Download Characterization of Silicon Photodetectors (Avalanche Photodiodes

Characterization of Silicon
Photodetectors (Avalanche Photodiodes
in Geiger Mode)
at Fermilab
G. Mavromanolakis, A. Para. N.Saoulidou
Novel PhotDetectors 07,
Kobe, June 27, 2007
Goals



Develop a complete characteristics of the detector response to
the external light signal
 As a function of the light source characteristics (intensity,
duration, time structure)
 As a function of the operating conditions (voltage,
temperature)
 light impact point onto the detector (inter/intrapixel
uniformity)
Develop algorithm for readout strategy and calibration procedure
(integration time, cross-talk, after-pulses tratement, etc..)
Feed back the information to the manufacturers
Step 1: Database of Static
Characteristics


Develop an automated procedure for static characterization
(breakdown voltage, resistance) as a function of the operating
temperature
 Keithley 2400 source-meter
 Dark box
 Peltier cold plate
 Labview controls/readout
Create a database of the samples, enter the static and image
data
I-V Characteristics at Different
Temperatures


Different
detectors have
quite different
operating point
Dark current and
the operating point
depend on
temperature
Breakdown Voltage: a Knee on the
I-V plot?
Linear or logarithmic
plot (derivative)?
 Two voltage scales with
about 4-5 V saparation
 Operating point
somewhere in the middle
(left half)

Step 2: Characterization of the
Detector Response to a Light Pulse


Light source:
 Short pulse duration (<1 nsec)
 Variable light intensity (0.1 – 1000 photons)
 Absolute light calibration (still in the works)
Readout strategy:
 Trans-impedance amplifier ( MITEQ amplifiers: AU-2A0159, AU-4A-0150, AM-4A-000110)
 Tektronix 3054B digital scope
 LabView DAQ and analysis program
 Root-based analysis environment
GMAPD Characterization Setup
Disclaimer





The following ‘results’ are primarily meant to
demonstrate the scope of the data collected.
Two analysis chain (root-based and Labview-based)
have been just established
The results and tentative conclusions are preliminary
and are meant as ‘food for thought’
Results shown will be for Hamamatsu MPPC 025U
detector
There are much more results coming, the analysis just
started last week.
Snapshot of Several Regimes at the
Same Time



-2.0 – 0 msec: ‘quiet state’ of the MPPC:
 Dark rate
 Gain
 Cross talk, afterpulses
‘Laser gate’:
 Response to the light input
 Cross talk
 Afterpulses
‘Post laser gate’
 Afterpulsing, recovery
Examples: 71.0 V
Examples: 72.0 V
Examples: 72.5 V
Salient Features: Detector
Instabilities




Often called cross-talk, afterpulsing, etc.
These instabilities determine the nature of the
response of the detector in a manner which depends
on the temporal characteristic of the measured light
and/or on the characteristics of the read-out
electronics
It is very important to understand their origin and to
reduce their incidence
Tens of years of R&D for the SPAD detectors should
be of great help
(Naïve Understanding) ofTwo Popular
Models



Photon-mediated cross talk: Infrared photons created in the
avalanche initiate a response in the neighboring pixels.
 Remedy: trenches for optical isolation
 Naïve expectation this cross-talk will be ‘in-time’ with the
original signal. This is probably a very small effect.
Carriers produced in the avalanche trapped in traps. Traps have
finite lifetime and release electrons which create subsequent
avalanches.
 Remedy: long recovery time of a pixel
There are likely more effects which need to be understood.
Operating voltage seems to be of critical importance.
Time arrival of avalanches


Response to the
initial light
impulse creates
a long chain of
‘afterpulsing’
spread over 0.51 msec
This effect
grows very
rapidly with the
bias voltage.
Shown in 0.5 V
steps.
‘Quiet Time’ – Thermal ElectronsInduced Avalanches?



Count pulses – ‘dark rate’
Look at the time-difference between pulses: evidence
for correlation at the scale of ~100 nsec
Fraction of single pulses + Poisson statistics =>
calculate afterpulsing probability
Dark Count Rate vs Voltage



Raw dark count rate rises dramatically with voltage
Afterpulsing probability rises almost linearly (in the
studied range) with voltage
All of the increase of the dark rate is consistent with
being the result of increased afterpulsing, whereas
the rate of the ‘true’ dark pulses is~75K, voltage
independent
Single (Isolated) Dark Pulses: SelfCalibration of the Detector




Detect pulses in the ‘quiet time’
Plot the peak value of the detected pulses:
 DV/V ~ 8-10%
Integrate the charge within some gate (8ns)
 To reduce impact of the afterpulsing require no
other pulse within 50 nsec
 DQ/Q ~ 10-15%
Width of the ‘calibration pulses’ represents
uniformity of the response over the front face of the
detector
Single (Isolated) Dark Pulses: SelfCalibration of the Detector
With longer gate or
higher voltage a long
tail and a double
avalanche peak
appear
Detector Gain Vs Voltage


Gain = Q/e
Q = C (V-Vbr)
Insights about the IV plot?
Break-down
voltage of the
detector
Increase of gain x
(mostly) increase of
afterpulsing
Afterpulsing
probability ~ 1,
run-away
Dark Counts: Comment About the
Rates


71.5 V, integration gate of 50
nsec
Dark count rate: what is the
reduction when cutting at 1.5
pe?? It depends on the
definition of ‘rate’:
 Factor of 30-50 if
measure the amplitude,
bias voltage dependent
 Factor of 5-10 if measure
integral within some gate
(gate dependent)
Dark Counts: Afterpulsing
Mechanisms?





71.5 V, integration gate of 50
nsec
An unique laboratory for studies
of afterpulses: starting with a
single electron avalanche, single
pixel
Evidence for ‘instantenous’ (like
photon mediated) cross talk at
few percent level, (depending on
the bias voltage)
Integrated charge: ‘fractional’ +
‘whole’ avalanche. Combination of
two effects:
 Smaller charges produced.
Avalanche in not-fullyrevered pixel?
 Cutting the pulse at the edge
of the gate
Merits further studies.
Amplitude/Peak Value in the Laser
Gate vs Voltage
‘Laser Gate’:Integrated Charge vs
Bias Voltage
Analysis of the ‘Laser Gate’ Data




Two independent methods:
 Take the peak value
 Integrate the charge within a given gate (30 nsec)
shown thereafter
Use Fourier analysis to determine the fundamental
frequency
Automatically partition the spectrum into 0-1st-2nd3th-etc… peak
Compare with the expected Poisson distribution. Any
additional contributions (like afterpulses) will shift
the distributions towards the higher values
Detection Efficiency vs Bias Voltage



Fractional content of the ‘zero’ bin -> average number
of photons detected
Good agreement between ‘charge’ and ‘amplitude’ –based
measurement
PDE increases by a factor of ~ 1.5 between 71 V and
72.75 V
Reconstructing the Poisson
Distribution (Charge and Amplitude)
Charge/amplitude Spectrum at 71 V

Both charge (integral in 30 nsec gate) and the
amplitude spectrum follows Poisson distribution
Charge/Amplitude Spectrum at 71.5 V

Amplitude spectrum follows Poisson distribution,
wheras integrated (30 nsec) charge distribution
shows deviations due to afterpulsing distribution
Charge/amplitude distribution at 72.5 V

Charge distribution completely dominated by
afterpulsing (to the point that automated Fourier
analysis fails miserably), whereas amplitude spectrum
still shows no deviations from Poisson
Lessons From the Charge/Amplitude
Comparison





Amplitude spectrum is not sensitive to afterpulsing but it is
sensitive to the instantaneous cross-talk (like photon-mediated).
Amplitude spectrum shows no deviation from Poisson, hence the
‘instantaneous’ cross-talk is a very small effe
Charge distribution is very sensitive to the integration gate.
Even with 30 nsec gate it shows huge effects (and deviations
from Poisson distribution).
This effect is a very strong function of the bias voltage. The
time scale (30 nsec) seems to indicate that these afterpulses
are generated in different pixels (no trapping model)
What is the mechanism of cross talk at the timescales of the
order of tens of nsecs???
Improvements ‘in the Works’


Environmetal
chamber for the
temperature control
and study the
temperature
dependence
Independent
measurement of the
input light intensity
Conclusions




Waveforms produced by GMAPD’s under different conditions
provide wealth of information about properties and behavior of
the detectors
Detector instabilities (cross-talk, afterpulsing) are major
contribution to the detector response. Their practical
consequences depend on the readout details and the
experimental conditions (temporal structure of the measured
light pulses)
Physics of afterpulsing is likely to be rather complicated
A lot more understanding can be achieved from the existing
setups/data, but (perhaps) special detector samples (different
recovery time, different number of pixels of a given size) could
be very helpful in disentangling various contributions.