High-Voltage
Monolithic Active Pixel Sensors for the
Mu3e Experiment
Niklaus Berger
Institut für Kernphysik, Johannes-Gutenberg Universität Mainz
October 2018
• The Mu3e Experiment
• High-Voltage Monolithic Active Pixel Sensors (HV-MAPS)
• The MuPix Sensors
• The Mu3e Data Acquisition
Overview
The Mu3e Experiment:
Searching for μ
+→ e
+e
-e
+with a sensitivity of 10
-16(2∙10
-15in phase I)
e +
e + e -
• μ+ → e+e-e+
• Two positrons, one electron
• From same vertex
• Same time
• Sum of 4-momenta corresponds to muon at rest
• Maximum momentum: ½ mμ = 53 MeV/c
The signal
• Combination of positrons from ordinary muon decay with electrons from:
- photon conversion, - Bhabha scattering, - Mis-reconstruction
• Need very good timing, vertex and momentum resolution
Accidental Background
e
+e
+e
-• Allowed radiative decay with internal conversion:
μ
+→ e
+e
-e
+νν
• Only distinguishing feature:
Missing momentum carried by neutrinos
Internal conversion background
• Need excellent
momentum resolution
• New: NLO available from Matteo Fael and Signer et al. - now 10-20% easier
Branching Ratio
10-12
10-16 10-18 10-14
105 106 104
103 102
101
Internal conversion background
Signal
Building the
Mu3e Experiment
aiming for a branching ratio sensitivity of 10
-162 Billion Muon Decays/s
50 ns, 1 Tesla field
• Apply magnetic field (e.g. 1 Tesla)
• Measure curvature of particles in field
• Limited by detector resolution and scattering in detector
Momentum measurement
• Apply magnetic field (e.g. 1 Tesla)
• Measure curvature of particles in field
• Limited by detector resolution and scattering in detector
Momentum measurement
• At ~ 30 MeV/c momentum: Scattering completely dominates
• Large pixels: 80 μm
• Very little material: 0.1% X0 per layer
• Treat hit measurements as arbitrarily precise
• Consider scattering in each detector plane
• Two hits, two helices:
Underconstrained problem
• Minimize scattering angles
• Use multiple scattering theory to define χ2
Nucl. Instrum. Meth. A 844C, 135 (2017)
Multiple Scattering Track Fit
x y
z s
x0
x1
x2
c1 c2
r1 r2
d01 d12
s01 s12
Φ1 Φ2 ΦMS
ϕ01
ϕ12 B
x0
x1
x2
z01
z12 ϑ1
ϑ2 ΘMS B
ΦMS
• 1 T magnetic field
• Resolution dominated by multiple scattering
• Momentum resolution to first order:
Σ
P/P ~ θ
MS/Ω
• Precision requires large lever arm (large bending angle Ω) and low multiple scattering θMS
Momentum measurement
Ω MS
θ
MSB
Precision vs. Acceptance
50 MeV/c 25 MeV/c 12 MeV/c B→
33 cm
Precision vs. Acceptance
50 MeV/c 25 MeV/c 12 MeV/c B→
Precision vs. Acceptance
50 MeV/c 25 MeV/c 12 MeV/c B→
Precision vs. Acceptance
50 MeV/c 25 MeV/c 12 MeV/c B→
Precision vs. Acceptance
50 MeV/c 25 MeV/c 12 MeV/c B→
Ω ~ π MS
θMS
B
Detector Design
muon beam
target
Detector Design
muon beam
target
Detector Design
muon beam
target
inner pixel layers
Detector Design
outer pixel layers
muon beam
target
inner pixel layers
Detector Design
scintillating fibres
outer pixel layers
muon beam
target
inner pixel layers
Detector Design
outer pixel layers
muon beam
target inner pixel layers recurl pixel
layers
scintillating fibres
Detector Design
outer pixel layers
muon beam
target inner pixel layers recurl pixel
layers
recurl pixel layers
scintillating fibres
Scintillating tiles
2
] [MeV/c m
rec96 98 100 102 104 106 108 110
2
Events per 0.2 MeV/c
−3
10
−2
10
−1
10 1 10 10
2at 10
-12→ eee µ
at 10
-13→ eee µ
at 10
-14→ eee µ
at 10
-15→ eee µ
ν ν
→ eee µ
muons/s muon stops at 10
810
15Mu3e Phase I
Performance Simulations: Mass reconstruction
Work in
progress
High-Voltage
Monolithic Active Pixel Sensors
High voltage monolithic active pixel sensors - Ivan Perić
• Use a high voltage commercial process (automotive industry)
Fast and thin sensors: HV-MAPS
P-substrate
N-well E field
High voltage monolithic active pixel sensors - Ivan Perić
• Use a high voltage commercial process (automotive industry)
• Small active region, fast charge collection via drift
Fast and thin sensors: HV-MAPS
P-substrate
N-well E field
High voltage monolithic active pixel sensors - Ivan Perić
• Use a high voltage commercial process (automotive industry)
• Small active region, fast charge collection via drift
Fast and thin sensors: HV-MAPS
P-substrate N-well
Particle E field
• Implement logic directly in N-well in the pixel - smart diode array
• Can be thinned down to < 50 μm
(I.Perić, P. Fischer et al., NIM A 582 (2007) 876 )
Developed a series of HV-MAPS prototypes
• Goal: Detection and signal processing with just 50 μm silicon
• 6th chip, MuPix7, is a full system-on-a-chip
• Well characterized, working very nicely
• Next step is going big: 2 x 1 cm2 MuPix8 under test
The MuPix Prototypes
MUPIX electronics (MuPix7)
MuPix7
3 m m
MuPix7
3 m m
Pixels with amplifier
40 x 32 pixels
80 x 103 μm pixel size
MuPix7
3 m m
Pixels with amplifier
40 x 32 pixels
80 x 103 μm pixel size
Comparator and digital pixel logic
• Hits are streamed out on a 1.25 Gbit/s LVDS link
• Up to 30 MHz hits
• Tested up to 2.5 MHz - no loss of effi- ciency beyond single pixel dead-time (~ 1 μs)
Fully digital output
Rate [1000/s]
0 500 1000 1500 2000 2500
Efficiency
0.99 0.991 0.992 0.993 0.994 0.995 0.996 0.997 0.998 0.999 1
χ
−
−
− χ
−
−
−
Tests done at
• CERN 250 GeV pions
• DESY 5 GeV electrons
• PSI 250 MeV pions
• Mainz 855 MeV electrons
• Thanks for all the beam time and support!
Beam tests
Introduction
Y
• X
Introduction
Y
• X
• Beam test at DESY with 4 GeV electrons
• 50 μm sensor, 90° incidence
• Using high-resolution EUDET-Telescope as reference
• All features well understood
MuPix7 Performance: Efficiency
column direction [mm]
0.5 1 1.5 2 2.5 3
row direction [mm]
0 0.5 1 1.5
2 2.5 3
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Digital readout: Resolution given by pixel size (plus reference telescope resolution)
MuPix7 Performance: Spatial Resolution
MuPix7 Performance: Time Resolution
Time difference [ns]
−200 −180 −160 −140 −120 −100 −80 −60 −40 −20 0
Number of entries / 2 ns
2000 4000 6000 8000 10000 12000
Sigma = 12 ns Mean = -110 ns
• Using 16 ns timestamps
• Relative to scintillator reference
• Sizeable tail: time-walk
MuPix7 Performance: Time resolution
ToT length [ns]
0 500 1000 1500 2000
• Single pixel with time-over-threshold signal (~ signal size)
• MuPix8 has signal size for all pixels and finer timestamps
• Can do time-walk correction
• Measurements of time delay
(At fixed threshold: proxy for signal size) with sub-pixel resolution
• Simulation using TCAD: All features can be reproduced
MuPix7 Simulation and Data
Column direction [µm]
0 50 100 150 200
hit - trigger time [ns]
−104
−102
−100
−98
−96
−94
Simulation Measurement
1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
charge collection time [ns]
0 50 100 150 200
112
− 111
− 110
−
−109 108
−
−107
Simulation Measurement
0.8 0.9 1 1.1 1.2 1.3 1.4
Column direction [µm]
charge collection time [ns]
hit - trigger time [ns]
• MuPix8, the first large sensor (2 cm x 1 cm) now available
• Currently under test
• Three sub-matrices with different signal transmission to periphery
• Results from matrix A with the Mupix7-like source follower
MuPix8
MuPix8 Architecture
MuPix8 Performance
MuPix8 Performance
MuPix8 Performance
• Charge sharing only at pixel edges
MuPix8 Performance
• Resolution given by pixel size (80 x 81 μm)
MuPix8 Performance
• 8 ns timestamps
• Some delays over the chip, large pixel- to-pixel variations: Need correction
• Further improvements possible, for matrix subset, 6 ns were obtained
MuPix8 issues
• Powering: Some voltage drop over chip, results obtained at 1.9 V or 2 V vs. 1.8 V nominal operation voltage
• Cross-talk: Long lines to the periphery have capacitive coupling
50 μm silicon is not self-supporting
• Need “no-mass” mechanics
• Also: “no-mass” connection to the outside world
See Joost’s talk
Chips are active: ~ 300 mW/cm2
• Need “no-mass” cooling
• Gaseous helium at very high flow speeds
• Prototype tests so far successful, full mock-up under construction
How to get to ~0.1 X
0per layer
• Note: The PANDA luminosity detector will operate MuPix in vacuum:
Cooling via diamond wafers
Data Acquisition
• 1.25 Gbit/s 8b10b encoded LVDS links
• Either three submatrices with a link each or one link multiplexing the sub-matrices
• Roughly 30 MHits/s per link maximum
• Hits are 32 bit: column, row, time, charge
• Hits are not strictly time sorted - see backup for the workings of the MuPix readout state machine
MuPix output
Phase I:
• 280 Million pixels (+ fibres and tiles)
• No trigger
• ~ 100 Gbit/s
• FPGA-based switching network
• 12 PCs with GPUs
Data Acquisition
2844 Pixel Sensors
up to 45 1.25 Gbit/s links
FPGA FPGA FPGA
...
86 FPGAs
1 6 Gbit/s link each
GPU
PC GPU
PC
GPU 12 PCs PC
4 12 Gbit/s links per
16 Inputs each
3072 Fibre Readout Channels
FPGA FPGA
...
12 FPGAs
6272 Tiles
FPGA FPGA
...
14 FPGAs
Data Collection
Server
Mass Storage Gbit Ethernet
Switching Board Switching
Board
Front-end(inside magnet)
Switching Board
Switching Board
Switching Board
Front end (Altera Arria V FPGAs):
• Receive and decode data
• Correct for time-walk
• Time sorting (most resources)
• Slow control and configuration
• Send data out via 6 Gbit/s optical link
Data Acquisition
2844 Pixel Sensors
up to 45 1.25 Gbit/s links
FPGA FPGA FPGA
...
86 FPGAs
1 6 Gbit/s link each
GPU
PC GPU
PC
GPU 12 PCs PC
4 12 Gbit/s links per
16 Inputs each
3072 Fibre Readout Channels
FPGA FPGA
...
12 FPGAs
6272 Tiles
FPGA FPGA
...
14 FPGAs
Data
Collection Mass Storage Gbit Ethernet
Switching Board Switching
Board
Front-end(inside magnet)
Switching Board
Switching Board
Switching Board
Switching (Altera Arria 10 FPGAs):
• PCIe40 board (Marseille, LHCb and ALICE)
• Merge datastreams
• Inject pixel configuration data
• Perform monitoring tasks
Data Acquisition
2844 Pixel Sensors
up to 45 1.25 Gbit/s links
FPGA FPGA FPGA
...
86 FPGAs
1 6 Gbit/s link each
GPU
PC GPU
PC
GPU 12 PCs PC
4 12 Gbit/s links per
16 Inputs each
3072 Fibre Readout Channels
FPGA FPGA
...
12 FPGAs
6272 Tiles
FPGA FPGA
...
14 FPGAs
Data Collection
Server
Mass Storage Gbit Ethernet
Switching Board Switching
Board
Front-end(inside magnet)
Switching Board
Switching Board
Switching Board
PCs (Altera Arria 10 FPGAs):
• DE5a-NET board (Terasic Inc.)
• Receive data, preprocess
• DMA to GPU
• Buffering
Data Acquisition
2844 Pixel Sensors
up to 45 1.25 Gbit/s links
FPGA FPGA FPGA
...
86 FPGAs
1 6 Gbit/s link each
GPU
PC GPU
PC
GPU 12 PCs PC
4 12 Gbit/s links per
16 Inputs each
3072 Fibre Readout Channels
FPGA FPGA
...
12 FPGAs
6272 Tiles
FPGA FPGA
...
14 FPGAs
Data
Collection Mass Storage Gbit Ethernet
Switching Board Switching
Board
Front-end(inside magnet)
Switching Board
Switching Board
Switching Board
• 280 Million pixels (+ fibres and tiles)
• No trigger
• ~ 1 Tbit/s
• Need to find and fit billions of tracks/s
Online reconstruction
• PCs with Graphics Processing Units (GPUs)
• Online track and event reconstruction
• 109 3D track fits/s achieved
• Data reduction by factor ~1000
• Data to tape < 100 Mbyte/s
Online filter farm
• Mu3e aims for μ → eee at the 10-16 level
• First large scale use of HV-MAPS
• Working full prototypes MuPix7 and MuPix8
• Reconstruct 100 million tracks/s in 100 Gbit/s on ~12 GPUs
• Start data taking in 2020
• 2 billion muons/s not before 2024
Conclusion
Backup Material
• Add no material:
Cool with gaseous Helium (low scattering, high mobility)
• ~ 250 mW/cm2 - total ~3 kW
• Simulations: Need ~ several m/s flow
Cooling
• Full scale heatable prototype built
• 36 cm active length
• Vibrations studied using Michelson-Interferometer
• Can keep temperature below 70°C
Cooling tests
Global helium stream
Local helium stream
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Readout
Niklaus Berger – μEDM October 2018 – Slide 93
Introduction
Y
• X
State Machine
StateSync Reset
sendcounter True StateSendCounter1 StateSendCounter2
sendcounter StatePD1 True
StatePD2
StateLDCol1
StateLDCol2
StateLDPix1
PriFromDet slowdownend
True
False False
StateLDPix2 True
StateRDCol1 PriFromDet False
True
StateRDCol1
PriFromDet True maxcycles
False
StateSync Reset
sendcounter True StateSendCounter1 StateSendCounter2
sendcounter StatePD1 True
StatePD2
StateLDCol1
StateLDCol2
StateLDPix1
PriFromDet slowdownend
True
False False
StateLDPix2 True
StateRDCol1 PriFromDet False
True
StateRDCol1
PriFromDet True maxcycles
True False
False
• 90° incidence angle
• 99% efficient for less than 10 Hz noise per pixel
MuPix7 Performance: Efficiency vs. Noise
Threshold [V]
0.7 0.71 0.72 0.73 0.74 0.75
Efficiency
0.95 0.96 0.97 0.98 0.99 1
Efficiency Noise
99 %
Preliminary
Noiserate per pixel [1/s]
−1
10 1 10 102
103
104
• 90° incidence angle
• 99% efficient for less than 10 Hz noise per pixel
• 45° incidence angle
• 99% efficient for less than 1 Hz noise per pixel
• MuPix8 has higher resistivity substrate:
45° signal at 90°
MuPix7 Performance: Efficiency vs. Noise
Threshold [V]
0.7 0.71 0.72 0.73 0.74 0.75
Efficiency
0.95 0.96 0.97 0.98 0.99 1
Efficiency Noise
99 %
Preliminary
Noiserate per pixel [1/s]
−1
10 1 10 102
103
104
Efficiency
0.975 0.98 0.985 0.99 0.995 1
Efficiency Noise
99 %
Preliminary
Noiserate per pixel [1/s]
−1
1 10