new the phase-1 upgrade of the atlas level-1 endcap muon trigger · 2018. 11. 22. · introduction...
TRANSCRIPT
The Phase-1 Upgrade of the ATLAS Level-1 Endcap Muon TriggerShunichi Akatsuka (Kyoto University, Japan) on behalf of the ATLAS Collaboration
RealTime 2018 @Williamsburg, 9th-15th June, 2018
Introduction◆ LHC Run 3: 2021~ 2023 ‣ √s = 14 TeV, instantaneous luminosity will be 3.0 × 1034 cm-2s-1
+50 % compared to Run 2 luminosity of ~2.0 × 1034 cm-2s-1 ‣ Integrated luminosity estimated to be ~150 fb-1 → Run 2 + Run 3 = 300 fb-1
◆ ATLAS Trigger System - basic scheme will remain the same in Run 3
◆ Phase-1 Upgrade for Level-1 Muon Trigger ‣ With Run 2 trigger algorithm, the rate for muon trigger with pT > 20 GeV (L1_MU20)
will become 30 kHz @ 3.0×1034 cm-2s-1 ‣ Run 3 requirement at this luminosity is 15 kHz (TDAQ TDR, ATLAS-TDR-023) ‣ More powerful trigger strategy is needed to reduce the trigger rate,
while keeping the trigger threshold and the efficiency
2
From Detector
Level-1 trigger
High Level Trigger
~ 40 MHz ~100 kHz, within 2.5 μs
~1 kHz, within a few sec.
Data Recorded
Fixed latency, Hardware-based trigger
L1 Muon Trigger in Run 2◆ Run 2 Level-1 muon trigger rate is dominated by:
① Triggers with no matching real muons (Fake triggers) ② Triggers by Low pT muons especially in |η| > 1 region (Endcap region).→ Strategy: Reject these triggers by introducing new “coincidence logic”
3
2− 1− 0 1 2η
0
1
2
3
4
5
6
7
8
9
10310×
entri
es /
0.06
= 13 TeVsData 2017, -1L dt = 2.9 fb∫L1 MU20 2017
rejected by Tile coincidenceexpected distribution at the end of Run 2
offline reconstructed muons 20 GeV≥
Toffline p
ATLAS Preliminary
muons with pT over threshold
①①
② ②
L1 Muon Endcap Trigger in Run 3 4
2 4 6 8 10 12 14 16 18 20
2
4
6
8
10
12 m
0
Large (odd numbered) sectors
BIL
BML
BOL
EEL
EML EOL
EIL
CSC
1 2 3 4 5 6
1
2
3
4
5
6
EIL4
0
1 2 3 4 5 6
1 2 3 4 5 6
TGCs
1
2
3
4
5
1
2
3
End-cap
magnet
RPCs
y
z
1
2
Small (even numbered) sectors
EES
EMSEOS
RPCs
y
2 4 6 8 10 12 14 16 18 20
2
4
6
8
10
12 m
0
BIS
BMS
BOS
EIS
CSC
0
TGCs
1
2
3
4
5
1
2End-cap
magnet
z
1
2
3
4
5
6
1 2 3 4 5 6
1 2 3 4 5 6 BEE7 81 2
1 2 3 4 5 6
2
1
Figure 2: Schematic side view of the ATLAS muon spectrometer depicting the naming andnumbering scheme; top: sector with large chambers; bottom: sector with small chambers.
2
NSW
2 4 6 8 10 12 14 16 18 20
2
4
6
8
10
12 m
0
Large (odd numbered) sectors
BIL
BML
BOL
EEL
EML EOL
EIL
CSC
1 2 3 4 5 6
1
2
3
4
5
6
EIL4
0
1 2 3 4 5 6
1 2 3 4 5 6
TGCs
1
2
3
4
5
1
2
3
End-cap
magnet
RPCs
y
z
1
2
Small (even numbered) sectors
EES
EMSEOS
RPCs
y
2 4 6 8 10 12 14 16 18 20
2
4
6
8
10
12 m
0
BIS
BMS
BOS
EIS
CSC
0
TGCs
1
2
3
4
5
1
2End-cap
magnet
z
1
2
3
4
5
6
1 2 3 4 5 6
1 2 3 4 5 6 BEE7 81 2
1 2 3 4 5 6
2
1
Figure 2: Schematic side view of the ATLAS muon spectrometer depicting the naming andnumbering scheme; top: sector with large chambers; bottom: sector with small chambers.
2
η=2.4
η=1.0
η=1.3
TGC Big WheelNew Small Wheel
real
Tile Calorimeter
fake
EI RPC BIS 7/8
◆ Run 3 Trigger scheme overview ‣ TGC BW hit position defines the RoI, which will be the trigger seed. ‣ TGC Big Wheel local dR/dPhi coincidence logic calculates the pT of the track. ‣ Require additional inner coincidence to reject fakes, and also to re-calculate the pT
I.P.
zdRdΦ
Φ
R
dΦ
dR
Toroidal Field
pT = ∞
muon track
M1
M2M3
z = 13
z = 14z = 14.5
z = 7
Detectors Inside the
Magnetic Field
Online hardware triggering◆ Main idea of the trigger logic: ‣ dR, dφ defined as the hit position difference between M1and M3 ‣ dR, dφ are handed over to a Look-Up-Table,
which immediately returns the trigger decision and the candidate’s pT ‣ Similar logic to take coincidence with detectors inside to reject fake/low-pT muons→ New LUT for the Inner Coincidence is needed for additional fake and low-pT rejection
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 2
1 1 1 1 1 1 1 1 1 1 1 1 2 2 2
1 1 1 1 1 1 1 1 1 1 2 2 2 2 2
1 1 1 1 1 1 1 1 1 2 2 2 2 2 2
1 1 1 1 1 1 1 1 1 2 2 2 2 2 21 1 1 1 1 1 1 1 1 2 2 2 2 2 11 1 1 1 1 1 1 1 2 2 2 3 3 1 11 1 1 1 1 1 1 1 2 2 3 3 2 1 11 1 1 1 1 1 1 3 3 3 3 3 2 2 11 1 1 1 1 1 1 3 3 3 3 3 3 1 11 1 1 1 1 1 1 3 3 4 4 4 1 1 11 1 1 1 1 1 4 4 5 5 5 4 1 1 11 1 1 1 1 4 5 5 5 6 5 4 1 1 11 1 1 1 1 1 5 5 6 6 5 5 3 1 11 1 1 1 1 6 6 6 6 6 6 3 1 1 11 1 1 1 1 6 6 6 6 6 6 1 1 1 11 1 1 1 6 6 6 6 6 6 6 1 1 1 11 1 1 1 6 6 6 6 6 6 6 1 1 1 11 1 1 1 6 6 6 6 6 6 6 1 1 1 11 1 1 1 6 6 6 6 6 6 6 1 1 1 11 1 1 4 5 5 5 6 6 5 1 1 1 1 11 1 1 4 4 5 5 5 5 5 1 1 1 1 11 1 1 4 3 4 4 5 5 5 1 1 1 1 11 1 1 3 4 3 4 4 5 5 1 1 1 1 11 1 1 3 3 3 3 3 3 3 1 1 1 1 1
1 1 1 2 3 3 3 3 3 3 1 1 1 1 1
1 1 1 2 2 2 2 2 2 1 1 1 1 1 1
1 1 2 2 2 2 2 2 2 1 1 1 1 1 1
1 1 1 2 2 2 2 2 2 1 1 1 1 1 1
1 1 1 1 2 2 2 2 2 2 1 1 1 1 1
φ∆
R∆
RoI 111
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
-15
-14
-13
-12
-11
-10
-9-8-7
-6
-5-4
-3-2
-1
012
34
567
8
910
11
12
13
14
15
RoI 111
example of Look-Up Table
What sort of LUTs can effectively
reduce fake/low-pT ?
5
Inner Coincidence Logic (1)◆ Position matching ‣ Matching between TGC BW and Inner station hit position. ‣ Fake triggers can be rejected dramatically ‣ Low-pT muons can also be rejected (with enough granularity at inner station)
6
7 m 13 m 14.5 m
Toroidal Magnetic Field
pT = ∞
NSW
TGC BW
M1
M3
High pT
calorimeter
Big Wheel dR has coarser granularity
NSW dη has finer granularity
Low pT
η�(
NSW
)�-
η (
BW
) NSW BW
dΦ dΦ
dR
Inner Coincidence Logic (2)◆ Angle matching ‣ Angle information at the inner station can be used to
further reject the low-pT triggers. ‣ Combine position and angle to tag muons with large incident angle
7
7 m 13 m 14.5 m
Low pT
pT = ∞
NSW
TGC BW
M1
M3
High pT
multiple scattering
calorimeter
coarser granularity
dη cannot distinguish
combining dη and dθ can distinguish
IP ~ 10 cm
Toroidal Magnetic Field
dθ
straight line from detector center
muon track
dθ definition
NSW
Trigger performance studies (1)◆ Fake rejection power estimation ‣ Fakes cannot be modeled in MC, so we used 2017 data for the estimation. ‣ MDT segments are used to emulate NSW, by smearing its position and angles
resolution to NSW resolution for the Level-1 trigger ‣ ~90% fake triggers are rejected with η- and (rough) phi- position coincidence
◆ Low-pT rejection power estimation ‣ Single muon MC is used to estimate low-pT rejection for NSW and RPC BIS7/8. ‣ Low-pT candidates are rejected effectively: -50% @10 GeV, -85% @5 GeV
8
[GeV]muonT
Offline p0 5 10 15 20 25 30 35 40
L1_M
U20
can
dida
te
02000400060008000
1000012000140001600018000
Phase I upgrade study = 13 TeV, 25 nssData
| < 2.4RoIη1.3 < |
ATLAS Preliminary
Run-2 (BW + FI))φ:dηBW + NSW(d
)θ:dη & dφ:dηBW + NSW(d
2− 1− 0 1 2η
0
1
2
3
4
5
6
7
8
9
10310×
entri
es /
0.06
= 13 TeVsData 2017, -1L dt = 2.9 fb∫L1 MU20 2017rejected by Tile coincidencerejected by RPC BIS 7/8 coincidence (estimation)rejected by NSW coincidence (estimation)expected distribution in Run 3offline reconstructed muons
20 GeV≥ T
offline p
ATLAS Preliminary
Trigger performance studies (2)◆ Fake rejection power estimation ‣ Fakes cannot be modeled in MC, so we used 2017 data for the estimation. ‣ MDT segments are used to emulate NSW, by smearing its position and angles
resolution to NSW resolution for the Level-1 trigger ‣ ~90% fake triggers are rejected with η- and (rough) phi- position coincidence
◆ Low-pT rejection power estimation ‣ Single muon MC is used to estimate low-pT rejection for NSW and RPC BIS7/8. ‣ Low-pT candidates are rejected effectively: -50% @10 GeV, -85% @5 GeV
◆ Rate reduction and efficiency ‣ Rate reduction
Run 2 extrapolation: 30 kHz Run 3 14.2 kHz (~53% rate reduction)
‣ Efficiency~ 95% for muons with pT > 20 GeV,relative to Run2 trigger(assuming NSW segment finding efficiency 97%)
9
2− 1− 0 1 2η
0
1
2
3
4
5
6
7
8
9
10310×
entri
es /
0.06
= 13 TeVsData 2017, -1L dt = 2.9 fb∫L1 MU20 2017rejected by Tile coincidencerejected by RPC BIS 7/8 coincidence (estimation)rejected by NSW coincidence (estimation)expected distribution in Run 3offline reconstructed muons
20 GeV≥ T
offline p
ATLAS Preliminary
L1 Muon Trigger implementationTGC Big Wheel New Small WheelBW1 BW2 BW3
New Trigger Processor
ηNSW : 8 bitsφNSW : 6 bitsdθ : 5 bits
flag, spare : 5 bits
Same as Run 2; 202-bit /25 ns ~8 Gbps
( ηBW, φBW )
muon
IP
( dR, dφ )
Trigger data to Central
Trigger Processor
vector information
10
dθ( ηNSW, φNSW )
max. 24 track candidates/ 25 ns
576 bit/ 25ns, ~23 Gbps
New Sector Logic
Tile, TGC EI, RPC BIS 7/8
Other detectors:
New Sector Logic Board design 11
New Sector Logic Board design 12
Optical interfaces to use GTX transceiver* at 6.4 Gbps
12 lanes × 6.4 Gbps = 76.8 Gbps for NSW and Tile, BIS7/8 **
Optical interfaces for G-Link connection at 0.8 Gbps 14 × 0.8 Gbps = 11.2 Gbps
for TGC BW and EI
795 × 36 Kb Block RAM → × 20 times the resource compared to the FPGA for the current trigger board
* GTX: multi-gigabit transceivers for Xilinx Kintex-7 FPGAs [Link] ** For Tile, we actually use 1.6 Gbps × 2 lanes
Xilinx Kintex-7 XC7K410T
Firmware design for Fixed Latency 13
optical fibers
sync. to LHC clock, but with different phase fiber-by-fiber
Firmware design for Fixed Latency 14
optical fibers
sync. to LHC clock, but with different phase fiber-by-fiber
LHC 40 MHz clock
Trigger decision
Align to same clock domain
◆ Trigger decision part ‣ Maximum 16 inner station candidates will be given for 1 BW trigger seed ‣ Regardless of the number of candidates, we need the trigger decision in
a fixed (and small) latency. What can we do? -> Try taking coincidence with all 16 candidates, choose the best one among them
‣ In a simple implementation, this requires 16 times the latency, or 16 same LUTs, which is not realistic in terms of latency/resource on FPGA.-> Use a faster clock to re-use the same LUT while keeping the latency small
‣
Firmware design for Fixed Latency 15
BW trigger seed pos. & pT
NSW candidates’ pos. & angle
0 to max. 16 candidates
LUT pT re-calc. selector
LUT pT re-calc. selector
All within 2 × 40 MHz clock 40 MHz320 MHz
Final pT
8-to-1 selection 2-to-1Schematic diagram
Test results◆ Firmware logic test for trigger part ‣ Simulation test for trigger part, on Vivado software (Xilinx compiler, Link) ‣ Shown below is an example of very simple test, where: ‣ 1 LUT is implemented to process 8 candidates one-by-one, on 320 MHz clock. ‣ The LUT simply returns least-significant 3 bits of the input address.
‣ Other tests are successfully done with different input patterns,not only on simulation but also on the actual SL board.
‣ Tests using more realistic LUTs (created from MC data), is ongoing.
16
input
40 MHz
320 MHz
Test results◆ Firmware logic test for trigger part ‣ Simulation test for trigger part, on Vivado software (Xilinx compiler, Link) ‣ Shown below is an example of very simple test, where: ‣ 1 LUT is implemented to process 8 candidates one-by-one, on 320 MHz clock. ‣ The LUT simply returns least-significant 3 bits of the input address.
‣ Other tests are successfully done with different input patterns,not only on simulation but also on the actual SL board.
‣ Tests using more realistic LUTs (created from MC data), is ongoing.
17
processing 8 NSW candidates one-by-one
Final pT Back to 40 MHz domain
40 MHz
320 MHzoutput
40 MHz
Summary◆ Upgrade of the muon trigger system is essential for Run 3: ‣ The main strategy is to take coincidence with NSW and other detectors,
to reject fake and low pT muons. ‣ New hardware is needed to combine data from current trigger chamber BW, NSW,
and several other detectors.
◆ Trigger Logic and Performance ‣ Taking position matching and angle matching between BW and NSW can reject low
pT muon candidates effectively. ‣ The estimated rate is 14.2 kHz @ L = 3.0×1034 cm-2s-1,
which meets the Run3 requirement of 15 kHz. (ATLAS-TDR-023)
◆ Hardware and Firmware development ‣ New trigger processor board, New Sector Logic, has been produced for Run3. ‣ Firmware is fully-designed with all the trigger LUTs implemented. ‣ Fast clock is used in the trigger logic to overcome the latency limitation
while keeping the FPGA resource usage reasonable.
18
backup slides
Physics Motivation◆ Run 3 trigger rate estimation
‣ Without the phase-1 upgrade, to keep the trigger rate to the require level, the pT threshold will need to be raised to ~40 GeV.
20
◆ Physics Acceptance
◆ If the threshold is raised to 40 GeV,the efficiency for muons from the decays of W boson produced in association with Higgs will be 61%.
threshold [GeV]T
p5 10 15 20 25 30 35 40 45
Muo
n le
vel-1
trig
ger r
ate
[Hz]
410
510
610= 14 TeVs -1s-2 cm34 10×L=3
Extrapolationwith 2012-run settingExtrapolationwith Pre-phase1Extrapolationwith Pre-phase1 + NSW
ATLAS
15 kHz
New Small Wheel◆ Consists of sTGC and Micromegas ‣ sTGC: small strip TGC ‣ TGC chamber with strip width of 3.2mm, smaller than the strip width of current TGC (> 15 mm)
‣ 4 wire-strip pairs are combined to make 1 module. ‣ position resolution 60~150 μm
‣ Micromegas: micro mesh gaseous structure ‣ position resolution ~90 μm ‣ 8 layers are sandwiched by sTGC 4-layer modules,to compose the New Small Wheel
21
sTGC 4 layers
Micromegas 8 layers
sTGC 4 layers
~22 cm~36 cm
Resolution: position ~30 μm angle ~0.3 mrad.
NSW Technical Design Report, ATLAS-TDR-020
Region of Interest◆ The smallest unit for the Level-1 Muon Trigger: ‣ Each side is divided into 72 parts, shown as green line in this figure → 72 ‘Sectors’ per side, 48 × Endcap sectors and 24 × Forward sectors
‣ Each Endcap (Forward) sector is divided into 148 (64) ‘Regions of Interest’ = RoI
‣ One Endcap (Forward) SL board handles 2 Sectors, i.e. 296 RoIs (128 RoIs)
‣ Trigger decision is performed RoI by RoI → 296 trigger decision logic should run in parallel on a single FPGA → 296 different LUTs need be implemented
22
Trigg
er S
ecto
r
2.5
2.0
1.5
η = 1.0
4 4 4 RoI = 16 x 4 = 6416
RoI = 37 x 4 = 148
RoIΔφ = 0.1
44 4 4 4
4
37
coverage per Endcap SL
coverage per Forward SL
New Sector Logic Board design 23
FPGA (Xilinx Kintex-7 XCK410T)
CPLD (XC2C256-7PQ208C) for VME control
BPI (PC28F256P30TF) for FPGA configuration
14 optical inputs (800 Mbps) from BW-TGC SFP RX + G-Link RX chip
RJ45 connector for readout (SiTCP)
LEMO IN/OUT
16-pin connector to receive trigger/ timing information
6 optical inputs for other detectors SFP+ with GTX RX
6 optical inputs (6.4 Gbps) from NSW SFP+ with GTX RX in FPGA
2 optical outputs (6.4 Gbps) to MUCTPI SFP+ with GTX TX
Input Data Format (1)◆ TGC BW ‣ G-LINK 16 or 17 bits/BC per fiber ‣ Endcap: 12 fibers per Sector Logic, 202 bits/BC ‣ Forward: 6 fibers per Sector Logic, 100 bits/BC
◆ TGC EI ‣ G-LINK 16 bits/BC per fiber ‣ Endcap only, 2 fibers per SL, 32 bits/BC
◆ New Small Wheel ‣ GTX 6.4 Gbps (= 128 bit/BC, with 8B/10B) per fiber ‣ Endcap: 6 fibers per SL, 768 bit/BC ‣ Forward: 8 fibers per SL, 1024 bit/BC
24
) (-452-452-452-452-452-452-452-452
31762 3 489 3
65013 6501365013 6501365013 65013
65013 6501365013 6501365013 65013
17550 . 17550 .
ATLAS Phase-I UpgradeLevel-1 Endcap Muon Trigger
Full Design ReportDraft 0.5, March 7, 2018 21:59
Data format, connection and protocol Sector Logic receives both position (⌘, �) and angle272
information (d✓) of the track from the NSW trigger processor. These information are calculated273
separately on sTGCs and MicroMegas local trigger processor, and are merged by the NSW274
trigger processors. The expected resolution, range, and the assigned bit width for each of the275
variables are summarized in Table 2. In addition to these variables, NSW trigger processor are276
expected to send some flag bits for sTGC and MM track qualities. The track information arepacked into 24 bits, as shown in Figure 3.277
Table 2: Resolution and bit assignment
variable resolution range for 1 NSW sector range/ resolution assigned bit width⌘ 0.005 1.2(1.3 ⇠ 2.5) 240 8 bits (256 patterns)� 10 mrad. 2⇡/12(2⇡/24) rad. 52.3 (26.2) 6 bits (64 patterns)d✓ 1 mrad. 32 mrad. (�15 ⇠ 15 mrad.) 32 5-bit (32 patterns)
Table 3: Bit assignment for one NSW track. Track information are encoded into 24 bits.
Bit Assigned information7-0 ⌘[7 : 0]13-8 �[5 : 0]
18-14 �✓[4 : 0]20-19 MM type22-21 sTGC type
23 spare
The data transfer from NSW trigger processor to the Sector Logic is performed throughoptical serial connections. The Xilinx GTX transceiver, which is one of Multi-Gigabit Transceiver278
(MGT) technology, is adopted as the serial-parallel interface. GTX enables the data transfer279
rate of maximum 12.5 Gb/s [8]. The most popular SFP+ optics are directly supported by the280
transceiver. The optical links will be operated with bit rate of 6.4 Gb/s, synchronous with the281
LHC bunch crossing clock provided from CTP via TTC systems. 8b/10b encoding are used for282
the data transfer, therefore 128-bit data per fibre can be transferred every BC. The data formatis shown in Figure 9. Sector Logic can receive 4 track candidate per fibre, every BC.283
) (-452-452-452-452-452-452-452-452
31762 3 489 3
65013 6501365013 6501365013 65013
65013 6501365013 6501365013 65013
17550 . 17550 .
Figure 9: Data format for data from NSW per BC, received using 1 fibre. 4 track candidate with24-bit information can be received.
In order to cover boundaries of NSW sectors by a Sector Logic, as shown in Figure 8, track
10 2 Physics Algorithm
Input Data Format (2)◆ Tile Cal. ‣ GTX 1.6 Gbps (= 32 bits/BC with 8B/10B) per fiber ‣ Endcap only, 1 fiber per SL
◆ RPC BIS 7/8 ‣ GTX 6.4 Gbps (= 128 bits/BC with 8B/10B) per fiber ‣ Endcap only, 1 fiber per SL
25
( ) (H[] &)H[] &*
78>9K,3)L) A[ ,K+3)L
9 G .NL 0D KL .NL
FF 3 ( A[ +K+3)L A[ *K+3)L A[ )K+3)L 8NOXRK*3)L
GA97K,3)L
) (563563563563563563563563
76124 076124 0
31762 3 489 3
17550 . 17550 .
76124 076124 076124 076124 076124 0
76124 076124 076124 076124 076124 0
0- 0
ATLAS Phase-I UpgradeLevel-1 Endcap Muon Trigger
Full Design ReportDraft 0.5, March 7, 2018 21:59
) (563563563563563563563563
76124 076124 0
31762 3 489 3
17550 . 17550 .
76124 076124 076124 076124 076124 0
76124 076124 076124 076124 076124 0
0- 0
Figure 14: Format for data from RPC BIS 7/8. 6.4 Gbps GTX.
Table 7: Bit assignment for RPC BIS 7/8 trac candidate. ⌘, � are the track position, and d⌘, d�are the track segment angle. 2/3 flag shows whether the track candidate passed 3-out-of-3 or2-out-of-3 locacl coincidence. 2 bits are assigned to this flag to show which 2 layers (layers 1-2,1-3 or 2-3) are used to calculate the position/angle of the track.
Bit Assigned information5-0 ⌘[5 : 0]11-6 �[5 : 0]
14-12 d⌘[2 : 0]17-15 d�[2 : 0]19-18 2/3 flag[1:0]23-20 spare
Secotr Logic will be connected to MUCTPI.343
Data format, connection and protocol Data format to MUCTPI board is well described in344
a document at MUCTPI PDR [9]. The number of muon candidate per sector will increase in345
Phase-I upgrade of MUCTPI. For each LHC bunch crossing, every Sector Logic will send four346
muon candidate information to MUCTPI. A specificpT value indicates if it is an empty candidate347
to be ignored by MUCTPI. The muon candidate will be ordered in terms of the pT-threshold348
value, the first muon candidate being the one with the highest pT-threshold value. 6.4 Gbps349
GTX protocol will be used for connection between the Sector Logic and the MUCTPI. The dataformat is shown in Figure 15, and the content of each track information is shown in Table 8.350
6BC6BC6BC6BC6BC6BC (6BC )6BC
10 54 234 54
4 B .9 9D 7 (4 B .9 9D 7 (4 B .9 9D 7 (
.5.7 F0 3
4 B .9 9D 7 (1 B:9 9 7 -.2 7
F-. 3 ( F.( ( ) F.( ( ) F.( ( )
Figure 15: Format for data to MUCTPI, per trigger sector. Up to four candidates will be sent,each muon candidate will have 16-bit data width.
16 2 Physics Algorithm
Output Data Format◆ Output to Muon-to-CTP Interface (MUCPTI) ‣ GTX 6.4 Gbps (= 128 bits/BC with 8B/10B) per fiber ‣ 1 fiber per trigger sector, 2 fibers per SL board ‣ Commas are sent at the end to gain latency.
26
6BC6BC6BC6BC6BC6BC (6BC )6BC
10 54 234 54
4 B .9 9D 7 (4 B .9 9D 7 (4 B .9 9D 7 (
.5.7 F0 3
4 B .9 9D 7 (1 B:9 9 7 -.2 7
F-. 3 ( F.( ( ) F.( ( ) F.( ( ) ATLAS Phase-I Upgrade
Level-1 Endcap Muon TriggerFull Design Report
Draft 0.5, March 7, 2018 21:59
Table 8: Muon candidate data format. 16 bits are assigned for each muon candidate. 8 bits areused to indicate the RoI of the candidate, 4 bits for the pT value, and 4 bits for other flags. 1 bitof the flag bits are used to indicate the sign of the muon candidate’s charge. The other flag bitscan be used for inner coincidence debug signals, during the commissioning period.
Bit Assigned Data7-0 RoI
11-8 pT value12 Candidate sign
15-13 NSW/Inner Coincidence flags
2.3 Algorithms in the Sector Logic351
In this section, the algorithm that will be implemented on the new SL is explained. A brief352
block diagram of the endcap Sector Logic trigger algorithm is shown in Figure 16. The Sector353
Logic algorithm can be divided into several sub-modules. The Decoder part receives the data354
and translates the data into position/ track information, and provide them to the downstream355
logics. TGC Big-Wheel coincidence part receives the dR and d� information of the TGC BW, to356
take dR - d� coincidence. Inner coincidence part receives the information of the inside-magnet357
detectors such as NSW, to take BW - Inner coincidence. The Track Selection part selects up358
to 4 track candidates, and the output data is sent to the MUCTPI. The implementation of thetrigger logic on the firmware will be discussed in detail at Section 4.359
2.3.1 Decoder360
The Decoder module retrieves �R and �� information from the TGC BW signal received viaG-LINK.361
The Decoder module converts raw data of tracks into the format of BW-Inner Coincidenceand ditribute to each logic.362
2.3.2 TGC Big Wheel coincidence363
The Sector Logic uses LUTs for the R–� coincidence of TGC Big wheel, where the �R in-364
formation with �� information from the HPT boards is combined to get pT information of the365
muon track. To implement LUTs in FPGA embedded memory (Block RAM; BRAM), one sector366
is divided into small blocks called Sub-Sector-Cluster (SSC). Figure 17 shows one SSC which367
consists of 8 ROIs (2 rows of wire by 4 columns of strips). Thanks to the specification of HPT368
board coincidence, one SSC includes a maximum of only one hit in R and two hits in �. Hit369
information is 19 bits in total and is used as the LUT address. The output of the TGC-BW R–�coincidence logic will be the pT of the track candidates in 4-bit resolution.370
2.3.3 Inner coincidence371
The main idea of the inner coincidence is simple: when the TGC Big Wheel finds a trigger372
candidate, confirm the decision by requiring hit in detectors inside the magnetic field. The373
simplest way to implement this algorithm is to require hit at approriate position in the inner374
detector (position matching). As shown in Fig. 18, the position matching algorithm will not only375
2.3 Algorithms in the Sector Logic 17
[GeV]muonT
Offline p0 5 10 15 20 25 30 35 40
Rel
ativ
e ef
ficie
ncy
0
0.2
0.4
0.6
0.8
1
Run-2 (BW + FI))φ:dηBW + NSW(d
)θ:dη & dφ:dηBW + NSW(d
Phase I upgrade study| < 2.4RoIη1.3 < |
ATLAS Simulation Preliminary
‣ New Inner LUTs uses NSW position & angle information ‣ Efficiency is calculated by simulation, for L1_MU20(L1_MU20: Level-1 trigger for muon with pT > 20 GeV)
‣ The track finding efficiency is assumed to be 97%
Trigger Performance 27
Efficiency > 95% for pT > 20 GeVSharper curve
acquired with new LUT
◆ Data reception Clock-Domain-Crossing part ‣ choose a phrase of 40 MHz to latch the received data, so that they will have enough
margin from the data transition point:
Firmware design for Fixed Latency 28
128-bit word
received data (phase unknown)
LHC 40 MHz
+ π/2
+ π
+ 3π/2use this phase to latch data
data transition
quite a lot of margin
less margin
Run 3 Readout system
SROD on PC‣Event Building ‣Formatting for ROS ‣Error handling
10GbE Switch‣ N-to-1 connection
‣Hit information ‣event ID
TTC Fan-out board
‣event ID ‣Trigger Information
Read Out System
S-LINK Card
GbE 10 GbE
GbE
(Trigger, Timing & Control)
New Sector Logic
TTC
SiTCP
SiTCP
SiTCP
TTCrq
… × 12
29
SiTCP[*] : Technology to connect FPGA to the Ethernet, and establish connection via TCP/IP
[*] Tomohisa Uchida, “Hardware-Based TCP Processor for Gigabit Ethernet”, IEEE Trans. Nucl. Sci. Vol.55, No.3, June 2008, [LINK]