Abstract:The combinatorics of track seeding has long been a computational bottleneck for triggering and offline computing in High Energy Physics (HEP), and remains so for the HL-LHC. Next-generation pixel sensors will be sufficiently fine-grained to determine angular information of the charged particle passing through from pixel-cluster properties. This detector technology immediately improves the situation for offline tracking, but any major improvements in physics reach are unrealized since they are dominated by lowest-level hardware trigger acceptance. We will demonstrate track angle and hit position prediction, including errors, using a mixture density network within a single layer of silicon as well as the progress towards and status of implementing the neural network in hardware on both FPGAs and ASICs.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the track reconstruction combination problem in high - energy physics (HEP), especially in the trigger and offline calculation of the High - Luminosity Large Hadron Collider (HL - LHC). Specifically, the paper explores how to use the next - generation pixel sensors to directly predict the charged - particle trajectory parameters and their uncertainties on the sensors, in order to reduce the combinatorial explosion problem and thus improve the efficiency and precision of physical research. ### Main problem points: 1. **Combination problem of track reconstruction**: Traditional track reconstruction methods will encounter the problem of combinatorial explosion when dealing with a large amount of data, which not only increases the computational burden but also limits the efficiency of physical experiments. 2. **Implementation of real - time trigger system**: In order to implement a real - time trigger system, efficient track parameter prediction at the hardware level is required in order to quickly screen out events of interest. 3. **Improvement of sensor technology**: By using finer - grained pixel sensors, the angular information of charged particles can be measured more accurately, but new algorithms are required to process these data. ### Solutions: - **Mixture Density Network (MDN)**: The paper proposes a method based on the Mixture Density Network, which can predict the trajectory angle and position of charged particles on a single - layer silicon sensor while providing error estimates. - **Hardware implementation**: This network can not only run on FPGA but also be synthesized into ASIC to achieve low - power consumption and efficient data processing. - **Data compression and feature extraction**: By performing data compression and feature extraction in real - time on the device, the amount of data transmission can be significantly reduced and the performance of the trigger system can be improved. ### Experimental results: - **Prediction accuracy**: The paper shows the prediction results on the test data set, with the average prediction errors being 5.7 micrometers in the x - direction, 1.1 micrometers in the y - direction, 3.8 degrees in the polar angle, and 1.7 degrees in the azimuth angle respectively. - **Search window size**: Through the predicted error, an efficient search window can be determined, so that the search space for matching hit points in the subsequent pixel layers is reduced by two orders of magnitude. ### Conclusion: This paper successfully demonstrates a highly compressed, quantized, and well - performing regression neural network that can be implemented in modern FPGA and silicon lithography processes for predicting track states from future silicon pixel sensors. This method not only improves the efficiency of track reconstruction but also makes it possible to implement a real - time trigger system. Future research will further optimize the power consumption and implementation method of the network and verify its performance in hardware.

Smartpixels: Towards on-sensor inference of charged particle track parameters and uncertainties

Smart pixel sensors: towards on-sensor filtering of pixel clusters with deep learning

Intelligent Pixel Detectors: Towards a Radiation Hard ASIC with On-Chip Machine Learning in 28 nm CMOS

Smart Pixels: In-pixel AI for on-sensor data filtering

Pixel data real time processing as a next step for HL-LHC upgrades and beyond

Charged Particle Tracking in Real-Time Using a Full-Mesh Data Delivery Architecture and Associative Memory Techniques

A Simulation System and Its Applications for Signal Readout of CMOS Pixel Sensors in High Energy Physics Experiments

Triplet Track Trigger for Future High Rate Experiments

Looking Forward: A High-Throughput Track Following Algorithm for Parallel Architectures

Graph Neural Networks for Charged Particle Tracking on FPGAs

Improved particle-flow event reconstruction with scalable neural networks for current and future particle detectors

Hybrid pixel detector development for the linear collider vertex detector

An Object Condensation Pipeline for Charged Particle Tracking at the High Luminosity LHC

3D Particle Track Reconstrution in a Single Layer Cadmium-Telluride Hybrid Active Pixel Detector

4-Dimensional Tracking with Ultra-Fast Silicon Detectors

Level-1 pixel based tracking trigger algorithm for LHC upgrade

Kalman-Filter-Based Particle Tracking on Parallel Architectures at Hadron Colliders

Pixel detector R&D for the Compact Linear Collider

Embedded FPGA Developments in 130nm and 28nm CMOS for Machine Learning in Particle Detector Readout

Beyond the CMOS sensors: the DoTPiX pixel concept and technology for the International Linear Collider A