Andrei Nomerotski 1 CCD-based Pixel Detectors by LCFI Andrei Nomerotski (U.Oxford) on behalf of LCFI collaboration Hiroshima2006, Carmel CA Outline  LCFI Collaboration  Pixel Sensors and their Readout u Column-Parallel CCDs u Storage Pixels : ISIS  Mechanical Studies  Summary

Andrei Nomerotski 2 LCFI : Linear Collider Flavour Identification Valencia Goals : Development of technologies and algorithms for the ILC vertex detector

Andrei Nomerotski 3 Pixel Sensors  Traditionally LFCI develops CCD-based sensors  Builds on successes of the SLD vertex detector which was based on CCDs and was the closest to the required ILC parameters ever built The VXD3 upgrade vertex detector: 96 large CCDs, 307 Million pixels (1996)

Andrei Nomerotski 4 Vertex Detector for ILC Main requirements:  Excellent point resolution (3-4 μm), ~1 Gigapixel 20x20 μm  Low material budget ( 0.1% X 0 per layer)  Low power dissipation  Moderate radiation hardness (  20 krad/year)  Tolerance to Electro-Magnetic Interference (EMI)  Operation in 5T magnetic field  Fast Readout – The Challenge

Andrei Nomerotski 5 The Challenge What readout speed is needed?  If read once per train : occupancy ~200 hits/mm 2 : too slow  Need to read once accumulated occupancy ~10 hits/mm 2  => 20 times per train = 50 µ s/MPixel u Fastest commercial CCDs ~ 1 ms/MPixel  Two approaches 1. Parallel Readout of traditional CCD: CPCCD – information leaves the sensor as fast as it can 2. Storage Sensors : each pixel has a ‘ memory ’ filled up during collisions and read out between trains at slow rate: ISIS technology – information is stored in the sensor 337 ns 2820x 0.2 s 0.95 ms LC Beam Time Structure : = one train

Andrei Nomerotski 6 “Classic CCD” Readout time  N  M/f out N M N Column Parallel CCD Readout time = N/f out M Simple idea : read out a vector instead of a matrix  Readout time shortened by orders of magnitude BUT Despite ‘parallel processing’ readout rate is still challenging : 50 MHz clock moves charge 2500 times in 50 µs. 2500x20 µm = 50 mm Every column needs own amplifier and ADC  requires readout chip Column Parallel CCD

Andrei Nomerotski 7 Column Parallel CCD  Readout Chip is a difficult but clearly feasible problem u Geometrically concept of columns is similar to the silicon strips - strip detectors have complex readout chips integrated in ladder u However density of channels and 0.1% Xo constraint requires bump- bonding – non-trivial anyway  Difference wrt Strips: to move the charge need to clock all columns of the CCD simultaneously u Simple exercise : s typical capacitance of CCD sensor : 100nF s 50MHz  10 ns rise time s Clock current = 100 nF x 2 V / 10 nsec = 20A ! s Voltage drops 20 A x 0.1 Ohm = 2 V s Inductance of 1mm long bond wire = 1 nH : corresponds to 0.3 Ohm at 50 MHz  Driving a full area CPCCD is a major challenge! u Need a special high current clock driver u Need to be extra careful with the design of clock distribution

Andrei Nomerotski 8 CPCCD : LCFI R&D Milestones  Established proof of principle for small area sensors : CPC1  Established proof of principle for readout chip : CPR1, developed and produced more sophisticated CPR2  Moved on to large area sensors : CPC2 u Need to handle the problem of clock driver 1. Design dedicated clock driver : CPD1 2. Find ways to reduce the CCD capacitance 3. Find ways to reduce the required clock voltage Next slides: Results from prototypes

Andrei Nomerotski 9 CPC1 : Two phase CCD, 400 (V)  750 (H) pixels, 20 μm square; CMOS readout chip (CPR1) designed by the Microelectronics Group at RAL:  0.25 μm process  Charge and voltage amplifiers matching the outputs of CPC1  Correlated double sampling  5-bit flash ADCs and 132-deep FIFO per column  Everything on 20 μm pitch  Size : 6 mm  6.5 mm  Manufactured by IBM Bump-bonded by VTT (Finland) using solder bumps Bump-bonded CPC1/CPR1 in a test PCB CPC1 Bump-bonded to CPR1

Andrei Nomerotski keV X-ray hits, 1 MHz column-parallel readout Voltage outputs, non- inverting (negative signals) Noise  60 e- Charge outputs, inverting (positive signals) Noise  100 e- First time e2V CCDs have been bump-bonded High quality bumps, but assembly yield only 30% : mechanical damage during compression suspected Differential non-linearity in ADCs (100 mV full scale) : addressed in CPR2 Bump bonds on CPC1 under microscope CPC1/CPR1 Performance K.Stefanov RAL

Andrei Nomerotski 11 CPR2 designed for CPC2 Results from CPR1 taken into account Numerous test features Size : 6 mm  9.5 mm 0.25 μm CMOS process (IBM) Manufactured and delivered February 2005 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 (2  2 kernel) Sparse readout circuitry FIFO Next Generation CPCCD Readout Chip – CPR2 Steve Thomas, RAL

Andrei Nomerotski 12 CPR2 Test Results Tim Woolliscroft, Liverpool U ● Tests on the cluster finder: works! ● Several minor problems, but chip is usable ● Design occupancy is 1% ● Cluster separation studies:  Errors as the distance between the clusters decreases  Reveal dead time  Many of the findings have already been input into the CPR2A design Tim Woolliscroft, Liverpool U Parallel cluster finder with 2  2 kernel Global threshold Upon exceeding the threshold, 4  9 pixels around the cluster are flagged for readout Test clusters in Sparsified output

Andrei Nomerotski 13 Next Generation CPCCD : CPC2 Three different chip sizes with common design:  CPC2-70 : 92 mm  15 mm image area  CPC2-40 : 53 mm long  CPC2-10 : 13 mm long Compatible with CPR1 and CPR2 Two charge transport sections Choice of epitaxial layers for different depletion depth: 100 .cm (25 μm thick) and 1.5 k .cm (50 μm thick)

Andrei Nomerotski 14 CPC2 + ISIS1 Wafer ISIS1 CPC2-70 CPC2-40 CPC2-10 5” wafers One CPC2-70 : 105 mm  17 mm total chip size Two CPC2-40 per wafer 6 CPC2-10 per wafer 14 In-situ Storage Image Sensors (ISIS1) 3 wafers delivered

Andrei Nomerotski 15 Clock monitor pads CPR1/CPR2 pads CPC2-40 in MB4.0 Johan Fopma, Oxford U  Transformer drive for CPC2  “Busline-free” CCD: the whole image area serves as a distributed busline  50 MHz achievable with suitable driver in CPC2-10 and CPC2-40  First clocking tests have been done Transformer

Andrei Nomerotski Fe spectrum from CPC2-10 at 1 MHz ● First 55 Fe spectrum at 1 MHz, -40  C, reset every pixel CPC2: First Results K.Stefanov RAL

Andrei Nomerotski 17 IC driver: CPD1 Transformer is bulky: IC driver could be a better solution; First CPCCD driver chip (CPD1) submitted in July 2006 CPD1: 2-phase CMOS driver chip for 20 Amp current load at 25 MHz (L2-L5 CCDs) 0.35 μm process, size  3 x 8 mm 2 32 W peak power but 0.5% duty cycle Thermal and electromigration issues seems to be under control Clock Drivers for CPC2 Transformer Driver Requirements: 2 V pk-pk at 50 MHz over 40 nF (half CPC2-40); Planar air core transformers on 10-layer PCB, 1 cm square Parasitic inductance of bond wires is a major effect – fully simulated; Brian Hawes, Oxford U

Andrei Nomerotski 18 Next Steps for CPCCD  Evaluate performance of CPC2 bump-bonded to CPC2/CPR2  Designing with e2V test devices to study how to reduce CCD capacitance and how to reduce clock voltage u Theoretically can achieve factor of 4 reduction in C  Design of CPC3 will depend on results of these tests

Andrei Nomerotski 19 Radiation Damage Effects in CCDs: Simulations Simulation at 50 MHz Operating window L. Dehimi, K. Bekhouche (Biskra U); G. Davies, C. Bowdery, A.Sopczak (Lancaster U) ● Full 2D simulation based on ISE- TCAD developed ● Trapped signal electrons can be counted ● CPU-intensive and time consuming ● Simpler analytical model also used, compares well with the full simulation ● Window of low Charge Transfer Inefficiency (CTI) between -40  C and 0  C ● Will be verified by measurements on CPC2 Signal density of trapped electrons in 2D

Andrei Nomerotski 20 Storage Pixels  Industry analogy is “Burst mode” : capturing a limited number of images at short intervals u Burst mode imagers are available commercially (ex. DALSA) with rates up to 100MHz : the rate is limited by the charge transfer between neighboring cells u Memory is implemented as a CCD register associated with an imaging pixel : whatever one can fit in an area of one pixel 2003: Dart bursting a ballon : 100 consecutive frames at 1M frame/sec

Andrei Nomerotski 21 Storage Pixels as Particle Detectors  ILC requirements : capture charge every 50 us  20kHz – no problem Challenges :  Used as particle (not visible light) detector – need efficient charge collection from the whole area  Need to fit 20 cell CCD register into 20x20 square micron pixel (together with photogate and some logic)  Need a more complicated than pure CCD process

Andrei Nomerotski 22 In-situ Storage Image Sensor : ISIS  Charge is collected into a photogate  Each pixel has its own 20-cell CCD register : store raw charge during collisions  Increased resistance to RF  Column-parallel readout during quiet time at ~1 MHz: much reduced clocking requirements

Andrei Nomerotski 23 In-situ Storage Image Sensor (ISIS) RG RD OD RSEL Column transistor Additional ISIS advantages:  ~100 times more radiation hard than CCDs – less charge transfers  Easier to drive because of the low clock frequency: 20 kHz during capture, 1 MHz during readout ISIS combines CCDs, active pixel transistors and edge electronics in one device: specialised process Development and design of ISIS is more ambitious goal than CPCCD “Proof of principle” device (ISIS1) designed and manufactured by e2V Technologies On-chip logicOn-chip switches 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

Andrei Nomerotski 24 Output and reset transistors Photogate aperture (8 μm square) CCD (5  6.75 μm pixels) The ISIS1 Cell OG RG OD RSEL OUT Column transistor 16  16 array of ISIS cells with 5-pixel buried channel CCD storage register each; Cell pitch 40 μm  160 μm, no edge logic (pure CCD process) Chip size  6.5 mm  6.5 mm

Andrei Nomerotski 25 Tests of ISIS1 Tests with Fe-55 source The top row and 2 side columns are not protected and collect diffusing charge The bottom row is protected by the output circuitry ISIS1 without p-well tested first and works OK ISIS1 with p-well has very large transistor thresholds, permanently off K.Stefanov RAL

Andrei Nomerotski 26 Mechanical Options Target of 0.1% X 0 per layer (100μm silicon equivalent)  Unsupported Silicon u Longitudinal tensioning provides stiffness u No lateral stability u Not believed to be promising  Thin Substrates u Detector can be thinned to epitaxial layer (~20 μm) u Silicon glued to low mass substrate for lateral stability u Longitudinal stiffness still from moderate tension u Beryllium has best specific stiffness  Rigid Structures u Foams look very promising u Will start to investigate shell structure supports

Andrei Nomerotski 27 Ladder testing with Be and Carbon Fibre  Beryllium substrate u Minimum thickness 0.15% X 0 u Good qualitative agreement from FEA models and measurement  Carbon Fibre substrate u Better CTE match than Be u ~0.09% X 0, no rippling to <200K  lateral stability insufficient Tension Silicon Glue Beryllium J.Goldstein RAL

Andrei Nomerotski 28 Rigid Structures: Foams  Properties: u Open-cell foam u Macroscopically uniform u No tensioning needed  3% RVC prototype u Sandwich with foam core u 0.09% X 0 u Mechanically unsatisfactory u Working on glue application  8% Silicon Carbide prototype u Single-sided: substrate + foam u 0.14% X 0 u 3-4% believed possible 20 µm silicon 1.5 mm silicon carbide

Andrei Nomerotski 29 Silicon Carbide Foam Glue “ pillars ” (right plot) are better than thin glue layer (left plot)

Andrei Nomerotski 30 Summary  LCFI is a viable and growing collaboration to develop technologies and algorithms for VD  First generation sensors extensively studied u Column parallel CCD principle proven  First results from the second generation of sensors and readout chips u Detector-scale CCDs u Sparsified readout u Developing advanced clock drivers u First prototypes of storage devices  Mechanics : 0.1% X 0 ladders seems achievable