A Fast General-Purpose Clustering Algorithm Based on FPGAs for High-Throughput Data Processing

TL;DR

This paper introduces a fast, FPGA-based clustering algorithm designed for high-throughput data processing, capable of real-time operation in high-density environments like high-energy physics detectors.

Contribution

It presents a novel linear-scaling clustering algorithm that can be implemented in FPGAs, enabling real-time, high-throughput clustering across various scientific fields.

Findings

Linear processing time scales with data size

Suitable for high-density, high-luminosity data environments

Applicable to multiple scientific and medical imaging fields

Abstract

We present a fast general-purpose algorithm for high-throughput clustering of data "with a two dimensional organization". The algorithm is designed to be implemented with FPGAs or custom electronics. The key feature is a processing time that scales linearly with the amount of data to be processed. This means that clustering can be performed in pipeline with the readout, without suffering from combinatorial delays due to looping multiple times through all the data. This feature makes this algorithm especially well suited for problems where the data has high density, e.g. in the case of tracking devices working under high-luminosity condition such as those of LHC or Super-LHC. The algorithm is organized in two steps: the first step (core) clusters the data; the second step analyzes each cluster of data to extract the desired information. The current algorithm is developed as a clustering device for modern high-energy physics pixel detectors. However, the algorithm has much broader field of applications. In fact, its core does not specifically rely on the kind of data or detector it is working for, while the second step can and should be tailored for a given application. Applications can thus be foreseen to other detectors and other scientific fields ranging from HEP calorimeters to medical imaging. An additional advantage of this two steps approach is that the typical clustering related calculations (second step) are separated from the combinatorial complications of clustering. This separation simplifies the design of the second step and it enables it to perform sophisticated calculations achieving online-quality in online applications. The algorithm is general purpose in the sense that only minimal assumptions on the kind of clustering to be performed are made.