The Research On CMS Phase-2 L1 Fast Track Trigger Based On FPGA Retina Algorithm

With the discovery of Higgs particles,the two general-purpose detectors ATLAS and CMS on the Large Hadron Collider(LHC)have achieved their design goals.Precision Higgs studies and the extended research for new physics present strong demands for higher luminosity of the LHC,therefore physicists propose the High Luminosity LHC(HL-LHC)upgrade,leading to a peak luminosity of 5×1034 cm-2s-1,with an average of about 140 proton-proton collisions(pileup,PU)per 25 ns bunch crossing(BX).High interaction rate associated with the pileup effects will be a particular concern for the CMS detector.A few years ago CMS has proposed the Phase-2 upgrade plan to prepare during LS3(2023-2025)for the upcoming HL-LHCThe CMS Phase-2 upgrade plans to make use of the track information from the Outer Tracker at L1 trigger to maintain the L1 acceptance rate below the maximum of 750 kHz and conserve an efficient discriminating power of the event selection.The total L1 latency is limited to at most 12.5 ?,of which the track trigger budget is set at 5?s.It seems that track triggering calls for pattern recognition algorithms with real-time performance.Retina is a fast track mapping and matching algorithm inspired by the early stage of the vision mechanism in mammals.In this thesis,my R&D work applies Retina algorithm to the CMS Outer Tracker for the first time and develops a FPGA-based Retina track finder to perform pattern recognition of charged particles in the magnetic field.The main purpose is to explore the feasibility of the highly parallel Retina algorithm application for the L1 track finding in CMS and provide guidance and evaluation for the scale of the tracking system.The detailed research and innovations are as follows:Firstly,to investigate the optimal solution of Retina algorithm using a simple ideal forward detector model without magnetic field,two full FPGA-based approaches are implemented:one of which uses a mixed Floating-Point and Fixed-Point and the other employing Fixed-Point and Look-Up Table(LUT),by comparing the hardware cost,latency and tracking performance.The ideal detector model is made of eight parallel planes with a typical pixel size of 15×15 ?m2 perpendicular to the z-axis plane in the space without magnetic field.The Retina parameter space contains 44944 cells with a granularity of 350?m.Both mixed Floating-Point and Fixed-Point,Fixed-Point and LUT solutions are successfully implemented on KC705.Two implementations can provide a track parameter resolution of 99 ?m.Since the Fixed-Point and LUT solution obtains considerable gains in resource(4 times)and latency usage(2 times),the Fixed-Point and LUT solution is a more advantageous choice.Secondly,we propose a Pre-judge method for for the CMS Outer Tracker detector in magnetic field,design and implement the Retina track finder of which the effectiveness and real-time performance are verified by single particle events.The so-called Pre-judge solution enables the Retina track finder to dynamically target regions of interest in the track parameter space,avoid storing and processing all the parameter space cells.This solution can reduce the computation scale by 16 times,improve the real-time performance of the algorithm and meet the track trigger latency.For a sub-sector of the detector barrel region,the Retina track finder is implemented on KC705 and has been tested with single particle Monte Carlo events.The results show that the Retina track finder can meet the trigger latency requirements.For the charged particles with 2<pT<8 GeV/c,the relative PT resolution is better than 10%and the?0 resolution is between 0.47 crad and 0.5 crad,which meets the expected goal.Thirdly,we implement the Retina track finder to perform pattern recognition for t t events.By optimizing the Pre-judge location and the threshold selection,the Retina track finder can process multiple tracks in parallel,ensuring high efficiency and high purity of t t events.For the same sub-sector of the detector barrel region,the Retina track finder is implemented on KC705 and has been tested with t t Monte Carlo events(PU140-200).The results show that the relative pT resolution is better than 10%with 2<PT<10GeV/c,and the ?0 resolution is between 0.52 crad and 0.55 crad for pT>2 GeV/c.The tracking efficiency can reach 96.08%,and the purity reaches 97,59%.The Retina track finder successfully perform pattern recognition for the CMS Outer Tracker detector with good track parameter resolution,purity and efficiency.Therefore,the FPGA-based Retina implementation solution in this paper can be used for CMS Phase-2 L1 fast track trigger.