Page 1
A MAPS-based readout for a A MAPS-based readout for a Tera-Pixel Tera-Pixel
electromagnetic calorimeter electromagnetic calorimeter at the ILC at the ILC
Marcel StanitzkiMarcel StanitzkiSTFC-Rutherford Appleton LaboratorySTFC-Rutherford Appleton Laboratory
Y. Mikami, O. Miller, V. Rajovic, N.K. Watson, J.A. WilsonUniversity of Birmingham
J.A. Ballin, P.D. Dauncey, A.-M. Magnan, M. NoyImperial College London
J.P. Crooks, B. Levin, M.Lynch M. Stanitzki, K.D. Stefanov, R. Turchetta, M. Tyndel, E.G. VillaniSTFC-Rutherford Appleton Laboratory
Page 2
Marcel Stanitzki2
The ILC
ILC calorimetry focused on Particle Flow Approach (PFA)
− Requirement of highly granular calorimeters
− Goal : Jet Energy resolution ~ 30 % /√E
ILC environment is very different compared to LHC
− Bunch spacing of ~ 300 ns (baseline)
− 2625 bunches in 1ms
− 199 ms quiet time
Occupancy dominated by beam background & noise
2625
Page 3
Marcel Stanitzki3
What are Particle Flow Algorithms (PFA)?
Calorimeter Clustering
Match Tracks with Calorimeter Clusters
Remove Photon Calorimeter Clusters
Track reconstruction
RemainingEM-only Calorimeter Clusters
RemainingCalorimeter Clusters
Remove associated Calorimeter Clusters
DONE
Charged particles
Neutral Hadrons
Photons
Page 4
Marcel Stanitzki4
SiW EM Calorimetry The baseline for the SiD &
ILD detector concepts
Sampling Calorimeter
− Silicon sensors embedded in tungsten sheets
− 30 layers
− 1.3 - 1.7 meters radius
1300- 2000 m2 silicon area
Analog read out (4x4-5x5 mm pixels)
Compact, has to fit inside the coil
ECALMODULE
COIL
HCAL
Page 5
Marcel Stanitzki5
Increasing the granularity
PFA based on
− track-shower matching
− clear shower separation
Granularity of 5x5 mm may not be sufficient for
− e.g. π0 identification from τ decays
− shower separation in dense jets
Digital Pixels with 50x50 microns
− basically a Particle Counter
− requires highly integrated sensor
− ideal for MAPS-> TPAC design
− but 1 TeraPixel system ...τ decay
Page 6
Marcel Stanitzki6
TPAC Sensor requirements
Sensitive to MIP signal
Pixels determine “hit” status (binary readout)
Store bunch crossing number & location of “hits”
Target noise rate 10-6 per Bunch crossing
Design to buffer data for up to 8192 bunch crossings
Readout in quiet time
Masking & trimming individual pixels
Minimize “dead space”
Page 7
Marcel Stanitzki7
The INMAPS process
• Additional module: Deep P-Well– Developed specifically for this
project– Added beneath all active circuits
in the pixel– Should reflect charge,
preventing unwanted loss in charge collection efficiency
• Device simulations using TCAD– confirm shielding effect
• Test chip processing variants– TPAC 1.0 manufactured
with/without deep p-well for comparison
• Standard 0.18 micron CMOSUsed in the TPAC 1.0 sensor• 6 metal layers• Analog & Digital @ 1.8 V & 3.3 V• 12 micron epitaxial layer
Page 8
Marcel Stanitzki8
The TPAC 1.0 Sensor
• 8.2 million transistors• 28224 pixels (168 x 168) ; 50
microns; 4 variants• Main variants
– PreShaper and PreSampler• Minor variants
– Capacitor variants• Sensitive area 79.4 mm2
− of which 11.1% “dead” (logic)
• Four columns of logic + SRAM– Logic columns serve 42 pixels– Record hit locations &
timestamps– Local SRAM
• Data readout– Slow (<5Mhz)– 30 bit parallel data output
Pre
Sha
per
Pre
Sam
pler0
1
2
3
Page 9
Marcel Stanitzki9
TPAC Architecture Details
Deep p-well
Circuit N-Wells
Diodes
The two main variants − PreSampler problematic
in array
− only the PreShaper worked well in the array
PreShaper− 4 diodes − 1 resistor (4 MΩ)− Configuration SRAM & Mask− Comparator trim (4 bits)
Two PreShaper variants − subtle changes to capacitors
Predicted Performance− Gain 94 μV/e− Noise 23 e-
− Power 8.9 μW
Page 10
Marcel Stanitzki10
Sensor testing
quad0
quad1
Test pixels• preSample pixel variant• Analog output nodes• IR laser stimulus (1064 nm)• 55Fe stimulus
Single pixel in array● preShape (quad0/1)● Per pixel masks● IR Laser Stimulus (1064 nm)● 55Fe stimulus
Full pixel array• preShape (quad0/1)• Pedestals & trim adjustment• Gain uniformity• Crosstalk
Page 11
Marcel Stanitzki11
Analog Test Pixel : Laser Using 1064 nm Laser
back-illuminate through substrate
2x2 μm spot, 2 μm steps
Take Profile through 2 diodes in test pixel
Page 12
Marcel Stanitzki12
Analog Test Pixel: 55Fe
55Fe main decay
− 5.9 keV photon
All energy deposited in approx 1 μm3 silicon
− Generates 1640 e−
If a photon hits a diode
− no diffusion
Absolute Gain calibration
Page 13
Marcel Stanitzki13
Array : Pixel Response to laser
Single active pixel with/without laser firing
Use same laser setup as for analog scans
Fire Laser at fixed point in pixel
Threshold Scan with and without Laser
Plateau due to memory saturation
Page 14
Marcel Stanitzki14
Array : Single Pixel comparison
F
B
Pixel profiles
Amplitude results from Laser Scan
− With/without deep p-well
Compare
− Simulations “GDS”
− Measurements “Real”
Page 15
Marcel Stanitzki15
Array: Single Pixel 55Fe response
use 55Fe source on Pixel Array
Do a threshold Scan
Need the derivative to reconstruct 55Fe peak
− Derivative approximated using bin subtraction
Single active pixel with/without source
Page 16
Marcel Stanitzki16
Array: Pixel Noise Threshold scan required to see pedestal
and noise
Comparator fires on signal going high across threshold level
− No hits when far above or below threshold
− Width of distribution equivalent to noise
RMS ~ 5.5 Threshold Units (TU) ~ 44 e− ~ 170 eV on average
− Minimum is ~ 4 TU ~ 32 e− ~ 120 eV
− Target level was ~ 90 eV
− No correlation with position on sensor
− Spread not fully understood
− Quad1 ~ 20% larger than Quad0
Threshold
Page 17
Marcel Stanitzki17
Array: Pedestal adjustments
Trimmed: Quad0; Quad1
Plot the distribution of pedestals
− Mean of Noise
Calculate necessary trim adjustment
− Per-pixel trim file
− uni-directional adjustment
Re-scan pixels with trims
− Re-plot the distribution of pedestals
Planned to have pedestal width ~ ½ Noise width
− have more trim bits
Trim=0: Quad0; Quad1
Mean (TU)
Mean (TU)
Page 18
Marcel Stanitzki18
Array: Pixel Gain
Use laser to inject fixed-intensity signal into many pixels
Relative position should be equivalent for each pixel scanned
Adjust/trim for known pixel pedestals
Results
− Gain uniform to 12%
− Quad1 ~ 40% more gain than Quad0
− Quad1 ~ 20% better S/N than Quad0
GAIN Quad0; Quad1
Threshold
Page 19
Marcel Stanitzki19
Array Pixel Cross-talk
Scan one pixel in the column, all others off.
scan entire pixel column
Effect of all pixels (other than the one being scanned) is to increase the general noise around zero.
Shared power mesh between comparator and and monostable prime culprit, will be fixed
Page 20
Marcel Stanitzki20
Future Plans – TPAC 1.1
Have received TPAC1.1 a week ago
− Only one pixel variant (preShaper quad1)
− Upgrade trim adjustment from 4 bits to 6 bits
− Compatible format: size, pins, PCB/DAQ etc.
− Minor bugs fixed (e.g. cross-talk)
− Additional test pixels & devices for further process characterization
Page 21
Marcel Stanitzki21
Conclusion
TPAC 1.0 has been a success
− See response to Laser, 55Fe
− Proved deep p-well approach for MAPS
− Only minor problems found
− Finishing characterization
TPAC 1.1
− will be evaluated in the upcoming months
We plan to make full-reticle size sensor after that
− 2.5 x 2.5 cm