Markovian Models Based Stochastic Communication in Networks-in-Package

摘要

Like networks-on-chip (NoC), a solution for inter intellectual properties (IPs) communication in system-on-chip (SoC), networks-in-package (NiP) is a solution for inter NoCs communication in system-in-package (SiP), where power dissipation, performance, and size of processing package are the major issues. NiP is a group of NoCs mounted on a single package. Developing a design methodology for local communication in NoC and global communication in NiP poses novel and exciting challenges to the electronic design and automation (EDA) community. This paper introduces a novel communication paradigm, which merges these two communications. Transfer of information in NoC and among NoCs can be represented using closed donor controlled based compartmental model, convertible into a stochastic model. This stochastic model is more realistic and enables to compute the transition probabilities for dynamic allocation of single buffer, energy saving from inactive routers, and expected delay in NoC and among NoCs. Moreover, at any particular instant of time (while the transition probability is maximum) we can track the packets among on chip routers or switching elements (SEs). These tracked packets are used in the real time applications such as robotics, real time data acquisitions, and control applications.