Close

Presentation

Reconfigurable Network-on-Chip Architecture for Intra-Cluster Communication on Spiking Neuromorphic Hardware
DescriptionA major challenge when designing hardware for large-scale, spiking neuromorphic architectures is connecting large number of neurons in a reconfigurable and efficient manner. Here, we propose a reconfigurable network-on-chip architecture for local spike communication among tightly packed neurons in a cluster. The proposed architecture is based on a circuit switching scheme and is reconfigurable to connect any two neurons within the cluster. We present the mathematical analysis for determining the parameters of this network. Furthermore, we show a comparison of the proposed network architecture against a traditional AER based approach in terms of communication latency, area overhead, and power consumption.
Event Type
Work-in-Progress Poster
TimeTuesday, July 11th6:00pm - 7:00pm PDT
LocationLevel 2 Lobby
Topics
AI
Autonomous Systems
Cloud
Design
EDA
Embedded Systems
RISC-V
Security