| |
Prafull Purohit and Rajit Manohar
Address-event representation (AER) is an event-driven,
neuromorphic inter-chip encoding and communication
protocol originally proposed to communicate location and timing
information of sparse neural events between neuromorphic
chips. The protocol is widely used in bio-inspired, event-based
vision sensors to communicate visual events. The same
approach has been explored for scientific imaging applications,
but the performance of existing encoding schemes degrades in
the presence of large sensor arrays and a wide range of event
rates. In this paper, we introduce a new AER encoding scheme
based on hierarchical token-rings (HTR), which blends event-based
and scanning approaches. We show that HTR offers
significant improvement in latency, throughput, and power
compared to existing tree-based approaches in the context of
scientific imaging applications.
|
|
