ARTICLE IN PRESS

0168-9002/$ - se

doi:10.1016/j.ni

E-mail addr

Nuclear Instruments and Methods in Physics Research A 579 (2007) 718–722

www.elsevier.com/locate/nima

Noise performance of the D0 layer 0 silicon detector

M. Johnson, For the D0 Collaboration

Fermilab, MS 357, P.O. Box 500, Batavia, IL 60510-500, USA

Available online 25 May 2007

Abstract

A new inner detector called Layer 0 has been added to the existing silicon detector for the DZero colliding beams experiment

[V.M. Abazoz et al., Nucl. Instr. and Meth. A 565 (2006) 463]. This detector has an all carbon fiber support structure that employs thin

copper clad Kapton sheets embedded in the surface of the carbon fiber structure to improve the grounding of the structure and a readout

system that fully isolates the local detector ground from the rest of the detector. Initial measurements show efficiencies greater than 90%

and 0.3 ADC count (240 e) common mode contribution to the signal noise. The total detector capacitance is 24 pF so this corresponds to

2mV of common mode voltage.

r 2007 Elsevier B.V. All rights reserved.

PACS: 29.40.Wk

Keywords: Silicon tracker; Carbon fiber; Detector grounding

1. Introduction

The original silicon detector for the D0 experiment hadfour layers of silicon. A new radiation hard inner layer (L0)[1] was added to improve the resolution for displacedvertices and also provide increased redundancy for failuresin the existing detector. Because the main barrel part of theexisting detector could not be removed, this new detectorhad to be designed for insertion into the existing detector.This placed severe constraints on the design. Table 1 showsthe detailed specifications.

The overall length and small diameter required the use ofvery high modulus carbon fiber. It also made it verydifficult to provide a dielectric break in the structure. Highmodulus carbon fiber is quite conductive at high frequen-cies so there is a classic ground loop formed by thegrounded electronics at each end, the carbon fiber supporttube and the rest of the detector. In order to eliminate thisground loop, we developed a readout system that isolatesthe local detector ground from the rest of the detector. Weachieved isolation greater than10O per end at 10MHz.

The small diameter of the detector did not allow thedirect mounting of the readout chips on the sensors so we

e front matter r 2007 Elsevier B.V. All rights reserved.

ma.2007.05.281

ess: mjohnson@fnal.gov

were forced to use a Kapton flex circuit cable (called theanalog cable) between 200 and 360mm long to bring thedetector signal to the readout electronics. The capacitanceof this cable (0.35 pF/cm) doubled the detector capaci-tance, which then doubles the intrinsic noise of thedetector.

2. Mechanical design

The L0 support structure can be divided up into threemajor regions (Fig. 1). The first region, occupying thecentral 760mm of the structure, is the silicon sensormounting outer shell with the precision sensor mountingsurfaces. The other two regions are the 450mm longhexagonal hybrid mounting outer shells at each end of thestructure. Cooling distribution manifold assemblies at eachend also act as the support and connecting points for L0.All of these components are connected together via a longdodecagonal inner shell.The inner shell, the sensor mounting shell and the hybrid

mounting shells are all made from unidirectional Mitsu-bishi Chemical K13C2U carbon fiber pre-impregnated withRS-24 MOD resin supplied by YLA Inc. This fiber has aYoung’s modulus 900GPa This material has a pre-cure

ARTICLE IN PRESSM. Johnson / Nuclear Instruments and Methods in Physics Research A 579 (2007) 718–722 719

weight of 66 g/m2 and resin content of 39wt%. Thenominal thickness is 63.5 mm.

The inner shell spans the full length of the supportstructure (1660mm) and has an inscribed diameter of31.7mm. It is made of three layers of pre-preg with a01/901/01 arrangement. In another words, fibers are alongthe long axis in the two outside layers and are in theazimuth direction in the middle layer. The nominal curedthickness is 195 mm. The sensor mounting and hybridmounting outer shells are both made of five layers of thesame carbon fiber material with a symmetrical fiberorientation of +201/�201/01/�201/+201. An outer layerof Kaptons with copper mesh and gold plated contacts isco-cured with the outer shells giving a combined curedthickness is 325 mm. The sensor mounting outer shell has asix-sided shape that allows for mounting of the siliconsensors at two different radial locations.

We increased the conductivity of the carbon fiber [2] bycovering the outer layer of carbon fiber on the sensormounting surfaces with a layer of copper clad Kapton film(25 mm Kapton, 5 mm copper). The copper is etched in amesh pattern (234 mm trace width, 1.45mm pitch, 30%copper coverage) to reduce the amount of added material.

This film is applied to the outer layer of carbon fiber onthe sensor mounting surfaces, and is bonded to the shellduring the cure process. The copper mesh is facing thecarbon fiber. This process forced the copper mesh intogood electrical contact with the conductive carbon fibers.

Fig. 1. Drawing of the L0 detector showing the sensor mounting

Table 1

Specifications for the L0 detector

Over all length 1660mm

Minimum diameter 31.7mm

Sensor pitch 71 or 81 mmLength of sensors 70 or 120mm

Number of sensors in longitudinal direction 8

Number of sensors in azimuth direction 6

Sensor thickness 0.3mm

We used the same techniques as we used for the plies ofcarbon fiber with special care to ensure that the Kapton/Cu-mesh conformed to the surface features of thecastellated shell with no corner air gaps. Contact to thecopper mesh is made by plated through holes in theKapton and gold pads on the outer surface. For the B-layersensors that mount on the top of the castellations a flexiblegrounding strap is bonded to the ground circuit on thebottom of the castellation to make contact with theunderside of the sensor.

3. Electrical design

The electrical design was dominated by the need toeliminate common mode noise. Because of the addedcapacitance of the analog cable and the use of 300 mm thickdetectors, even a small amount of common mode noisecould seriously affect the performance of the detector. Inaddition the analog cable can act as a good antenna forpicking up common mode noise. The overall philosophy istwo fold. First, we isolate the detector grounds from therest of the world to prevent any ground loops through thebody of the detector. Second, we provide low impedanceconnections between the sensor and the SVX 4 [3] chip.So that there is little relative voltage between them at allfrequencies in the pass band of the SVX 4.Fig. 2 shows a simplified schematic of the detector

readout electronics. Every circuit has an input path and areturn path. The input path to the SVX 4 is from thereversed bias diode through the analog cable and into thepreamp in the SVX 4. There are no local ground traces onthe analog cable (done to minimize input capacitance) sodetector pulses will couple into neighboring channels.However, the time constants for this AC coupling are muchsmaller than the integration time for the SVX 4 so almostall the charge flows back into main channel before the endof the charge integration time. There is a trade off betweenthe capacitance between the cable and the ground plane

shell, the hybrid mounting shell and the inner support shell.

ARTICLE IN PRESS

Fig. 2. Simplified schematic of the L0 electrical circuit. The hybrid refers to the BeO hybrid mounted on the carbon fiber substrate, the link is the

combination of the analog cable and the carbon fiber support structure and the sensor includes the bias filter board.

Fig. 3. Sketch of copper tab connection between the ground plane and the

bias voltage board.

M. Johnson / Nuclear Instruments and Methods in Physics Research A 579 (2007) 718–722720

and the loop area between the cable and the ground plane.The larger the spacing the smaller the capacitance but theloop area increases which may pick up noise. In this designwe emphasized smaller capacitance so all cables areseparated by a 200 mm spacer.

The return circuit to the bias side of the silicon diode isas important as the input circuit. The return side of thesilicon diode is at the bias voltage so the bias voltage mustbe connected to the amplifier ground with a capacitor.In the L0 design this is done at both the hybrid andthe sensor.

The bias voltage is transmitted from the hybrid to thesensor over a trace at the edge of the analog cable.The return current flows on an adjacent trace. These traceshave a resistance of 20O and are also quite inductive. Thisimpedance is too large for the traces to be a good path forthe amplifier return current. Because of this, the capacitorat the hybrid is not very important for the return current.It is mainly used as part of a local RC filter for the biasvoltage supply. The primary return path is through thecopper mesh on the surface of the carbon fiber supportstructure and a capacitor mounted directly on the sensor.This has the additional advantage of forcing the structureto be at the same voltage as the SVX 4 ground so that thereis little or no noise pickup from the shell.

The capacitor at the sensor is on a small printed circuitboard that is glued to the top of the sensor. This cardreceives the bias voltage directly from the analog cable.Since the main return path is through this capacitor, itsconnection to the carbon fiber is very important. We usedcopper plated Kapton tabs, which were connected to goldplated pads on the copper ground mesh with conductiveepoxy (Fig. 3).

The beryllium oxide hybrid is 380 mm thick and has agold plated strip on one side. This stripe is connected to thehybrid ground plane by plated through holes in the BeO.The Kapton mesh on the carbon fiber support structure hasa corresponding gold plated section. The hybrid is thenglued to this support structure with conductive epoxy.This creates a connection with very low inductance andalmost no loop area. The outer layer hybrids are connectedto the ground plane by wide copper clad Kapton tabs.Each of the two SVX 4 chips on a hybrid was connected

to the hybrid ground plane with a single point ground.This was done to minimize ground currents inside the chip.All the SVX 4 ground bonds are connected to a pad that is

ARTICLE IN PRESS

Fig. 4. Pedestal and noise signal for 1 sector of L0. The black (upper)

trace is the average pedestal. The light gray trace is 10 times the standard

deviation for that channel and the dark gray trace is 10 times the

differential noise (see text).

M. Johnson / Nuclear Instruments and Methods in Physics Research A 579 (2007) 718–722 721

the size of the SVX 4 chip. This pad is connected to thehybrid ground plane at one point.

The ground loop through the body of the detector iseliminated by creating an isolated ground on the detectorside of the electronics. We do not need large voltageisolation so we chose a fairly simple method. We convertedall single ended signals to differential ones and sent thesignals across the ground barrier to a differential receiver.The power supplies use local regulators that were chosenon the basis of their AC isolation specifications. However,the isolation was not adequate so we added additionaldifferential and common mode filtering outside thedetector.

On an individual basis these ideas worked quite well. Butthere are about 500 parallel connections at each end so theoverall ground isolation is about 10O at 10MHz which isadequate for this device.

The power supplies need an external DC reference point.Otherwise the zero voltage point in the isolated system maydrift above the mid point voltage of the external differentialdriver. If this happens, signals may not be receivedproperly on the isolated side. This reference is providedthrough the high voltage bias system. The bias voltagereturn line has a 10K Ohm resistor to external ground thatprovides both ground isolation and the proper DCreference point.

4. Noise calculations

We define the noise on the output of channel n as thestandard deviation of the digital output, a, in ADC countsfor that channel

s2a ¼ a2n � an

2. (1)

We also define ‘‘differential’’ noise on a given channel as

ðsdiffa Þ2¼ðanþ1 � anÞ

2� ðanþ1 � anÞ

2

2. (2)

Here sdiffa is estimated by computing the root mean squarevalue of

anþ1 � anffiffiffi2p (3)

over a large number of events.Differential noise would be a good estimator of common

mode noise if there were no correlations between adjacentchannels. However, adjacent channels are correlated boththrough the interstrip capacitance and the capacitancebetween traces in the analog cable following Ref. [4] wecalculate (2). The total noise charge, Qn, in channel n is

Qn ¼ Cs Un�1 � ð2Cs þ Cg þ CAÞUn þ CsUnþ1

� �

¼ Cs Un�1 � ð2þ rÞUn þUnþ1ð Þ ð4Þ

where U is the noise amplitude, Cs is the capacitancebetween adjacent channels including the analog cable, Cg isthe capacitance to ground, CA is the total amplifiercapacitance and r ¼ ðCg þ CAÞ=Cs. Next, form the ratio

of the differential noise to the noise in a single channel

ðQnþ1 �QnÞ=2

Qn

¼Un�1 � ð3þ rÞUnð Þ þ ð3þ rÞUnþ1 �Unþ2ð Þ=2

Un�1 � ð2þ rÞUn þUnþ1. ð5Þ

Assuming that all the noise amplitudes are identical andconverting to expectation values we get

ðQnþ1 �QnÞ=2� �2

Qn

� �2 ¼10þ 6rþ r2

6þ 4rþ r2(6)

CA is estimated [4] from the noise performance of the SVX4 (380+41 e/pF) to be 9.3 pF. From sensor and analogcable measurements, Cs ¼ 11 pF and Cg ¼ 1 pF givingr ¼ 0.93 and

Qn

� �¼ :8 Qnþ1 �Qn

� �=2

� �. (7)

That is, we would expect the single channel noise to be 80%of the differential noise. If the single channel noise is largerthan this, there is a contribution form common mode noise.

5. Performance

Fig. 4 shows the pedestal (blue), ten times the total noise(cyan) and ten times the differential noise (yellow) for theentire L0 detector in normal running mode. The averagedifferential noise is about 1.8 ADC counts and the averagetotal noise is about 1.7 ADC counts. From Eq. (7) wewould expect the total noise to be 1.4 ADC counts so weconclude that there are about 0.3 ADC counts of commonmode noise. One ADC count corresponds to 800 e so this is

ARTICLE IN PRESSM. Johnson / Nuclear Instruments and Methods in Physics Research A 579 (2007) 718–722722

a noise level of 240 e. Two times ðCs þ CgÞ ¼ 24 pF (factorof 2 is for capacitance to both adjacent sides) so thecommon mode voltage is about 2 mV.

We also connected the detector ground to the isolatedground and measured the change in noise. There was only asmall increase which indicates that the major contribution tothe low noise is from the overall grounding of the detector.

6. Conclusion

The L0 detector shows that good noise performance canbe achieved with minimal impact on the mechanical design.

However, it is important that the electrical design be doneat the same time and on an equal footing with themechanical design.

References

[1] V.M. Abazoz, et al., Nucl. Instr. and Meth. A 565 (2006) 463.

[2] W. Cooper, et al., Nucl. Instr. and Meth. A 550 (2005) 127.

[3] B. Krieger, et al., IEEE Trans. Nucl. Sci. NS-51 (2004) 1968.

[4] G. Lutz, Nucl. Instr. and Meth. A 309 (1991) 545.