Draft version May 17, 2023

Typeset using LATEX twocolumn style in AASTeX631

Scientific CMOS sensors in Astronomy: IMX455 and IMX411

Miguel R. Alarcon ,1, 2 Javier Licandro ,1, 2 Miquel Serra-Ricart ,1, 2 Enrique Joven ,1 Vicens Gaitan ,3
and Rebeca de Sousa4
1 Institutode Astrofı́sica de Canarias (IAC), C/ Vı́a Láctea, s/n, E-38205, La Laguna, Spain
2 Departamento de Astrofı́sica, Universidad de La Laguna (ULL), E-38206 La Laguna, Canarias, Spain
arXiv:2302.03700v2 [astro-ph.IM] 16 May 2023

3 Aplicaciones en Informática Avanzada (AIA), E-08172 Sant Cugat del Vallès, Catalonia, Spain
4 Departamento de Historia y Filosofı́a de la Ciencia, la Educación y el Lenguaje, Universidad de La Laguna (ULL), E-38206 La Laguna,

Canarias, Spain

(Received March 29, 2023; Accepted April 26, 2023)

Accepted for publication in PASP

ABSTRACT
Scientific complementary metal-oxide-semiconductor (CMOS) detectors have developed quickly in
recent years thanks to their low cost and high availability. They also have some advantages over charge-
coupled devices (CCDs), such as high frame rate or typically lower readout noise. These sensors started
to be used in astronomy following the development of the first back-illuminated models. Therefore,
it is worth studying their characteristics, advantages, and weaknesses. One of the most widespread
CMOS sensors are those from the Sony IMX series, which are included in large astronomical survey
projects based on small and fast telescopes because of their low cost, and capability for wide-field and
high-cadence surveys. In this paper, we aim to characterize the IMX455M and IMX411M sensors,
which are integrated into the QHY600 and QHY411 cameras, respectively, for use in astronomical
observations. These are large (36 × 24 and 54 × 40 mm) native 16 bit sensors with 3.76 µm pixels
and are sensitive in the optical range. We present the results of the laboratory characterization of
both cameras. They showed a very low dark current of 0.011 and 0.007 e− px−1 s−1 @–10◦ C for the
QHY600 and QHY411 cameras, respectively. They also show the presence of warm pixels, ∼0.024% in
the QHY600 and 0.005% in the QHY411. Warm pixels proved to be stable and linear with exposure
time, and are therefore easily corrected using dark frames. Pixels affected by the Salt & Pepper noise
are ∼2% of the total and a method to correct for this effect is presented. Both cameras were attached
to night telescopes and several on-sky tests were performed to prove their capabilities. On-sky tests
demonstrate that these CMOS behave as well as CCDs of similar characteristics and (for example)
they can attain photometric accuracies of a few milli-magnitudes.

Keywords: Astronomical detectors (84) — Astronomical instrumentation (799)

1. INTRODUCTION of <0.1% made it imperative to develop more effec-

Historically, observational astronomy has been deter- tive alternatives. The turning point came in 1969 with
mined by the development of imaging technology. In the appearance of charge-coupled devices (CCDs, Boyle
the early twentieth century, the introduction of the & Smith (1970)), which are silicon-based sensors with
photographic plate marked an unprecedented revolu- two-dimensional arrays of photosensitive units (pixels).
tion, but their very limited quantum efficiency (QE) These devices produced improved linearity and quan-
tum efficiency over photographic plates and simplified
data analysis thanks to their analog-to-digital conver-
Corresponding author: Miguel R. Alarcon sion (ADC). The first astronomical image that was taken
[email protected] with a CCD appeared in 1975 when scientists from
NASA’s Jet Propulsion Laboratory obtained an image
2

of the planet Uranus at a wavelength of 8900 Å (Janesick out noise (1.6 e− ), good linearity (99.7%), and a stable
et al. 1987). The development and refinement of CCDs bias. This sensor family has a dual-amplifier structure
have increasingly made these sensors the most widely in which two 12 bit images are taken simultaneously and
used option for the manufacture of astronomical instru- are then merged into a single 16 bit high dynamic range
ments. image. However, this mechanism is susceptible to jumps
Over the last two decades, the rise of alternative and instabilities around the transition region between
technologies has undermined the prevalence of CCD high and low gain. In addition, edge glowing and charge
sensors. Complementary metal-oxide-semiconductor persistence over long periods of time is also present. The
(CMOS) image sensors started to be developed in the capabilities of these sensors decrease considerably with
1990s (Fossum 1997). However, they still had a num- exposure times longer than several seconds, which is a
ber of disadvantages when compared with CCDs, such consequence of these effects and the increase of dark
as lower dynamic range (DR), and poorer linearity and current; however, they are still suitable instruments for
sensitivity (Bigas et al. 2006). Although CMOS sen- high frame rate observations.
sors soon established themselves in the consumer mar- Many leading manufacturers are now developing in-
ket, their inherent constraints restricted their applica- strumentation based on next-generation sCMOS sen-
tion in certain fields, especially those related to sci- sors, but their suitability for general use in astronomy
ence. To overcome the typical limitations of CMOS, is still largely unexplored. One of the most widespread
the so-called scientific CMOS (sCMOS) were introduced sensors are those from the Sony IMX series, which are
in 2009 as a result of a collaboration between An- being included in large projects such as the Argus Op-
dor Technology, Fairchild Imaging (BAE Systems) and tical Array (Law et al. 2022), Large Array Survey Tele-
PCO Imaging (Coates et al. 2009). This new genera- scope (Ofek et al. 2023) or the next generation of tele-
tion of sensors combined high frame rates, reasonable scopes for the ATLAS project (Tonry et al. 2018) , such
pixel and sensor sizes, and quantum efficiencies com- as the one that will be installed at Teide Observatory,
parable to CCDs—especially back-illuminated (BI) sC- ATLAS-Teide (Licandro et al. 2023a), because of their
MOS (Princeton Instruments 2016)—and a considerable low cost and capability for wide-field and high-cadence
reduction in the noise levels that are traditionally asso- surveys.
ciated with CMOS. In this paper, we present the results of the laboratory
Scientific CMOS are beginning to be accepted in as- characterization of the sCMOS BI Sony IMX455M and
tronomy. The first analyses in this respect mainly IMX411M sensors when integrated into the QHY600M
involve two sensors: CIS2051 (later rebranded as and QHY411M cameras. In Section 2, the two de-
CIS2521), which was developed by Andor Technol- vices are tested, and the laboratory set-up and the tele-
ogy and integrated into the Neo series detectors; and scopes that we used are presented. Sections 3.1 and
GSENSE, which was developed by Gpixel and integrated 3.2 describe the tests performed under dark conditions:
by several companies, e.g., Finger Lakes Instrumenta- spatio-temporal variation of bias and dark and contam-
tion, QHYCCD, and Andor Technology. Experiments inating effects, such as random telegraph noise and the
performed with Andor Neo showed low readout noise presence of warm pixels. In Section 3.3, the main op-
(∼1 e− ) and linearity deviations up to the saturation erating features such as the gain fix pattern noise and
point within the expected range of ±1%. However, two linearity are verified using the photon transfer curve.
features were noticed: the transfer curve was observed The quantum efficiency measures are shown in Section
to bend, even at low signal levels, which was possibly 3.4 and the charge persistence effect reported in other
caused by the non-linearity of an amplifier stage; and sCMOS sensors is reviewed in Section 3.5. Finally, sev-
the signal shows certain irregularities with high vari- eral approaches to processing telescope data and on-sky
ance in the transition region between high and low gain results based on this analysis are discussed in Section 5.
modes (Schildknecht et al. 2013). This, along with a
limited QE due to being front-illuminated and a lower
fill factor (Qiu et al. 2013), means that this sensor is not
the most suitable for regular observations, and its use-
fulness is mostly constrained to bright objects and high
frame rates. The back-illuminated GSENSE2020 sensor,
integrated in the Andor Marana, was recently tested for
its application in astronomy by Qiu et al. (2021) and 2. METHODS
Karpov et al. (2021). This sensor showed a low read- 2.1. Instruments
 3

Figure 1. Diagram of the optical test bench set-up, with the (1) Newport 68945 digital power supply, (2) Newport M-66881
QTH lamp, (3) Newport 76994 shutter, (4) Newport Oriel Cornerstone monochromator, (5) Hamamatsu S1336-5B1 photodiode,
(6) Labsphere SC6000, (7) Labsphere US-080-SF/SL integrating sphere, (9) Hamamatsu S2281 photodiode, and (10) QHY600M
Pro / QHY411M

Table 1. Technical data of QHY600M Pro and QHY411M sCMOS cam-

eras provided by the manufacturer.

Feature QHY600M Pro QHY411M

Sensor Sony BI IMX455M Sony BI IMX411M

Sensor size (diagonal) 43.3 mm 66.7 mm

Pixel size 3.76 × 3.76 µm

Pixel area 9600 × 6422 14304 × 10748

Effective pixels 61.17 Mpx 151 Mpx

Max full frame rate

(USB 3.0 port, full frame
and 16-bit output) 2.5 fps 1 fps

A/D sample depth

(1×1 binning) 16–bit
Shutter type Rolling shutter

Cooling system
(temperatures below ambient) Air cooling (-30C) Air cooling (-35C)
Water cooling (-45C)
4

The QHY600M1 camera is based on the back- integrating sphere, placed next to the monochromator,
illuminated IMX455 monochrome sensor that is man- had incorporated a Hamamatsu S1336-5B1 photodiode,
ufactured by SONY, which is a full-frame (35 mm for- which was in turn connected to a Keysight B2980A pi-
mat) sensor with 9576 × 6388, 3.76 µm square pixels. coammeter. Next, a dark box was located, followed by
The QHY411M2 camera is based on SONY’s IMX411 the test of the camera. A second photodiode (Hama-
monochrome sensor, which is also backilluminated but matsu S2281) was connected to the exit port of the dark
with a larger sensor size, 14304 × 10748 (equivalent to box. The conversion factors between the intensity mea-
medium-format cameras, 54 × 40 mm). Both of these sured by the picoammeter on the first photodiode and
sensors include an overscan region of 33 and 91 rows, the radiant power received on the second photodiode at
respectively, and they are native ADC sampled at 16 the exit of the dark box were characterized, in addition
bit, which is a significant change from previous genera- to the correction to the distance between the photodiode
tions of sCMOS sensors that were based on 12 bit im- and the sensor back focus. During the tests, the labora-
age merging (see Karpov et al. (2021)). The main fea- tory temperature was about 23◦ C and the humidity was
tures of the cameras are listed in Table 1. We used the 40–50%. Both cameras were air-cooled, with operating
QHY600M Pro version, which allows a faster 2×10 Gbps temperatures of −5◦ C for the QHY600M Pro and 0◦ C
fiber connection to a frame grabber with an additional 4 for the QHY411M.
GB of DDR3 memory—although in this paper all tests
were done via USB 3.0 connection—and triggering the 2.3. Telescopes
rolling shutter via an external GPS with an accuracy of The sky tests were performed with two of the robotic
more than a microsecond. It is worth mentioning that telescopes (Telescopios Abiertos Robticos, TAR) of
both sensors, especially the IMX455, are used in cam- Teide Observatory (Tenerife, Canary Islands, Spain).
eras from other manufactures, e.g., Atik Apx60, ZWO The QHY600M Pro was installed on the prime focus
ASI6200MM Pro. of TAR03, 0.46-m f /2.8 C18 reflector telescope on a
These cameras can operate in several modes and gain Planewave L500 altaz mount. An ROI of 4096×4096 was
settings, which essentially change their gain, readout used, giving a FOV of 410 .1 × 410 .1 with 000 .6 px−1 . The
noise (RON), and full-well capacity (FWC). In this pa- QHY411M was mounted on a Meade LX200-ACF 1600
per, we have focused on those that are considered most 0.406-m f /10, on an APM GE-300 Direct Drive equa-
appropriate for use in astronomy because they maintain torial mount, with UV/IR-Cut/L, SDSS g0 , r0 and i0 50
a good balance between these characteristics, which are mm filterswhich were manufactured by Baader Plane-
Mode#1 (High Gain Mode) and gain setting 0 (here- tarium GmbH. The sensor was trimmed to 40 × 40 mm,
after, #1@0) on the QHY600M Pro and Mode #4, and giving a FOV of 340 .2 × 340 .2 arcmin with 0.4000 px−1 .
gain setting 0 (hereafter, #4@0) on the QHY411M, al- Both cameras were connected via USB 3.0 and oper-
though some results for other modes are also shown. ated with air cooling at −10◦ C. After the first tests, the
2.2. Optical test bench QHY411M was installed in one of the 80 cm telescopes
of the Two-meter Twin Telescope (TTT) project, which
The evaluation of the sensors was performed using is a 0.80-m f /6.85 Ritchey–Chrétien altaz telescope that
existing experimental equipment, which are available was manufactured by ASA Astrosysteme GmbH, with a
at the Laboratory of Imaging and Sensors for Astron- total FOV of 330 .4×250 .0 and a plate scale of 000 .14 px−1 .
omy (LISA), at the Instituto de Astrofı́sica de Canarias
(Tenerife, Spain). The set-up is schematically shown in 3. RESULTS
Figure 1. A Newport M-66881 QTH lamp was connected
3.1. Bias stability and ”Salt & Pepper” effect
to A Newport 68945 digital power supply, with inten-
sity stabilization. In the Newport Oriel Cornerstone The bias frame stability in both cameras was tested
monochromator, the option to apply an order sorting in the laboratory under conditions of complete darkness.
filter was selected. This filter blocked higher diffraction In all of the tests that are presented here, including the
orders from interfering with the selected wavelengths. telescope runs, the offset setting was fixed at 10. First,
A Newport 76994 shutter was placed between the lamp 21 continuous unbinned full frames bias were obtained
and the monochromator to control the light beam before and stacked with a 3σ-clipping median to obtain the
taking each dark frame. A Labsphere US-080-SF/SL master bias. The result is shown in Figure 2. Both
cameras show dark signal non-uniformity (DSNU), with
a notable column-to-column pattern and pixel-to-pixel
1 https://www.qhyccd.com/scientific-camera-qhy600pro-imx455/ variations. Gradients can be observed in the QHY600M
2 https://www.qhyccd.com/scientific-camera-qhy411-qhy461/ Pro toward the upper region, with variations of less than
 5

instead of taking the three-clipped median, the un-

clipped standard deviation and mean signal are plotted
against each other, for every single pixel, in the left-hand
plot of Figure 3. This shows that most of the pixels are
found in a circular zone with a well-determined average
signal between 170 and 175 ADU, corresponding to the
median bias value, and with a standard deviation below
4 ADU. However, about 11% of the pixels show a tem-
poral standard deviation that is higher than this value
and also present deviations in the mean signal, with a
kite-shaped dispersion. If this scatter were caused by
pixels exhibiting higher values of read noise only, then
the points would be distributed vertically, like a gra-
dated column. If, however, the pixels had a different
threshold voltage, i.e., bias level, but maintained the
RON, then the points would be distributed horizontally.
Instead, the kite-shape indicates that either both occur
at the same time for these pixels or that there is an
additional anomalous behavior at play.
For comparison, the same number of frames was taken
with a CCD, the ANDOR iKon-L 936 BEX2-DD3 , and
the standard deviation and temporal average of each
pixel was also obtained. The kite-shaped dispersion is
Figure 2. Full-frame master bias of the QHY600M Pro not visible in this case. Furthermore, the histograms
(top) and QHY411M (bottom) obtained by 3σ-clipping me- along both axes show the expected Gaussian shape: on
dian stacking of 21 bias frames taken consecutively. the standard axis, a peak centered on the RON value;
and on the mean signal axis, a peak centered on√the
1 ADU, especially in the QHY411M, with more than 2 average bias level and a width equal to the RON/ 21.
ADU of difference between the upper left-hand and the In the case of the QHY411M—the same behavior is ob-
lower right-hand corner. served in the QHY600M Pro—the histograms show a
In the readout process of typical CCDs, the charge narrow peak and tails toward high standard deviation.
is transferred to one—or several in the case of large This implies that most of the pixels show a low RON,
sensors—output channels, each with a high-quality am- but some others have a skewed signal on both sides of
plifier and an analog-to-digital converter (ADC). In con- the average bias level.
trast, in CMOS, each pixel has its own built-in electron- With this distribution, defining a nominal value for
ics, which contain an amplifier and ADC. This allows a RON is not straightforward. Looking at the standard
very high frame rate by parallelizing the readout process deviation histogram, the peak is located at 3.0 ADU
but exposes the sensor to these kinds of pixel-to-pixel (3.1 e− ), and the rms at 3.74 ADU (3.83 e− ), close to
variations, both in darkness (with different readout or the 3.72 e− reported by the manufacturer. From here
thermal noises) and illuminated (with variations in gain on, the rms value will be used as a reference for the
or quantum efficiency). In sCMOS, each pixel must be RON. For the QHY600M Pro, an rms value of 3.48
understood as an independent sensor, so all corrections e− was obtained, compared to the 3.67 e− reported by
over them, such as subtraction of the bias level, must be the manufacturer. The values for the other modes of
done while bearing in mind that there is no single value operation are included in Table 3.
that characterizes all of the pixels simultaneously. For
this reason, although the QHY600M Pro and QHY411M Returning to the standard versus average plot for the
have an overscan region, it is more accurate to use a full sCMOS in Figure 5, the pixels are now grouped in ranges
frame master bias or dark that takes into consideration of std. σADU and the signal of each obtained in all of the
these non-uniformities between pixels.
Once the spatial behavior of each pixel in darkness has 3 https://andor.oxinst.com/assets/uploads/
been revised, the same 21 bias frames of the QHY411M products/andor/documents/andor-ikon-l-936-specifications.pdf
are used to see its temporal performance in detail. Now,
6
0.0010 ADU > 30.0
(CCD) Andor iKon-L 936 0.0005
40 (sCMOS) QHY411M
0.002 25.5 < ADU 30.0
0.001
30
(ADU)

0.0050 21.0 < ADU 25.5

0.0025
ADU

20 0.02 16.5 < ADU 21.0

Std.

0.01

Frequency (%)
10 0.050 12.0 < ADU 16.5
0.025
0.2 7.5 < ADU 12.0
160 180 200 220 0.01 0.1 1 10 0.1
10 Frequency (%)
Frequency (%)

1 10 3.0 < ADU 7.5

5
0.1
10 ADU 3.0
0.01
160 180 200 220 0 120 140 160 180 200 220
Mean signal (ADU) Pixel signal (ADU)
Figure 3. Left-hand panel: Standard deviation vs. mean signal of each QHY411M pixel (warm colors) in 21 consecutive bias
frames. Color indicates data density on a logarithmic scale. The histogram of the distribution of these two variables is included
below and to the right-hand side. For comparison, the same information is shown in blue tones for a CCD, the ANDOR iKon-L
936 BEX2DD. The QHY411M pixels have been grouped according to their standard deviation over the 21 frames and, for each
of them, the signal measured in these frames was taken individually. The distributions of each group are given in the right-hand
plot.

21 frames is retrieved. The distribution of each of these a column, given that the average of the signal would
groups is shown in the right-hand plot of Figure 5. For always lie in the central range. To show a kite-shaped
pixels showing a temporal deviation of the order of the dispersion, there must be some pixels that tend to
RON (σADU ≤7.5), the distribution of the measured sig- exhibit more deviations toward one of the Gaussians,
nal is Gaussian, similar to that of a conventional CCD. either the upper or the lower Gaussian, thus skewing
The large majority of the pixels are in these first two the average to either side and resulting in the widen-
ranges (45% of the pixels with σADU ≤3 and 98% with ing of the tails of the mean signal distribution, which
σADU ≤7.5). For the other ranges (with larger devia- is shown in yellow in the lower left-hand plot of Figure 3.
tions), the pixels begin to show a triple Gaussian dis-
tribution, with two smaller peaks that are centered on For a more detailed insight into the behavior of these
both sides of the central peak and with similar widths. pixels with high standard deviation, 1000 consecutive
For larger values of σADU the separation of the peaks is bias frames were obtained with the QHY600M Pro.
greater. In these anomalous pixels, the pixel distribu- The temporal average and standard deviation of all
tion is indicative of the following behavior: while most the frames has been taken in a small central region of
of the time these pixels return values that are around the 20 × 20, which is shown in the upper part of Figure 4.
mean value of the bias level with a normal noise equiva- Most of the pixels have an average signal around 171
lent to readout, their returned values occasionally jump ADU and a dispersion below the RON, as expected.
toward larger or smaller values with defined separations However, some of them show anomalous patterns in the
and different probabilities for each pixel. average value, the standard deviation, or both at the
The symmetry in the three Gaussian distributions same time. Several pixels have been selected as samples,
does not necessarily mean that all of the anomalous showing the time evolution of the signal in the lower
pixels jump between higher and lower levels with the plots. As a reference, a pixel with normal values of
same probability. In that case, the standard versus average signal and deviation has been taken, following
average plot would not show a kite-shaped scatter but a normal distribution over the 1000 frames, with mean
 7

the same effect more often, e.g., S&P 3, where 19.3%

of the 1000 frames show an anomalous value of ±60
ADU. This is revealed by a high standard deviation
of 23.6 ADU. Although this can be understood as a
high RON in its electronics, it should be noted that its
distribution is not a wide Gaussian but a set of three
normal distributions—the main distribution is centered
on 171 ADU, and two smaller ones corresponding to
these random leaps to higher and lower values at 110
and 230 ADU. The S&P 2 pixel shows jumps at the
same levels but with a higher proportion of values in
the upper level (10.6% of the total) than in the lower
level, 3%. Consequently, the signal that was obtained in
the average frame, 174.5 ADU, deviates from the other
pixels, as shown in the image.

These jumps between above- and below-average sig-

nal values are observed when blinking between images,
regardless of their exposure time and temperature. A
pattern of bright and dark pixels that appear and disap-
pear from one frame to the next is observed. This effect,
which is sometimes referred as Salt & Pepper, is random
telegraph noise (RTN). RTN is the fluctuation of the sig-
nal between discrete levels as a consequence of the cap-
ture and emission of charges by defects or traps located
very close to the Si–SiO2 interface (Uren et al. 1985).
In scaled CMOS detectors, this trapping process causes
a shift in the relationship between the drain current of
the MOS transistor and the gate voltage, discretely in-
creasing or decreasing the offset level, which fluctuates
as random trapping and de-trapping of charges, either
electrons or holes (Martin-Martinez et al. 2020).
This random noise must be addressed when process-
ing images taken with these detectors. Unusually low
Figure 4. Top: temporal mean signal (left-hand) and stan-
or high pixel values can cause deviations in photomet-
dard deviation n terms of readout noise (right-hand) ob-
tained in a 20 × 20 pixel central region of 1000 bias frames ric measurements of sources in low light levels. Simply
taken consecutively with the QHY600M Pro. Bottom: signal averaging images is not an optimal method because it
vs. frame number for some relevant pixels identified in the does not avoid, but may reduce, RTN. For instance, the
images above. Points within the average value of the master pixel “Salt” in the Figure 4 shows fluctuations only to-
bias ±3× RON are shown in black, with outliers identified ward higher, not lower, values, and therefore any aver-
in red. The signal distribution of these pixels is displayed on aging that is done by including any of the anomalous
the right-hand side.
points (25.5%) will be inaccurate. The way to deal with
these outliers is to take several images and stack them
170.4 ADU and standard deviation 3.69 ADU. The pixel using a more robust statistic. This technique will be dis-
labeled S&P 1 also has an average signal similar to the cussed further in Section 5. It is also possible to identify
others, with a slightly higher standard deviation of 6.42 some pixels with higher than average signals but with
ADU but still close to the RON obtained in the previ- low standard deviation. Some of them are independent
ous section. However, the temporal distribution reveals of exposure time, such as the pixel “High” in the fig-
how some points, 5.5% of the total, appear both above ure, which can be corrected by subtracting the master
and below the average signal, with a gap of about 20 bias. However, others depend on exposure time and are
ADU, which is more than three times the RON (see treated as warm pixels in the next section.
the red dots in the plots below). Other pixels show
8

5 C 5 C
10 0 C 10 10 C
5 C 15 C
10 C 20 C
1 1
Frecuency (%)

Frecuency (%)
0.1 0.1

0.01 0.01

0.001 10 3 10 2 10 1 100 0.001 10 3 10 2 10 1 100

Dark current (e /px/s) Dark current (e /px/s)

Figure 5. Dark current distributions for the QHY600M Pro (left-hand panel) and the QHY411M (right-hand panel). They
were obtained as the median stacked set of five dark frames with an exposure time of 1000 s.

3.2. Dark current and warm pixels S = (0.007±0.007)(ADU/s) texp(s) + (171.6±0.6)(ADU)

S = (0.029±0.007)(ADU/s) texp(s) + (171.4±0.5)(ADU)
Dark current (DC) refers to the unwanted leakage S = (0.405±0.006)(ADU/s) texp(s) + (171.9±0.5)(ADU)
current that is generated in photosensitive devices in S = (0.839±0.012)(ADU/s) texp(s) + (171.3±0.9)(ADU)
the absence of incoming light, which is mainly due to the
thermal generation of charges in the silicon layer, and Normal 1
325 Normal 2
is strongly dependent on the temperature and exposure Warm 1
time. To characterize the DC, five dark frames of 1000 Warm 2
300
s exposure time were taken, a master bias taken just
before was subtracted, and they were median stacked. 275
Signal (ADU)

This was done for various temperatures: from 5◦ C to

−10◦ C on the QHY600M Pro and from −5◦ C to −20◦ C 250
on the QHY411M, on which water cooling was installed
for these tests. The DC distribution in electrons per 225
pixel and seconds of exposure is shown in Figure 5. The
distribution curves have similar shapes, with a shift to- 200
ward higher DC with increasing temperature. When the
175
peak is reached, the DC drops rapidly toward a smooth
hump, from which it drops back down again. This hump 0 25 50 75 100 125 150 175
Exposure time (s)
corresponds to a set of pixels with exceptionally higher
DC, which are seen in the images as pixels with a higher Figure 6. Signal of two warm and two normal pixels of
signal that is steady from frame to frame, unlike the the QHY411M as a function of exposure time, obtained as
Salt & Pepper effect described in the previous section, the median value of these pixels over sets of 21 frames taken
which increases with exposure time. At −10◦ C, only consecutively, with the standard deviation included as its
0.024% of the QHY600M Pro pixels and 0.005% of the uncertainty. The parameters of the least squares fitted lines
QHY411M have a DC greater than RON. These warm are shown at the top.
pixels have signals lower than the saturation level, and
therefore they can be corrected with an appropriate reported for other sCMOS sensors (Karpov et al. 2021).
master dark. The median value of the dark current for
each camera and temperature is shown in Table 2. It
is worth mentioning here that no glow has been seen Some warm pixels in a central region of the QHY411M
in any of the darks taken in either of the cameras, as have been identified and revised in more detail. First,
sets of 21 dark frames taken consecutively with different
 9

Normal 1 subtracted from the science images, it should be noted

210 Normal 1 that warm pixels, having a higher thermal signal, will
Warm 1 be noisier than those with a lower DC.
Warm 2 For more than 3 weeks, sets of 21 darks of 30 s ex-
200 posure were taken every night. The median value in
these four pixels as a function of time is shown in Fig-
Signal (ADU)

ure 7, where the error bars again show their standard

190 deviation. With this test, we have tried to see if there
is a substantial variation in the values of these pixels
over longer periods of time. Taking into account that
180
the noise in these pixels is a combination of the RON
and the thermally generated electrons, which follow a
170 Poissonian distribution, it can be seen that the possi-
ble variations in the pixel dark signal over at least these
16-10-2022 23-10-2022 30-10-2022 three weeks is not significant with respect to the total
Date
noise. Therefore, meaningful night-to-night variations
of the master dark or bias are not expected.
Figure 7. Signal of two warm and two normal pixels of
the QHY411M obtained daily for more than three weeks,
3.3. Photon Transfer Curve and linearity
obtained as the median of the value of these pixels in sets of
21 frames of 30 s exposure taken consecutively, including the The photon transfer curve (PTC) is used to charac-
standard deviation as its uncertainty. terize the response of the sensors to homogeneous illu-
mination, obtaining main features such as the gain—
conversion factor between electrons and digital counts
Table 2. Median dark current (DC) for
(ADU), the full-well capacity (FWC), or the contribu-
different cooling temperatures for each
camera. tion of the different sources to the total noise. The lab-
oratory set-up that is described in Figure 1 was modi-
T (◦ C) DC (e− px−1 s−1 ) fied by removing the monochromator, so that the light
coming from the QTH lamp could directly enter the in-
QHY600M Pro QHY411M tegrating sphere. The power supply was kept stable.
5 0.025 Sets of three images of uniformly illuminated exposures
0 0.018 were then taken while increasing the exposure time un-
-5 0.014 0.009 til the saturation turn-off point was reached. Three bias
-10 0.011 0.007 frames were taken before and after each series, stacked
-15 0.005 by median 3σ-clipping and subtracted from all of the il-
-20 0.004 luminated images. A central region of interest (ROI) of
4096 × 4096 unbiased pixels was used for both cameras.
The signal was obtained as the mean of the three images
stacked. The mean of the standard deviation across the
three frames was taken to obtain the noise value.
exposure times have been stacked. The median signal of
The total noise under illumination is given by the
two warm and two normal pixels is shown in Figure 6,
quadrature sum of three main components: RON, shot
where the standard deviation is included as uncertainty
noise, and fixed pattern noise (FPN):
bars. It is clear that the signal of these warm pixels
scales linearly with the exposure time in the range of 1– 
S
1/2
2 2
300 s. The result of the linear least squares fit is shown σTOTAL (ADU) = σRON + + (PFPN S) (1)
G
at the top of the plot, all of them with R2 > 0.998. Note
that in this case a master bias has not been subtracted, where σRON (ADU) is the readout noise, S (ADU) the
and therefore the value of the offset is different from 0. signal, G (e− /ADU) the gain and PFPN the FPN factor
This result implies that it is possible to correct the warm (Janesick 2007).
pixels in images of a given exposure time if the slope of The relationship between the noise and signal for the
this line for each pixel is known, which can be obtained two sensors is shown in Figure 8. The RON was fixed
with other exposure times (e.g., at the beginning of the with the value defined in the previous section, while the
night). Although a scaled master dark is intended to be gain and FPN parameters were obtained by fitting the
10

Total noise Total noise

Readout noise Readout noise
Shot noise Shot noise
Fixed pattern noise Fixed pattern noise
Gain = 0.763 ± 0.003 e /ADU Gain = 1.024 ± 0.010 e /ADU
FPN = 0.55 ± 0.07 % FPN = 0.31 ± 0.02 %
102
102
Noise (ADU)

Noise (ADU)
101 101

100 101 102 103 104 105 100 101 102 103 104 105
Signal (ADU) Signal (ADU)
Figure 8. Photon transfer curves obtained from sets of three bias-subtracted images, median stacked, and taken under uniform
illumination with increasing exposure times. The QHY600M Pro Mode #1@0 (left-hand panel) and the QHY411M Mode #4@0
(right-hand panel). The principal components of the noise have been distinguished as colored lines, whose characteristic values
are also included. The gain and FPN has been obtained from the fitting to the total noise function.

60000 Saturation level = 65532±1 ADU 60000 Saturation level = 65523±1 ADU
FWC = 50.0±0.2 ke FWC = 67.1±0.7 ke
50000 50000
Signal (ADU)

Signal (ADU)

40000 40000
30000 30000
20000 20000
10000 10000
0 0
100 100
Linearity (%)

Linearity (%)

99 99
98 98
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 160
Exposure time (ms) Exposure time (ms)
Figure 9. Average signal in sets of three images taken with increasing exposure time (top) for the QHY600M Pro Mode #1@0
(left-hand panel) and the QHY411M Mode #4@0 (right-hand panel). The saturation level is indicated by a horizontal-dashed
line. A straight line was fitted to points with values between 100 and 60,000 ADU. The relative rms of the differences between
the measurements and the fitted line is shown in the lower plots.

expression (1) on a logarithmic scale. This procedure component between 100 and 10,000 ADU. Thereafter,
was repeated in windows of 200 × 200 pixels along the the contribution of the FPN becomes significant and
ROI to obtaining a curve for each of them, 360 in to- a deviation from Poissonian behavior is observed. In
tal. The rms of the parameter distributions was used the QHY600M Pro, this source is the main contributor
to estimate their uncertainty. In both plots, three re- above 50,000 ADU, with a factor of 0.55% ± 0.02% ,
gions can be distinguished. At low illumination, below being lower, 0.31% ± 0.02%, in the QHY411M, where
10 ADU, RON, which does not depend on the signal, is the shot noise is dominant almost up to the saturation
the dominating noise source. From then on, shot noise point. A PTC has also been obtained for other operat-
starts to become important, and is essentially the main
 11
100
Table 3. Main features of the different operating modes of QHY600M PRO
the QHY600M Pro (top) and QHY411M (bottom), obtained QHY411M
in laboratory tests. including the gain, readout noise (RON),
fixed pattern noise (FPN) and full-well capacity (FWC). 80

Quantum Efficiency (e-/ph, %)

Mode Gain RON FPN FWC
60
(e− /ADU) (ADU) (%) (ke− )
QHY600M Pro
0@26 0.405±0.003 6.28 0.521±0.005 25.3±0.2 40
1@0 0.763±0.003 4.56 0.55±0.07 50.0±0.2
QHY411M
1@0 0.979±0.008 3.03 0.29±0.02 27.2±0.2 20
4@0 1.024±0.010 3.74 0.31±0.02 67.1±0.7
5@0 0.332±0.002 4.65 0.32±0.02 22.8±0.1
6@0 0.776±0.003 4.48 0.32±0.02 50.6±0.2
0 400 500 600 700 800 900 1000 1100
7@0 0.251±0.001 5.69 0.34±0.02 16.50±0.05 Wavelength (nm)
Figure 10. Absolute quantum efficiency curves of the
QHY600M Pro (blue-open circles) and the QHY411M (yel-
ing modes, whose resulting values are included in Table low dots). The systematic uncertainty is estimated to be
3. around 2%
The linearity of the sensors is evaluated by plotting the
signal (adding the bias level) against the exposure time, with the picoammeter for approximately 30 s, taking an
as is done for the standard operating modes in Figure average value. The observed fluctuations in this value
9. A straight line was fitted to the points between 100 were always less than 2%. The systematic uncertainty
and 60,000 ADU (top plot). The relative rms difference of the method is estimated to be around 2%. To avoid
between the points and the line is shown below. In both vignetting effects at the edges of the dark box-camera
cases, the deviation from linearity is less than 2% up to junction, a central ROI of 1000 × 1000 binned pixels was
the saturation point, which was determined for all of the used to obtain the median signal S, after checking that
operating modes and included in Table 3. there were no inhomogeneities in the sensor illumination
in that area. The exposure time texp had to be varied
3.4. Quantum efficiency throughout wavelength steps to keep all of the measure-
ments in the shot noise dominated region between 1000
Using the optical setting described in Figure 1, central
and 40,000 ADUs. The absolute quantum efficiency for
wavelengths in the range between 350 and 1100 nm with
each wavelength is given by the expression:
25 nm steps were selected in the monochromator. The
grating configuration was set to have an outcoming light S(ADU) · G(e− /ADU)
with ±1 nm bandwidth. The QHY600M Pro was placed QE(%) = · 100 (2)
texp (s) · F (e− /s)
at 13.3 mm from the exit of the dark box, which is sepa-
rated by 30 mm from the Hamamatsu S2281 photodiode. The QE curves for the sensors are shown in Figure
The total distance between the IMX455 sensor and the 10. Both show very similar behavior, with a peak ef-
photodiode, considering the back focus distance of the ficiency of 80% at 475 nm, a steep drop at shorter
camera, was 66.8 mm. In the case of the QHY411M, it wavelengths, and a gradual decrease toward the red-
could be placed in contact with the box, so the total dis- der ones, with a QE of 40% at 700 nm and 10% at
tance between the photodiode and the IMX411 sensor 900 nm. Back-illuminated sCMOS sensors with reduced
was 58.5 mm. Both cameras were binned 4 × 4 to have pixel size, such as the IMX455 and IMX411, have a typ-
enough signal with exposure times shorter than 10 s at ical silicon substrate thickness of around 3 µm (Yoko-
those wavelengths where they are less efficient. gawa et al. 2017). This optimizes the photon absorption
Three images were taken at each wavelength step, a in the visible range but makes the less energetic pho-
master bias that was created at the beginning of the tons, which have a higher penetration capability, less
series was subtracted from all of them and they were likely to be detected. Without any additional red en-
stacked with the 3σ-clipped median. Simultaneously, hancement technology, improving efficiency in the red
the output intensity of the photodiode placed at the and near-infrared requires a thicker substrate that, with
secondary port of the integrating sphere was measured such small pixels, would lead to image degradation ow-
12

Figure 11. Etaloning pattern observed when illuminating the QHY600M Pro in a central ROI of 1000 × 1000 pixels 4 × 4
binned with ±1 nm bandwidth light at different wavelengths. Similar behaviour is observed in the QHY411M.

ing to crosstalk between adjacent pixels. This is the nm and above. This is a known effect in thinned back-
reason for the poor performance at longer wavelengths. illuminated devices, resulting from multiple reflections
It should be noted that the curves that are obtained produced inside the depletion region by a mismatch be-
here represent an overall reduction of 9% over the full tween its refractive indices and the adjacent layers. This
absolute QE curve published by QHYCCD, which has effect has not been observed in sky images taken with
a peak of 92% at 450 nm, 46% at 700 nm, and 15% TAR04 using a wider bandwidth, such as SDSS i0 (690–
at 900 nm. Nonetheless, the conditions and configura- 850 nm).
tion used by them, as well as the uncertainties in their
measurements, are unknown and the calibration method 3.5. Charge persistence
to obtain the QE curves is different from the one used Charge persistence is an effect that occurs when a por-
here, which has been used to calibrate multiple astro- tion of the signal remains in the detector element after
nomical instruments before. The same procedure and the sensor has been read out. It is a consequence of the
test bench have been used, for instance, to calculate the creation of traps at the interface between the photodi-
QE in sCMOS cameras such as Andor Marana, FLI- ode and the transfer gates, which capture free electrons
Kepler, and ORCA-Hamamatsu. They have also been and gradually release them, resulting in a decay of the
used in several deep depletion CCDs, such as the well- residual signal in the subsequent images after the illu-
known Teledyne e2v 4482 and 231-84 BI. In all cases, minating source has been removed. This effect has been
QE fits rather well with data supplied by the manufac- observed with tests performed in the laboratory with the
turers. Gill et al. (2022) have very recently presented FLI Kepler KL400 and Andor Marana cameras, both
a low-cost method to obtain, among other features, the with GSENSE400BSI sensors, showing a behavior simi-
absolute QE of detectors being applied to the IMX455 lar to that reported by Karpov et al. (2021). Although
sensor. Their results also show a lower performance than they did not observe any effect in sky images, previ-
that presented by QHYCCD, especially in the red part, ous tests performed with these cameras on the same
although they had a similar peak efficiency, with 93% at telescopes used in our work did show a smearing ef-
480 nm, 42% at 700 nm, and 10% at 900 nm, making fect in pixels that had previously been exposed close to
an overall deviation from our results of 5%. Finally, Be- the FWC, which remained visible up to tens of minutes
toule et al. (2023) have studied the quantum efficiency later.
of the IMX411 sensor in great detail and obtained a re- This effect has been tested in the laboratory for both
sult that is very similar to Figure 10, with a peak of 83% sensors. To do so, following the same set-up used in
at 490 nm, 37% at 700 nm and 6% at 900 nm. Section §3.3, i.e., removing the monochromator and ex-
During the tests, optical etaloning was observed at posing the cameras to the QTH lamp light, a pinhole
longer wavelengths, as shown in Figure 11 for the was placed inside the black box, just in front of the sen-
QHY600M Pro. Similar behavior was observed also for sor. Consecutive 1 s images were taken with the shutter
the QHY411M and reported by Betoule et al. (2023). closed. The shutter was then opened for about 10 s, sat-
The maximum variation over the average frame mean urating the illuminated pixels, and closed again. This
value is around 1% at 650 nm, and reaches 10% at 850 sequence is shown in Figure 12. On the top are the
 13

16.85
16.90
16.95

Magnitude
17.00
17.05
17.10
17.15 Period = 3.60 ± 0.03 hr
Amp = 0.13 mag
Figure 12. Average signal (black dots) in 1 second frames 0.0 0.2 0.4 0.6 0.8 1.0
Phase
taken continuously during sensor flooding (central shaded
area). Images before (left-hand), during (center), and after
(right-hand) are shown at the top. The region in which the Figure 13. Phased light curve of the asteroid (3200)
average signal has been calculated is indicated by dashed Phaethon, obtained with the QHY600M mounted at the
rectangles. prime focus of a 0.46-m f /2.2 telescope, no filter installed.

images taken with the QHY600M Pro before the shut-

ter was opened (left-hand), during the exposure to the 10
light coming from the pinhole (center), and just after 5
the shutter was closed again (right-hand). In the cen-
Relative flux (ppt)

0
tral region of the illuminated spot area, indicated by a
dashed square, the mean signal has been measured and 5
is shown in black dots below. The signal before and after 10
the flooding shows a stable trend, with very small oscil-
lations, not a sharp increase in the signal right after the 15
exposition followed by a slow decay to the original level, 20
as shown in Figure 10 of Karpov et al. (2021). Hence, 25 0.00
we conclude that the IMX455M and IMX411M sensors 0.05 0.10 0.15 0.20 0.25 0.30
JD - 2459936.3319
do not exhibit charge persistence.
Figure 14. Transit of the exoplanet TOI-1135 observed
4. ON-SKY TESTS with QHY411M mounted on a Nasmyth focus of the TTT-2
Both cameras have been extensively tested on images telescope at 0.80-m and f /6.85, SDSS g0 filter.
taken with telescopes, obtaining photometric accuracies
as expected for the characteristics described above. The low readout noise and negligible readout time, it is pos-
QHY600M Pro was installed at the prime focus of a 0.46- sible to take continuous short frames and combine them
m f /2.2 telescope. Figure 13 shows the phased light by aligning with the object or the stars, thus allowing
curve of asteroid (3200) Phaethon, which was observed fainter objects to be reached with very little time lost.
for 6 consecutive hours. At the time of observation, the This also allows the study of, for instance, very fast ro-
object had an apparent magnitude of V = 17 and was tating objects with good temporal sampling (Licandro
moving at a speed of 100 .6 minute−1. The photometric et al. 2023b) or using “shift-and-add” techniques, such
uncertainties that we obtained are in the order of hun- as synthetic tracking (Shao et al. 2014), to improve the
dredths of a magnitude and the rotation of the asteroid, detection performance of faint fast-moving objects.
with an amplitude of 0.13 mag, is clearly detectable.
This camera is a very suitable choice for very fast tele- Figure 14 shows a transit of the exoplanet TOI-1135
scopes with primary focus because its small size and that was observed with the QHY411M mounted on one
compact shape drastically reduce obscuration. With a of the Nasmyth foci of the TTT-2 telescope, 0.80-m
pixel size of 3.76 µm, it also allows us to obtain a plate f /6.85, SDSS g0 filter, with a plate scale of 000 .14 px−1 .
scale that is very suitable for sites with excellent seeing, Groups of four images of 5 s exposure were stacked to
such as the Teide Observatory. Furthermore, given its improve the SNR. The standard deviation of the differ-
14

ential aperture photometry measurements in pre-transit, tometric measurements. Figure 15(a) shows an image
during, and post-transit is 2.2, 3.3, and 2.0 mmag, re- of the galaxy M33 taken with the QHY411M on the
spectively (M. Mallorquin et al., in preparation). With TAR04 telescope, with a UV/IR-Cut/L filter, in a sin-
such a small plate scale, the psf was oversampled, with gle exposure of 60 s. When zooming in on a small area of
approximately 8 px of FWHM, spreading the flux of the 100×100 pixels (b), several pixels are observed with val-
star over a larger number of pixels. For very bright tar- ues that are clearly higher than those of their neighbors,
gets, such as this V = 9.6 mag star, this allows slightly as described in Section 3.1. This effect could influence
longer exposures to be taken without reaching the satu- the photometry of faint objects because the amplitude
ration point, thus reducing the random and flat-fielding of this random fluctuation may be comparable to the
errors of the telescope that, with a larger scale, would source signal in the pixel.
perhaps need some defocusing (Southworth et al. 2009). SSeveral strategies can be pursued to mitigate this
In addition, by taking short exposures, the time lost on problem, especially by using several consecutive frames.
a CCD could be equivalent to, or even longer than, the Figure 15(c) shows a sequence of 3 × 20s, where the
exposure time, which makes it very inefficient. With the S&P is clearly visible in each of them. If simple av-
QHY411M, the exposure time could be extended, thus eraging were performed, then the outliers would skew
improving photometric accuracy, and the readout time the signal obtained. Algorithms such as 3σ clipping
is almost non-existent. Although such small plate scales that use outlier-sensitive dispersion measures generally
are generally undesirable, they allow these cameras to do not work either because these metrics are biased by
be used in other scientific applications (e.g., fast-moving these fluctuations and define a very wide clipping range.
object astrometry or lucky imaging). This is seen in Figure 15(d) (top plot), where the three
frames have been combined with an average after 3σ-
5. DISCUSSION clipping. Some of the pixels that exhibited S&P also
show deviating values in the stacked frame.
The two instruments with sCMOS sensors that are
It is common to use spatial filters for this kind of
analyzed here present characteristics that are compat-
localized noise. The most typical for S&P is a median
ible with their use in astronomy: they are linear over
filter, where the value of each pixel is replaced by the
the whole dynamic range, have a high full-well capac-
median of the values of this and its close neighbors. This
ity, and are slightly affected by dark current, despite
is done in the second row of Figure 15(c). Note that
being able to work at higher temperatures than CCDs.
the image has been smoothed and the anomalous pixels
Regarding the quantum efficiency, although the curve
may have disappeared. However, this has been done at
obtained here is slightly lower than that reported by the
the expense of: (1) changing the signal and noise pat-
manufacturer, 80% at 500 nm is an acceptable perfor-
tern of the image and (2) correlating the nearby pixels.
mance for many scientific programmes and is in general
Although this is a very useful filter for improving the
similar or better than other CCD sensors in the same
cosmetics of the images, the photometry measurement
cost range. An improvement in efficiency toward redder
in the resulting image may be highly biased because the
wavelengths should be achieved in the next few years,
fluxes of each pixel have been altered by its neighbors.
so that sCMOS sensors can be used on a wider variety
of observational targets.
In this work, a solution based on convolutions and
The small pixel size means that these sensors are gen-
these two previous ideas is proposed to try to mitigate
erally not the best solution for slow focal length systems,
the S&P effect. First, it should be considered that the
except for dedicated programmes such as high spatial
S&P effect mostly impacts low noise areas, such as the
resolution or lucky imaging. Binning in sCMOS sensors
sky background or faint sources. For bright sources, the
is done after exposure, so it does not improve the read-
shot noise becomes higher than the random telegraph
out noise or frame rate. In general, in case it is needed,
noise and dominates all of the other fluctuations. The
it is better to do this by software after the exposure, so
median filter can be used to obtain a reference frame to
that the statistics can be preserved and the 16 bit limit is
identify outliers because, in well-sampled fields, they are
not reached. Nevertheless, they can be very valuable in
a good approximation to a smoothed frame. This can be
fast telescopes with larger fields and higher plate scales,
seen in Figure 15(c), where the third row shows the dif-
which allows better sampling of the PSF. In addition,
ference between the raw frame and the one filtered with
their manageability, and small size and weight are very
a 3 × 3 median kernel. The residual pattern is generally
interesting, e.g., for prime focus telescopes.
homogeneous, both in the sky area and in the vicinity of
The S&P effect is one of the main issues that affect
the sources. To identify the S&P, a threshold of 12 ADU
the use of these IMX455 and IMX411 sensors for pho-
 15

Figure 15. (a) Image of M33 taken with the QHY411M at the TAR04 telescope with an exposure time of 60 s. (b) Zoomed
region 100 × 100 pixels where the Salt & Pepper effect is visible. (c) From top to bottom: 3 × 20 s sequence of the same
zoomed region; convolution with a 3 × 3 median filter; difference between the frames in the previous rows, showing the S&P
contaminated pixels in red; and raw frames with those pixels masked in red. (d) Result of stacking the three frames with an
average 3σ-sigma clipping (top) and with an average after masking the S&P affected pixels.
16

has been set because it is at this point that the distribu- 2. Their dark current is very low, as is the number of
tion of three Gaussians in Figure 3 (right-hand) starts warm pixels. They are stable for at least several
to be revealed. Hence, all of the pixels whose absolute weeks and their signal scales linearly with expo-
difference between the raw value and that resulting from sure time, so they may be quite easily removed
the convolution with the median filter is greater than 12 with dark subtraction. This is an improvement
ADU are masked. In the bottom row of Figure 15(c), from previous sCMOS sensors and makes it pos-
the raw frames are shown with the pixels masked in red. sible to take images with longer exposure times
By having a sequence of frames, the average of the un- without being affected by dark current.
masked values of each pixel can be taken to get a stacked
3. Its quantum efficiency peaks at 80% at 475 nm
image, which is shown on the bottom right-hand of the
and then drops rapidly at longer wavelengths, with
figure, where the S&P contamination has been highly
40% at 700 nm and 10% at 900 nm.
reduced. In cases where a pixel shows S&P in all of the
frames of the sequence and is therefore masked, the as- 4. They do not exhibit charge persistence or edge
tronomer has to decide, for instance, either not to take glow.
that pixel into account in the photometry or to replace
its value by an approximation, such as an interpolation 5. These sensors are affected by random telegraph
of the neighboring pixels or their median. In the exam- noise, which can introduce non-negligible de-
ple included here, this happens in only two pixels out viations in photometric measurements of low-
of 104 . It should be noted that this method may be brightness sources. Simple frame averaging, even
less accurate in fields with critically sampled sources. A with algorithms such as σ-clipping, is not enough
further review is currently underway for future work. to mitigate its effect.
Developing new algorithms or even using those already 6. They show promising performance on photomet-
available in common packages such as IRAF (Tody 1986) ric observations done with both fast and slow tele-
or Astropy (Astropy Collaboration et al. 2013) may not scopes.
be trivial when working with these cameras. The size of
Their low cost, power consumption, and replicability
the raw 16 bit images of the QHY600M Pro are about
make both cameras a very suitable solution for astro-
120 MB, while those of the QHY411M are 300 MB. If
nomical applications, especially with regard to their
a sequence is taken at a high frame rate, then the data
high frame range, near-zero readout time, and low read-
set may not be manageable with commonly available
out noise. These sensors are very good options for fast
CPU capacities. The advantage of convolution-based
small telescopes with large fields because of their small
approaches and simple arithmetical operations, such as
pixel size and large formats. The combination of such
median filtering or frame differencing, is that they are
telescopes and cameras permit very large field-of-view
easily deployable on GPUs, which allows the data to be
images with plate scales with reasonably good sampling
processed more efficiently and faster. The development
of the PSF. For instance, a 1100 f /2.2 telescope such
of GPU algorithms for astronomical image processing is
as the Celestron RASA11 with a QHY600 camera can
essential for further progress in the use of large sensors,
produce images covering a FOV of 7.5 deg2 with a plate
such as these sCMOS.
scale of 100 .27 px−1 , which are excellent options for sur-
6. CONCLUSION veys such as ATLAS-Teide (Licandro et al. 2023a). Even
so, having the necessary computational tools to process
In the previous sections, the key features of the
the data, especially GPU developments, is essential to
QHY600M Pro and QHY411M cameras as scientific in-
take advantage of the full performance of these cameras.
struments have been discussed in detail. For astronomy,
they have characteristics that make them very suitable
The authors declare no conflict of interest or rela-
for general use, although certain issues need to be ad-
tionship with the manufacturers of the cameras tested.
dressed. Our main conclusions are that:
M.R.A, M.S-R and J.L. acknowledge support from the
1. The built-in electronics in the pixels of sCMOS ACIISI, Consejerı́a de Economı́a, Conocimiento y Em-
sensors require that each pixel should be consid- pleo del Gobierno de Canarias and the European Re-
ered as an individual detector and this should gional Development Fund (ERDF) under grant with ref-
be taken into account when performing processes erence ProID2021010134 and from the Agencia Estatal
such as bias or dark subtraction. Spatial inhomo- de Investigacion del Ministerio de Ciencia e Ińnovacion
geneities in darkness are detectable all over the (AEI-MCINN) under grant ”Hydrated Minerals and Or-
frame. ganic Compounds in Primitive Asteroids” with reference
 17

PID2020-120464GB-100. This research has been par- presented here. This article includes observations made
tially funded by Light Bridges, SL. which provided the in the Two meter Twin Telescope (TTT) at the IAC’s
QHY411M and Andor iKon-L 936 cameras for the tests Teide Observatory that Light Bridges, SL, operates on
the Island of Tenerife, Canary Islands (Spain).

REFERENCES
Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., Licandro, J., Popescu, M., Tatsumi, E., et al. 2023b,
et al. 2013, A&A, 558, A33, MNRAS, 521, 3784, doi: 10.1093/mnras/stad708
doi: 10.1051/0004-6361/201322068 Martin-Martinez, J., Rodriguez, R., & Nafria, M. 2020,
Betoule, M., Antier, S., Bertin, E., et al. 2023, A&A, 670, Advanced characterization and analysis of random
A119, doi: 10.1051/0004-6361/202244973 telegraph noise in CMOS devices (Springer), 467–493
Bigas, M., Cabruja, E., Forest, J., & Salvi, J. 2006, Ofek, E. O., Ben-Ami, S., Polishook, D., et al. 2023, arXiv
Microelectronics J., 37, 433, e-prints, arXiv:2304.04796,
doi: https://doi.org/10.1016/j.mejo.2005.07.002 doi: 10.48550/arXiv.2304.04796
Boyle, W. S., & Smith, G. E. 1970, Bell Syst. Tech. J., 49, Princeton Instruments. 2016, New Scientific CMOS
587, doi: 10.1002/j.1538-7305.1970.tb01790.x Cameras with Back-Illuminated Technology. Technical
Coates, C., Fowler, B., & Holts, G. 2009, White Paper: Note., Available at: https:
//www.princetoninstruments.com/wp-content/uploads/
Scientific CMOS technology, a high-performance imaging
2020/11/TechNote sCMOSBackIlluminatedTech.pdf
breakthrough, Available at:
(accessed July 8, 2022)
https://www.pco.de/fileadmin/user upload/db/
Qiu, P., Mao, Y.-N., Lu, X.-M., Xiang, E., & Jiang, X.-J.
download/scmos white paper 2mb.pdf (accessed Oct 3,
2013, Research in Astronomy and Astrophysics, 13, 615,
2022)
doi: 10.1088/1674-4527/13/5/012
Fossum, E. R. 1997, Nucl. Instrum. Methods Phys. Res. A,
Qiu, P., Zhao, Y., Zheng, J., Wang, J.-F., & Jiang, X.-J.
395, 291, doi: 10.1016/S0168-9002(97)00812-7
2021, RAA, 21, 268, doi: 10.1088/1674-4527/21/10/268
Gill, A. S., Shaaban, M. M., Tohuvavohu, A., et al. 2022, in Schildknecht, T., Hinze, A., Schlatter, P., et al. 2013, in
Society of Photo-Optical Instrumentation Engineers Proceedings of 6th European Conference on Space
(SPIE) Conference Series, Vol. 12191, X-Ray, Optical, Debris, Darmstadt, Germany
and Infrared Detectors for Astronomy X, ed. A. D. Shao, M., Nemati, B., Zhai, C., et al. 2014, ApJ, 782, 1,
Holland & J. Beletic, 1219114, doi: 10.1117/12.2627564 doi: 10.1088/0004-637X/782/1/1
Janesick, J. R. 2007, Photon Transfer DN → λ Southworth, J., Hinse, T. C., Jørgensen, U. G., et al. 2009,
(Bellingham, WA: SPIE Press) MNRAS, 396, 1023,
Janesick, J. R., Elliott, T., Collins, S., Blouke, M. M., & doi: 10.1111/j.1365-2966.2009.14767.x
Freeman, J. 1987, Optical Engineering, 26, 692, Tody, D. 1986, in Society of Photo-Optical Instrumentation
doi: 10.1117/12.7974139 Engineers (SPIE) Conference Series, Vol. 627,
Karpov, S., Bajat, A., Christov, A., Prouza, M., & Beskin, Instrumentation in astronomy VI, ed. D. L. Crawford,
G. 2021, arXiv e-prints, arXiv:2101.01517. 733, doi: 10.1117/12.968154
https://arxiv.org/abs/2101.01517 Tonry, J. L., Denneau, L., Heinze, A. N., et al. 2018, PASP,
Law, N. M., Corbett, H., Galliher, N. W., et al. 2022, 130, 064505, doi: 10.1088/1538-3873/aabadf
PASP, 134, 035003, doi: 10.1088/1538-3873/ac4811 Uren, M., Day, D., & Kirton, M. 1985, Applied physics
letters, 47, 1195
Licandro, J., Tonry, J., Alarcon, M. R., Serra-Ricart, M., &
Yokogawa, S., Oshiyama, I., Ikeda, H., et al. 2017, Scientific
Denneau, L. 2023a, arXiv e-prints, arXiv:2302.07954,
Reports, 7, 3832, doi: 10.1038/s41598-017-04200-y
doi: 10.48550/arXiv.2302.07954