Characterization of a CID-38 Charge Injection Device
Brian Backer, Zoran Ninkov, and Massimiliano Corba
Chester F. Carlson Center for Imaging Science
Rochester Institute of Technology
54 Lomb Memorial Dr., Rochester, NY 14623
e-mail: bsb5301@cis.rit.edu
ABSTRACT
Charge coupled devices have been the dominant
solid state detector array in the visible due to their relatively
simple design and easy implementation. With recent advances in
lithographic techniques, arrays having smaller photosite dimensions
and an increased number of pixels have become available. Further
advances in large format CCDs have been limited by charge transfer
efficiency (CTE) of photoelectrons to the readout amplifier. The
increased number of pixel transfers in large arrays can degrade image
quality and MTF unless even higher CTEs are achieved.
Multiplexer designs that remove the need for thousands of charge
transfers can bypass these CTE limitations. One such focal plane
architecture is the CID or charge injection device. This paper
presents results obtained with one particular CID based system. The
array is housed in a dewar capable of liquid nitrogen operation. The
output signal from the array is amplified with a nearby low noise
preamplifier before digitization. Results on injection efficiency,
readout noise, and other pertinent CID parameters, are presented
obtained from this device preamplifier as well as specific
experiments.
Keywords: Charge Injection Device (CID), device characterization,
injection, low noise techniques.
Most applications of solid state sensors to low light imaging have
used CCDs. An example is the near dominance of CCDs in astronomical
imaging. However there are many situations where an alternate focal
plane architecture such as that of the CID has distinct advantages
over the CCD. CIDs have been proposed for application in space
environments since the device is inherently radiation hard due from
the high resistivity of the P-channel process used to create the
imager. In addition, since no charge transfer between pixels is
needed for readout, defective pixels do not result in more that a
single pixel being corrupted. In addition to this application, CIDs
have been designed and used in spectroscopy and medical imaging.
CIDs use row and column electrodes to select a single pixel and
sense the stored photoelectrons from the perimeter of the light
sensitive area. In this way there is only one charge transfer (from a
storage well to a sense well) per pixel for each pixel on the array.
In addition, CIDs have advantages of non-destructive readout for
noise reduction, individual pixel injection, and multiple subarray
definition and readout. Due to the minimal amount of electrical
structure on the surface, the array has good blue wavelength response
extending into the UV.
The array and pixel layouts of a CID-38 are shown in Figure 1. This front illuminated device has 512x512 square pixels on 28mm pitch. Each pixel consists of a pair of orthogonal polysilicon electrodes that creates two MOS capacitors in n-doped silicon for storage and sensing of photogenerated charge. These electrodes also connect the rest of the pixels on the column or row to the scanners on the periphery.
The CID-38 has a preamp per row (PPR) architecture. All pixels along each row are buffered by a single FET and hence the PPR architecture requires calibration of the row FETs to minimize nonuniformities between them. A thin metal strip is layered on top of the row polysilicon to reduce readout noise. A drawback to this metal structure is a small amount of obstruction of the incoming light. Further reduction of read noise will be achieved using improved row amplification and with preamp per pixel (PPP) structures currently being manufactured by CIDTEC.
Figure 1: Structure of a
CID-38 (a) and of an individual pixel (b). Pixel boundaries are
created by the field oxide which reduces the electric field in the
epitaxial region under the polysilicon runs and creates pixels.
Integration of photogenerated holes occurs in the positively biased epitaxial region. Since the substrate is grounded, a reverse biased p-n junction is created inside of every pixel. This provides excellent antiblooming protection when overexposed since excess holes outside the well are swept through the junction into the substrate. Negative or slightly positive voltages on the column and row electrodes create depletion wells for storage of holes. In PPR devices such as the CID-38, columns are biased more negatively and holes collect under this electrode called the "collection pad".
Figure 2: Photon collection and the readout scheme of a single CID-38 pixel.
Two types of readout techniques are shown in Figure 2. During
readout, a "zero level" is sensed on the sense pad along the row by
allowing the pad to float and digitizing. The sole charge transfer
from the collection well to the sense well is performed by driving
the column high and all stored charge moves to the sense well. The
amount of collected charge sensed on the row electrode modulates the
drain-source current of the output FET amplifier. In non-destructive
readout (NDRO) the low potential on the collection pad is
reestablished and accumulation continues. In destructive readout
(DRO), the pixel is injected.
Pixels along a row are read by shifting the column register until
the last pixel has been read. The column register is then reset and
the row register shifted to the next row. This row read operation is
repeated for all of the rows. This method of readout differs from
"true" random access. Any single CID-38 pixel can be randomly read
but the addressing of that pixel is achieved by quickly shifting the
row and column shift registers to get to the pixel. A true random
access array with multiplexing, the CID-35, is under development and
will be tested at RIT soon.
Pixel injection can be performed with two different methods. The
entire array can be cleared in one step with "global" injection. This
is done by simultaneously driving all columns and rows to
Vinj, the prescribed injection voltage. Each pixel in the
array then has both pads biased such that holes are injected through
the junction. The other injection method removes only the charge in a
subarray in a method similar to array readout. In "subarray"
injection, each pixel is addressed sequentially as during readout and
the injection voltage is applied to the row electrode. Horizontal
shifting during subarray injection (0.2mm/pixel) is substantially
faster than readout (62mm/pixel).
The commercial CID control electronics, or SiCam, generate the readout and injection clocking sequences with a sensor control board. These clocks drive the shift registers on the periphery of the CID and produce the pixel readout and injection results described above. The electronics also perform dual slope sampling of the analog input and interface the digitized data to a PC.
A modified preamplifier scheme was implemented to further reduce
readout noise. A two channel preamp design was designed since the
CID-38 has output FETs for both even and odd rows. Each channel
consists of three low noise operational amplifiers. The first
amplification stage consists of a current-to-voltage converter. The
second gain stage is a simple inverting amplifier configuration. A
third op-amp buffers the biasing voltage applied to the CID output
FETs. Feedback capacitors on the gain stages filter high frequency
above the maximum clocking frequency. Large ground planes and input
guard shielding were used in the design and construction of the
amplifier board.
The preamplifier design was validated with SPICE analysis. The DC Gain was simulated at 623 and the gain at 16.5kHz readout frequency is 580. The -3dB point for high frequency cutoff was determined to be about 100kHz.
The layout for all of the experiments is shown in Figure 3. An
incandescent white light source is powered with a stabilized voltage
supply and focused onto a pinhole. The spatially filtered light is
wavelength bandpassed into the integrating sphere which produces a
uniform illumination field on the detector array. An external shutter
on the dewar provides control of integration times. A close up of the
devices inside the evacuated dewar is shown in the insert. The CID
package makes thermal contact against a cold finger which keeps the
array near liquid nitrogen temperatures (77K). The pins of the CID
make electrical connection to the fanout board which is also held at
LN2 temperatures to reduce thermal conductivity to the
array. External clocks and bias voltages are fed to the fanout board
through fine wires and a connector not shown in the figure. The
fanout board contains no active components.
The preamplifier is placed inside the dewar to reduce the length of the wires between the CID output pins and amplifier input. The op-amps are thermally connected to the dewar wall with metal based adhesive tape. The odd and even row output pins of the CID are connected to the input of the preamplifier with low resistance fine wire. The amplifier outputs are fed through a military connector in the dewar wall and supplied to the SiCam electronics for further amplified and digitization. The digital data is then transferred to a PC and displayed with modified commercial imaging software.
Figure 3: Physical layout of the experimental optics and electronics.
Figure 4: Photon transfer curve for the CID-38. Both noise and gain measurements were made at a pixel read rate of 16kHz.
The gain and noise of the system were characterized with the
photon transfer technique. Two frames of equal exposure are
subtracted and the resultant frame has no bias shift and any pattern
noise is removed. The average exposure level is plotted vs. the
variance read noise of a subarray in Figure 4.
The system gain, 1/g, and the read noise Nr can be determined with the equation below.
The gain was graphically determined with the reciprocal of the
regression slope. The read noise was determined from the system gain
and the intercept of the regression fit to the data points. From the
figure, the system gain and read noise are ~20 e-'s/ADU
and ~170e-'s respectively. This read noise figure is a 15%
improvement over previously published read noise figures of
~200e-'s5.
When not in operation, the output of the CID-38 can be controlled
with a simple hardware circuit and a software driven TTL pulse. A pin
on the package can be used to turn the output FETs on and off. This
enables temperature control of the op-amps by reducing their power
consumption. These modifications were determined to be unnecessary
since temperatures fluctuated by only 1K with the op-amps thermally
attached to the inside of the dewar wall.
The efficiency of an injection cycle depends on three key parameters shown in Figure 5. One variable is the injection voltage applied to the row and column electrodes. Voltages less than the positive biasing of the n-doped epitaxial region do not completely remove the depletion wells and not all of the charge is injected. Exploiting this procedure is called "pixel knockdown" and can be used to increase the effective dynamic range of the array.5 The second parameter for injection is the duration of the injection pulse. Longer pulses give holes more time to diffuse to the junction and be removed from the pixel. The third parameter is the number of injection pulses that are applied to the pixel. If a short pulse duration is chosen, many pulses can be applied to the pixel to fully clear the accumulated charge.
Figure 5: Representation of a complete
injection cycle. The delay time (~4ms) and minimum pulse width (~1ms)
are limiting factors due to the SiCam electronics and delays in
software processing.
A series of measurements were made to determine the efficiency of an injection cycle. The array is exposed to a level near full wells, S, and the digital values readout. The array is injected and another readout, sinj, is performed. The efficiency is determined by the following ratio.
The experiment was repeated for two different injection voltages and are shown in Figure 6 and Figure 7. From the plots, higher injection voltages remove charge much more efficiently than lower injection voltages as expected. Higher injection voltage is also more effective parameter in removing charge than longer injection pulses.
Figure 6: Plot of the injection efficiency for Vinj = 8.15V. The 'Injection Cycles' axis is not a direct correlation to injection time since the pulse duration and delay times vary for each of the curves.
Figure 7: Plot of the injection
efficiency for Vinj = 8.15V. Again the 'Injection Cycles'
axis in not a direct correlation to injection time since the pulse
duration and delay times vary for each curve. The efficiency scale
for both graphs is the same to show the effects that injection
voltage has on efficiency.
One goal of the experiments was to determine the best combination
of injection parameters for the fastest global injection of the
array. For example, Figure 6 shows that one 640ms injection pulse
completely clears the array. However, seven 2ms pulses (and 6 delay
times, tdelay) also clears all of the charge and this
procedure only takes a total of 38ms. A trade off situation exists
between the number of pulses and the pulse duration.
A table of the total injection times obtained by experiment is
shown below. From the table, the fastest way to totally clear the
array at 8.15V is with three 4ms injection pulses; the total
injection time then becomes 20ms. Similar experiments with a 8.10V
injection voltage produced a best injection method of nine 3ms pulses
and a total injection time of 59ms. No final injection time is
available for Vinj=8.10V:1ms pulse duration since the
array wasn't fully injected after many cycles.
|
pulse duration |
640 |
320 |
160 |
80 |
40 |
20 |
10 |
5 |
4 |
3 |
2 |
1 |
|
8.15V Total Injection time |
640 |
320 |
160 |
80 |
40 |
44 |
24 |
23 |
20 |
24 |
26 |
61 |
|
8.10V Total Injection time |
640 |
644 |
324 |
248 |
172 |
116 |
80 |
68 |
60 |
59 |
62 |
--- |
The main limitation to the above measurements is the delay time between pulses and the minimum pulse duration of 1ms. Both of these are related to hardware limitations that are currently being modified. It is thought that the total injection time at 8.15V can be as small as 1ms with a modified system that removes these limitations.
The linearity of a CID in low light situations was determined as follows. Incremental time exposures of low intensity were given to the array and the bias subtracted signal level is plotted vs. the exposure time in Figure 8.
Figure 8: A plot representing low
light level linearity of CID arrays. The lack of response in the
ultra low exposure times is classically referred to as reciprocity
failure. The array was determined to be >99% linear in the
midrange of the A/D.
The array shows minimal reciprocity failure at ultra low light
levels. To find the optimal injection settings necessary to reduce
this effect, a series of measurements varying the three injection
variables were made. Three injection voltages, 8.15V, 8.125V, and
8.1V were used. Plots from these experiments were created and the
intercept of the regression line drawn from the linear region was
determined. From the regression intercepts, the "precharge" levels,
or number of photons necessary to overcome the insensitivity, were
determined and these values are shown in the table below.
|
Voltage |
8.15V |
8.15V |
8.15V |
8.15V |
8.15V |
8.125V |
8.125V |
8.10V |
|
Pulse Duration |
640ms |
320ms |
160ms |
4 ms |
1ms |
640ms |
320ms |
640ms |
|
Number of cycles |
1 |
1 |
1 |
5 |
20 |
1 |
1 |
1 |
|
precharge |
1671.177 |
1602.241 |
1718.562 |
1238.2 |
648.681 |
1836.417 |
2167.712 |
3521.274 |
Lower injection voltages have higher precharge levels. Different pulse durations used with a 8.15V injection voltage have little effect on the precharge level. The main improvement comes from using multiple injection pulses. The 300% improvement with multiple injection reduces the precharge level from ~1600 photons to ~650 photon. Again the limitation to more precharge reduction lies in our inability to create smaller pulse durations and delay times. These are hardware and software programmable and are being modified for further investigation.
All of the injections studies described above were performed with
global injection techniques. It is important to characterize the
performance of subarray injection as well. To do this and perform
different CID clocking schemes, a high speed Pulse Instruments based
array testing system has been acquired. The system supplies
electronics for biasing voltages, A/D conversion and a VMEbus
backplane interface. A DSP board with two 40Mhz C-40 DSPs is used to
produce the clocks. Preliminary tests with this system are under
way.
An application using the CID-38 to correct for the random image
motion of stellar objects through the atmosphere has been discussed.
A subarray on the CID is defined in software and used for object
tracking. Error signals are derived from motions of the object
centroid and sent to a fast steering mirror which compensates the
random motion induced by the atmosphere. The rest of the array
continues integration until the exposure is complete. The resultant
point spread functions of the objects in this field are sharper as a
result of the image correction.
A second application uses a CID as the imager in a photon counting
system. Different regions of the field of view contain both bright
and dim objects and a separate subarray is defined for each.
Subarrays containing brighter objects can be read faster than the
subarrays with dimmer objects. This type of control effectively
increases the dynamic range of the system. In addition to the Pulse
Instruments electronics described above, event detection and event
centroids are calculated by a custom VMEbus centroider board.
The authors acknowledge support received from a NSF Support for
the Center for Electronic Imaging Systems grant. We would like to
thank Joe Carbone, Steve VanGordon and Jeff Zarnowski at CIDTEC and
Bob Slawson at RIT for their support with the above research and data
collection.