Indian Journal of Science and Technology
Year: 2017, Volume: 10, Issue: 16, Pages: 1-7
S. Santhosh Kumar1*, S. Vidhya2 and M. M. Shanmugapriya3
1Department of Mathematics, KG College of Arts and Science, Coimbatore – 641035, Tamil Nadu, India; santhoshkumar.s@kgcascom 2Department of Information Technology, KG College of Arts and Science, Coimbatore – 641035, Tamil Nadu, India; [email protected] 3Department of Mathematics, Karaqpagam University, Coimbatore – 641035, Tamil Nadu, India; [email protected]
*Author for correspondence S. Santhosh Kumar Department of Mathematics, KG College of Arts and Science, Coimbatore – 641035, Tamil Nadu, India; santhoshkumar.s@kgcascom
Hardware-based computer vision accelerators are going to be an important part of future mobile devices to satisfy the low power and data processing demand. In order to comprehend a high power potency and high turnout, the accelerator design will be massively parallelized and tailored to vision process that is a plus over software-based solutions and all-purpose hardware. In this research Spiking neural networks (SNNs) arrange to emulate scientific discipline within the class brain supported neurons parallel arrays that communicate through spike events. The opportunity to perform embedded neuromorphic circuits is supplied by SNNs, with low power consumption and high correspondence in comparison with the normal laptop paradigms of John von Neumann. Even so, the poor property and modularity shortage as shown in ancient neuron cell interconnect implementations supported shared bus topologies is barring climbable hardware operations of SNNs. In order to effectively apply SNN traffic patterns and neighborhood among neurons in the current design the Hybrid Network on Chip (H-NoC) design integrates a spike traffic compression technique, thus dropping traffic overhead and up turnout on the network provides world traffic hundreds to sustain turnout underneath bursting activity. The planned system reduces overhead and improves the performance through native routing of the neutron cell facilities that are the gifts within constant tile facility. This will increase the potency of the system. The scalability of the adopted H-NoC approach under completely different situations is shown by analytical results show, while synthesis and simulation analysis reveal, area of low-cost, and delay for each cluster severally. This methodology finds its application in various sector such as medical image processing and bio signal processing.
Keywords: Hybrid Network, Hybrid Network on Chip, Neural Network, Scalable Spiking, Spiking Neural Network
Subscribe now for latest articles and news.