An Efficient Framework for Floor-plan Prediction of Dynamic Runtime Reconfigurable Systems

TL;DR

This paper introduces a machine learning-based framework for predicting resource requirements and optimizing floorplans in dynamic reconfigurable systems, improving over static allocation methods.

Contribution

It presents a novel adaptive methodology that learns from historical data to accurately estimate resources and layout for reconfigurable hardware applications.

Findings

The system improves resource prediction accuracy over time.

It enables near-optimal floorplanning for dynamic reconfigurable systems.

The approach adapts to unseen applications through learned knowledge.

Abstract

Several embedded application domains for reconfigurable systems tend to combine frequent changes with high performance demands of their workloads such as image processing, wearable computing and network processors. Time multiplexing of reconfigurable hardware resources raises a number of new issues, ranging from run-time systems to complex programming models that usually form a Reconfigurable hardware Operating System (ROS). The Operating System performs online task scheduling and handles resource management. There are many challenges in adaptive computing and dynamic reconfigurable systems. One of the major understudied challenges is estimating the required resources in terms of soft cores, Programmable Reconfigurable Regions (PRRs), the appropriate communication infrastructure, and to predict a near optimal layout and floorplan of the reconfigurable logic fabric. Some of these issues are specific to the application being designed, while others are more general and relate to the underlying run-time environment. Static resource allocation for Run- Time Reconfiguration (RTR) often leads to inferior and unacceptable results. In this paper, we present a novel adaptive and dynamic methodology, based on a Machine Learning approach, for predicting and estimating the necessary resources for an application based on past historical information. An important feature of the proposed methodology is that the system is able to learn and generalize and, therefore, is expected to improve its accuracy over time. The goal of the entire process is to extract useful hidden knowledge from the data. This knowledge is the prediction and estimation of the necessary resources for an unknown or not previously seen application.