LBNL Leo Greiner , Eric Anderssen, Giacomo Contin,

L. Greiner 1ATLAS Workshop LBL - September, 2013

STAR HFTSTAR HFT

LBNLLeo Greiner, Eric Anderssen, Giacomo Contin,Thorsten Stezelberger, Joe Silber, Xiangming

Sun, Michal Szelezniak, Chinh Vu, Howard Wieman, Sam Woodmansee

UT at AustinJerry Hoffman, Jo Schambach

IPHC StrasburgMarc Winter CMOS PICSEL group

Assembly and testing of a MAPS based vertex detector for the STAR

experiment at RHIC

2ATLAS Workshop LBL - September, 2013L. Greiner

STAR HFTTalk Outline

•Short overview of detector design.•Design of ladders, sectors, insertion mechanism.•Detector assembly, yields, integration and installation for engineering run.•In progress production, assembly and QA for the primary detector.

The primary focus of this talk is technical and on instrumentation.


STAR HFTPXL in STAR Inner Detector UpgradesTPC – Time Projection Chamber(main tracking detector in STAR)

HFT – Heavy Flavor Tracker SSD – Silicon Strip Detector

r = 22 cm IST – Inner Silicon Tracker

r = 14 cm PXL – Pixel Detector

r = 2.7, 8 cm

We track inward from the TPC with graded resolution:

TPC SSD IST PXL~1mm ~300µm ~250µm <30µm

Direct topological reconstruction of Charm

vertex


STAR HFTPXL Detector Requirements and Design Choices

• -1 ≤ Eta ≤ 1, full Phi coverage (TPC coverage)• ≤ 30 µm DCA pointing resolution required for 750 MeV/c kaon

– Two or more layers with a separation of > 5 cm.– Pixel size of ≤ 30 µm– Radiation length as low as possible but should be ≤ 0.5% / layer (including support structure). The goal is 0.37% / layer

(limits MCS)• Integration time of < 200 μs (limit pile-up, occupancy of ~250 hits/inner sensor)• Sensor efficiency ≥ 99% with accidental rate ≤ 10-4.• Survive radiation environment.

• Air cooling• Thinned silicon sensors (50 μm thickness)• MAPS (Monolithic Active Pixel Sensor) pixel technology

– Sensor power dissipation ~170 mW/cm2

– Sensor integration time <200 μs (L=8×1027)• Quick detector installation or replacement (1 day)

Req

uire

men

tsD

esig

n C

hoic

es


STAR HFTPXL Detector Design

Mechanical support with kinematic mounts (insertion side)

Insertion from one side2 layers (tiled)5 sectors / half (10 sectors total)4 ladders/sector

Aluminum conductor Ladder Flex Cable

Ladder with 10 MAPS sensors (~ 2×2 cm each)

carbon fiber sector tubes (~ 200 µm thick)

20 cm


STAR HFT

2 m (42 AWG TP)11 m (24 AWG TP)

100 m (fiber optic)

Highly parallel system

4 ladders per sector 1 Mass Termination Board (MTB) per sector 1 RDO board per sector 10 Sector chains in the PXL system

RDO motherboard w/ Xilinx Virtex-6 FPGA

DAQ PC with fiber link to RDO board

Mass Termination Board (signal buffering) + latch-up protected power

PXL Detector Basic Unit (RDO)

Clk, config, data, powerClk, config, data

PXL built events

Trigger, Slow control,Configuration,etc.

Existing STAR infrastructure


STAR HFTDetector Characteristics

Pointing resolution (12 19 GeV/pc) m

Layers Layer 1 at 2.7 cm radiusLayer 2 at 8 cm radius

Pixel size 20.7 m X 20.7 m

Hit resolution 6 m

Position stability 6 m rms (20 m envelope)

Radiation length per layer X/X0 = 0.37%

Number of pixels 356 M

Integration time (affects pileup) 185.6 s

Radiation environment 20 to 90 kRad / year2*1011 to 1012 1MeV n eq/cm2

Rapid detector replacement ~ 1 day

356 M pixels on ~0.16 m2 of Silicon


STAR HFTPXL Detector Highlights• The HFT upgrade to STAR received DOE CD-2/3 approval in September

2011 and became a construction project.• We installed a 3 sector engineering run detector into the STAR experiment

on May 8, 2013 and ran the detector until June 10, 2013. • The production detector is scheduled to be installed in December of 2013.• We will build 2 complete detectors and enough good ladders to build a third.

These serve as spares for rework if necessary.

I will give a brief overview of the PXL detector design details, components and assembly in the following areas:

• Sensors• Ladder assembly and QA• Sector assembly and QA• Detector Half• Insertion mechanics• RDO System• Engineering run


STAR HFT

• Sensors


STAR HFT

• Standard commercial CMOS technology

• Room temperature operation• Sensor and signal processing are

integrated in the same silicon wafer• Signal is created in the low-doped

epitaxial layer (typically ~10-15 μm) → MIP signal is limited to <1000 electrons

• Charge collection is mainly through thermal diffusion (~100 ns), reflective boundaries at p-well and substrate.

• High resistivity epi for larger depleted region → more efficient charge collection, radiation tolerance.

• 100% fill-factor • Fast readout• Proven thinning to 50 micron

MAPS pixel cross-section (not to scale)

Monolithic Active Pixel Sensors


STAR HFTPXL detector Ult -1/2 Sensor• Reticle size (~ 4 cm²)

• Pixel pitch 20.7 μm • 928 x 960 array ~890 k pixels

• Power dissipation ~170 mW/cm² @ 3.3V• Short integration time 185.6 μs• In pixel CDS• Discriminators at the end of each column

(each row processed in parallel)• 2 LVDS data outputs @ 160 MHz• Zero suppression and run length encoding on

rows with up to 9 hits/row.• Ping-pong memory for frame readout (~1500

hits deep)• 4 sub-arrays to help with process variation• JTAG configuration of many internal

parameters.• Individual discriminator disable, etc.• Built in automated testing routines for sensor

probe testing and characterization.• High Res Si option – significantly increases

S/N and radiation tolerance.

Developed by IPHC, Strasbourg France


STAR HFTUltimate 1 efficiency vs. fake hit rate

Tested for STAR Operating conditions

Developed and tested by IPHC Marc WinterCMOS GroupIPHC, Strasbourg, France


STAR HFT

• Ladder assembly• Sector assembly• Half detector assembly


STAR HFTLadder Design• Ladder assembly is the primary task for the detector fabrication.• We assemble ladders by edge bumping sensors on a vacuum

chuck (the soft acrylic adhesive provides mechanical decoupling for CTE mismatches)

• DRIE needed for precise sensor size. (most engineering run sensors were saw cut and this gave significant variations in the assembled length of the 10 sensor array, after DRIE, length variation in butted edge sensor array is ~50um)

• Wire bonding is very sensitive to small sensor position variations (small displacements cause the bonding pads on the ladder to displace wrt bonding pads on the sensors and wire bonds then require steep angles).


STAR HFTLadder Design

Ladder cable concept


STAR HFTLadder assembly work flow chart

Probe tested sensors

Electrically tested low mass cables

Electrically tested driver boards

Dimensionally checked composite backer

Ladder assembly

Ladder wire bonding

Wire bond encapsulation

Quick test

Full functionality test

Complete ladder

• Ladder characterization

• Reworking and troubleshooting

• Quality assessment• Initial validation

1 day without problem

sFull functionality testbias optimizationThreshold scanNormal readout mode testAccidental hit rate scan

Quick testThreshold scan @ nominal bias settings


STAR HFTSector/half-detector work flow chart

TestedLadders

SizedSector Tubes

machined Dovetail/D-tube

Elect testedMTB/cables/insertion

Sector assembly

Full functionality test

Quick test

Half detector headassembly

Sector metrology

½ HFT PXL

Quick test

Half detector head metrology

Full Half detector assembly

Quick test


STAR HFTOverall detector design factors (metrology)

• We intend to insert a fully mapped detector with all pixel positions known and stable.

• We have gone to great lengths to ensure that the ladder/sector/half detector mechanics are stable to 30um for all forces (gravity, vibration and deflection due to air cooling (~10 m/s), thermal distortion, mechanical stress from cable loading, etc.) anticipated during running.

• We map each sector using fiducials on the sensors and relate pixel positions to tooling balls on the sectors.

• We then assemble half detectors and map the sector tooling balls to give the pixel positions for each half detector.


STAR HFTSensor Thinning• We use sensors thinned to 50 µm thickness.• We use DRIE processing on the wafers to give accurate sensor dimensions

and edge quality.• Wafers are delivered with DRIE trenches 70 µm deep. The wafers are back-

thinned and polished at Aptek to release the dies from the wafer (yield started quite low, but is now significantly improved).

• First 25 production wafers, 5 were lost completely during thinning as process was refined. Last 17 wafers thinned had >99% yield.

• Die position on the wafer and wafer # are preserved through the thinning process and this identifies the die in our database.

• We select dies based on probe testing of thinned sensors.• 50 um thick sensors are curved, up to 2 mm out of plane, this has

implications.


STAR HFT

• Probe card with readout electronics – derived from individual sensor test card

• Analog and digital sensor readout• Full speed readout at 160 MHz• Full sensor characterization at full speed

– Test results used for initial settings in ladder testing and PXL detector configuration

• 2nd generation probe card for production testing– only digital readout pins loaded

Yield modeling makes probe testing critical to the goal of assembling functional 10 sensor ladders.We test thinned and diced 50 µm thick sensors (curved). This is not easy.

Assembling sensors into ladders – Probe Testing


STAR HFTAssembling sensors into ladders – Probe Testing• Vacuum chuck for testing 20 thin

sensors (50 µm)– Thin sensors are sensitive to chuck

surface imperfections– Need a special probe pin design and

to work in clean environment• Testing up to 18 sensors per batch

– Optimized for sensor handling in 9-sensor carrier boxes

• Manual alignment (18 sensors ~ 1 hr)– Testing time is ~ 15 minutes per

sensor (3 supply voltages) (4.5 hr testing time)

Thr

esho

ld D

AC

cou

nts (

x5)

Column number

LabW

indo

ws

GU

I

Dis

crim

inat

or T

hres

hold


STAR HFTExamples of failures

D07_w_12E01_w_8 (@3.0V)

E05_w_12 E06_w_12

E05_w_08 C04_w_12

OK

Faultycolumns

Sub-arrays

Faulty rows

3.3 or 3.0 V


STAR HFTPXL Sensor selection• Automated interface to a database (18 config + 68 result parameters)• Sensors are binned according to performance

Example of wafer maps for Ultimate-2 wafers with DRIE processing (ion etching)

~65% - 70% yield (Tier 1-3)

40% Tier 1-2


STAR HFT

• Building sensors into ladders


STAR HFTAssembling sensors into ladders – Hybrid Cable• The cable is needed to deliver power and ground and provide signal routing to and from

the sensors on the ladder with minimal radiation length.• Production aluminum conductor flex cables are being fabricated in the CERN PCB shop.

(difficult to produce, only one real vendor)

Hybrid Copper / Aluminum conductor flex cable design concept

Low mass region calculated X/X0 for Al conductor = 0.079 %Low mass region calculated X/X0 for Cu conductor = 0.232 %

• 20 differential signal output pairs• JTAG, 2 temp diodes, CLK, START• VDD, VDA, GND - ~1.8 A / ladder

Layout (in copper)


STAR HFTAssembling sensors into ladders – Assembly

• We use precision vacuum chuck fixtures to position sensors and assemble ladders.

• Sensors are positioned with butted edges. Acrylic adhesive mechanically decouples sensors from the cable and prevents CTE difference based damage.

• Weights taken at all assembly steps to track material and as QA.

Reference pins for cable/sensor alignment


STAR HFTAssembling sensors into ladders – Assembly

• Hybrid cable with carbon fiber stiffener plate on back in position to glue on sensors.

• The cable reference holes are used for all aspects of assembly of ladders and sectors.


STAR HFTQA/QC for Sector Production

Completed ladder in anti-static carrying box with follower and check off list.

Assembled ladder with driver board, wire bonded and encapsulated


STAR HFTEngineering Run Ladder Yield• The assembly of ladders and sectors for the PXL Engineering Run has been

crucial for process optimization and we have dealt with a number of unexpected issues.

• Thanks to this experience, we were able to refine the assembly procedure and to develop effective tools for troubleshooting ladders.

• The final ladder assembly statistics for the engineering run detector are:

Quality assessment (03/20/2013) # %Fully operating ladders 15 55.6

Lower quality operating ladders (>8 operating chips) 4 14.8

Recoverable not operating ladders (need fix or workaround)

4 14.8

Dead ladders (not recoverable) 4 14.8

Total 27 100.0• The primary lessons learned had to do with the need for coverlay on the

cable, modification of tooling that was damaging sensors, adjustment of adhesive sizes and backing plates, proper adjustment of sensor array on the cable.


STAR HFT

• Ladders to sectors


STAR HFTLadders to sectors


STAR HFTLadders to sectorsSector in optical metrology machine

• Sensor fiducial positions on sector are measured optically and related to tooling balls.

• Surface profiles of sensors (sensors are not flat) are measured using feather touch probe CMM and pixel positions are related to sensor fiducials using thin plate spline fit. Now we have pixel positions related to the tooling balls.

• After touch probe measurements, sectors are tested electrically for damage from metrology.


STAR HFTsectors to detector half (ER)

• Sectors are mounted in dovetail slots on detector half.• Metrology is done to relate sector balls to each other and

to kinematic mounts. We now have pixel positions for the half detector object.


STAR HFT

• PXL detector mechanics


STAR HFTMechanical Development

Silicon power: tested at 170 mW/cm2 (~ power of sunlight) 350 W total in the ladder region (Si + drivers)

computational fluid dynamics

Air-flow based cooling system for PXL to minimize material budget.

Detector mockup to study cooling efficiency

ladder region


STAR HFT

Power ~340 W Airflow 10.1 m/s

Thermal camera image of sector 1 (composite image):

Mechanical Development

unsuported endmid-section fixed endSolid – inner layerOpen – outer layer

340 W with ambient air = 26.8 C

NTC thermistors(averaged at same “Z” position)

• Measurement results agree with simulations and meets calculated stability envelope tolerance.

• Air flow-induced vibrations (10 m/s) are within required stability window.


STAR HFTPXL detector – insertion mechanics

• Unusual mechanical approach.• Cantilevered and insertable (1 day)• Pre-surveyed and mechanically stable to a level that preserves the survey.• Detector inserts and initiates a clamshell closing around beam pipe, and is

locked into position on kinematic mounts.


STAR HFTPXL insertion mechanics

Interaction point view of the PXL insertion rails and kinematic mount points

Carbon fiber rails

Kinematic mounts


STAR HFTPXL insertion mechanics

PXL detector half with complete insertion mechanism


STAR HFTEngineering Run

Left detector half being inserted

Right detector half being inserted



PXL eng detector inserted, cabled and working in 1 day access



• PXL system working with 3 sectors

• Integrated with STAR trigger, DAQ, slow controls, online event pool.

• Software and firmware for performing data quality checks were modified and optimized.

Hit display for PXL engineering run detector.


STAR HFTPreliminary results from Engineering Run• First tracking results show

good matching of TPC tracks to hits on PXL sensors

• Residuals compatible with TPC track resolutions on the sensors (~1-2 mm)

• 3mm beam shift reflects the PXL&TPC relative position


STAR HFTLessons learned• “Full dress rehearsals” are critical to success.• We discovered a mechanical conflict in the ladder driver

board section that necessitated a sector tube redesign.• Tooling that would fracture sensors during assembly

was modified.• Our ladder temperature monitoring system didn’t work

properly and was re-designed.• Other electronics / sensor issues are still under

investigation and will require further updating of the existing systems.


STAR HFTProduction detector

• We are currently building ladders for the production detector.

• Mechanical thinning yields are now close to 100%.• Current ladder yields are much better.• 36 ladders produced, 26 tested, 1 ladder has 1

damaged sensor, all others tested are fully functional.

• Production sector assembly will start in mid-September.


STAR HFT

• end


STAR HFT

• Extra slides


STAR HFT


STAR HFTRead-out Electronics

1. The detector is 10 parallel sector chains.2. Each sector is configured and delivers data to the RDO boards

continuously into a data pipeline.3. The receipt of a trigger opens an event buffer for 1 frame of data.

(2 actual frames with address selection performed to give one compete frame based on the time the trigger is received)

4. The data is formed into an event and shipped to the DAQ receiver PCs.

Full PXL RDO fits into 1 crate


STAR HFT

Tier1 yield Vs position in wafer

0

5

10

15

Chip Position

Coun

ts

Tier2Tier1

First 25 production wafers.

Radiation Length of active area

NOTE: Does not include sector tube side walls


STAR HFTHFT PXL status – fabrication and tooling

L. Greiner 53ATLAS Workshop LBL - September, 2013

STAR HFT

2 2

GEANT: Realistic detector geometry + Standard STAR trackingincluding the pixel pileup hits at RHIC-II luminosity

Hand Calculation: Multiple Coulomb Scattering + Detector hit resolutionPXL telescope limit: Two PIXEL layers only, hit resolution only

Mean pT

30 m

Expected DCA resolution performance


STAR HFTPXL Detector Production Path (post CD – 2/3)

Sensor development

Readout development

Mechanical development

Cable development

Pre-production prototype

UltProbe tests

Infrastructure development

Ult sensorprototype

10 sensorCu+kapton pre-prod

Eng run Sector

prototype

Final Ladder readout

Final individual readout

Update mech for

size

Prod readout

production sensors

ProdProbe tests

10 sensorAl+kapton

prod

prod Sector

Eng run detector

prod detector

May 2013

Dec 2013

Culmination of >9 years of development


STAR HFTPixel signal extraction and operation

GND

VDD

select

outputoutputin equilibrium

time

pote

ntia

l on

the

char

ge c

olle

ctin

g no

de

chargecollection

chargecollectingdiode

Capacitor Storage for Correlated Double Sampling (CDS)

~ 100s ms

Discriminator

Zero suppression and memory

LVDS outputs 160 MHz

1 2 3

(Sample 1 – Sample 2) > threshold => hit(Sample 2 – Sample 3) < threshold => no hit

CDS

Simplified description

O

Self-biasedconfiguration


STAR HFT

Driver section ~ 6 cm

Sensor section ~20 cm

Kapton cables with copper traces forming heaters allow us to dissipate the expected amount of power in the detector

6 NTC thermistors on each ladder

Sector 1 was equipped with 10 thinned dummy silicon chips per ladder with Pt heaters vapor deposited on top of the silicon and wire bonded to heater power.

Power ~340 W Airflow 10.1 m/s

Thermal camera image of sector 1 (composite image):

Mechanical Development


STAR HFTAlternate Technologies Considered

• Hybrid– X0 large (1.2%)– Pixel Size large (50 m x 450 m)– Specialized manufacturing - not readily available

• CCDs– Limited radiation tolerance– Slow frame rate, pileup issues– Specialized manufacturing

• DEPFET– Specialized manufacturing– very aggressive unproven technology

MAPS sensors are the technology selected

Documents

LBNL Leo Greiner , Eric Anderssen, Giacomo Contin,