elementary particle physics seminar, oxford, 6 th november 2007sonja hillert (oxford) p. 0 ilc...

Elementary Particle Physics Seminar, Oxford, 6th November 2007 Sonja Hillert (Oxford) p. 1

ILC vertex detector R&D:

optimising the detector design for physics

Sonja Hillert (Oxford)

Elementary Particle Physics Seminar

Oxford, 6 November 2007


Outline of this talk

Role of the vertex detector for extracting ILC physics

Measuring quark charge: vertex charge and charge dipole procedure

Sensitivity of vertex charge reconstruction to detector design: fast MC results

LCFI Vertex Package: software tools for full MC studies

Optimising the vertex detector design

Recent results of LCFI vertex detector R&D on sensors & mechanical support


Typical event processing at the ILC

reconstruction

of tracks, CAL-cells

energy flow objects

first order

jet finding

b-jets

c-jets

uds-jets

gluon-jets

tune track-jet

association

for tracks

from SV or TV

contained in

neighbouring

jet

associate with parent jet in some cases

classify B

as charged

or neutral

classify D

as charged

or neutral

charged B

charged D

neutral B

neutral D

charge dipole,

protons, charged

kaons or leptons

from SV, TV

charged kaons

or leptons

b

bbar

b

c

cbar

c

cbar

bbar

flavour

identification


Dependence of physics reach on detector performance

Flavour tag needed for event selection and reduction of combinatoric backgrounds

Quark charge sign determination used for measurement of ALR,

angular correlations ( top polarisation) – vertex detector performance crucial

Examples:

• Higgs branching ratios:

classical example of a process

relying on flavour tag

• e+e- ZHH:

4 b-jets in final state requiring

excellent tagging performance;

could profit from quark charge

sign selection

M. Battaglia


Processes requiring quark sign selection: e+e- b bbar

e+e- bb: indirect sensitivity to new physics, such as extra spatial dimensions, leptoquarks,

Z´, R-parity violating scalar particles (Riemann, LC-TH-2001-007, Hewett PRL 82 (1999) 4765);

quark charge sign selection to large cos needed to unfold cross section and measure ALR:

without quark sign selection with perfect quark sign selection

Sensitivity to deviations of extra-dimensions model from SM predicition (S. Riemann):


b

e+ e-

t

t

W

W+b

e.g. cq’ q’

s

qqe.g. s c

a typical e+e- t t event

e+e- tt demanding for vertex detector:

• multijet event: final state likely to include soft jets

some of which at large polar angle

• flavour tag needed to reconstruct the W bosons and

top-quarks

• quark charge sign selection will help to reduce

combinatoric backgrounds

• top decays before it can hadronise: polarisation of top quark

can be measured from polarisation of its decay products;

best measured from angular distribution of s-jet (quark charge)

Processes requiring quark sign selection: e+e- t tbar


Requirements

To measure quark charge efficiently one needs:

an excellent vertex detector:

• pixel-based system

• few micron point resolution (< 5 m)

• small inner layer radius (~ 15 mm)

• good polar angle coverage

• low mass support structure (<= 0.1 % X0)

• mechanical stability

appropriate high-level reconstruction software, e.g.:

• topological vertex finding

• flavour tagging

• vertex charge reconstruction (charged hadrons)

• charge dipole reconstruction (neutral B hadrons)


Vertex charge reconstruction b-jets contain a complex decay chain, from which the charge has to be found

in the 40% of cases where b quark hadronises to charged B-hadron,

quark sign can be determined by vertex charge

need to find all stable tracks from

B decay chain:

• define seed axis

• cut on L/D (normalised distance

between IP and projection of track POCA

onto seed axis)

• tracks that form vertices other than IP

are assigned regardless of their L/D

need vertex finding as prerequisite (definition of seed axis)

in most analyses, only calculate charge for jet of specific flavour: need flavour tagging

probability of mis-reconstructing vertex charge is small for both charged and neutral cases


Charge dipole procedureSLD, NIM A 447 (2000) 90-99 For some neutral vertices, quark charge can be obtained from

the charge dipole formed by B- and D-decay vertex.

“ghost track” vertexing algorithm (aka ZVKIN)

developed at SLD, was shown to yield higher purity for

charge dipole than standard ZVTOP code (ZVRES, cf p12)

advantage: one-prong vertices identified by vertex finder

increased efficiency, especially at short B decay lengths

at ILC, charge dipole procedure still to be explored


Performance of charge reconstruction: leakage rates

define leakage rate 0 as probability of reconstructing neutral hadron as charged

performance strongly depends on low momentum tracks:

largest sensitivity to detector design for low jet energy, large cos

preliminary results from fast MC study (SGV)


250 mm

15 mm

60

mm

100 mm

8 mm 25 mm

e+e- BEAM

Using vertex charge for detector optimisation

preliminary results from fast MC study (SGV)

Using fast MC SGV, studied dependence of leakage rate on vertex detector design:

• compared detectors with different inner layer radii ( beam pipe radii)

• varied amount of material per detector layer (factor 4 compared to baseline)

translated into integrated luminosity required to obtain physics results of same significance

for processes requiring independent quark charge measurement in 2 jets,

increase of beam pipe from 15 to 25 mm has sizeable effect (factor 1.5 – 2)


The LCFI Vertex Package

The LCFIVertex package provides, in a full MC and reconstruction framework:

• vertex finder ZVTOP with branches ZVRES and ZVKIN (new in ILC environment)

• flavour tagging based on neural net approach (algorithm: R. Hawkings, LC-PHSM-2000-021);

includes full neural net package; flexible to allow change of inputs, network architecture

• quark charge determination, currently only for jets with a charged ‘heavy flavour hadron’

first version of the code released end of April 2007:

code, default flavour tag networks and documentation available from the ILC software portal

http://www-flc.desy.de/ilcsoft/ilcsoftware/LCFIVertex

next version planned to be released this Friday:

• minor corrections, e.g. to vertex charge algorithm; further documentation

• diagnostic features to check inputs and outputs

• new vertex fitter based on Kalman filter to improve run-time performance


ZVTOP vertex finder, Pt-corrected mass

two branches: ZVRES and ZVKIN (already mentioned when discussing charge dipole)

The ZVRES algorithm (D. Jackson, NIM A 388 (1997) 247)

very general algorithm that can cope with arbitrary multi-prong decay topologies

• ‘vertex function‘ calculated from Gaussian ´probability tubes´ representing tracks

• iteratively search 3D-space for maxima of this function and minimise 2 of vertex fit

central for flavour tag:

pT-corrected vertex mass

• this kinematic correction is applied

to partly account for undetected

neutral particles


Flavour tagging approach

Vertex package provides flavour tag procedure developed by R. Hawkings et al

(LC-PHSM-2000-021) as default

number of vertices found determines which

NN input variables are used:

• if secondary vertex found: MPt , momentum

of secondary vertex, and its decay length and

decay length significance

• if only primary vertex found: momentum and

impact parameter significance in R- and z for the

two most-significant tracks in the jet

• in both cases: joint probability in R- and z (estimator of

probability for all tracks to originate from primary vertex)

flexible: permits user to change input variables, architecture and training algorithm of NN

b

c

uds


Flavour tagging performance

open: SGV fast MC

full: MARLIN, SGV-input

cc (b-bkgr)

b

Identical input events

open: SGV fast MC

full: MARLIN, SGV-input

Identical input events

c

c (b-bkgr)b

bc (b-bkgr)

c

open: BRAHMS, LC-note

full: MARLIN (Mokka), Si-only track cheater

open: BRAHMS, LC-note

full: MARLIN (Mokka), Si-only track cheater

b

c (b-bkgr)

c

Z-peak

Z-peak

500 GeV

500 GeV


Diagnostic features plan to make available inputs and outputs for ZVRES & flavour tag (later: vertex charge)

nearly complete: LCFIAIDAPlot – module for flavour tag diagnostics based on AIDA

• input and output variables of the flavour tag neural nets, separately for b-, c-, light jets

• graphs of purity vs efficiency and flavour leakage rates (i.e. efficiencies of wrong flavours)

vs efficiency separately for the 1-, 2- and 3-vertex case

• JAS3-macro to plot these easily

• optionally: raw numbers of jets vs NN-output, AIDA tuple with flavour tag inputs written out

example:

inputs to

flavour tag


New vertex fitter: Kalman filter

Motivation: improve run time performance by replacing the “space-holder”

Least-Squares-Minimisation (LSM) fitter of first release

Kalman filter code by S. Gorbunov, I. Kisel interfaced to Vertex Package

successfully tested:

find same flavour tagging performance

as with LSM-fitter

resulting improvement in run time performance:

overall run time of Vertex Package is

reduced to ~ 25% of the 1st-release value


Towards a realistic simulation

Current simulations are based on many approximations / oversimplifications.

The resulting error on performance is at present unknown and could be sizable,

especially when looking at particular regions in jet energy, polar angle (forward region!)

Issues to improve:

• Vertex detector model: replace model with cylindrical layers by model with barrel staves

• GEANT4: switched off photon conversions for time being (straightforward to correct)

• hit reconstruction: using simple Gaussian smearing at present; realistic code exists only

for DEPFET sensor technology, not for CPCCDs and ISIS sensors developed by LCFI

• track selection:

• KS and decay tracks suppressed using MC information

• tracks from hadronic interactions in the detector material discarded using MC info –

only works for detector model LDC01Sc (used for code validation) at present

• current default parameters of the code optimised with fast MC or old BRAHMS (GEANT3) code

• default flavour tag networks were trained with fast MC


open: hadronic interaction effects

full: no hadronic interactions

b

c (b-bkgr)

track cheater (04/07)

500 GeV

Examples of impact of simplifications

open: photon conversions, uncorrected

full: photon conversions off

c

c (b-bkgr)

b

track cheater, Z-peak

c

c

b

c (b-bkgr)

full LDCtracking

Z-peak

open: VXD geometry with ladders

full: VXD geometry with cylinders

effects of simplifications can be sizeable;

note: photon conversions and hadronic

interactions in detector material can

efficiently be corrected for

currently making initial checks needed for

implementing these corrections


Areas of relevance for wider user community:

integration into ALCPG software framework org.lcsim: drivers under development in the US,

to be released as soon as possible (N. Graf)

consistent IP treatment, based on per-event-fit in z and on average over N events in R

Vertexing:

• explore use of ZVKIN branch of ZVTOP for flavour tag and quark charge determination:

• optimise parameters

• study performance at the Z-peak and at sqrt(s) = 500 GeV

• explore how best to combine output with that of ZVRES branch for flavour tag

• use charge dipole procedure (based on ZVKIN) to study quark charge determination for

(subset of) neutral hadrons

Further development of the Vertex Package


Areas of relevance for wider user community contd:

Flavour tagging: explore ways to improve the tagging algorithm, e.g. through use of

• different input variables and/or different set-up of neural nets that combine these

• improvements to MPt calculation using calorimeter information, e.g. from high-energy 0

• vary network architecture (number of layers & nodes, node transfer function), training algorithm

• explore new “data mining” and classification approaches (e.g. decision trees, … )

Vertex charge reconstruction:

• revisit reconstruction algorithm using full MC and reconstruction (optimised with fast MC)

Functionality specifically needed for vertex detector optimisation:

Correction procedure for misalignment of the detector and of the sensors will need to be

developed, adapted or interfaced (see optimisation of the detector)

Improvements and extensions


Optimising the detector design

Simulations serve to estimate performance of

• benchmark quantities: impact parameter resolution, flavour tag, vertex charge reconstruction

• reconstruction of physics quantities obtained from study of benchmark physics processes

Vertex detector-related software cannot be developed in isolation:

• vertexing, flavour tag, vertex charge rec´n performed on a jet-by-jet basis (depends on jet finder)

• strong dependence on quality of input tracks (i.e. hit and track reconstruction software)

• physics processes to optimise calorimeter also depend on tagging performance (e.g. ZHH)

Study of benchmark physics processes

• performed in close collaboration with the two main ILC detector concept study groups:

SiD and ILD (formerly GLD / LDC)

• ILCSC has issued call for Letters of Intent, to be submitted 1 October 2008

• These will be followed by more detailed Engineering Design Studies to be completed by 2010


Benchmark Physics Studies Benchmark physics processes should be typical of ILC physics and sensitive to detector design.

A Physics Benchmark Panel comprising ILC theorists and experimentalists has published

a list of recommended processes that will form the baseline for the selection of processes

to be studied in the LoI- and engineering design phases.

Following processes were highlighted as most relevant by the experts (hep-ex/0603010):

particularly

sensitive to

vertex detector

design


Parameters and aspects of design to be optimised The Vertex Package, embedded into full MC and reconstruction frameworks,

permits the following aspects of the vertex detector design to be optimised:

• Beam pipe radius

• Sensor thickness, material amount at the ends of the barrel staves

• Material amount and type of mechanical support (e.g. RVC, Silicon carbide foams)

• Overlap of sensors: linked to sensor alignment, tolerances for sensor positions along

the beam & perpendicular to it

• Arrangement of barrel staves

• Long barrel vs short barrel plus endcap geometry

Study of trade-offs, involving variations of more than one parameter, should be aimed at

Physics simulation results will be only one of the inputs that determine the detector

design – the more decisive input may well be provided by what is technically feasible.


LCFI pursuing development of two sensor technologies for ILC vertex detector:

Column Parallel CCD (CPCCD)

• CPC1 (first CPCCD)

• CPC2 – second generation, large area device

In-situ Storage Image Sensor (ISIS)

• proof-of-principle device (ISIS1)

• effort towards ISIS2

CPC1 CPC2 ISIS1

LCFI sensor development: introduction


“Classic CCD”Readout time NM/fout

N

M

N

Column Parallel CCDReadout time = N/fout

M

Main sensor technology developed by LCFI

Every column has its own amplifier and ADC

Readout time shortened by ~3 orders of magnitude compared to “classic” CCD

All of the image area is clocked, complicated by the large gate capacitance

Optimised for low voltage clocks to reduce power dissipation

Column Parallel CCD: principle


ISIS1Busline-free CPC2

104 mm

CPC2-10

CPC2-40

CPC2-70

CPC2 devices

Three device sizes: 10, 40 and 70 mm length (requirement for inner layer device: 100 mm)

Some devices designed to reach high speed (up to 50 MHz operation):

2-level metallisation permits using whole image area as distributed bus line

Charge to voltage conversion can happen on CCD (50% channels) or on readout chip (other 50%)


10 MHz System noise = 75 e- RMS

20 MHz System noise = 110 e- RMS

CPC2 test results

short (CPC2-10) device, high-speed design, tested with 55Fe signal (1620 e-, MIP-like)

• X-ray hits seen up to 45 MHz: important milestone

• CPC2 works with clock amplitude down to 1.35 Vpp

• in standalone tests (w/o readout chip, using 2-stage source follower outputs):

at 10 MHz achieve noise level of 75 e- RMS (CMOS driver chip)

• tests continuing


both chips made on 0.25 μm CMOS process (IBM)

front-end amplifiers matched to the CCD outputs

additional test features in CPR2

CPR2 includes data sparsification

Bump bond pads

Wire/Bump bond pads

CPR1

CPR2

Voltage and charge amplifiers 125 channels each

Analogue test I/O

Digital test I/O

5-bit flash ADCs on 20 μm pitch

Cluster finding logic (22 kernel)

Sparse readout circuitry

FIFO

Readout chips: CPR1 and CPR2


Readout chips: results

Charge outputs

inverting (positive signals)

Voltage outputs

non-inverting (negative signals)

CPR1, 1 MHz:

Sparsified output

Cluster finding with X-ray signals

CPR2: sparsification

CPR1: in charge channels expected gain observed, constant over device;

in voltage channels gain drops from edge to the centre of the device (not understood)

CPR2: same gain drop in voltage channels as in CPR1, charge channels don’t work;

errors in sparsification for cluster distances < 60 – 100 pixels: extensive tests with measured

and simulated input led to major re-design of next generation chip CPR2A


Clock driver for CPCCD Challenge: providing 2 Vpp clocks at 50 MHz for

CPCCD (capacitance 40 nF / phase): 20 A peak current

Further requirements:

• low power dissipation

• must be close to CCD (to reduce parasitic inductance)

• must not add too much to material budget

• may have to work at low temperature (down to -100 oC)

2 approaches: transformer (fallback), custom ASIC (baseline)

performed tests with 16:1 transformer integrated into PCB (above)

and with first version of driver chip CPD1 (right):

• 1 chip drives 2 phases, up to 3.3 V clock swing

• 0.35 m CMOS process, chip size 3 x 8 mm2

• 8 independent clock sections

• careful layout on- and off-chip to cancel inductance

• bump-bondable


Integration of CPC, CPR and CPD

CPC2-40CPR2

CPD1

a lot more work needed to arrive at ladders

(design of ladder end: left)

but an integrated system of CPC, CPR and CPD

exists and works up to frequency of 9 MHz

(speed limited by CPR2)


LCFI Mechanical Studies

LCFI mechanical work comprises:

• support structure prototyping

(RVC-, SiC foam, carbon fibre, shells)

• cooling studies

• conceptual design

SiC, samples of 8% and 6% relative density obtained

Plot: deviation under temperature cycling

example design of a foam ladder (cross section)


Summary

ILC physics will require an excellent vertex detector as well as adequate reconstruction software.

Studies of the performance of vertexing, flavour tagging, quark charge reconstruction,

and studies of benchmark physics processes are used to optimise the vertex detector design.

The LCFI Vertex Package provides software tools to perform such optimisation using

GEANT4-based simulation and full reconstruction software (including e.g. pattern recognition).

Further development of these tools is needed for realistic detector assessment and comparison.

The development of CPCCD-sensors, CPR readout chips and CPD drivers within the

LCFI collaboration is far advanced. In parallel, mechanical work is progressing well.

In lab tests, CCDs were driven with amplitudes as low as 1.35 Vpp (design spec: 2 Vpp)

and signals observed up to frequencies of 45 MHz (design spec: 50 MHz)

A combined system of CPC2-CPR2-CPD1 was successfully operated up to 9 MHz.


Additional Material


The ZVTOP vertex finderD. Jackson,

NIM A 388 (1997) 247 two branches: ZVRES and ZVKIN (also known as ghost track algorithm)

The ZVRES algorithm: very general algorithm

that can cope with arbitrary multi-prong decay topologies

• ‘vertex function‘ calculated from Gaussian

´probability tubes´ representing tracks

• iteratively search 3D-space for maxima of this function

and minimise 2 of vertex fit

ZVKIN: more specialised algorithm to extend coverage to b-jets with

1-pronged vertices and / or a short-lived B-hadron not resolved from the IP

• additional kinematic information

(IP-, B-, D-decay vertex approximately

lie on a straight line) used to find

vertices

• should improve flavour tag efficiency

and determination of vertex charge

37Konstantin Stefanov, STFC Rutherford Appleton Laboratory 37Vertex Detector Review, Fermilab 2007

CPC1 did not have optimal drive conditions due to the single level metal

Novel idea from LCFI for high-speed clock propagation: “busline-free” CCD:

50 MHz achievable with suitable driver in CPC2-10 and CPC2-40 (L1 device)

Transformer or ASIC driver

Φ1 Φ2

Φ1 Φ2

Level 1 metal

Polyimide

Level 2 metal

Φ2 Φ1

Φ2 Φ1

To multiple

wire bonds

To multiple

wire bonds

1 mm

CPC2 Clock Driving



Operating principles of the ISIS:

1. Charge collected under a photogate;

2. Charge is transferred to 20-cell storage CCD in situ, 20 times during the 1 ms-long train;

3. Conversion to voltage and readout in the 200 ms-long quiet period after the train (insensitive to beam-related RF pickup);

4. 1 MHz column-parallel readout is sufficient;



RG RD OD RSEL

Column transistor

The ISIS offers significant advantages:

Easy to drive because of the low clock frequency: 20 kHz during capture, 1 MHz during readout

~100 times more radiation hard than CCDs (less charge transfers)

Very robust to beam-induced RF pickup

ISIS combines CCDs, active pixel transistors and edge electronics in one device: non-standard process

Presently discussing with 3 vendors for the ISIS2 development:

Looking at modified 0.18 μm CMOS process with additional buried channel and deep p+ implants

Proof-of-principle device (ISIS1) designed and manufactured by e2V Technologies

On-

chip

logi

c

On-

chip

sw

itche

s

Global Photogate and Transfer gate

ROW 1: CCD clocks

ROW 2: CCD clocks

ROW 3: CCD clocks

ROW 1: RSEL

Global RG, RD, OD

5 μm


Tests of ISIS1

Tests with 55Fe X-ray source:

ISIS1 without p-well tested first and works OK:

Correct charge storage in 5 time slices and consequent readout

Successfully demonstrated the principle

New ISIS1 chips with p-well have been received, now under tests

ISIS1 without p-well is now in a test beam in DESY

elementary particle physics seminar, oxford, 6 th november 2007sonja hillert (oxford) p. 0 ilc...

Documents