Single Photon Imaging Using a CCD and Electron Multiplication
Dr Mark RobbinsPredevelopment Technology Manager, Imaging Division
Invited TalkFrontiers in Electronic Imaging 15th –16th June 2009
Slide 2
Introduction
CCD have been around for 40 yearsSignificant advances in manufacture, performance and architecturesElectron Multiplying CCDs (EMCCDs) commercially available for 10 years
EMCCDs are currently the technology of choice for many photon starved applications demanding
Sub electron readout noiseLow noise factor Achievable at high data/frame rates
High quantum efficiency and high fill factor (up to 100%)
This talk presents the EMCCD and associated technology together with some novel application areas
Slide 3
EMCCD Concept
For imaging in a photon starved environment sources of noise must be minimised
Shot noise on input signalShot noise on dark signalNoise from charge to voltage conversionSpurious noise from video chain
Shot noise on input minimised by efficiently converting photons to signal electrons.
Shot noise on dark signal minimised by reducing dark signal
EMCCD technology reduces the effect of charge to voltage conversion noise and noise from the video chain
Slide 4
Amplifier Noise
Noise introduced by the charge to voltage conversion process
Resetting the node introduces “kTC” noiseEliminated by the use of correlated double sampling (CDS)
Left with noise of the source follower FETs. Noise Equivalent Signal (NES) given approximately by
NES (e rms) ≈ 0.5 (1 + f/f0)1/2 Cn1/2
f = pixel rate, f0 ~ 150 kHz and Cn = node capacitance (fF)
readout register
Reset FET
Output FET
outputnode
φR RD OD
readout register
Reset FET
Output FET
outputnode
φR RD OD
Slide 5
Amplifier Noise
To get low noise from a conventional device need to run in slow scan mode.
Can non-destructively sample the output many times (skipper)
Can use column parallel approach to increase frame rate.
Difficult to reduce noise for single photon imaging.
0
5
10
15
10 100 1,000 10,000
Readout Rate (kHz)
inpu
t ref
erre
d no
ise
(ele
ctro
ns) CCD30 Spectroscopic - Measurement
CCD30 Spectroscopic - Theory
CCD44 Astronomy - Measurement
CCD44 Astronomy - Theory
Slide 6
Conventional CCD
Iφ1Iφ2Iφ3
image/store areas
Rφ1 Rφ2 Rφ3
readout register
charge to voltageconversion
Slide 7
EMCCD CCD
Rφ1 Rφ2 Rφ3
Iφ1Iφ2Iφ3
multiplicationregister
Rφ2HV
Slide 8
Single Photon Imaging: EMCCD technology
Signal clocked through many gain elements, each with a probability of inducing impact ionisation, α.
( ) elements gain of number the is 1 NG Nα+=
φ2-φDC (Volts)
40.5 41.0 41.5 42.0 42.5
Mul
tiplic
atio
n G
ain
0
200
400
600
800
1000
1200
10 oC 25 oC
CCD97536 Multiplication Elements
If N = 536 1000x gain reached with an α of 1.3%
α is clock amplitude and temperature dependant
Slide 9
Single Photon Imaging: EMCCD technology
Single carrier impact ionisation is a very low noise processcan “noiselessly” increase the signal level above the amplifier noise, σamp
effective noise given by
( ) 2
22
GF amp
dsσ
μμσ ++=
µs = mean signal
µd = mean dark signal
G = multiplication gain
F = “Noise Factor”
0.01
0.1
1
10
38 39 40 41 42 43 44Rφ2HV-RφDC (Volts)
NES
(rm
s e)
10
100
1000
10000
Mul
tiplic
atio
n G
ain
Slide 10
Single Photon Imaging: EMCCD technology
Noise from a BI CCD97 operating at 35 frames per second
Data courtesy of Andor Technology
Slide 11
EMCCD technology: The F factor
The Noise Factor, Fcaused by statistical fluctuations in the gainmuch lower than conventional APDs or intensifiers etc.
( ) ( )( )
GGG
GG
F
NN 112
1121
/1
2
+−=
⎟⎠⎞
⎜⎝⎛
+−+
=
+−
αα
Slide 12
EMCCD technology: The F factor
The effect of the noise factor
Increases the width of the ideal shot noise distribution by 1.4
For very low photon fluxes can be eliminated by “photon counting” (counting output when above threshold)
Noise from the EMCCD will be less than a CCD if Signal < (Amp Noise)2
Not always the whole story as no account taken of noise on background or object visibility.
Output Signal (electrons)0 2000 4000 6000 8000 10000
Prob
abili
ty D
ensi
ty (e
-1)
0.0000
0.0001
0.0002
0.0003
0.0004
5e 4e 3e 2e 1e
1000x gain, F2 = 2Mean Input
Output Signal (electrons)0 2000 4000 6000 8000 10000
Pro
babi
lity
0.0
0.1
0.2
0.3
0.4
5e 4e 3e 2e 1e
1000x gain, F2 = 1Mean Input
Slide 13
EMCCD technology: The F factor
Rose criterion used to quantify visibility of a feature in a noisy image
Assumes noise on background = noise in featureRose criterion must be modified to compare emccd with conventional ccd
probability
ΔSfeature
Sbackground
5 ifcertainty 100% withvisible >Δ
= RSNRbackground
feature
σ
Slide 14
Figure of Merit for uniform feature visibility is proposed as
N is the number of pixels in a feature and accounts for the “averaging” performed by the eye.
A more sophisticated approach would take into account different noise distributions visibility due to noise difference.
However, useful for a comparison.
EMCCD technology: The F factor
backgroundfeature
featureSNVσσ +Δ
=2
Slide 15
EMCCD technology: The F factor
For a conventional device with no gain we have
For an emccd where gain >> amp noise
The visibilities, Vccd and Vemccd, are equal when
( ) ( )22
2
ampbackgroundampbackgroundfeature
featureccd
SSS
SNVσσ ++++Δ
Δ=
( ) backgroundbackgroundfeature
featureemccd
SSSSNV
222
++Δ
Δ=
2
2
83
22
⎟⎟⎟
⎠
⎞
⎜⎜⎜
⎝
⎛
++Δ
Δ−
+Δ=
backgroundbackgroundfeature
featurebackgroundfeatureamp
SSSSSS
σ
0 when8 2 ==Δ backgroundampfeature SS σ
Slide 16
EMCCD technology: The F factor
Amplifier noise versus mean signal in feature for equal “visibilities”
ΔSfeature (electrons)
0.01 0.1 1 10 100 1000
Am
plifi
er N
oise
(r.m
.s. e
lect
rons
)
0.01
0.1
1
10
100
zero background0.01e background0.1e background1e background
Conventional CCD Beneficial
EMCCD Beneficial
Slide 17
EMCCD technology: The F factor
Mean Signal in spots = 1 electron
No background signal
emccd 1000x gain ccd σ = 1.5 e ccd σ = 1.0 e
ccd σ = 0.7 e ccd σ = 0.35 e ccd σ = 0.2 e
Visibility equal
S = σ2
Slide 18
EMCCD technology: The F factor
Mean Signal in spots = 0.1 electrons
No background signal
emccd 1000x gain ccd σ = 0.5 e ccd σ = 0.3 e
ccd σ = 0.2 e ccd σ = 0.1 e ccd σ = zero
Visibility equal
S = σ2
Slide 19
Optimising Quantum Efficiency
Electrode structure for normal front face devices limits QE to about 40%
Apertures can be created as in “Open Electrode” or virtual phasedevices. QE increases to around 60%.
For ultimate sensitivity back illumination is required. Removes absorbing layersAllows anti reflection coatings to be appliedQE over 95% can be achieved
Slide 20
Optimising Quantum Efficiency
QE as a function of wavelength (midband AR coated CCD97)
Slide 21
Optimising Quantum Efficiency
CCD216 back thinned 2/3” format device in L3C216 Camera
Overcast Starlight Illumination.
Lens f1.4. ~10 µLux faceplate illumination
Spatial (Gaussian 3x3 window) and k4 Temporal Filtered.
40ms integration
Slide 22
Optimising Quantum Efficiency
QE in the NIR can be increased further by using thicker silicon
standard thickness is ~16 µm.Our “Deep Depletion” devices are
40 µm thick.depletion depth must increase else
MTF will be degraded.1500 Ωcm silicon used for good
depletion.
Electron multiplying deep depletion CCDs have been demonstrated
Wavelength (nm)200 400 600 800 1000 1200
Abs
orpt
ion
Coe
ffici
ent, α
(µm
-1)
10-4
10-3
10-2
10-1
100
101
102
103
Abso
rptio
n Le
ngth
(µm
)
10-3
10-2
10-1
100
101
102
103
104
Absorption CoefficientAbsorption Length
Slide 23
Optimising Quantum Efficiency
Even thicker (100 to 300 µm) 8 kΩcm Si used for specially designed “HiRho” devices.
“Over depleted” with high negative substrate bias (>70V) to maximise the MTF.
Amplifier structures biased with local substrates to increase design options and minimise noise
low substrate potential high substrate potential
Slide 24
Optimising Quantum Efficiency
CCD217 100µm thick HiRho device (NIR antireflection coating)
Wavelength (nm)
400 600 800 1000
Qua
ntum
Effi
cien
cy (%
)
0
20
40
60
80
100
Temperature = +20oCTemperature = -100oC
measurements
Slide 25
Example EMCCD Applications
EMCCD technology + back illumination employed where very high sensitivity with high frame rate is required.e.g. high end surveillance, special forces etc.Life sciences
Single molecular imagingFluourescence microscopy (TIRF, Confocal etc)
COTs sensors have been available for several yearsCustom sensors now being developed
no gain 1000x gain
Slide 26
Example EMCCD Applications: LIDAR
Light Detection and Ranging (LIDAR)A pulsed laser beam is transmitted from the satellite (100 Hz rep. freq.).light will be backscattered from aerosols and clouds and detected at the
focal plane.
Slide 27
Example EMCCD Applications: LIDAR
e2v LIDAR sensor
Slide 28
Example EMCCD Applications: LIDAR
Measurements from simulated laser return and ground echo.
Time (µs)
0 50 100 150 200 250
Inpu
t Ref
erre
d Si
gnal
(ele
ctro
ns)
0.1
1
10
100
Mean of 16 Atmospheric ReturnsSingle Artificial Atmospheric Return
GR
OU
ND
EC
HO
0.2e
0.9e
2.5e
6e12e
31e
69e
Slide 29
Example EMCCD Applications: Adaptive Optics
The atmosphere limits the performance of ground-based telescopes. Adaptive optics can correct for the effects of atmospheric turbulence. A real or artificial guide star is focused on a sensor to sample the wavefront. A spot is focused onto different parts of the image sensor. The wavefront error is calculated and a deformable mirror adjusted
significantly greater sampling in both spatial and temporal domains required for next gen 8-10m telescopes.
Greater sampling implies fewer photons per pixel. A read noise << 1 electron for centroiding to the required accuracy. 240x240 pixels at a frame rate of 1.5 kHz have to be sampled
Application ideally suited to EMCCD technology.
Slide 30
Example EMCCD Applications: Adaptive Optics
e2v funded by The European Southern Observatory and JRA2 OPTICON to develop sensor
8 Outputs for high frame rate. BI and deep depleted silicon for high QE.Novel integral shutter being evaluated
Slide 31
Example EMCCD Applications: Adaptive Optics
Packaged with a 2-stage peltier pack for cooling and reduction of dark signal.
Slide 32
Example EMCCD Applications: Adaptive Optics
Test results operating at 1300 fps, 1000x multiplication gain.
Mean of 2000 frames3.5 electrons peak white
Single frame3.5 electrons peak white
Single frame2.5 electrons peak white
Single frame1.3 electrons peak white
Slide 33
Conclusion
EMCCDs utilise established CCD technology
Addition of new readout register structure provides 1000+ gainApplies very low noise gain before the charge to voltage conversionAmplifies signal above the CCD readout and video chain noise
Amplification can be applied at high pixel rateNoise equivalent signals < 1e possible up to video rates and above
The combination with back thinning provides an image sensor ideal for many applications requiring the ultimate sensitivity.