+ All Categories
Home > Documents > The Track-Finding Processor for the Level-1 Trigger of the CMS Endcap Muon System D.Acosta,...

The Track-Finding Processor for the Level-1 Trigger of the CMS Endcap Muon System D.Acosta,...

Date post: 27-Dec-2015
Category:
Upload: jonah-baldwin
View: 214 times
Download: 0 times
Share this document with a friend
17
The Track-Finding Processor for the Level-1 Trigger of the CMS Endcap Muon System D.Acosta, A.Madorsky, B.Scurlock, S.M.Wang University of Florida A.Atamanchuk, V.Golovtsov, B.Razmyslovich St. Petersberg Nuclear Physics Institute
Transcript

The Track-Finding Processor for the Level-1 Trigger of the CMS

Endcap Muon System

D.Acosta, A.Madorsky, B.Scurlock, S.M.Wang

University of Florida

A.Atamanchuk, V.Golovtsov, B.Razmyslovich

St. Petersberg Nuclear Physics Institute

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 2

Muon Track-Finding

• Perform 3D track-finding from trigger primitives

• Measure PT , , and

• Transmit highest PT candidates to Global Level-1

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 3

Track-Finder Requirements

• High efficiency with low PT threshold

• Single muon trigger rate < few kHz at L=1034 cm-2s-1

– Should have large safety factor

• PT resolution 20%

– Require , information from 3 muon stations

• Multi-muon capability 3 muons per 60° azimuthal sector

– Best 4 muons overall sent to Global Level-1 Trigger

• Pipelined and deadtime-less– 40 MHz B.X. frequency

• Minimal latency – < 16 B.X. (400 ns)

• Programmable– FPGA and RAM implementation

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 4

Trigger Regions in

Cathode Strip Chamber System

Drift-Tube systemOverlap 1.2 > > 0.9

• 4 muon stations in each endcap• Each station contains 6 layers of CSC chambers

• PT measured from fringe field of 4 T solenoid using 3D track segments in each station

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 5

Trigger Regions in

ME1/3

MB2/2

MB2/1

Illustration of overlap region

Track-Finding performed in independent 60° sectors

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 6

Why Two Muon Track-Finders?

• The Barrel Track-Finder:– intrinsically 2D

• road-finding in only (expect low rates)

• uniform magnetic field in central region

– large number of neighbor interconnections

• chambers are staggered, non-projective

• The Endcap Track-Finder:– intrinsically 3D

• road-finding in and to reduce backgrounds

• non-uniform magnetic field in endcap

– No interconnections between trigger sectors

• chambers are projective in • small bending in endcap

• Therefore, different needs in each region imply two different designs

• Must ensure that fake double triggers do not occur in region of overlap

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 7

Level-1 Trigger Scheme

Strip FE cards

Wire FE cards

Motherboard

TMB

Port Card

PC

Sector Receiver Sector Processor

OPTICAL

SR SP

CSC Track-Finder

CSC Muon Sorter

Global Trigger

DTRPC

FE

FE

Global L1

2 / chamber 3 / port card 3 / sector

4

4

4 4

LCT

LCT

Strip LCT card

Wire LCT card

On chamber On periphery

In counting house36

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 8

Track-Finder Architecture

• Track-Finder implemented as 12 Sector Processors

• Each Sector Processor: – Implemented on a 9U VME card

– Processes 15 CSC segments and 8 DT segments

– Identifies 3 muons per 60°

• CSC data received by 3 Sector Receiver cards– Each receives 6 track segments on optical links

– Reformats data to , b ,

– Applies alignment corrections

– Communicates to Sector Processors via custom point-to-point backplane

– Presently under development at UCLA

• DT data sent to transition board at back of crate

• Custom point-to-point backplane:– Delivers ~600 bits every 25 ns (3 GB/s)

– Operates at 280 MHz to reduce connections:

• National Channel Link 28:4 serialization

– Presently being prototyped in Florida

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 9

Sector Processor Layout

P1

CCB

SR4 st.

SR1,3 st.

SR1,2 st.

SRBarrel

Ch

ann

el L

inks

2 B

unch

Cro

ssin

g A

nal

yzer

VMEInterface

ControlLogic

Ext

rap

olat

ion

Un

its

Glo

bal

Bu

ffer

(F

IFO

s)

FinalSelection

Unit

TA

U1

(ove

rlap

)

Pt-

assi

gnm

ent

Un

its

(FP

GA

s)

OutputData

Storage

Ch

ann

el L

inks

FPGADownload

Logic

Control Logic(Clock distribution, SRAM read/write

and other devices)

LE

DD

rive

rs

CustomBackplane

TA

U2

(en

dcap

)

Pt-

assi

gnm

ent

Un

its(

LU

Ts)

TransitionModule

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 10

Sector Processor Logic

• Latch input and hold for possibly more than one B.X.– Allows for timing errors from trigger primitives

• Perform all possible station-to-station extrapolations in parallel– Simultaneously search roads in and

• Assemble 3- and 4-station tracks from 2-station extrapolations

• Cancel redundant short tracks if track is 3 or 4 stations in length

• Select the three best candidates

• Calculate PT , , and send to CSC muon sorter

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 11

Two Bunch Crossing Mode

• Input data can be latched for 2 B.X. to accommodate timing errors from trigger primitives

• Sector Processor still reports trigger at correct B.X.

Title:(dblecnt.eps)Creator:Adobe Illustrator(R) 8.0Preview:This EPS picture was not savedwith a preview included in it.Comment:This EPS picture will print to aPostScript printer, but not toother types of printers.

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 12

Extrapolation Unit

FF3

FF3

FF6

LUT64x2

LUT64x1

SB

LUT64x1

LUT64x7low

LUT64x7med

LUT64x7high

SB

FF5

ABS

FF5

FF5

FF5

FF1

FF6

FF10

FF10

FF1

FF2

FF7

FF1

FF10

FF7

FF7

FF7

FF1

LUT32x5

1min

LUT32x5

1max

LUT32x5

min

LUT32x5

max

FF5

FF5

NOTAND

FF1

FF1

Amb1

Amb2

1

2

1

2

1

2

Q1

Q2

LUT128x1

GlobalClock

SB

1

SB

1

SB

2

SB

2

LUT8x1

SBh

SBm

SBl

enable

LUT32x2

enable

6 -inputAND

3 -inputAND

FF2

ExtrapolationQuality

FF1

FF1

FF1

FF1

FF1

FF1

FF1

FF1

FF1

FF1

FF1

FF1

FF2

road finder

Course Pt assign

road finder

Z

1Bx 1Bx 1Bx

Xilinx Virtex FPGA XCV150

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 13

Track Assembly Procedure

Title:(streams.eps)Creator:Adobe Illustrator(R) 8.0Preview:This EPS picture was not savedwith a preview included in it.Comment:This EPS picture will print to aPostScript printer, but not toother types of printers.

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 14

Track Assembler Unit

SRAM256Kx16

IDT

LINK33

LINK32

LINK31

Fro

m E

xtra

pol

atio

n U

nits

12

3

3

12

3

3

12

3

3

ME33 – ME4

ME33 – ME2

ME33 – ME1

ME32 – ME4

ME32 – ME2

ME32 – ME1

ME31 – ME4

ME31 – ME2

ME31 – ME1

To

Fin

al S

elec

tion

Uni

t

9

6

9

6

9

6

6 bit Ranking &9 bit hit i.d. :

LINK23

LINK22

LINK21

12

3

3

12

3

3

12

3

3

ME23 – ME4

ME23 – ME3

ME23 – ME1

ME22 – ME4

ME22 – ME3

ME22 – ME1

ME21 – ME4

ME21 – ME3

ME21 – ME1

9

6

9

6

6

9

LINK23

LINK22

LINK21

4

6

4

6

4

6

ME23 – ME1

ME23 – MB2

ME23 – MB1

ME22 – ME1

ME22 – MB2

ME22 – MB1

ME21 – ME1

ME21 – MB2

ME21 – MB1

9

6

9

6

6

9

8

8

8

Extrapolation output small enough to address SRAM for quick decision

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 15

The Final Selection Unit A Sorter with Cancellation Logic

“Final Selection Unit”FPGA XCV50 (XILINX)

Fro

m T

rack

Ass

emb

ler

Un

itSt

ream

1 –

ME

2 ke

y st

atio

nSt

ream

2 –

ME

2 ke

y st

atio

nSt

ream

3 –

ME

3 ke

y st

atio

n

Ext

rapo

lati

ons

Qua

lity

Com

para

tors

(36

Uni

ts)

Hit

Num

ber

Com

para

tors

(36

Uni

ts) FinalDecision Unit

9

9

9

9

9

9

6

6

6

6

6

6

36

36

13

13

13

Hit

Num

ber

Part

Ext

rapo

latio

n Q

ualit

y Pa

rt

17 bits:1st stub number – 5 bits;2nd stub number – 5 bits;3rd stub number – 5 bits;2-stn or 3-stn mode – 1 bit;overlap or endcap track – 1 bit.

4 4 4

Track 1

Track 3

Track 2

Track 6

Track 5

Track 4

Track 1

Track 3

Track 2

Track 6

Track 5

Track 4

9

9

9

ConverterEnable

ConverterEnable

ConverterEnable

4

4

4

9

9

9

Stre

am 2

Station ME2

Station MB1Station MB2Station ME1

3

446

Fro

m 2

Bx

An

alyz

erB

unch

cro

ssin

g ID

1

1

1

To

Ass

ignm

ent

Uni

t (F

PG

As)

4 bits:1st stub Bx ID – 1 bit;2nd stub Bx ID – 1 bit;3rd stub Bx ID – 1 bit;4th stub Bx ID – 1 bit.

MUX

Sel1 Sel2 Sel3

MODE

MUX

Select

MUX

Select

MUX

Select

AND

AND

AND

1

1

1

1

1

1

1

1

1

1

1

1

Converter

Converter

Converter

17

17

17 To

Ass

ignm

ent

Uni

t (F

PG

As)

To prohibitDouble Count

To select3 best tracks

out of 9

Track 7

Track 8

Track 9

56

6

6

9

9

9

9

9

9

Track 7

Track 8

Track 9

Stre

am 3

Stre

am 1

Stre

am 2

Stre

am 3

Stre

am 1

Station ME3 3Station ME4 3

0 – no 1 – a Match bit unset (only for 3-stn)2 – 1-2-33 – 1-2-44 – 1-3-45 – 2-3-46 – 1-27 – 1-38 – 2-39 – 2-410 – 3-4

11 – MB1-MB2-ME212 – MB1-ME1-ME213 – MB2-ME1-ME214 – MB1-ME215 – MB2-ME2

VHDL code written

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 16

Assignment Unit

• Determines , , PT of the selected 3 best muons

• PT assignment uses , measurements from 2 or 3 stations PT/PT ~ 30% with only 2 stations

PT/PT ~ 20% with 3 stations improves Level-1 rate reduction

• Implemented with FPGA preprocessing followed by large SRAM look-up table

2Mb16 SRAM

FPGA

LEB99 Workshop, September 21, 1999

D. Acosta, University of Florida 17

Summary of CSC Track-Finder

• Conceptual design complete• 12 Sector Processors cover CSC and CSC/DT

overlap– 1.0 < < 2.4 and = 60° on one board

• Track-finding algorithms are three-dimensional– Improves background suppression

• PT assignment includes , measurements from 3 stations PT/PT ~ 20% (30% with only 2 stations)

– Significantly improves rate reduction at Level-1

• Inputs can be latched for 2 B.X.– Tolerates timing errors from trigger primitives

• Latency expected to be only 14 B.X.• Fully re-programmable• Xilinx Virtex FPGAs and SRAM used• Board layout and backplane design started


Recommended