Performance Analysis of Compact On-Chip Operators for Cluster Feature Extraction

In the International Linear Collider (ILC), charged particles that come from the beam background may hit on the vertex detector and generate elongated clusters, leading to the increasing of useless data in the detector system. In previous work, we conducted the feasibility study of a CMOS pixel sensor with ON-chip artificial neural networks (ANNs) in order to remove the elongated clusters. The principle component analysis (PCA) is a good approach, which was developed in the FPGA implementation for feature extraction. However, the PCA method is challenging for the ASIC implementation. In this article, we propose four features and corresponding two operators for feature extraction. A total of 4500 frames of raw data at nine angles of incidence are sampled from MIMOSA-18. The clustering procedure is executed on raw data to gather clusters. Two operators are achieved by the C code and tested on the experimental clusters. Distributions of four features of these clusters are analyzed. Results show that these features can represent the pattern of clusters with different incident angles. The ratio of feature Summin to Summax is decreased as the incident angle increases, which verifies that the clusters are elongated at the large incident angle.