Emulated L1Track resolution of the d0 track parameter as a function of the truth track η for single muons with pT= 5 GeV. The ”Offline” scenario (black markers) uses unmodified offline tracks. L1 track reconstruction is emulated starting from offline tracks and dropping hits in pre-defined ITk layers. The selection of the hits is done independently per sub-system and it favors hits in the inner most part of the ITk. In addition, a minimum of two pixel hits is requested, whenever possible. The hits are then refitted using only eight ITk layers. According to which ITk layers are used, two scenarios are defined: ”TDR” (red open markers) and ”Strip only” (blue markers). The TDR scenario includes pixel layer 4, ring layers 2-3 and inclined layers 2-3. In the strip only scenario, no pixel layers are used. It represents the worst case scenario where no information from the pixel detector is available for L1Track. These results were derived using Step 3 geometry (ATLAS-P2-ITK-17-00-01). An additional factor of 2 is applied to account for differences between online and offline track reconstruction algorithms.

Emulated L1Track resolution of the z0 track parameter as a function of the truth track η for single muons with pT= 5 GeV. The ”Offline” scenario (black markers) uses unmodified offline tracks. L1 track reconstruction is emulated starting from offline tracks and dropping hits in pre-defined ITk layers. The selection of the hits is done independently per sub-system and it favors hits in the inner most part of the ITk. In addition, a minimum of two pixel hits is requested, whenever possible. The hits are then refitted using only eight ITk layers. According to which ITk layers are used, two scenarios are defined: ”TDR” (red open markers) and ”Strip only” (blue markers). The TDR scenario includes pixel layer 4, ring layers 2-3 and inclined layers 2-3. In the strip only scenario, no pixel layers are used. It represents the worst case scenario where no information from the pixel detector is available for L1Track. These results were derived using Step 3 geometry (ATLAS-P2-ITK-17-00-01). An additional factor of 2 is applied to account for differences between online and offline track reconstruction algorithms.

Background rejection versus Z→μμ signal efficiency for events with at least 2 muons with |η| < 2.5, using emulated L1Track tracks. For background, Pythia dijet JZ0-2 samples are used. Signal and background samples were simulated with √s = 14TeV and <μ> = 200. The discriminant variable is the difference between the z0 of the tracks matched to the muons. Events are required to have at least 2 offline muons, reconstructed using information from the muons spectrometer alone. The 2 leading muons are matched to the leading track within a cone of ∆R < 0.1. For signal, events are required to have 2 truth-matched muons from the Z boson with offline pT > 10 GeV, in order to quantify signal efficiency with respect to offline selection. For background, the Level-0 (L0) trigger efficiency is emulated by applying an event weight based on the efficiency for a 10 GeV pT muon trigger, in order to quantify background rejection with respect to the L0 acceptance. The ”Offline” scenario uses unmodified offline tracks with pT(trk) > 1 GeV, whilst the ”TDR” and ”Strip-only” scenarios use offline tracks with pT(trk) > 4 GeV and for which the z0 track parameter resolution has been smeared according to the relevant emulated L1Track scenario resolutions multiplied by an additional smearing factor of 2. The added factor is to account for the differences between online and offline track reconstruction algorithms. In addition, 5% of tracks are randomly dropped to simulate a 95% reconstruction efficiency with respect to offline track reconstruction. These results are used to quantify the relative performances for an L1Track-like system only, a sthey are based on a manipulation of offline reconstructed tracks rather than a realistically simulated system.

Light-flavour jet rejection versus b-jet efficiency for the IP3D [1] (dashed lines) and DIPS [2] (solid lines) b-tagging algorithms trained on jets from t ̄t events with pT > 40 GeV and |η| < 2.5, using emulated L1Track tracks. Signal and background samples were simulated with √s = 14 TeV and <μ> = 200. DIPS is a neural network based algorithm which outperforms the IP3D tagger by utilizing extra track-based kinematic inputs and by exploiting correlations between tracks. This version of DIPS, unlike the one described in [2], also uses the jet kinematics as inputs to the network. Both taggers have been modified to not use track hit content knowledge, as a realistic hit content has not been simulated for L1Track-like tracks. The ”Offline” scenario uses unmodified offline tracks with pT(trk) > 1 GeV, whilst the ”TDR”and ”Strip-only” scenarios use offline tracks with pT(trk )> 4GeV and for which the impact parameter (IP) resolutions have been smeared according to the relevant emulated L1Track scenario resolutions multiplied by an additional smearing factor of 2. The added factor is to account for the differences between online and offline track reconstruction algorithms. In addition, 5% of tracks are randomly dropped to simulate a 95% reconstruction efficiency with respect to offline track reconstruction.These results are used to quantify the relative performances for an L1Track-like system only, as they are based on a manipulation of offline reconstructed tracks rather than a realistically simulated system. Calibrated offline jets and offline reconstructed primary vertices were used and the contribution from fake tracks for the online system was neglected in this study. For the DIPS algorithm, the performance based on offline tracks is not shown since the necessary samples were no longer available.

Background rejection versus HH→4b signal efficiency for events with at least 3 jets with pT > 65 GeV and |η| < 3, using emulated L1Track tracks. For background, Pythia dijet JZ0-2 samples are used. Signal and background samples were simulated with √s = 14 TeV and <μ> = 200. The discriminant variable is the output of a BDT that combines ten variables based on z0, d0 and track multiplicity inside jets. The z0 and d0 jet parameters are computed from the pT weighted sum of tracks in the jets. The ”Offline” scenario uses unmodified offline tracks with pT(trk) > 1 GeV, whilst the ”TDR”and ”Strip-only” scenarios use offline tracks with pT(trk) > 4 GeV and for which the impact parameter (IP) resolutions have been smeared according to the relevant emulated L1Track scenario resolutions multiplied by an additional smearing factor of 2. The added factor is to account for differences between online and offline track reconstruction algorithms. In addition, 5% of tracks are randomly dropped to emulate a 95% track reconstruction efficiency with respect to offline. These results are used to quantify the relative performances for an L1Track-like system only, as they are based on a manipulation of offline reconstructed tracks rather than a realistically simulated system. Calibrated offline jets were used and the contribution from fake tracks for the online system was neglected in this study.

The impact of different curvature resolutions on the RoI-to-truth matching for muons from L1_MU20 RoIs, showing the maximum p of the matching true muon after smearing of the curvature variable, q/p by values shown in the figure.

The fraction of |η| <1.3 L1_MU20 RoIs that remain after matching to a truth muon with various increasingly tight matching requirements; The “All RoIs” bin shows the fraction remaining (1.0) with no matching requirement as a reference; The “TruthMatch” bin shows the fraction after a match to a true muon with any pT; The “TruthMatch pT> 15” bin shows the fraction after a match to a true muon with pT > 15 GeV; and the “SmTruMatch > 15” shows the fraction after a match to true muon with pT > 15 GeV after smearing the muon curvature, q/pT by the resolutions shown in the figure.

The arrival time at the end-of-stave of the final R3 packet following an R3 request, for barrel hybrid 0 in the innermost ITK Strip Tracker layer. The L0 and L1 accept rates are 500 kHz and 200 kHZ with 10% R3 detector occupancy and 160 Mbps bandwidth from the HCC. In the simulation there are 10 ABC chips, arranged in 2 daisy chains, each of 5 chips attached the the HCC. R3 data is prioritised on the HCC. The separation between the peaks is determined by the time taken to transfer packets between adjacent chips in each daisy chain.

The time required to read out all the R3 data packets for 95% of all R3 requests as a function of the Level 1 accept rate for the discrete event simulation of the Phase II ITK Strip Tracker. The different curves in each of the two groups correspond to all the hybrids in the inner most Strip Tracker Barrel layer. The Level 0 accept rate is 500 kHz and the Regional occupancy is 10%. The bandwidth from the HCC is 160 Mbps. In the simulation there are 10 chips per hybrid in 2 daisy chains of 5 chips. Shown are the latencies both with, and without prioritisation of the R3 data on the HCC.

The time required to read out all the R3 data packets for 95% of all R3 requests as a function of the Level 1 accept rate for the discrete event simulation of the Phase II ITK Strip Tacker for all the hybrids in the Endcap petal furthest from the interaction point. The Level 0 accept rate is 500 kHz and the Regional occupancy is 10%, the bandwidth from the HCC is 160 Mbps. The number of chips per hybrid varies with the hybrid number - hybrid 6 has 12 chips. Shown are the latencies including prioritisation of the R3 data on the HCC with the solid lines corresponding to the latencies using a FIFO with a maximum depth of 32 packets to receive from each daisy chain, the dotted lines are for a FIFO with unlimited depth.

The time required to read out all the R3 data packets for 95% of all R3 requests as a function both of the Level 1 accept rate and the R3 rate (occupancy×L0 rate) for the discrete event simulation of the Phase II ITK Strip Tracker for the highest occupancy hybrid in barrel layer 0. In the simulation, the bandwidth from the HCC is 160 Mbps and the number of chips attached to the hybrid is 10, in 2 daisy chains of 5 chips. The latencies including prioritisation of the R3 data on the HCC. For reference, the dotted lines represent the baseline 200 kHz L1 rate and 500kHz × 10% occupancy = 50 kHz R3 rate.

The time required to read out all the R3 data packets for 95% of all R3 requests as a function both of the Level 1 accept rate and the R3 rate (occupancy×L0 rate) for the discrete event simulation of the Phase II ITK Strip Tracker for hybrid 6 in the endcap petal furthest from the interaction point. In the simulation, the bandwidth from the HCC is 160 Mbps and the number of chips attached to the hybrid is 12, in 2 daisy chains of 6 chips. The latencies including prioritisation of the R3 data on the HCC. For reference, the dotted lines represent the baseline 200 kHz L1 rate and 500kHz × 10% occupancy = 50 kHz R3 rate.

The time required to read out all the R3 data packets for 95% of all R3 requests as a function both of the Level 1 accept rate and the R3 rate (occupancy×L0 rate) for the discrete event simulation of the Phase II ITK Strip tracker for hybrid 6 in the endcap petal furthest from the interaction point. In the simulation, the bandwidth from the HCC is 320 Mbps and the number of chips attached to the hybrid is 12, in 4 daisy chains of 3 chips. The latencies including prioritisation of the R3 data on the HCC. For reference, the dotted lines represent the baseline 200 kHz L1 rate and 500kHz × 10% occupancy = 50 kHz R3 rate.

Detector map showing the latency within which 95% of L0-Priority requests for regional detector readout after an L0 request can be completed. The full system delivers a rate of 1MHz of full detector data of which 10% are L0-Priority requests corresponding to a Regional Readout Request (R3) .The data from the R3 requests will be processed by the L1 Track system. The latencies are estimated with the L1Track discrete event simulation. The readout bandwidth from the Hybrid Chip Controllers (HCC) for each hybrid is 320 Mbps. The data format for the cluster data within the system is the same for both L0 and L0-Priority requests. The chip hit occupancies correspond to a mean inclusive pileup interaction multiplicity, <μINCL> of 196 interactions per bunch crossing, the upper limit which is expected with a bunch separation of 25 ns at an instantaneous luminosity 7× 10^34 cm-2 s-1.

Detector map showing the latency within which 99% of L0-Priority requests for regional detector readout after an L0 request can be completed. The full system delivers a rate of 1MHz of full detector data of which 10% are L0-Priority requests corresponding to a Regional Readout Request (R3) .The data from the R3 requests will be processed by the L1 Track system. The latencies are estimated with the L1Track discrete event simulation. The readout bandwidth from the Hybrid Chip Controllers (HCC) for each hybrid is 320 Mbps. The data format for the cluster data within the system is the same for both L0 and L0-Priority requests. The chip hit occupancies correspond to a mean inclusive pileup interaction multiplicity, <μINCL> of 196 interactions per bunch crossing, the upper limit which is expected with a bunch separation of 25 ns at an instantaneous luminosity 7× 10^34 cm-2 s-1.

Detector map showing the latency within which 95% of non-prioritised requests for full detector readout, at either L0 or L1, can be completed. The full system delivers a rate of 1MHz of full detector data of which 10% are L0-Priority requests corresponding to a Regional Readout Request (R3). The data from the R3 requests will be processed by the L1 Track system. The latencies are estimated with the L1Track discrete event simulation. The readout bandwidth from the Hybrid Chip Controllers (HCC) for each hybrid is 320 Mbps. The data format for the cluster data within the system is the same for both L0 and L0-Priority requests. The chip hit occupancies correspond to a mean inclusive pileup interaction multiplicity, <μINCL> of 196 interactions per bunch crossing, the upper limit which is expected with a bunch separation of 25 ns at an instantaneous luminosity 7× 10^34 cm-2 s-1.

Detector map showing the latency within which 99% of non-prioritised requests for full detector readout, at either L0 or L1, can be completed. The full system delivers a rate of 1MHz of full detector data of which 10% are L0-Priority requests corresponding to a Regional Readout Request (R3) .The data from the R3 requests will be processed by the L1 Track system. The latencies are estimated with the L1Track discrete event simulation. The readout bandwidth from the Hybrid Chip Controllers (HCC) for each hybrid is 320 Mbps. The data format for the cluster data within the system is the same for both L0 and L0-Priority requests. The chip hit occupancies correspond to a mean inclusive pileup interaction multiplicity, <μINCL> of 196 interactions per bunch crossing, the upper limit which is expected with a bunch separation of 25 ns at an instantaneous luminosity 7× 10^34 cm-2 s-1.