DCE3 - An universal real-time clustering engine

Abstract

In this paper we describe an universal algorithm for data clustering on hardware level. By utilization of a software-inspired hardware architecture, the clustering task can be executed by a data clustering engine (DCE) in a pipelined data stream mode with a latency of only one frame. The scalable architecture of the clustering core allows a quick adaption of the engine to the specific needs of the target application. Furthermore, a prototype implementation in a TSMC 65nm low power CMOS process and an outlook on the final design for the Belle II experiment at KEK/Japan will be presented.