power at early stages of development. Conventional power estimation tools provided by FPGA vendors usually rely on detailed low-level design information and simulation activity, which are often unavailable during early design phases.

This paper presents an AI-based FPGA power prediction system that employs machine learning techniques to estimate dynamic and total power consumption using high-level design and implementation features. The proposed system extracts relevant design attributes such as resource utilization, clock frequency and logic structure statistics, and uses supervised learning models to predict power values. A complete workflow for dataset generation, feature extraction, model training and evaluation is presented. Experimental results demonstrate that the proposed approach can predict FPGA power consumption with high accuracy and significantly reduced estimation time, making it suitable for early-stage design exploration and optimization.