Download - Module studies at IC - University of California, Santa …hep.ucsb.edu/people/affolder/cms1/apv/raymond_1001.pdf · higher noise for channels at detector edge (see later) October,

October, 2001 CMS Tracker Electronics 1

Module studies at IC

OUTLINE

laboratory setup description, APV I2C settingspulse shape studies (dependence on ISHA, VFS)results with β source (effect of varying det. bias)noise performance (PA resistance contribution)on-chip CM subtraction explanation for unbonded channel behaviourconclusions

emphasis on verifying APV performance and understanding any unexpected behaviourDCU not studied (yet)

Mark Raymond([email protected])


Laboratory test setup

Scintillator

CollimatedSr-90 beta source

PM

start

stop

TDC

SEQSI

trig

T1

CK/T1CK

modulemodule/UTRI setup adaptedto in-house DAQ

allows use of extensive LabViewsoftware, previously used toevaluate individual APVperformance

TDC incorporated for use with beta sourceto record interaction time w.r.t. APV samplingclock edge


APV bias settings

~50VPSP

60VFS

30VFP

34IMUXIN

55IPSP

34ISSF

80ISHA

34IPSF

52IPCASC

98IPRE

value (decimal)register

*

Values used according to most recent user manual V.2.2 (www.te.rl.ac.uk/med)

*note ISHA value larger than “rough” guide range in manual

operation at different temperatures will affectchoice of values here – module not mountedon heat sink so hybrid running warm

T gm and analogue stages speed upISHA


120

100

80

60

40

20

0

AD

C u

nits

VFS=0

VFS=120

ideal CR-RC ISHA=80,VFS=60

100

80

60

40

20

0

-20

AD

C u

nits

100806040200

3.125 ns steps

VFS=0

VFS=120

ISHA=80,VFS=60

Pulse shape dependence on VFS

Pulse shape controlled by ISHA and VFS

other bias parameters will affect shape(eg IPRE) but may end up with unreasonable power consumption

For fixed ISHA, Peak mode fall time and amplitude strongly dependent on VFS

Deconvolution mode less sensitive, only some over/undershoot


Pulse shape dependence on ISHA

For fixed VFS, Peak mode pulse shape only weakly dependent on ISHA

But Deconvolution mode amplitude quite sensitive to peak mode rise time (and consequently ISHA)

Remainder of results here use

ISHA=80, VFS=60

120

100

80

60

40

20

0

AD

C u

nits

ISHA=30

ISHA=100 ideal CR-RC ISHA=80,VFS=60

120

100

80

60

40

20

0

AD

C u

nits

100806040200

3.125 ns steps

ISHA=30

ISHA=100

ISHA=80,VFS=60


200

150

100

50

0

AD

C c

ount

s

1009080706050403020100

TDC steps [ns]

50 ns50 ns

HT=100 V200

150

100

50

0

AD

C c

ount

s

1009080706050403020100

TDC steps [ns]

50 ns50 ns

HT=250 V

Effect of detector bias voltage on pulse shape (in deconvolution)

plot single strip samples vs. TDC value for all scintillator triggers-> pulse shape for real detector signals (internal cal. gives impulse response only)

effective signal pulse shape depends on detector biaslonger charge collection time results in reduced signal amplitude and broader pulse widthmore significant when operating in deconvolution mode


Effect of detector bias voltage on signal

Beta pulse height spectrum acquired in deconvolution mode

detector depleted at ~ 100V

100-150 V over-voltagerequired for max S/N

1000

800

600

400

200

0

coun

ts

120100806040200ADC units

Deconvolution Mode 250V 200V 150V 100V


Sr-90 beta pulse height spectra

single strip spectrum acquired in Peak (Deconvolution) modedetector bias = 250 Vrms noise 2.2 (3.2) ADC units -> 930 (1500) electronshit included if > 6 (9) and neighbouring channels < 6 (9)TDC cut 10 (5) ns windowS/N ~ 27 (16.5)

1000

800

600

400

200

0

coun

ts

150100500ADC units

PEAK MODE

150100500ADC units

DECONVOLUTION


Noise performance10

8

6

4

2

0

rms A

DC

uni

ts

120100806040200

APV0 channel number

peak deconvolution

120100806040200

APV1 channel number

peak deconvolution

above pictures show raw noise – no software CM algorithm appliedsome across chip variation – PA contribution (next slide)shorted channels and shorted detector capacitors -> lower noise as expected (preamp O/Ps saturated)unbonded channels show high noise (see later)higher noise for channels at detector edge (see later)


Noise performance – calculations of pitch adapter contribution

pitch adapter shortest – longest strips RPA 0 24 60 [ohms]room temp. noise spectral density VPA 0 0.63 1.0 [nV/ Hz]relative noise contribution at preamp O/P 35 37.2 40.3% increase 0 6.3% 15%

so expect to see ~ 8 % difference (37.2 -> 40.3) between chans bonded to shortest and longest PA strips

vFETvPA

CFET

preamp

CDET

RPA

~ 1.4 nV/ Hz

~ 5pF~ 20pF

Cf

O/P noise due to preamp I/P FET

VFET*(CFET+CDET)/Cf

O/P noise due to PA resistance

VPA*CDET/Cf


5.0

4.5

4.0

3.5

3.0

2.5

2.0

1.5

1.0

0.5

0.0rm

s AD

C u

nits

120100806040200

APV0 channel number

8%

8%

peak deconvolution peak, no PA resistance decon, no PA resistance

APV0

Noise performance – PAcontribution

APV0 PA geometry -> longest line for ch0shortest for ch127

expect to see slope across chip

effect just about visible, but 8% effectnot dramatic anyway


On-chip CM subtraction

Occurs because of external resistor supplying power to preamp output inverter stage (introduced for stabilityafter 1st prototype hybrid tests)

CM signal appears on externalresistor – NOT on internalinverter output nodes

V125

V250

VSS

V250

R (external)

vIN+vCM

vCM

vOUT = -vIN

this node common to all 128 invertersin chip (other 127 have CM only)

preamp


On-chip CMsubtraction

vIN+vCM vCM

1 channel withsignal + CM 127 channels with CM only

R (external)

V250

vR

small signal model Rgm(vIN+vCM-vR)

gm(-vOUT)

127*gm(vCM-vR)

vCM vCM

vR

vOUT

vOUT

vRR = gm(vIN+vCM-vR) + 127*gm(vCM-vR)

(vIN +128 vCM) gm R1+128 gm R

vR = ≈ (vIN +128 vCM) gm R128 gm R

= vIN128

+ vCM ≈ vCM

sum currents into node vR:

currents down left hand branch: gm(vIN+vCM-vR)gm(-vOUT) =

but if vR = vCM, then: vOUT = -vIN


-100

0

10016

-100

0

10032

-100

0

10048

-100

0

10064

-100

0

10080

-100

0

10096

-100

0

100112

-100

0

100

100500

128

On-chip CM subtraction – can see effect using internal calibrate

no. of callines fired

cal signal in everychannel -> flatline=> CM rejection


Implications of on-chip CM subtractiondetector bias line noise suppressed, but only for bonded channels

=> unbonded channels show “noise” after on-chip CM subtraction (not actually noise but CM signal itself)=> should be correlation between unbonded channels

can verify by doing scatter plot of pedestal samples from one channel vs. another for many triggers(i.e. look for correlations in the noise)

-30-20-10

0102030

chan

nel 9

4

-30-20-10 0 102030channel 98

-30-20-10

0102030

chan

nel 6

0

-30-20-10 0 102030channel 61

2 normal channels 2 unbonded channels

CM effects also explain edge channel noise since edge channels see less CM signal (nothing coupling in from neighbour strips on one side)

=> anti-correlation between edge channel and unbonded channel

-30-20-10

0102030

chan

nel 0

-30-20-10 0 102030channel 94

edge channel vs. unbonded channel


1.2

0.8

0.4

0.0

Vol

ts

1.2x10-6 0.80.60.40.20.0-0.2-0.4

time

digital headeramplitude forAPVs 1 & 2

1.2x10-6 0.80.60.40.20.0-0.2-0.4

time

dig. head.amp. APV3

dig. head.amp. APV4

Strange behaviour of APV4 on this module

~ 30 % amplitude reduction of digital header for APV4similar reduction for signal amplitudenot consistent with wafer test results for the respective chipsno obvious explanation (bonding looks ok)


Conclusions

1st opportunity for us (at IC) to examine APV performance with full size CMS detectors

no nasty surprises, module performance (pulse shape, noise) appears good consistent with predictions from individual chip measurementsand consistent with detectors produced by others

unbonded channels behaviour understood in terms of on-chip CM subtractionnote: on-chip subtraction only takes care of CM occurring in or previous

to preamp

Download - Module studies at IC - University of California, Santa …hep.ucsb.edu/people/affolder/cms1/apv/raymond_1001.pdf · higher noise for channels at detector edge (see later) October,

Top Related