Future silicon trackers: 4D tracking, very high fluences, very small - - PowerPoint PPT Presentation
N. Cartiglia, INFN. Terascale meeting - 27-Nov-2019 Future silicon trackers: 4D tracking, very high fluences, very small pixels Nicol Cartiglia INFN - Italy 1 Outline N. Cartiglia, INFN. Terascale meeting - 27-Nov-2019 A brief history of
1
Nicolò Cartiglia INFN - Italy
2
3
The beginning of the Silicon detector era is set in the period 1978-1982, and the NA11/NA32 experiments are credited to be the first one to have used a silicon tracker Shortly after, successful tests of silicon strip detectors with VLSI readouts were carried out in 1985. During the 1990s, CDF and the LEP experiments were instrumented with Silicon trackers, with the electronics at the edges. Here at DESY, we even manufactured a curved silicon detector, to be placed near the proton beam. The ZEUS experiment was also instrumented with the silicon vertex detectors
4
This incredible evolution was made possible by the development of the “silicon” industry and by the collaboration of our community with several silicon foundries
5
Taken from Doris Eckstein
Incredible development of manufacturing capability Very good understanding of the silicon properties under irradiation: modelling of silicon detectors and the effect of irradiation is well modelled. Similar development in read-out capability HL-LHC: the CMS-ATLAS upgrades are very large, however, they are in spirit similar to the present LHC detectors. Higher radiation levels, more channels and much more performing electronics. One novel request: need to measure the time of each track, to bundle correctly the tracks of each vertex.
6
There are many futures in Silicon trackers: some are clever redesigns of existing systems, some requires much higher radiation tolerance, some extremely good position resolution. One of the most challenging design: the Future Circular Collider tracker Tracker requirements: position: 7.5 - 9.5 μm time resolution = 5 ps Radiation levels: up to ~1E17 n/cm2 Note: there are many R&D directions in Silicon detectors. This presentation is not a review but it is about a possible future.
7
First question: Can we design a single detector that can concurrently measure (a) time with ~ 10 ps precision (b) position with ~ 10 micron precision This is an extraordinary challenge in sensor design and ASICs Second question: can we make silicon detectors able to work at fluences about 1E16 – 1E17 n/cm2? A lot has been understood regarding the design of radiation hard silicon detectors, with a key contribution from Hamburg, however, currently we don’t know how to do design a sensor for extreme fluences, F = 1E16 – 1E17 n/cm2
8
Silicon sensors were never considered accurate timing devices However, in the last 10 years there has been a very intense R&D At present, silicon sensors are the ONLY detector able to provide excellent timing capability (~ 30 ps) , good radiation hardness (fluence ~ 1E15 n/cm2), good pixelation (10um – 1 mm), and large area coverage (many m2) Important: Sensors provide the current signals, read-out chips use them Timing is the to combination of these two parts, that succeed and fail together
9
The inclusion of track-timing in the event information has the capability of changing radically how we design experiments. Timing can be available at different levels of the event reconstruction, in increasing order of complexity: 1) Timing in the event reconstruction è Timing layers (time, position)
2) Timing at each point along the track è 4D tracking (time, position)
3) Timing at each point along the track at high rate è 5D tracking (time, position, and rate)
chip and data output organization
Timing
10
Signal shape is determined by Ramo’s Theorem:
Drift velocity Weighting field
The charge carriers motion induces variable charge on the read-out electrode. The signal ends when the charges are collected
Induced charge
++++ ++++++
11
The goal of a sensor designer is to minimize the differences in the sensor’s output, providing well defined, uniform current signals to the electronics. The prerequisite for this goal is the capability of simulating the physics of the particle-sensor interaction. Chip designers need to test their solutions on a realistic sets of current signals that reproduce the full variability of the sensor’s output. Good sensor simulation is necessary to achieve excellent time resolution
12
Available at: http://personalpages.to.infn.it/~cartigli/Weightfield2/Main.html It requires Root build from source, it is for Linux and Mac. It will not replace TCAD, but it helps in understanding the sensors response
13
Weightfield2:
It crashes occasionally Other simulators: KDrtSim, https://indico.desy.de/indico/event/12934/session/3/contribution/26/material/slides/ TRACS https://indico.desy.de/indico/event/12934/session/3/contribution/29/material/slides/
14
Calculating the correct weighting field for a variety of situations is a very difficult task. Most of the time we rely on simulator to do it. Please note a series of papers that are approaching this problem analytically:
silicon sensors” Nucl.Instrum.Meth. A940 (2019) 453-461 arXiv:1812.07570 Academic training at CERN: https://indico.cern.ch/event/843083/ Joern Schwandt, Robert Klanner, On the weighting field of irradiated silicon detectors, https://arxiv.org/abs/1905.08533
The timing capabilities are determined by the characteristics of the signal at the output of the pre-Amplifier and by the TDC binning.
15
Time is set when the signal crosses the comparator threshold
(a simplified view)
Strong interplay between sensor and electronics
16
Usual “Jitter” term Here enters everything that is “Noise” and the steepness of the signal Amplitude variation: variation in the total charge Shape distortion: non homogeneous energy deposition total current electron current hole current total current electron current hole current
Need large dV/dt
!"
# =
%&'() *+/*"
#
+ ∆'&/'01"'&/ # + ∆(213) # + 456 #
Sensor design Subleading, ignored here
17
#
Signal shape is determined by Ramo’s Theorem:
Drift velocity Weighting field The key to good timing is the uniformity of signals: Drift velocity and Weighting field need to be as uniform as possible Basic rule: parallel plate geometry
18
The initial current for a silicon detector does not depend
−6 A
Number of e/h = 75/micron Weighting field velocity
è Initial current = constant
(Simplified model for pad detectors)
D d +
However, the weighting field is 1/d, so each charge contributes to the initial current as:
19
(Simplified model for pad detectors)
D d +
i(t)
Thin detector Thick detector
S
We need to add do something about this problem…
20
The study of the signal in silicon sensors has highlighted a few crucial aspects:
steeper using thinner/thicker sensors
capacitor
21
Gain in silicon detectors is commonly achieved in several types of sensors. It’s based
Silicon devices with gain:
Charge multiplication Gain:
( ) ( )
÷ ÷ ø ö ç ç è æ- ¥ = E b E
h e h e h e , , ,
exp * a a
a = strong E dependance a ~ 0.7/µm for electrons, a ~ 0.1/um for holes
+ +
+ +
+
Concurrent multiplication of electrons and holes generate very high gain
22
The LGAD sensors, proposed and manufactured for the first time by CNM (National Center for Micro-electronics, Barcelona): High field obtained by adding an extra doping layer E ~ 300 kV/cm, closed to breakdown voltage LGAD optimized for timing applications are often called Ultra Fast Silicon Detector (UFSD)
zoom
Drift area with gain 0.5 – 2 um long
Gain implant Gain layer
E field Traditional silicon detector Low Gain Avalanche Diode E field
23
+
+
absorbed immediately Gain holes: long drift home Initial electron, holes
Electrons multiply and produce additional electrons and holes.
è No holes multiplications
24
The rate of particles produced by the gain does not depend on d (assuming saturated velocity vsat)
Particles per micron Gain
Gain layer
è Constant rate of production è Gain current ~ 1/d However the initial value of the gain current depends on d (via the weighing field)
A given amount of new carriers has much more effect on thin detectors
Gain Layer
+ +
25
è Go thin!!
(Real life is a bit more complicated, but the conclusions are the same)
300 micron: ~ 2-3 improvement with gain = 20
Significant improvements in time resolution require thin detectors
26
i(t) The rise time depends only on the sensor thickness ~ 1/d thin medium
t t1 t2 t3 thick
27
The UFSD advances via a series of productions. For each thickness, the goal is to obtain the intrinsic time resolution Achieved:
Resolution without gain
UFSD1 UFSD2, 3
The higher tail of the Landau distribution is populated by events with very high ionization. These events contain a strong secondary ionization component, such as that caused by delta rays.
28
The shape of the events in these bins varies a lot, so they have worse time resolution
signals generated by 120GeV/c pions crossing a 50 micron thick UFSD
29
Vholes never saturates, so higher the voltage, better dV/dt is
The rise time depends
velocity
The amplitude depends on:
!"##$%
&
= ()*+$ ,-/,#
&
~ ()*+$ 0/#%*+$
&
trise A
The combination of gain and bias determines dV
Bias: 150V 50 micron 530V
30
5 10 15 20 25 30 35 40 100 200 300 400 500 600 700 800
Gain Bias [V]
30-35 ps 35-40 ps 40-45 ps 45-50 ps 50-60 ps
At lower bias, higher gain is needed to achieve a resolution of 30-35 ps HPK 3.2: too doped, very poor time resolution
Time resolution for new UFSD FBK & HPK sensors in the bias-gain plane
31
Currently there are 6 companies that either have produced, or are about to produce LGADs FBK, Italy CNM, Spain Hamamatsu, Japan Up coming: Brookhaven National Lab, USA NDL China Micron, England Maybe more
32
Current Amplifier Charge Sensitive Amplifier
Current signal in a 50 mm sensor Energy deposition in a 50 mm sensor
bumps
signal smoothing WF2 simulation WF2 simulation
33
Pads with gain Current due to gain holes creates a longer and higher signal
2 sensors
Pads with no gain Charges generated uniquely by the incident particle
Simulated Weightfield2
Oscilloscope
Gain 50 micron Much easier life!
To fully exploit UFSDs, dedicated electronics needs to be designed. The signal from UFSDs is different from that of traditional sensors
No Gain, 300 micron No 300 micron WF2 simulation
34
very complex operation
nm, 65 nm, 28 nm, SiGe, monolithic etc)
35
Particles hitting a pad create charges underneath the multiplication layer: no delay between the passage of the particle and the start of multiplication
p-stop
p bulk
Gain implant n++
+
create charges far from the multiplication layers: they generate late signals
p-stop
p bulk
Gain implant n++
+
~50 um As in every n-in-p sensor, the pads need to be isolated Unwanted consequence: late signals
36
Solution: add a deep n implant to collect the charges in the interpap
p-stop
p bulk
Gain implant n++
+
multiplications:
interpad signals 10 um = 100 ps No gain area We solved the “isolation” and the “late signals” problems, however we have created a “no gain” area of ~ 30-40 micron: impossible to make small pixels!
37
No gain area JTE + p-stop design Trench design R&D goal Current version
Trenches (the same technique used in SiPM):
FBK run @ RD50
38
Key points to achieve excellent position and timing performances.
39
Irradiation causes 3 main effects:
LGAD are particularly sensitive to the doping creation/removal, as it changes the electric field and therefore the gain.
40
Nicolo Cartiglia, INFN, Torino – Tracking in 4D
The amount of doping in the gain implant strongly affect the gain value +3% doping doubles the collected charge
The bias can be adjusted to keep the charge constant as the doping in the GL changes.
41
Unfortunate fact: irradiation de-activate p- doping removing Boron from the reticle
Boron Radiation creates interstitial defects that inactivate the Boron: Si_i + B_s è Si_s + B_i B_i might interact with Oxigen, creating a donor state Gallium From literature, Gallium has a lower probability
Carbon Carbon competes with Boron and Gallium in reacting with Oxigen
Two possible solutions: 1) use Gallium, 2) Add Carbon
42
Defect Engineering of the gain implant
Modification of the gain implant profile
are less prone to be inactivated
Boron + Carbon Gallium + Carbon Gallium Boron (High Diff) Boron (Low Diff)
Doping Concenration (a.u).
Boron profile
Acceptor removal is no understood from a microscopic point of view
43
44
~ 16000 sensors:
7 m2 of sensors
First detector for precision timing in Silicon
45
Very large volume with fluence in the range 1E16-1E17 n/cm2
46
Irradiation causes 3 main effects:
(F~5E15 n/cm2) ) VFD >> VBias
750V
(New) VFD << VBias
200V 500V
(F~1E15 n/cm2) VFD ~ 0.5 VBias
Irradiation models developed in the fluence range 1E14 – 1E15 n/cm2 predict standard silicon detectors (~ 200 um thick) almost impossible to operate è Mission impossible Partially depleted at very high Vbias Fully depleted at low Vbias Fully depleted at high Vbias
47
At high fluence the damage might saturate since clusters of damage start overlapping Overlapping clusters
48
Probability of hitting an empty square of area 1 Å2
At 1E16 n/cm2 only 30% of particles will hit an “empty square” Note: Silicon lattice has a cube of 5 Å; every cell has already been hit at 1E15. Damage on damaged Silicon probably has different consequences.
49
There is a consensus building that:
Marko Mikuž at AIDA2020 Topical Workshop on Future
Charge trapping
Saturation is a key aspect of the R&D in the next few years, we should learn how to take advantage of this effect The bottom line is: Silicon detectors irradiated at fluences 1E16 – 1E17 n/cm2 do not behave as expected, they behave better
50
What does it happen to a 25-micron sensor after a fluence = 5E16 n/cm2?
However: Charge deposited ~ 0.25 fC è Need a gain of at least ~ 5 in order to provide enough charge
51
Use thin (25 um) LGAD sensors: as the irradiation deactivate the gain layer, increase bias to obtain gain in the bulk Below 5E15 n/cm2 è Use LGAD design to obtain a gain of ~ 5 without breakdown è Vbias controls gain
Above 5E15 n/cm2
è is the gain still there? è Is the mobility decreasing to a point where no gain is possible? è Damaged bulk acts as a quenching resistor? è No holes multiplications?
+++ +++++
n-in-p
Gain:
+++++ +++++
n-in-p
Gain:
Trenches
(F~1E16 n/cm2) ) VFD < VBias
500V
52
Future trackers need to provide:
A very simplified design:
Ø Position, thin sensors, with small gain
Ø timing layers Ø Position, thick sensors Note: Limited number of timing layer: probably, they require too much power
53
Silicon sensors for tracking come in many shapes, fitting very different needs:
Likewise, silicon sensors for time-tracking are being developed to fit different needs with respect of time and space precision. Depending on the amount of optimization, several resolution ranges can be identified
54
There are no real alternatives: hopefully, as it happened in the past, Silicon detectors will be the enabling technology to new discoveries
55
We kindly acknowledge the following funding agencies, collaborations: Ø INFN - Gruppo V Ø Horizon 2020, grant UFSD669529 Ø Horizon 2020, grant no. 654168 (AIDA-2020) Ø U.S. Department of Energy grant number DE-SC0010107 Ø Dipartimenti di Eccellenza, Univ. of Torino (ex L. 232/2016, art. 1, cc. 314, 337)
56
57
The electrons’ drift velocity saturates at about ~ 150V. The holes’ drift velocity never
Higher bias è higher dV/dt
Bias = 500V, gain = 9 Bias = 170V, gain = 9 Bias = 120V, gain = 9 low bias high bias medium bias
Equal rise time: saturated electron drift velocity
For equal gain, better resolution at higher voltage
58
To some extent, the gain layer disappearance might be compensated by increasing the bias voltage
Acceptor removal, Gain layer deactivation
59
Acceptor removal coefficient
Puzzle: the removal of acceptor depends on their density è the removal is slower for higher densities
60
10% t1 t2 Vth t t V V
Constant Fraction Time-over-Threshold
(a) (b)
On paper both seem feasible, in practice ToT is much easier to implement What is the influence of the sensor on the level of the CFD or of Vth?
10% t V V
Constant Fraction
(a)
Amplitude My favorite: ToA and Amplitude è The tail of the signal is prone to large changes due to charge trapping ToA
(C)
61
However the shape of the signal depends on the thickness d: thinner detectors have higher slew rate D + - d + -
(Simplified model for pad detectors)
Let’s consider one single electron-hole pair. The integral of the current is equal to the electric charge, q:
i(t) t
Thin detector Thick detector
è One e/h pair generates higher current in thin detectors Weighting field
62
Common RD50 run with FBK: preproduction run demonstrated the proof of principle: nice isolation and nice gain
63
Gain happens when the Efield is near the critical values, 300 kV/cm 3 methods to increase Efield: 1. Doping in the bulk 2. Doping in the gain layer 3. Bias