1
Optically sensitive Medipix2 detector for adaptive optics
wavefront sensing
John Vallerga
a*
, Jason McPhate
a
, Anton Tremsin
a
, Oswald Siegmund
a
, Bettina
Mikulec
b
and Allan Clark
b
a
University of California, Berkeley, 7 Gauss Way, Berkeley, CA, 94720-7450 USA
b
University of Geneva, 24, quai Ernest-Ansermet, 1211 Geneva 4, Switzerland
Abstract
A new hybrid optical detector is described that has many of the attributes desired for the next generation
adaptive optics (AO) wavefront sensors. The detector consists of a proximity focused microchannel plate (MCP)
read out by multi-pixel application specific integrated circuit (ASIC) chips developed at CERN (“Medipix2”)
with individual pixels that amplify, discriminate and count input events. The detector has 256 x 256 pixels, zero
readout noise (photon counting), can be read out at 1kHz frame rates and is abutable on 3 sides. The Medipix2
readout chips can be electronically shuttered down to a temporal window of a few microseconds with an accuracy
of 10 nanoseconds. When used in a Shack-Hartmann style wavefront sensor, a detector with 4 Medipix chips
should be able to centroid approximately 5000 spots using 7 x 7 pixel sub-apertures resulting in very linear, off-
null error correction terms. The quantum efficiency depends on the optical photocathode chosen for the bandpass
of interest.
Keywords: adaptive optics; wavefront sensors; MCP; Medipix; GaAs photocathodes
———
*
Corresponding author. Tel.: 1-510-643-5666; fax: 1-510-643-9729; e-mail: jvv@ssl.berkeley.edu.
1. Introduction
Ground-based astronomy in the optical and
infrared has the distinct disadvantage of observing
through the atmosphere. Though mostly transparent
over much of the bandpass, the atmosphere is a
constantly changing and dynamic medium. The index
of refraction of air is a function of density and
temperature and the vertical spatial profile of these
parameters change with time due to windshear
induced turbulence. Hence, a plane wave of light
from a distant star will be distorted across the
constant phase wavefront, leading to blurry images in
the focal plane of a telescope or twinkling stars to the
human eye.
CERN-OPEN-2006-064
01/04/2005
2
Ideally, the angular resolution of a telescope
would be limited only by the diffraction limit of the
primary mirror, which improves linearly with
diameter. The typical ~ 1 arc second blurring caused
by the atmosphere corresponds to the diffraction limit
of a 20 cm diameter mirror in the optical bandpass.
Therefore, making a ground-based telescope larger
than 20 cm does not improve its angular resolution
(though it does improve its light gathering power).
Space-based telescopes, like the Hubble Space
Telescope, do not suffer from atmospheric distortion,
but they are expensive to build, launch and operate,
and therefore difficult to make much larger.
In the past two decades, techniques have been
developed to remove the effects of the atmosphere on
the light from distant sources. Adaptive optics is the
method of using fast deformable mirrors [1] to
conjugate and therefore cancel any phase errors
introduced in the light path between the object of
interest and its image at the focal plane (Fig. 1).
Before correcting the phase of the light, it must be
measured using a wavefront sensor that samples the
wavefront across the pupil. One method (among
many) is the Shack-Hartmann sensor [1] (Fig 2). The
wavefront is sampled by a lenslet array creating
individual images focused on an imaging array. If the
light is perfectly collimated (e.g. a plane wave from a
distant star), the images would be spots of light at the
regular spacing of the lenslet array. If the plane wave
is distorted, the centroids of the spots would spatially
shift depending on the local slope of the wavefront.
By measuring the centroids of these spots in real
time, one can determine the wavefront error and feed
this error signal back to the deformable mirror to
“close the loop” and correct all time variable
wavefront distortions.
Figure 1. Simple schematic of an adaptive optics system where the
aberrated wavefront is reflected off the deformable mirror resulting
in a corrected diffraction limited wavefront passed on to a
downstream camera. The beamsplitter takes a sample of the
corrected wavefront to monitor the phase and send updates to the
control electronics to “close the loop” to the deformable mirror.
Figure 2. Cartoon to show how a Shack Hartmann wavefront
sensor works. An input plane wave (top) is focused by a lenslet
array resulting in a linear grid of stellar images on the 2d detector.
If the wavefront is aberrated (bottom), the centroids of the stellar
image move away from the original grid. The amount of
movement is a function of the slope of the wavefront; hence a
Shack-Hartmann wavefront sensor measures the first derivative of
the phase of the wavefront.
Each centroid determination measures the slope of
the constant phase wavefront at that particular
location on the wavefront at that particular time.
Larger telescopes with larger pupils use deformable
mirrors with more actuators and hence more phase
3
measurements, therefore requiring detectors with
many pixels. The atmosphere’s variability on most
nights necessitates wavefront sampling rates on the
order of 100 to 1000 Hz. The number of photons
from the guide stars used as the reference beacon are
almost always limited in number since bright stars
are rare, so the wavefront sensor detector should have
high quantum efficiency and low readout noise to
improve the signal to noise ratio of the centroid
determinations.
A recent white paper on the AO instrumentation
needs for future large (> 30 m diameter) telescopes
entitled “A Roadmap for the Development of
Astronomical Adaptive Optics” [2] specified that
wavefront sensor detectors should have:
• Quantum efficiencies > 80%
• Pixel formats of 512 x 512
• Frame rates of 1 kHz or faster
• Readout noise less than 3 electrons rms.
The last three are not simultaneously achievable with
the current generation of CCDs. Larger detectors
require faster clocking rates to read out at fast frame
rates. But higher clocking speeds increase the readout
noise because of the increase in bandwidth required
of the amplifiers. Below we describe a detector that
can meet the last three of the above requirements, but
with a quantum efficiency approaching 40%.
2. Photon counting MCP detectors
Photon counting detectors, unlike charge
integrating arrays like CCDs, register each detected
photon as one count and so have no “readout noise”.
A variety of readout anodes are used in imaging
microchannel plate (MCP) detectors (Fig. 3) to report
the location and time of arrival of every detected
photon. Imaging MCP detectors can have large area
(100 x 100 mm), high spatial resolution (20 µm
FWHM), low background dark count, and event
timing resolution less than 1 nanosecond
[3]. Their
QE is determined by the characteristics of the
photocathode material that absorbs the initial photon
and releases the photoelectron. The optical QE of
these photocathodes is now approaching 50%, in
particular gallium arsenide (GaAs) (Fig. 4).
Figure 3. Microscope image of a Burle Industries glass MCP with
2 µm pore diameters and 3µm pore spacing.
Figure 4. Quantum efficiency of GaAs type photocathodes made
by ITT
Most of the readout anodes used in photon
counting MCP detectors are serial in nature, i.e. they
analyze single events in a short but finite time. Even
if this analysis “deadtime” per event could be reduced
to 50 ns with the current generation of fast electronics
(analog to digital converters etc.), the best counting
rates achieved would be 2 MHz at 10% deadtime
losses. For a future Shack-Hartmann wavefront
sensor with 5000 spots of 1000 photons at 1 kHz
sampling rate, the total detector event rate would be
5000x1000x1000 or 5 GHz, a factor of 2500 higher
rate than the current state of the art in photon
counting! To take advantage of the noiseless
4
operation of a photon counting detector using MCPs
therefore requires a specialized readout that can
localize and count the events in a massively parallel
way, i.e. pixelated counters. We envisioned
fabricating a specialized ASIC mounted behind an
MCP that could count and integrate events from the
MCP at each pixel and which could be read out
digitally and therefore very fast and noiselessly.
Photon
e
-
Q = 10
4
e
-
Pij = Pij + 1
Window
Medipix2
MCP
Photocathode
Figure 5. Schematic of a sealed microchannel plate image sensor.
The MCP amplifies a single photoelectron and the resultant charge
cloud event causes the ASIC pixel counter, Pij, to increment by
one.
3. Medipix2 as a readout anode for MCPs
Fortunately, we soon discovered that such a
device already existed, being designed and
constructed by the Microelectronics Group at CERN
for the multi-national MEDIPIX collaboration
(http://www.cern.ch/medipix).
Our novel detector scheme should achieve the
first three of the specific goals listed above with a QE
approaching 40%. The detector (Fig. 5) is an MCP
image tube with a GaAs photocathode and a new
pixelated CMOS readout chip called the “Medipix2”
[4, 5]. Photons interacting with the photocathode
release a photoelectron that is proximity focused to
the MCP input face. The MCP amplifies this single
photoelectron with a gain on the order of 10
4
. The
resultant charge cloud exits the MCP and lands on the
input pad of a Medipix2 pixel where it is counted as
one event. The counter will integrate until it is read
out in a digital, noiseless process. Also, because the
data is digital, it can be read out at ~246 MHz pixel
rates, which corresponds to a frame readout time of
266 µs for the current Medipix2 chip.
The deadtime per event for the Medipix2 pixel is
approximately 500ns, so each pixel can handle rates
up to 200 kHz without too much of a deadtime
correction. Over the whole 256x256 chip, count rates
can approach a 13 GHz event rate.
3.1. First results
Our first set of Medipix2 chips came mounted on
a printed circuit board used for testing by the
Medipix collaboration. The board has space for
bypass capacitor and resistors and a 68-pin connector.
For the eventual vacuum tube application, these parts
and connector would be incorporated behind the tube,
mounted externally and coupled to the Medipix2
through hermetic feedthroughs. We also designed an
MCP detector body that could hold 1 to 3 MCPs at
various heights above the Medipix2. The spatial
extent of the charge cloud at the Medipix2 input
surface is directly proportional to this distance. We
could not make this gap smaller than 300µm as the
electrical connection to the Medipix chip is through
bonding wires on the input surface. There is a voltage
bias across this gap to accelerate the electron charge
cloud to the Medipix2 to minimize charge spreading.
We could also adjust independently the voltage
across the MCP(s) to vary the single event gain
(previously calibrated with a simple copper anode).
This detector assembly was mounted to a 10-inch
vacuum flange mounted on a vacuum chamber and
actively pumped down to a pressure of 10
-6
Torr. We
used a Hg pen-ray lamp to illuminate the bare MCP
and could adjust the lamp intensity to get 10 cts. s
-1
to
500 million cts. s
-1
.
The Medipix2 chip was controlled and read out
using the “MUROS2” control electronics [6]
developed for the Medipix consortium by the
NIKHEF group and the control software “Medisoft
Ver. 4.0” developed by Univ. of Naples Federico II
[7]. The only extensive optimization of the Medipix2
settings at Berkeley involved the equalization of the
pixel lower charge thresholds per the techniques
discussed in reference 5.
5
Figure 6. Output images of MCP-Medipix2 test detector. The two images on the left show single photon events from short 100 µs
integrations. On the far left, the MCP gain was 200,000 and the rear field voltage was low (300V) resulting in large charge clouds that excited
many pixels per photon. In the center image, the MCP gain was reduced and the rear field increased (50ke
-
, 1500V) resulting in a smaller
charge cloud detected on average by only 2.4 pixels. The image on the right is a longer integration (100s) with the bias settings of the center
image revealing the Air Force test pattern illuminating the input MCPs.
The initial results (Figs. 6-9) are from a chevron
set of two, 33 mm diameter Photonis MCPs, 10-µm
channel diameter pores on 12 µm centers, 40:1 L/d
thickness, and low resistance (22 MΩ per plate). The
output end spoiling was 2 channel diameters. For a
bias of 1430V across the two plates, the average gain
was 20,000, and we tested up to a gain of 200,000
(1680V). The rear field between the output of the
MCPs and the Medipix2 could be varied from 0 to
1600V. At the low gains that we operated the MCP,
the lamp brightness for full field illumination did not
matter much, as we could use the shutter to take very
short exposures of a few microseconds to see single
photon events or integrate for many seconds to
acquire very deep flat fields.
Figure 6 shows two short integrations and one
long integration, which contrast the different data
collection modes. The first two are both short enough
to see individual photon events coming from the
MCP. For the higher gain, lower rear field case (left),
the charge clouds are large and extensive enough to
trigger many pixels, each recording 1 count (except
for event overlaps, where pixels might count 2
events). By lowering the gain and increasing the rear
field (middle), the single photon events trigger, on
average, 2.4 pixels and have many less overlapping
events at this input rate and integration time. Keeping
the same settings as the middle, the right image is a
long integration where individual events are no
longer distinct. This image reveals the Air Force test
pattern (AFTP) mask we mounted on the input
surface of the MCP. Pattern 3-2 is just
distinguishable (Fig. 9), corresponding to a resolution
of 9 lp/mm, exactly the Nyquist limit of a 55µm pixel
device.
Figure 7. Average event size in pixels vs. lower charge threshold
for 4 MCP gains. The rear field is 200V, leading to rather large
event sizes.
Our initial tests probed the extremes of the
parameter space for optimizing the detector for AO
performance. Fig. 7 shows the average number of
pixels excited per event at a 200V rear field at
different lower level charge thresholds for various
MCP gains. In effect, we are mapping the spatial
charge distribution of the average MCP output charge
cloud. When normalized by gain, these curves fall
