Queued-Stack Dataflow Processing Element for a Cognitive Sensor Platform

Abstract

This paper describes a Queued-Stack (QS) Dataflow Processing Element (DPE) that is used in a cognitive sensor platform. The queued-stack is used for buffering input data to the DPE and for storage of variables and results. The queuing mechanism and dataflow protocol provides the capability to compose multi-node computational systems where communication between elements is via non-blocking FIFO channels. System composition is achieved using synchronous dataflow tools such as SDF3 or Ptolemy. The dataflow-processing element is implemented using single cycle micro-coded engine where the ratio of datapath transistors to control logic is optimized for programmable energy-performance sensitive applications.