A z-Vertex Trigger for Belle II

TL;DR

This paper proposes a neural network-based z-vertex trigger for the Belle II experiment, capable of suppressing background events in real-time by estimating the event vertex position within trigger latency.

Contribution

It introduces a novel neural network approach for real-time z-vertex estimation in high-background conditions, implemented on FPGA hardware for the Belle II trigger system.

Findings

Feasibility of neural network-based z-vertex estimation demonstrated.

Implementation on FPGA shows promise for real-time processing.

Significant background suppression potential in Belle II experiment.

Abstract

The Belle II experiment will go into operation at the upgraded SuperKEKB collider in 2016. SuperKEKB is designed to deliver an instantaneous luminosity $\mathcal{L}=8\times10^{35}\,\mathrm{cm}^{-2}\,\mathrm{s}^{-1}$. The experiment will therefore have to cope with a much larger machine background than its predecessor Belle, in particular from events outside of the interaction region. We present the concept of a track trigger, based on a neural network approach, that is able to suppress a large fraction of this background by reconstructing the $z$ (longitudinal) position of the event vertex within the latency of the first level trigger. The trigger uses the hit information from the Central Drift Chamber (CDC) of Belle II within narrow cones in polar and azimuthal angle as well as in transverse momentum ("sectors"), and estimates the $z$-vertex without explicit track reconstruction. The preprocessing for the track trigger is based on the track information provided by the standard CDC trigger. It takes input from the 2D track finder, adds information from the stereo wires of the CDC, and finds the appropriate sectors in the CDC for each track. Within the sector, the $z$-vertex is estimated by a specialized neural network, with the drift times from the CDC as input and a continuous output corresponding to the scaled $z$-vertex. The neural algorithm will be implemented in programmable hardware. To this end a Virtex 7 FPGA board will be used, which provides at present the most promising solution for a fully parallelized implementation of neural networks or alternative multivariate methods. A high speed interface for external memory will be integrated into the platform, to be able to store the $\mathcal{O}(10^9)$ parameters required. The contribution presents the results of our feasibility studies and discusses the details of the envisaged hardware solution.