Dynamic GNN inference exhibits strong capability to model interactions over time, such as complex particle collision events in High Energy Physics (HEP) experiments at the High Luminosity Large Hadron Collider (HL-LHC). With a much larger scale of collision data captured in future HEP experiments, and limitations in both offline compute capacity and storage, revamped trigger systems require FPGAs to run ultra-low-latency machine-learning models with low power consumption for online filtering of useful events. Many state-of-the-art GNN accelerators rely on static graph structures, but this assumption breaks down in HL-LHC trigger systems and other edge-based dynamic GNN applications where edge embeddings can change in place based on neighbour node embeddings during runtime. We propose DGNNFlow, a novel streaming dataflow architecture for real-time edge-based dynamic GNN inference, with three key contributions: a hardware enhancement for edge-embedding dynamic computation; a Node Embedding Broadcast that alleviates data dependencies in the dynamic dataflow; and input dynamic graph construction for full support of graphs without pre-defined edge embeddings. Deployed on an AMD Alveo U50 FPGA at 200 MHz, DGNNFlow achieves 2.59×–4.36× speedup over an NVIDIA RTX A6000 GPU with 3.59×–3.70× less power, and 2.29×–3.54× speedup over an Intel Xeon Gold 6226R CPU.

A search for supersymmetric scenarios leading to compressed mass spectra is presented, using events with at least two low-pT electrons or muons and missing transverse energy, in proton-proton collisions at √s=13 TeV. This analysis is based on data collected with the CMS detector at the CERN LHC and corresponds to a total integrated luminosity of up to 139 fb−1. The analysis expands similar past searches through novel reconstruction techniques, enabling the use of electrons with pT as low as 1 GeV. The results are interpreted in terms of electroweakino pair production with a small mass difference between the produced supersymmetric particles (˜χ02, ˜χ±1) and the lightest neutralino (˜χ01). Two simplified models are considered: a wino-bino model and a higgsino model. Exclusion limits at 95% confidence level are set on ˜χ02 masses up to 250(120) GeV at a mass difference of 3(1) GeV in the wino-bino model, and up to 170(100) GeV at a mass difference of 3(1) GeV in the higgsino model.

A combination of the results of several searches for the electroweak production of the supersymmetric partners of standard model bosons, and of charged leptons, is presented. All searches use proton-proton collision data at √s = 13 TeV recorded with the CMS detector at the LHC in 2016-2018, corresponding to an integrated luminosity of up to 137 fb⁻¹. The results are interpreted in terms of simplified models of supersymmetry. Two new interpretations are added with this combination: a model spectrum with the bino as the lightest supersymmetric particle together with mass-degenerate higgsinos decaying to the bino and a standard model boson, and the compressed-spectrum region of a previously studied slepton pair-production model. Improved analysis techniques are employed to optimise sensitivity for the compressed spectra in the wino and slepton pair-production models. The results are consistent with expectations from the standard model. The combination provides a more comprehensive coverage of the model parameter space than the individual searches, extending the exclusion by up to 125 GeV, and also targets some of the intermediate gaps in the mass coverage.

A search for new physics beyond the Standard Model with compressed mass-spectra is performed, motivated by natural solutions to the hierarchy problem or the observed Dark Matter relic abundance. Compressed mass-spectra scenarios are particularly challenging to probe as they are characterized by final states with only low visible and missing energy. These signatures are hard to distinguish from the overwhelming background of Standard Model processes in the LHC environment, and therefore could have escaped experimental observation thus far. The analysis targets final states with multiple low-momentum prompt or displaced leptons and missing transverse energy, relying on 138 fb^-1 of proton-proton collision data at a center-of-mass energy of √s = 13 TeV, collected by the CMS experiment during LHC Run 2. The results of the search are interpreted in terms of a simplified wino-bino model of R-parity conserving Supersymmetry. The search sets exclusion limits at 95% confidence level as function of the lifetime and masses of the new particle states. For promptly decaying winos, the signal exclusion extends to mass-splittings as low as 900 MeV for a wino mass of 100 GeV. This result hereby closes the sensitivity gap at mass-splittings between 0.9 and 2 GeV, implying that LHC searches now exceed the mass-bounds for compressed electroweakinos from LEP experiments over the full range of the mass-splitting. For long-lived scenarios the maximum exclusion is obtained for a N2 lifetime of cτ = 100 mm, mass = 400 GeV and a mass-splitting = 20 GeV. The search categories targeting final states with prompt leptons are furthermore included in the legacy Run 2 combination of searches for electroweak Supersymmetry, covering the production of charginos, neutralinos and sleptons. The combined search expands the excluded parameter space by as much as 125 GeV, depending on the model, with respect to the most sensitive component search and also covers some of the intermediate sensitivity gaps between mass-compressed and uncompressed signal scenarios. Finally, to ensure feasibility of future physics analyses even under the harsh data-taking conditions of the High Luminosity LHC era, a new electron reconstruction and identification algorithm for the Phase-2 upgrade of the CMS Level-1 Trigger is developed. The new approach increases the electron selection efficiency by up to 10% with respect to the baseline algorithm, making use of machine learning techniques, tracks reconstructed at the 40 MHz collision rate, high granularity Calorimeter information and state-of-the-art FPGAs. It furthermore paves the way for the implementation of identification algorithms for low-momentum electrons that could help expand even further the analysis sensitivity at the HL-LHC.

In preparation for the High Luminosity LHC (HL-LHC) run, the CMS collaboration is working on an ambitious upgrade project for the first stage of its online selection system: the Level-1 Trigger. The upgraded system will use powerful FPGA processors connected by a high-bandwidth network of optical fibers. The new system will access highly granular calorimeter information and online tracking: their combination for identifying physics objects is pivotal to coping with the harsh HL-LHC environment without compromising on physics acceptance. In particular, track matching is exploited to identify calorimeter deposits originating from electron particles, reducing the background rate and thus allowing trigger thresholds as low as those used for electroweak physics during the LHC Run-2. Furthermore, tracks are used to compute pile-up resilient isolation discriminators. This DPS illustrates the current status and performance of the electron identification algorithms, focusing in particular on new approaches w.r.t what was presented in the Level-1 Phase-2 TDR.

The High-Luminosity LHC (HL-LHC) will open an unprecedented window on the weak-scale nature of the universe, providing high-precision measurements of the standard model as well as searches for new physics beyond the standard model. The CMS experiment is planning to replace entirely its trigger and data acquisition system to achieve this ambitious physics program. Efficiently collecting those data sets will be a challenging task, given the harsh environment of 200 proton-proton interactions per LHC bunch crossing. The new Level-1 trigger architecture for HL-LHC will improve performance with respect to Phase-1 through the addition of tracking information and subdetector upgrades leading to higher granularity and precision timing information. In this work, we present a multitude of trigger algorithms for the upgraded Phase-2 trigger system, which benefit from the finer information to reconstruct optimally the physics objects. Dedicated pile-up mitigation techniques are implemented for lepton isolation, particle jets and missing transverse energy to keep the rate under control. The expected performance of the new trigger algorithms will be presented, based on simulated collision data of the HL-LHC. The selection techniques used to trigger efficiently on benchmark analyses will be presented, along with the strategies employed to guarantee efficient triggering for new resonances and other new physics signals.

Supersymmetric models are characterized by a strong diversity of experimental signatures. Since general-purpose searches have not yet given any clear indication of new physics, dedicated methodologies and tools have been developed to target the regions of the parameter space where the analysis is most challenging and SUSY might still lie undetected. This contribution will describe relevant examples among searches performed by the CMS Collaboration using the full dataset of proton-proton collisions collected during the Run 2 of the LHC at a center-of-mass energy of 13 TeV.

A search for supersymmetry in events with two or three low-momentum leptons and missing transverse momentum is performed. The search uses proton-proton collisions at √s = 13 TeV collected in the three-year period 2016-2018 by the CMS experiment at the LHC and corresponding to an integrated luminosity of up to 137 fb⁻¹. The data are found to be in agreement with expectations from standard model processes. The results are interpreted in terms of electroweakino and top squark pair production with a small mass difference between the produced supersymmetric particles and the lightest neutralino. For the electroweakino interpretation, two simplified models are used, a wino-bino model and a higgsino model. Exclusion limits at 95% confidence level are set on χ̃⁰₂/χ̃±₁ masses up to 275 GeV for a mass difference of 10 GeV in the wino-bino case, and up to 205 (150) GeV for a mass difference of 7.5 (3) GeV in the higgsino case. In the top squark interpretation, exclusion limits are set at top squark masses up to 540 GeV for four-body decays and up to 480 GeV for chargino-mediated decays.