electron bombarded back-illuminated ccd sensors for low light
TRANSCRIPT
Electron bombarded back-illuminated CCD sensors for low light level imaging applications
George M. Williams Jr. and Alice L. Reinheimer, Scientific Imaging Technologies, Inc.
Beaverton, OR 97075
Verle W. Aebi and Kenneth A. Costello Intevac EO Sensors, Inc.
Palo Alto, CA 94304
Abstract
Significantly higher performance, reduced form factor, low light level surveillance cameras relative to
present state of the art are critical for many commercial and military applications. To achieve this goal, a
new approach to low light level cameras was successfully demonstrated. In a cooperative research and
development effort between Scientific Imaging Technologies, Inc. of Beaverton, OR and Intevac EO
Sensors of Palo Alto, CA, back illuminated, electron bombarded CCD (EBCCD) sensors were designed
and fabricated. Experiments demonstrated the EBCCD's sensitivity and contrast resolution superior to
conventional intensified CCD (ICCD) approaches. Low light level signal to noise (STN) and contrast
transfer function (CTF) data are presented. A model is derived that describes the performance of the
EBCCD and the back-illuminated CCD relative to conventional approaches to nighttime imaging. A
design and simulated performance of a video rate 2/3 inch, back-illuminated, electron bombarded CCD
currently under development for low light imaging applications is also described.
1.0 INTRODUCTION
During the last decade, CCD based camera systems have made great strides in achieving a low light
imaging capability. Standard black and white RS-170 security cameras achieve 100% video at 0.01
footcandles faceplate illumination that, with a fast, low f/# camera lens, corresponds to deep twilight
scene illumination. The highest performance low light level cameras available utilize a 'Gen-III' image
intensifier optically coupled to a standard CCD chip (Image Intensified CCD or ICCD). These camera
systems provide usable video at light levels as low as 10-5 footcandles -- starlight or lower scene
illumination. A conventional ICCD sensor is shown in Figure 1. In most high performance ICCD
systems, a fused fiberoptic images the output of the image intensifier onto the CCD array. In this
approach, a low light scene is imaged on the image tube photocathode generating photoelectrons which
are proximity focused onto a microchannel plate (MCP) where they are multiplied. A voltage potential
accelerates the amplified electron signal from the MCP output onto a phosphor screen where the image is
converted back to light. A glass fiberoptic element couples the phosphor screen image out of the tube.
Additional fiberoptic elements coherently relay the intensified image to a CCD chip where the optical
signal is converted back to an electrical signal and read out for image processing and display. At each
stage of the process, as light is converted to electrons, back to light, and finally once again to electrons,
image quality is lost .
For the ICCD system described above, the image is sampled at four interfaces: 1) at the microchannel
plate, 2) at the phosphor screen on the fiberoptic output window, 3) at the interface between the image
tube's fiberoptic window and the fiberoptic coupler, and 4) at the interface between the fiberoptic
coupler's output and the CCD. The optical quality of each interface is strongly dependent upon the fiber
size, the orientation, and the position of the fiberoptic array. The combined degradation of the electro-
optics, the microchannel plate, the phosphor screen, and the fiberoptic elements compromises resolution.
Moiré patterns, blemishes, and fiber array discontinuities ('chicken-wire') accumulate in the electro-
optical path and are imaged as 'fixed pattern' noise by the CCD. Moreover, scattering in the optical
interfaces and within the fiberoptic further degrades the modulation transfer (MTF) capability of the
sensor. This leads to 'washed-out', poor quality images.
Figure 1. Cross-sectional drawing of a conventional Figure 2. Cross-sectional drawing a proximity fiberoptically coupled ICCD. focused EBCCD.
The transmission loss of the fiberoptics, the inefficient collection by the fiberoptic of the near-lambertian
phosphor output of the intensifier tube, the inefficient of image tube phosphor, and the mismatch between
the spectral emission from the phosphor and the CCD spectral responsivity all decrease the sensor gain.
These losses require that the image intensifier operate at high gain. Operating the image tube at high gain
reduces the STN performance of the tube and increases the scintillation noise or 'snow' which degrades
image quality under low light conditions. Another ICCD sensor limitation is that, due to unreliable
PHOTOCATHODE
CCD
adhesion of the glass fiberoptic element to the CCD surface, the fiberoptic element may delaminate from
the CCD .
An Electron Bombarded CCD (EBCCD) eliminates the complicated image transfer chain by inserting a
thinned back-illuminated CCD into the image intensifier tube. Figure 2 shows a drawing of an EBCCD
sensor. The back-illuminated CCD forms the anode of the EBCCD sensor. It replaces the MCP, the
phosphor screen, and the fiberoptic coupler found in conventional image intensifier tubes.
The photoelectrons emitted from the EBCCD photocathode are proximity focused directly onto the
electron sensitive CCD, the silicon dissipates the incident photoelectron energy in the form electron-hole
pairs, and electron bombarded semiconductor (EBS) gain occurs. The EBS process is significantly lower
in noise than the electron gain obtained using a MCP. By imaging the electrons from the photocathode
directly with the CCD, the EBCCD avoids the inefficient and image degrading process of converting
visible light into electrons at the photocathode, back into light at the phosphor screen and then back into
electrons in the CCD. Due to the reduction in the number of image conversion steps and the significantly
greater signal to noise performance (STN) the EBCCD has higher contrast and resolution than does the
ICCD.
Eliminating the fiberoptic couple of the image intensifier tube to the CCD reduces the EBCCD sensor size
and weight. Excluding the weight of the high voltage power supply, a typical 25 mm ICCD weighs in
excess of 110 grams. In contrast, the 25 millimeter EBCCD typically weighs 39 grams. Whereas the
EBCCD requires only a single voltage supply (approximately 1.8 kV) to operate, the ICCD requires three
high voltage supplies (approximately 0.9 kV, 0 to 1.2 kV, and 7 kV). When one considers the weight of
each sensor's high voltage power supply, the benefits of the EBCCD's reduced weight and size are further
advanced. Figure 3 communicates the reduction in size and weight of the EBCCD over the ICCD
approach.
EBCCD sensor tests demonstrated significant advantages over standard intensified CCD sensors. These
advantages include:
1. Increased sensitivity that allows for greater resolution under low light conditions;
2. Superior contrast and resolution that allow for better target identification;
3. Increased dynamic range that allows for better contrast and less blooming;
4. Reduced size and weight that allow for more covert imaging and helmet mounting;
5. Increased mechanical integrity and reliability allow for longer lifetime; and
6. Lower cost.
The term EBCCD is commonly used to describe the detector that embodies the image tube and the back-
illuminated CCD as well as to describe the back-illuminated CCD in the image tube vacuum. Whenever
possible the authors use the term 'EBCCD' or 'EBCCD sensor' to describe the hybrid consisting of the
image tube and the back-illuminated CCD, and have used the term 'electron-bombarded CCD' or 'back-
illuminated CCD' when referring to the actual CCD array.
Figure 3 . Photograph showing the advantages in form factor available to the 25mm EBCCD approach over the conventional 25mm ICCD approach.
2.1 EBCCD PERFORMANCE: EBS Gain
The EBCCD achieves nearly 'ideal', noiseless gain through the electron bombarded semiconductor (EBS)
cascade process. Electron gain in an EBCCD sensor results when a high energy primary electron
dissipates its energy in the silicon of the CCD. Every 3.6 eV of energy lost by the primary electron
generates approximately one electron-hole pair. Diffusion in the silicon separates the electron-hole pairs.
The substrate connection collects the holes. The potential wells formed by the applied gate voltages
collect the amplified signal electrons. To the first order, the signal gain in the EBCCD is proportional to
the kinetic energy of the photoelectrons prior to their impinging on the CCD back surface. This
mechanism provides a convenient mean of controlling the overall gain of the tube by varying the
acceleration potential. Figure 4 displays an energy loss profile for the back-illuminated CCD [1].
In order to maximize the EBS gain, the 'active' material must dissipate the energy from the incident
electron and the CCD pixel must efficiently collect the electrons. To obtain EBS gain in the active
material, it is necessary for the photoelectron to enter the pixel from the back surface of the CCD -- away
from the gate structures that dissipate the electron energy. The back surface is typically thinned to a
thickness of 10 to 15 microns to optimize signal electron collection efficiency. Because the back surface
contains only a thin layer of epitaxial silicon and no device structures, the incident electron is able to enter
the active material with sufficient energy to allow high EBS gain.
Critical to the performance and stability of the device is the recombination phenomena at the back
surface. Proper back surface passivation (accumulation) is required to increase the collection efficiency
and to prevent 'surface trapping'. Figure 4 illustrates that below 2 keV a majority of the incident electron
energy is dissipated within a tenth of a micron from the back surface. To prevent recombination from
interfering with the gain process, it is critical that the surface be properly accumulated.
Penetration depth (um)
Nor
mal
ized
ene
rgy
loss
(dE
/dx)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0 0.2 0.4 0.6 0.8 1
4 keV
6 keV
8 keV
10 keV
Figure 4. Incident high energy electron energy loss profile as a function of distance from the back surface of a back-illuminated CCD [1].
At acceleration voltages sufficient to overcome the surface 'dead-layer', the back surface region where a
majority of electron-hole pairs recombine, the following equation approximates the EBS gain:
Gb Vacc Vdl
( )( )
.
1
3 6 (1)
where: Vacc is the accelerating voltage applied to the tube between the photocathode and the CCD; Vdl
is the voltage equivalent of the loss due to electron-hole pair recombination in the CCD back surface dead
layer; and b is the proportion of back scattered electron energy. The average energy required to create an
electron-hole pair in silicon is about 3.6 eV. In practice, 6 kV accelerating voltages obtain gain greater
than one thousand. For accelerating energies less than about 1 keV, the dead voltage model breaks down
and EBS gain is no longer linear. In fact, non-zero gains are measurable at virtually all accelerating
voltage potentials [1].
The EBS multiplication process exhibits low fluctuations. The gain variance, 2, is expressed as:
2 = F x G (2)
where F, the Fano factor, is 0.12 for silicon [2]. The low noise EBS process allows the EBCCD to obtain
much larger STN than a standard Gen-III image intensified CCD. To describe the degradation in signal-
to-noise attributable to the gain process, one uses the 'noise figure' coefficient, NFC. The ratio of the
input signal STN to the image tube's output STN defines the NFC and quantifies the system signal-to-
noise degradation due to the sensor. A perfect sensor has a 1.0 noise figure and introduces no additional
noise to the input signal. The EBCCD noise figure is expressed as:
NFC = 1 +
FG
1 - b . (3)
As silicon backscatters approximately 16% of the incident primary electrons, a noise figure of 1.09 is
attainable. In comparison, a noise figure of 2.0 or greater is typical for a standard Gen-III image
intensifier [3].
The Gen-III filmed MCP electron multiplication noise dominates the noise figure of a Gen-III image
intensifier tube. By not using an MCP, the EBCCD takes advantage of the high quantum efficiency of
the GaAs photocathode without suffering the degradation in STN due to the MCP. The EBCCD tube has
almost double the STN performance of a standard tube. In fact, for EBS gains greater than 10, the noise
figure will be almost entirely determined by b, the electron backscatter coefficient.
Because the first stage gain inversely reduces the noise figure contribution of subsequent stages, the first
amplification stage typically dictates the overall noise figure of the system. Low noise CCD amplifiers
exist with noise performance equivalent to as little as 35 electrons per pixel at RS-170 bandwidths. A
first stage gain of 200 will sufficiently eliminate further signal degradation by noise. An EBCCD with
35 electron readout noise and a 200 EBS gain is capable of single photon sensitivity.
Figure 5 shows the EBS gain curve measured on a SITe model SI502AB back-illuminated CCDs using a
Hitachi S4000 scanning electron microscope as an electron source. To optimize the back surface
accumulation process, CCDs manufactured with two different passivation processes were tested. To
calculate EBS gain, the beam currents on a Faraday cup were compared to the currents measured in the
back-illuminated CCDs. Due to the process dependent effects of electron-hole pair recombination at the
back surface, one passivation process shown in Figure 5 has significantly higher gain characteristics than
the other. The better process approaches theoretical gain performance.
Figure 5. SITe SI502AB back-illuminated CCD EBS gain versus incident electron energy measured in a Hitachi Model S4000 scanning electron microscope
The EBCCD gain performance was verified in an operational mode with two EBCCD sensors fabricated
using SI502AB back-illuminated CCDs and GaAs photocathodes. The CCDs' thinned back surfaces were
accumulated using the process shown in Figure 5 to result in higher gain characteristics. The EBCCD
EBS gain was tested using two methods. First, the ratio of the average signal in the electron bombarded
CCD pixel to a calibrated input light signal was used to compute the EBS gain. Second, the EBS gain
was calculated from the ratio of the variance to the mean of the signal in the CCD. The two methods used
to calculated the EBCCD EBS gain are consistent, and demonstrate agreement with the measurements
made using the electron beam of the SEM. Plotted in Figure 6 is the experimental device gain measured
at various acceleration voltages.
Incident Energy (eV)
EB
S G
ain
0
100
200
300
400
500
600
700
0 500 1000 1500 2000 2500 3000
Process 'A'
Process 'B'
Theoretical (350eV dead layer)
Figure 6. EBS gain measured on EBCCD #B versus acceleration voltage
2.2 EBCCD PERFORMANCE: Resolution and CTF
Three low light imaging sensors were fabricated and tested: 1) A SITe model SI502AB back-illuminated
CCD; 2) An Intensified CCD fabricated using a SITe SI502AF front-illuminated CCD array
fiberoptically coupled with a Schott 1:1 magnification 6 micron pore, type 32A glass window to an
Intevac 45 lp/mm GaAsP photocathode image intensifier tube; and 3) An Electron bombarded CCD
fabricated using a SITe SI502AB back-illuminated CCD separated by a .055 inch spacing from an Intevac
'extended-blue' GaAs photocathode. Experiments were conducted to determine the ability of each sensor
type to resolve various spatial frequency bar targets under photon shot-noise limited performance
conditions. Figure 7 exhibits each of the sensor's responsivity characteristics. Also shown in the Figure
is the responsivity of a high performance scientific grade front-illuminated CCD.
Figure 7. Responsivity characteristics of a SITe SI502AB back-illuminated CCD, a SITe SI502AF front-illuminated CCD, an Intevac GaAsP photocathode, and an Intevac GaAs photocathode.
The experiments used a diffuse 590 nanometer light source to image a 'multi-burst' target on each device.
Availability, rather than experiment design, governed the choice of a GaAs photocathode for the EBCCD
and a GaAsP photocathode for the ICCD. As is seen in Figure 7, although the GaAsP has 65 percent
greater responsivity at 590 nanometers than does the GaAs photocathode, both photocathode materials
Accelerating V oltage
Gai
n
0
50
100
150
200
250
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Wavelength (nanometers)
Res
pons
ivit
y (A
/W)
0
0.1
0.2
0.3
0.4
0.5
0.6
350 550 750 950
Front CCD
Back-IlluminatedCCD
GaAs
GaAsP
have good sensitivity. Although not optimal, 'ease of manufacturing' dictated the 0.055 inch
photocathode to the EBCCD spacing. A closer spacing, less than .018 inches, will result in higher
resolution and modulation transfer characteristics.
So that a 'true' comparison could be made between the sensor types, each sensor manufactured for the
experiments incorporated the SITe SI502A family of CCD imagers . The SI502A family is of a 512x512
format with 24 micron square pixels. Multi-burst bar target images were obtained using the SITe model
SMEC SI502 camera electronics. The SME Series of camera electronics is a versatile low noise (10
electrons rms/sample), 100 kHz, 14-bit module that optimally operates all of SITe's CCD products. The
same electronics module tested all three sensors. Figure 8. Measured CTF for an SI502AB back-illuminated CCD, a GaAs SI502AB EBCCD, and a GaAsP SI502AF ICCD
Figure 8 depicts each sensor's measured CTF. So that photon statistics would dominate the sensors' noise,
the measurements in Figure 8 used 'high light' signal levels. The light levels were chosen for each device
to corresponded to 80 percent of the CCD pixel 'full well' and were obtained while operating the device
for maximum sensitivity. As was anticipated, the back-illuminated CCD's CTF is superior to the
EBCCD as well as the ICCD over all spatial frequencies. The EBCCD's CTF is only slightly lower in
contrast than the back-illuminated CCD's CTF and, depending upon the spatial frequency of interest, is
20% to 100% higher in contrast than the ICCD's CTF. At their 'limiting' spatial resolution, unlike
CCDs, MCP image intensifier tubes have very low contrast. Moreover, the image intensifier tube's non-
linear intra-scene dynamic range degrades the contrast of the ICCD system.
Figure 9 contains a multi-burst target image obtained using an EBCCD and an ICCD for 16.67
millisecond exposures and a 6.6*10-7 footcandle faceplate illumination. It is apparent from the images
Spatial Frequency (lp/mm)
CT
F
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20 25
CCD
EBCCD
ICCD
that the EBCCD has much higher signal to noise and higher CTF than does the ICCD. As it is difficult to
determine the absolute contrast range from printed images, the comparative CTF for the EBCCD and the
ICCD for a 6.6*10-7 footcandle faceplate illumination and a 16.67 millisecond exposure is shown in
Figure 10.
The figures show that the superior EBCCD signal to noise characteristics will result in a better contrast
modulation and limiting resolution. In this respect, the low light EBCCD imagery is comparable in
contrast and resolution to daylight CCD imagery, and does not contain the 'washed-out', low contrast
characteristics of ICCD sensors. In particular, the low and medium spatial frequencies, frequencies found
to be critical for nighttime navigation, are dramatically superior in the EBCCD. As previously was
mentioned, a closer photocathode-to-CCD spacing will further improve the EBCCD CTF performance
over that of the ICCD.
The square-wave transfer function of the sensor is a physical property of the information link between the
scene and the observer. Unlike image intensified systems which obtain high resolution information at the
expense of gray level information, the EBCCD has high contrast as well as high resolution information.
Although the psycho-physics of scene interpretation is beyond the scope of this paper, if one considers
the total area beneath the CTF curve as a measure of the amount of scene information transmitted by the
sensor to the user, Figure 10 reveals that the EBCCD has greater than 95% more contrast resolution
information, as defined herein, than does the ICCD. Figure 11 contains a plot of the ICCD's CTF
measured at various exposure levels. Figure 12 contains a similar plot for the EBCCD and shows that at
low light levels the information content advantage of the EBCCD over that of the ICCD is even greater.
Figure 13 depicts the effects of the acceleration voltage on the EBCCD's CTF. For low acceleration
voltages, the electron's radial emission energy is a larger percentage of the acceleration energy. At low
accelerating voltages, the biplanar, proximity focused, electron lens optics dictate a wider distribution of
collected electrons at the back surface of the EBCCD. Increasing the accelerating voltage decreases the
spread of the distribution of accelerated electrons at the back surface and increases the EBCCD's CTF
performance.
A. Electron bombarded CCD B. Intensified CCD
Figure 9. Images of multi-burst target for GaAs SI502AB EBCCD and SI502AB ICCD using a 16.7 millisecond exposure and 6.6*10-7 footcandles faceplate illumination using a 590 nanometer wavelength source. Figure 10. Measured CTF for GaAs SI502AB EBCCD #A (1.8 keV) and SI502AB ICCD using a 16.7 millisecond exposure -6.6*10-7 footcandles 590 nanometer wavelength source.
Spatial Frequency (lp/mm)
CT
F
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20 25
EBCCD
ICCD
Figure 11. Measured CTF for GaAsP SI502AB ICCD for various light levels (footcandles of faceplate illumination using a 590 nanometer wavelength source)
Figure 12. Measured CTF for GaAs SI502AB EBCCD for various light levels (footcandles of faceplate illumination using a 590 nanometer wavelength source)
Figure 13. Measured CTF for GaAs SI502AB EBCCD for various acceleration voltages @ 1*10-4 footcandles faceplate illumination using a 590 nanometer wavelength source. The CTF for a back-illuminated SI502AB CCD is also shown.
Spatial Frequency (lp/mm)
CT
F
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20 25
1.32E-05
2.76E-06
6.60E-07
1.60E-07
Spatial Frequency (lp/mm)
CT
F
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20 25
1.32E-05
2.76E-06
6.60E-07
1.65E-07
Spatial Frequency (lp/mm)
CT
F
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20 25
500 eV
1000 eV
1500 eV
2000 eV
CCD (highlight)
2.3 EBCCD RELIABILITY AND MANUFACTURABILITY
The accelerated photoelectrons that produce the electron-hole pairs in the CCD also produce soft X-rays
as they bombard the silicon. Although the kilovolt electrons incident on the CCD's back surface do not
travel far enough into the silicon to affect the front surface gate structures (see Figure 4), the gate
insulator may absorb the X-rays and cause the production of fixed charge in that region as well as
'trapping' states at the Si-SiO2 interface. Damage to the gate structure oxide layers from keV level X-rays
will likely result in increased dark current and decreased full-well capacity. The intensity and dose of the
X-rays produced in the CCD are proportional to the number of incident electrons and their accelerating
energies.
Because of the X-ray's dependency on the incident electron's energy, a reduction in the accelerating
potential will minimize the intensity of the X-rays produced. However, reducing the accelerating
potential undesirably decreases the sensor's gain and contrast resolution. Similarly, as X-rays produced
further from the gate structure have lower probability of traveling to the gate structures, increasing the
back surface's thickness will result in fewer X-rays reaching the front surface. Albeit, at the expense of
contrast resolution.
As multi-pinned phase (MPP) CCDs operate in an 'inverted' mode, they are less susceptible to the effects
of ionizing radiation than are standard modes of CCD operation . Operating a CCD in MPP mode does
somewhat compromises the pixel's capacity to hold charge. Using accelerating voltages below 1.8 keV
(K-alpha X-ray producing energy in silicon), using radiation compatible gate dielectrics, and using an
MPP operating mode will minimize the effects of radiation damage in the electron bombarded CCD.
To determine the increase in CCD dark current attributable to high energy electron bombardment, a SITe
model SI502AB CCD, thinned to a less than a 15 micron epitaxial thickness, was tested in a Hitachi
model S4000 scanning electron microscope. Using a 1*10-2 C/cm2 electron dose, the experiment varied
the electron accelerating energies incident on the back-illuminated CCD and measured the resulting
increase in CCD dark current. Figure 14 shows the increase in dark current versus incident electron
energies. On the graph's right y-axis is the sensor's relative lifetime operating with a 1*10-4 footcandle
faceplate illumination. The back-thinned CCD's 'lifetime' is defined, herein, to be a 50 electron per pixel
increase in CCD dark current per 16.7 millisecond integration period. A 50 electron increase in dark
current will result in low light performance limited by the CCD's dark current shot noise and not the
CCD's readout noise and will thus begin to degrade contrast and sensitivity. The experiment
demonstrated that at accelerating voltages below 1.8 keV, the lifetime of both MPP and non-MPP CCDs
will exceed the operation life of military Gen-III image intensified tubes, which is typically specified at
15,000 hours.
Accelerating Energy, Operating Mode.
Dar
k cu
rren
t inc
reas
e (p
A/c
m^2
)
.
0
50
100
150
200
250
300
3keV,non-MPP
2keV,non-MPP
1.5keV,non-MPP
3keV,MPP
2keV,MPP
1.5keV,MPP
0
10
20
30
40
50
60
Hou
rs o
f O
pera
tion
at 1
0e-4
fC
(x1
000)
.
Figure 14: Dark current increase versus acceleration energy and operating mode for 1*10-2 C/cm2 electron dose. 1*10-2C/cm2 is equivalent to one year at 1*10-4 footcandles of faceplate illumination assuming 1200 micro-amps/lumen photocathode sensitivity. Also shown is lifetime (hours of operation x 1000) at 1*10-4 footcandles where lifetime is defined by an increase in dark current of 50 electrons per pixel per 1/60th second at room temperature.
Gallium arsenide photocathodes are particularly susceptible to contamination from residual gases within
the tube. Not having to use an MCP for gain removes over 0.240 square meters of surface area from the
image tube vacuum. The SI502AB used in the EBCCD sensor has only 0.00015 square meters of active
surface area. Since the EBCCD image tube has a relatively small volume compared to traditional image
intensified tubes and does not have the MCP's surface area to contend with, the EBCCD's photocathode
will have a far longer lifetime. Optimization and characterization of the EBCCD's lifetime will be an area
of continuing research.
Furthermore, the EBCCD's back surface thinning process must be compatible with high temperature
semiconductor processing. As organic materials degrade image tube lifetime, the adhesive used in the
back-illuminated CCD's supporting structure must contain no organic materials. Until recently, these
requirements have limited the maturation of thinned, back-illuminated CCDs for EBCCD applications.
The EBCCD image tube is inherently easier to process than standard image tubes. Because it does not
require a microchannel plate, a phosphor screen, or fiberoptics, the device is very simple to manufacture.
In addition to the pure material costs, the manufacturing tolerances required for the spacing between these
components increases the complexity and the cost of manufacturing conventional, high resolution, image
tube ICCDs.
3.0 EBCCD, ICCD, and CCD LOW LIGHT LEVEL PERFORMANCE MODEL
On the basis of a previously published performance model for ICCD devices [3], a model was derived for
each of the three sensor types. The model uses the modulation transfer characteristics and the signal-to-
noise of the ICCD's components to predict the sensor's low light level limiting resolution. Although CTF,
an optical system's response to a square wave pattern, is easier in practice to measure than MTF, MTF
has an advantage. In the case of components in cascade, the optical chain's MTF response is the product
of each component's MTF. The relationship between the MTF sine-wave response and the CTF square-
wave response is obtained by Fourier analysis.
As ambiguity arises when applying a single-valued isoplanatic MTF value to data from a discontinuous
non-isoplanatic discrete pixel CCD array, a wide variation in experimentally obtained MTF data will
result for small sample sizes. The model approximates the average anticipated value for MTF for large
sample sizes.
When used with visible light, the contrast resolution of large pixel devices, with thin epitaxial layers, is
not degraded by charge spreading at the CCD's back surface. Thus, a Sinc function approximates a back-
illuminated CCD's optimal geometric modulation transfer. A factor, , set equal to 1.6 to approximate
experimentally obtained mean MTF values, was introduced into the Sinc function.
MTF = ccd sin( )
( )
W f
W f
(4)
In contrast to visible light which is absorbed near the CCD's depletion region, electrons are absorbed
within several tenths of microns of the CCD's back surface (see Figure 4). The research team anticipated
that the 'electron cloud' resulting from the EBS gain process would have a point spread function larger in
area than the CCD pixel. To test the effect of electron bombarded charge spreading on MTF, difference
frames were obtained by subtracting two uniform 'flat field' images. Difference frames remove fixed
spatial pattern noise from the analysis. The charge from non-overlapping regions of pixels was then
summed to form images of statistically independent 'super-pixels'. For photo-electron shot noise limited
operation, the variance of a difference frame divided by the mean of the difference frame will calculate
the EBS gain. Figure 15 depicts the gain (variance/mean) plotted as a function of the number of pixels
used to form each of the image's super-pixels.
In the absence of charge spreading, pixels of a difference frame are statistically independent, and the
resulting variance/mean (gain) curves' slopes are flat with respect to the number of binned pixels. The
curves' offsets along the y-axis are proportional to the acceleration voltage. Analysis reveals that cross-
talk among the pixels due to charge spreading causes an inflection point in the gain curves at
approximately 16 binned pixels -- a 4 x 4 'super-pixel'. For the case of the SI502AB CCD used in the
experiments, this approximates a 96 micron diameter region of correlation in the variance.
Linearity of the curves' slopes below 16 pixels suggests a gaussian distribution of charge spreading due to
diffusion in the epitaxial layer. The inflection point at approximately 16 'binned' pixels on all the curves
is consistent with electron absorption relatively close to the back surface. The shape of the energy
distribution is dependent primarily on the distance between the back surface and the edge of the CCD's
depletion region. The effects of the distance between the back surface and the depletion region, the
acceleration voltage, and the 'dead-region' thickness on EBCCD performance will be items of further
research.
Figure 15. Plot of variance/mean of difference frames obtained from subtracting two 'flat field' images at various acceleration voltages and 'binning' the charge of pixels to form images of 'super pixels'
An exponential, whose curve is primarily dependent on the thinned epitaxial region's thickness with
regard to the depletion region, models charge spreading in the epitaxial layer's effects on the back-
illuminated CCD's MTF. Having quantified the effect of electron charge spreading due to the EBS gain
process, the theoretical EBCCD MTF characteristics are modeled by convolving the exponential term
used to describe the gaussian effects of the EBS electron diffusion in the CCD (the third term in Equation
5), with the MTF of a biplanar electron lens [4] ( the first term in Equation 5), and with the MTF term for
the CCD array (the middle term in Eq. 5):
#pixels binned
Var
/mea
n
0
50
100
150
200
250
300
1 10 100
300 eV
500 eV
1500 eV
1800 eV
MTF = e *eebccd (-4 (
VimVs (- ( s-Xd 2
2 22)( ) ) ) ( ) )*
sin( )
( )
fL X fW f
W f (5)
where, Vim is the maximum radial emission energy (eV), Vs is the acceleration voltage (eV), f is the
spatial frequency (lp/m), L is the spacing of the proximity focused diode, W is the linear dimension of the
CCD pixel (m), Xs is the thinned CCD epitaxial layer's thickness, and Xd is the CCD depletion layer
thickness.
An expression was derived for the MTF of an image intensifier tube by fitting an exponential curve to the
MTF measurements of a high performance 1600 micro-amp/lumen, 47 lp/mm 18 millimeter Gen-III GaAs
image intensifier tube. The first term in Equation 6 shows the resulting correction. The MTF of the
image intensifier convolved with the Bessel function of the first order, used to model the MTF of the
fiberoptic coupler [5], and the CCD MTF term (Equation 4) forms an expression for the ICCD MTF.
MTF = eiccd f( / . ) .
*[( )
( )] *
sin( )
( )2000 1 1 215 2J V f
V f
W f
W f
(6)
where: J1 is a Bessel function of the first order, and V is the fiberoptic coupler's core-to-core pitch.
The modeled MTF performance curves are shown in Figure 16, and when expressed as a square wave
response, reasonably approximate the data obtained experimentally. To test the EBCCD model, the
effect of the acceleration voltage on the EBCCD's MTF was modeled. The results plotted in Figure 17
are consistent with the experimental data shown in Figure 13.
Figure 16. Modeled MTF: EBCCD (1.8 keV), ICCD, and back-illuminated SI502AB CCD using the design parameters of those devices manufactured for the above described experiments.
Spatial Frequency (lp/mm)
MT
F
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20 25
EBCCD
ICCD
CCD
Figure 17. Modeled CTF for GaAs SI502AB EBCCD for various acceleration voltages (eV) - 590 nanometer light.
In a 100% contrast square-wave grating the light passes through the white lines without significant
attenuation; no light passes through the black lines. When this pattern is imaged by an optical sensor; the
image contrast transfer Ci is given by
CN N
N Ni
w b
w b =
-
+, (7)
where Nw and Nb are the number of signal electrons from the white and black lines respectively.
Because the contrast loss is caused by the light spread from the white lines into the black lines, the sum
of Nw and Nb is essentially constant for all spatial frequencies . The equation used to represent the signal
from the white lines of an imaged square-wave pattern is:
SIG I ART
eGainc s
f = .10 76 , (8)
where: 10.76 is the conversion from ft2 to m2; Ic is the faceplate illumination (footcandles); A is the
detector's pixel area at the focal plane (square meters); Rs is the sensor's responsivity(A/W); Tf is the
sensor's field integration time (seconds); e is an electron's charge (1.602*10-19 C/e); and Gain is the
sensor's electron gain (-).
The signal from a white line of the bar target is expressed as:
S SIG CTF N pixspix f = ( )( )( )( ) , (9)
Spatial Frequency (lp/mm)
CT
F
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 2 4 6 8 10 12 14 16 18 20
500
1000
1500
2000
where: CTF is the sensor's square-wave response; Nf is the number of frames integrated by the eye; and
pixs is the number of pixels imaging the white line.
The equation
Noise SIG Gain N N N N pixspcdk ccdk a f = ( + + + ))( ) ( )( )2 2 2 2 (10)
models the noise of the system. Npcdk is the number of electrons per pixel per frame generated in the
CCD by photocathode dark current (in the case of the unmodified back-illuminated CCD, Npcdk = 0),
Nccdk is the dark current signal attributed to the CCD, and Na is the sensor's readout noise.
The signal to noise ratio is thus given by the equation
STN dBNFC
S
Noise
pix( ) log = (
1) 20 , (11)
where: NFC is the noise factor due to the gain mechanism ( a value of 1.00 for a back-illuminated CCD;
approximately 1.09 for an EBCCD, and approximately 2.00 for an ICCD).
Equation 1 defines the expression for the EBCCD's Gain, Gain(ebccd). Alternatively the model may use
the experimental EBS gain from the curve in Figure 6. Equation 3 defines the EBCCD's noise figure,
Noise(ebccd), to be approximately 1.09 for acceleration voltages greater than 400 V.
The expression for the gain of an ICCD is:
Gain ICCD G T Va Vo Ep SccdMCP fo( ) ( )( ) = , (13)
where: GMCP is the microchannel plate's electron gain (-); Tfo is the fiberoptic coupler transmission (-);
Va (eV) is the acceleration voltage from the output of the MCP to the phosphor screen (eV); Vo is the
dead voltage attributed to the phosphor screen (eV); Ep is the phosphor screen's efficiency (-); is the
spectral matching factor between the phosphor screen and the CCD (-); and Sccd is the peak quantum
efficiency of the CCD (-). The noise figure of the ICCD is dominated by the microchannel plate's noise
figure and is typically greater than NFC(ICCD) = 2.
For the purposes of simulation, the gain and the noise figure of a standard back-illuminated CCD are
set to unity : Gain CCD( ) = 1 and NFC(CCD) = 1.
Parameters used to manufacture the SI502AB back-illuminated CCD, ICCD, and EBCCD were used to
test the model against the experimental data. The parameter values used in the simulation are listed in
Appendix A. Figure 18 presents simulated performance curves of a back-illuminated CCD, an ICCD,
and an EBCCD. The curves show signal-to-noise, expressed in decibels, versus spatial frequency,
expressed in line pairs per millimeter, for 6.6*10-7 footcandles of faceplate illumination.
Approximately six decibels (dB) of signal with respect to noise is necessary for the human eye to resolve
a three-bar target [3]. A comparison of the spatial frequency at which each device STN curve crosses the
six dB point with the point that corresponds to three percent modulation (limiting resolution) on the
Figure 10's CTF curve, shows that the model accurately predicts the limiting resolution measured for each
of the three device types. For example, the Figure 18 calculated STN curve for the ICCD crosses over the
six dB point at 13 lp/mm and also crosses the Figure 10 three percent modulation point t 13 lp/mm. The
Back-illuminated CCD achieves limiting resolution at approximately 7 lp/mm for both the experimental
and the simulated data . The EBCCD's Nyquist frequency limits its resolution in both the simulated and
experimental cases. The model was verified at three other light levels, and was found to approximate the
measured device performance.
Figure 18. Modeled signal to noise (dB) versus target spatial resolution SI502AB EBCCD, ICCD, and back-illuminated CCD measured at 6.6*10-7 footcandles of faceplate illumination
Spatial Frequency (lp/mm)
ST
N (
dB)
-10
0
10
20
30
40
50
0 5 10 15 20 25
Back-illuminatedCCD
ICCD
EBCCD
PerceptiveThreshold
4.0 VIDEO RATE 2/3 INCH EBCCD DESIGN
A video rate, low light level, EBCCD surveillance camera development effort is in progress. The design
uses a back-illuminated CCD, full frame-transfer architecture consisting of a 652 x 488 pixel array of 13.5
micron square pixels. 'Binning' in the serial register during readout forms a 13.5 micron (H) x 27 micron
(V) pixel. Shifting the binning operation's centriod by a single horizontal line during alternate fields
interlaces the video. As designed, the CCD runs at 13.5 MHz with 35 electrons rms readout noise. The
2/3 inch CCD format allows standard 18 mm image intensifier photocathode, image tube ceramics, and
manufacturing tooling to be used in the EBCCD design.
Figure 19 depicts the modeled performance of a 2/3 inch back-illuminated CCD, a 2/3 inch ICCD, and a
2/3 inch EBCCD, at 1.0*10-4 footcandles -- deep twilight lighting condition, where conventional CCDs
cannot operate. The Figure shows the EBCCD to be superior in STN performance at most spatial
frequencies. Surprisingly, at this light level, at all spatial frequencies the back-illuminated CCD is
superior in STN to the ICCD.
A low light imaging device's performance is proportional to its gain, its modulation transfer
characteristics, its responsivity, and its noise figure. A low light imaging system must embody all of
these factors and cannot maximize one at the expense of another. The modeled performance of the three
distinct devices clearly shows that ICCD's gain is achieved at the expense of MTF as well as at the
expense of the noise figure. The ICCD is ,therefore, limited in its ability to transmit information from the
scene to the user. In fact, the data shows that a back-illuminated CCD, despite an absence of a gain
stage, is a higher performance alternative to an ICCD for moderately low light levels.
Due to nearly ideal gain characteristics and good modulation transfer, the EBCCD is superior to the ICCD
over all spatial frequencies and in all light conditions. Figure 20 illustrates the three sensor types'
modeled performance at 2.2 *10-6 footcandles of faceplate illumination. As is seen in the model, at this
light level the limited number of photons in the target scene requires that a gain stage be used to
overcome the system noise. The back-illuminated CCD is useful only at very low spatial frequencies.
The 2/3 inch EBCCD has useful response almost out to the Nyquist limit, and has more than 60% greater
limiting resolution ( the point at which the curves crosses the 6 dB line) than does the ICCD.
Figure 19. Modeled Signal to Noise Ratio versus Target Spatial Resolution 13.5 micron x 27 micron 2/3" Video EBCCD, ICCD, and back-illuminated CCD with 1.0*10-4 footcandles illumination Figure 20. Modeled Signal to Noise Ratio versus Target Spatial Resolution 13.5 micron x 27 micron 2/3" Video EBCCD, ICCD, and back-illuminated CCD with 2.2*10-6 footcandles faceplate illumination
Figure 21 illustrates the faceplate illumination level required by each sensor to resolve spatial frequency
targets. The Figure shows the EBCCD's performance to be superior at all spatial frequencies and at all
illumination levels. For targets that exceed a 15 lp/mm focal plane spatial frequency, a bare back-
illuminated CCD is superior in its low light imaging capability to an ICCD. These results suggest that
the EBCCD, in combination with a back-illuminated CCD, will fully span the performance range
previously occupied by video front-illuminated CCDs and ICCDs. Moreover, at moderate light levels,
Spatial Frequency (lp/mm)
ST
N (
dB)
0
10
20
30
40
50
60
0 5 10 15 20 25 30 35 40
PerceptiveThreshold
Back-thinnedCCD
ICCD
EBCCD
Spatial Frequency (lp/mm)
ST
N (
dB)
-10
0
10
20
30
40
50
0 5 10 15 20 25 30 35 40
PerceptiveThreshold
CCD
ICCD
EBCCD
approximately 1*10-4 footcandles, such is characteristic of many night vision applications, the back-
illuminated and the EBCCD, provide better performance than the ICCD.
Figure 21. low light illumination required to resolve various spatial resolution targets, 13.5 micron x 27 micron 2/3" video EBCCD, ICCD, and back-illuminated CCD.
5.0 SUMMARY
Although the electron bombarded CCD concept was proposed early in the development of the CCD,
back-illuminated CCDs compatible with image tube vacuums have not been available. The EBS gain of
the SITe SI502AB back-illuminated CCD was successfully characterized in a scanning electron
microscope and in a proximity focused vacuum diode. Tests verified that nearly noiseless gain exceeding
240 is obtainable at acceleration voltages below 1.8 keV. A first stage gain of 240 is sufficient to
virtually eliminate further system noise degradation of the signal and achieve single photon sensitivity.
Achieving adequate gain at acceleration voltages below 1.8 keV avoids generating K-alpha X-rays and,
thus prolongs the lifetime of the EBCCD's back-illuminated CCD. Preliminary lifetime experimental
results show that using radiation hardened gate dielectrics and operating the devices in MPP mode allows
for negligible back-illuminated CCD performance degradation over the operational lifetimes expected of
night vision imaging devices. Experiments demonstrated that superior signal to noise and contrast
transfer characteristics allow the EBCCD sensor (shown in Figure 22) to be more sensitive and higher
resolution than conventional ICCD approaches. In the future, EBCCDs manufactured with closer spacing
of the photocathode to the back-illuminated CCD and optimized back surface passivation will further
advance the benefit of the EBCCD over the ICCD. On the basis of data gained in the experiments, a
model was developed that was shown to predict the low light performance of back-illuminated CCDs,
ICCDs, and EBCCDs. The research team used the model to design and simulate the performance of a
video rate back-illuminated CCD sensor and an EBCCD sensor. The development of these new devices
will realize a new generation of night vision devices.
Spatial Frequency (lp/mm)
Fac
epla
te I
llum
inat
ion
(foo
tcan
dles
, 286
4de
g T
ungs
ten
Lam
p)
1.00E-07
1.00E-06
1.00E-05
1.00E-04
1.00E-03
0 5 10 15 20 25 30 35
EBCCD
BCCD
ICCD
6.0 REFERENCES [1] Reinheimer, Alice L. and Blouke, Morley M., "A simple model of electron-bombarded CCD gain" in Charge-Coupled Devices and Solid State Optical Sensors IV, Morley M. Blouke, Editor, SPIE 2172 [2] W. van Roosbroeck, "Theory and yield and Fano factor of electron-hole pairs generated in semiconductors by high energy particles," Physical Review, Vol. 139 No. 5A, pp.A1702-16, Aug. 30, 1965 [3] Williams, George M. , "A high-performance LLLTV CCD camera for nighttime pilotage: in Electron Tubes and Image Intensifiers, C.B. Johnson and Bruce N. Laprade, Editors, SPIE 1655 [4] Eberhardt, E.H. Applied Optics, Vol. 16, No. 8, 2127 (1977)
[5] Eberhardt, E.H. , in ITT Technical Note # 126.
Figure 22. A 25 millimeter SI502AB, 512 x 512, GaAs EBCCD manufactured in a cooperative research and development
effort by Scientific Imaging Technologies, Inc. and Intevac EO Sensors.
Appendix A.
EBCCD, ICCD, CCD Model Parameters
Rs(CCD) = 5850 (microamps/lumen,2854 deg tungsten lamp) - Back-illuminated CCD responsivity
Rs(GaAs EBCCD, ICCD) = 1600 (microamp/lumen, 2854 deg tungsten lamp) - GaAs photocathode responsivity
NFC(CCD) = 1.00 (-) - CCD noise figure
NFC(ICCD) = 2.00 (-) - ICCD noise figure
NFC(EBCCD) 1.09 (-) EBCCD noise figure
Gmcp = 500 (-) - Electron gain of the microchannel plate
(Va-Vo) = 2,500 (V) - The difference of the phosphor screen voltage and its dead layer voltage
Ep = 0.08 (-) - The phsophor screen's efficiency
= 0.35 (-) - The spectral matching factor between the P-20 phsophor and the CCD
Tfo = 0.75 (-) - The transmission of the fiberoptic
A = 5.76E-10 (meters2) (24 micron square pixels - SI502AB) - Area of pixel
Tf = .0167 (seconds) - The video, field integration time
Nf = 3.5 (frames) - The number of frames integrated by the eye
pixs= 4/(5*f^2)/A (pixels/line-pair) - 1951 Air Force three bar targets.
e = 1.602*10-19 (C/e)
Npcdk(EBCCD) = (100*10-11amps/m2)(A)(Gain(EBCCD))(Tf)/e
Npcdk(ICCD) = (100*10-11amps/m2)(A)(Gain(ICCD))(Tf)/e
Npcdk(CCD) = (0 amps/m2)(A)(Gain(CCD))(Tf)/e
Nccdk = (amps/m2)(A)(Tf)/e 108 electrons/pix/frame (non-MPP)
(amps/m2)(A)(Tf)/e 9 electrons/pix/frame (MPP)
Na= 40 (electrons rms, 13 MHz, 23 deg C) - CCD readout noise