Pixel-Level Clustering Algorithm for Data Reconstruction in ILC

A high amount of hits on the vertex detector of the International Linear Collider (ILC) are generated by the beam background, which leads to an increase in the data flow of the detector system. Charged particles coming from the beam background have low momentum, resulting in the generation of elongated clusters. The CMOS pixel sensor (CPS), which integrates pre-processing functions and on-chip artificial neural networks (ANNs), could remove these elongated clusters. Clustering is the first step for data pre-processing and is used to collect clusters from raw data. In this article, a pixel-level clustering algorithm with a
$5\times5$
window executed in real time is proposed. The algorithm is tested using 4500 frames (500 frames for each angle of incidence) of raw data (12 bits/pixel) from a MIMOSA-18 sensor and compared to conventional clustering algorithms. The clustering implementation for an example array of
$5\times5$
pixels is synthesized for different frequencies (100 and 200 MHz) and analog-to-digital converter (ADC) resolutions (4 and 8 bits). The power dissipation and occupied area of the different implementations are analyzed. The hardware implementation of the algorithm provides the possibility to integrate the clustering function into the CPS.