Infrared Physics & Technology 42 (2001) 1±16

www.elsevier.com/locate/infrared

Extended wavelength InGaAs infrared (1.0±2.4 lm) detector

arrays on SCIAMACHY for space-based spectrometry of the
Earth atmosphere
Ruud W.M. Hoogeveen *, Ronald J. van der A, Albert P.H. Goede
Space Research Organization Netherlands (SRON), Sorbonnelaan 2, 3584 CA Utrecht, Netherlands
Received 22 August 2000

Abstract
Space quali®ed InGaAs solid-state array detectors, covering the near-infrared 1±2.4 lm wavelength range, have been
developed for application in space-based spectroscopy of the Earth atmosphere. The SCIAMACHY atmospheric
chemistry instrument on the ESA ENVISAT satellite (2002±2005) will be equipped with a series of these novel de-
tectors. Detectors are arranged in linear arrays of 1024 pixels of 25  500 lm2 dimension and meet requirements on
modestly low operating temperature (150 K) and low levels of dark current and noise. In this paper the underlying
physics of dark current and noise of the detector system is studied on the basis of a theoretical model in combination
with measurements. At 2.4 lm wavelength the dark-current performance achieved is 20±100 fA at an operating tem-
perature of 150 K and a bias voltage of ÿ2 mV. This corresponds to a ®gure of merit for detector resistance R0 times
detector pixel area A of R0 A ˆ 2:5±12.5 MX cm2 . This result has required the development of a customised multiplexer
for parallel detector read-out at near-zero bias voltage. Further reduction of the operating temperature will not result in
lowering the dark current and noise of the InGaAs detectors which are shown to be limited by tunnelling current. A
route to future improvement is discussed. Ó 2001 Elsevier Science B.V. All rights reserved.

PACS: 07.87.+v; 95.55.Qf; 85.60.Gz; 85.60.Bt

Keywords: Space-based remote sensing; Atmospheric research instrumentation; Near-infrared detector development; Extended-
wavelength InGaAs

1. Introduction quirements favour modest cooling power de-

mands, which is particularly true for low earth
Advanced methods for observation of the Earth orbiting, sun-synchronised satellites for Earth ob-
from space requires the development of spec- servation research purposes, where the heat load
trometers and hyper-spectral imagers that are from the Sun and the Earth complicates the use of
sensitive in the near-infrared wavelength region, cryogenically cooled detectors operating at near
ranging from 1 to 2.4 lm. Satellite resource re- absolute zero temperatures. For the ultraviolet,
visible and short infrared wavelength region
*
Corresponding author. Tel.: +31-30-253-5703; fax: 31-30-
ranging from 0.2 to 1 lm, silicon-based detector
254-0860. arrays allow acceptable performance at 240±300 K
E-mail address: [email protected] (R.W.M. Hoogeveen). operation temperature requiring modest to no

1350-4495/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved.
PII: S 1 3 5 0 - 4 4 9 5 ( 0 0 ) 0 0 0 6 1 - X
2 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16

cooling power. However, detector materials ment of low dark current and noise is the custom
currently available for the 2.4 lm wavelength range designed multiplexer (MUX) chip for the read-
need to be cooled to temperatures below 100 K. out of the InGaAs detectors at near-zero bias
For example, the commonly employed HgCdTe voltage. These MUXes are produced at Honeywell
composite detector material for the mid-infrared on the basis of a design by Kosonocky [9]. The
may be extended to the 1±2.4 lm range, but re- general concept of InGaAs detector and MUX
quires cooling to below 100 K for acceptable dark- system for SCIAMACHY was proposed by Geary
current and noise performance. [1]. The development of the space-quali®ed DMs,
In this paper we present for the ®rst time the including electronics, thermal and mechanical
development of space quali®ed near-infrared array system as well as system engineering and perfor-
detectors based on the solid state material InGaAs mance optimisation was carried out in-house by
with acceptable performance in the 1±2.4 lm the Space Research Organization Netherlands
wavelength range at 150 K operation temperature. (SRON) [10±14]. The instrument performance for
These detectors serve on the SCIAMACHY the detection of the atmospheric trace gases CO,
(SCanning Imaging Absorption spectroMeter for N2 O and CH4 on the basis of measured detector
Atmospheric CartograpHY) atmospheric chemis- performance is derived from an end-to-end simu-
try instrument [1±3] on board the European Space lation [15,16]. It is shown that the 2.3±2.4 lm
Agency environmental satellite ENVISAT, with wavelength InGaAs detectors will be able to
operations scheduled for the 2002±2005 time measure the column amounts of CO to better than
frame. The instrument design is based on a con- 10% accuracy, CH4 to an accuracy better than 1%
ventional double spectrometer concept employing and N2 O with accuracy in between.
a prism for pre-dispersion of the incoming radia- In this paper a theoretical model is developed
tion into eight spectrometer channels and subse- for the analysis and optimisation of dark-current
quent dispersion by a re¯ection grating to a ®nal and noise performance of the InGaAs detectors.
spectral resolution of 0.2±2 nm in each individual After a brief introduction on the InGaAs detector
channel [4]. Each spectrometer channel is equipped and read-out electronics development in Section 2,
with a detector module (DM) containing individ- a theoretical model of dark current is developed in
ual detector arrays of 1024 pixels, detector read- Section 3 for application to the SCIAMACHY
out and ampli®cation electronics, the interface detectors. In Section 4 measurements are inter-
to the digital electronics data processing unit, preted on the basis of this model. In Section 5 a
the mechanical support and thermal control and theoretical model for detector system noise is de-
connection to the external cooling system. The veloped, which is interpreted in Section 6. The
modest cooling power demands of the InGaAs- theoretical model presented has a more gen-
based detector system allow the use of a passively eral applicability for the further development of
cooled deep space radiator on SCIAMACHY. InGaAs solid state detectors and space-based re-
Thus, the need for a Stirling engine cooler is mote sensing of the Earth.
avoided, o€ering advantages in no electrical power
consumption, no mechanical vibrations and high
reliability for long duration mission operation.
The eight channel SCIAMACHY spectrometer 2. Detector and multiplexer
comprises ®ve channels for the 240±1000 nm
wavelength range, based on space quali®ed silicon 2.1. The InGaAs detector material
detectors of type RL1024SR, manufactured by
EG&G Reticon, Sunnyvale, CA, USA. The near- Indium-gallium-arsenide (Inx Ga1ÿx As) o€ers
infrared channels 6±8 covering the 1000±2400 nm the possibility to tune the band gap and therefore
wavelength range are based on the novel InGaAs the cut-o€ wavelength, through variation in the
detector arrays developed by Epitaxx Inc., West composition x of the InGaAs material. A compli-
Trenton, NJ, USA [5±8]. Essential for the achieve- cation in this scheme is that also the lattice constant
 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16 3

of the atomic layer is changed. This creates a mis- layers, bounded by sharp interfaces in between the
match with the InP substrate on which the InGaAs layers. In this case, the lattice dislocations are bent
material is epitaxially grown. It will give rise to into the plane of growth, rather than extending
dislocations in the photo-sensitive material and this into the active layer [5±8].
is a source of dark or leakage current that degrades On top of the bu€er layers the active layer of
detector performance. The problem is illustrated InGaAs is grown, with the required indium con-
by Fig. 1. Here, it is shown that for x ˆ 0:53 the tent. This layer is then covered by a layer of InAsP
In0:53 Ga0:47 As composite is perfectly matched to of equal composition to the graded layer directly
the InP substrate. This is the standard ``lattice- under the active InGaAs layer. This decreases
matched'' InGaAs photo-sensitive material for the surface recombination and thus enhances quan-
1±1.6 lm wavelength range, widely employed tum eciency. Mechanical protection of the device
in ®ber-optics telecommunication applications. It is provided by a passivating SiN layer, which also
yields the best InGaAs performance in terms of serves as an anti-re¯ection coating. Individual
dark or leakage current and responsivity. pixels are created by Zn di€usion through open-
For responsivity above the cut-o€ wavelength ings etched in the SiN layer, creating p-doped
of lattice-matched InGaAs, i.e. above 1700 nm at pixels. The geometry of the pixels is 500  13 lm2 ,
room temperature (or 1630 nm at 200 K), a higher on a 25 lm pixel grid spacing. The sensitivity of a
indium content value x is required in order to pixel extends beyond the 13 lm Zn-di€usion p±n
lower the band gap and thus extend the photo- junction area due to the charge-carrier di€usion
sensitive wavelength. It can be seen from Fig. 1 length. For this reason, the pixel area is de®ned as
that no binary substrate exists with a matched A ˆ 500  25 lm2 ˆ 1:25  10ÿ4 cm2 .
lattice constant. The practical approach chosen by Graded layers, active layer and top layer are all
Epitaxx is to adjust the lattice constant in small sulphur n-doped. The doping density demands a
steps, from the value of the InP substrate to the delicate balance between reduction of the deple-
required value of the Inx Ga1ÿx As, by growing in- tion region (and thus the dark current) and the
termediate bu€er layers of InAsP. The InAsP onset of tunnelling through the p±n junction, also
material has a larger band gap and therefore leading to dark current. This will be discussed in
produces less thermally activated electron±hole more detail in Section 4.
pairs and consequently less dark current. It has Dark current and quantum eciency are the
been found that the best performance can be ob- most important factors in determining detector
tained by deposition of InAsP in discrete stepped performance. The impact of the dark current on

Fig. 1. Lattice parameter vs. energy band gap for III±V compounds and alloys.
4 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16

system performance is twofold. Firstly, during Two interleaved MUXes sample the signals of
exposure the full-well capacity is ®lled with dark all 1024 pixels simultaneously. Subsequently, the
signal, thereby reducing the dynamic range. This stored analog signal of the pixels is read out
e€ect can be reduced by operating the detector at a sequentially and digitised (16 bits) by the DM
bias voltage close to zero. Secondly, dark current electronics and passed on to the science data
increases the noise of the system. Since the detec- processing unit where the information of all eight
tor is operated at essentially zero bias voltage it DMs is processed and controlled. The method of
makes more sense to describe the impact of dark simultaneous read-out of all detector pixels as
current in terms of detector resistance, giving rise opposed to sequential read-out o€ers an impor-
Johnson noise, being inversely proportional to the tant advantage in space-based remote sensing. For
square root of the diode resistance. example, the Reticon detectors are read out se-
quentially, therefore pixel number 1 correspond-
ing to spectrometer wavelength number 1 is read
2.2. Multiplexer out at time t1 and thus observes a di€erent scene
compared with pixel number n, corresponding to
The requirements of near-zero bias-voltage spectrometer wavelength n read out at a later
operation, in addition to the 25 lm detector- time tn when the spacecraft or scan mirror has
pixel spacing, prompted the development of a moved on to a di€erent position. As a result, an
dedicated MUX read-out system. For this purpose ill-de®ned spectral signature is produced of the
a customised charge integrator and MUX has scene observed.
been designed and produced employing silicon
CMOS technology [1,8,9]. The features of this
MUX are: 2.3. Detector package

· two interleaved rows of 512 individual pixel ca- The detector packaging is shown in Fig. 2.
pacitive trans-impedance ampli®ers (CTIA), It consists of a 24-pins, dual-in-line (DIL) In-
· bias voltage between ‡2 and ÿ16 mV, switch- var package containing two MUXes, and the
able by in-¯ight telecommand, InGaAs sliver (which in case of channel 6 and
· low noise through application of correlated 6‡ is composed of two butted parts). Internally, a
double sampling (CDS), SiO2 thick-®lm substrate connects the inputs
· high linearity (P99.9%), and outputs of the two MUXes (one for odd and
· built-in address generators, one for even pixels). Two SiN substrates carry
· two-directional read-out (count up and count the MUXes and a SiO2 substrate carries the de-
down), tector sliver. The lid of the package contains a
· integration times ranging from 31.25 ms up to cylindrical lens which compresses the optical
10 min, beam of the system to the speci®c detector ge-
· short integration times (28 ls < tint < 32 ms) ometry in the cross-dispersion direction. This
for test purposes (``hot mode''), allows reduction of detector pixel area and ac-
· radiation hardness up to 10 krad (Si based). cordingly dark current is reduced. The lid material
is matched in thermal expansion with the lens
An important feature of any detector system is material: For Ch. 6 the Covar lid matches to a
linearity over the dynamic range. During the de- ZnSe lens; for Ch. 7 & 8 the Invar lids match to a
velopment of the MUX special attention has been Silicon lens. In addition there are 1024 wire bonds
devoted to this aspect. Normally, the feedback from the InGaAs sliver to the two MUXes, 48
capacitor has a non-linear charge±voltage behav- wire bonds from the MUXes to the thick-®lm
iour. In this case highly linear polysilicon-on-N‡ substrate and 24 bonds from the substrate to the
capacitors have been employed, resulting in a pins. The inside of the lid is coated black to reduce
system linearity of better than 99.9%. stray light.
 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16 5

Fig. 2. Cross-section of the SCIAMACHY channels 6±8 detector DIL package. Inside the Invar housing (A) a SiO2 thick ®lm
substrate (B), connects the MUX (D) to the 24-pins outside connectors. The 1024 InGaAs detectors (C) are wire-bonded to the MUX
(D). The cover (F) contains a lens (E) for f-matching the detector pixel size.

3. Theory of thermal-background radiation and (black-body) radiation of the instrument itself and
dark current by dark or leakage current of the detector diodes.
The thermal-background signal depends on the
In the process of converting the incoming temperature of the instrument and, to a lesser ex-
photons into electron±hole pairs and subsequently tent, on the detector temperature through the
into a measurable voltage, several noise processes quantum eciency of the detectors. The dark
can be identi®ed in the detector system: Johnson current depends on the temperature of the detector
noise of the diode, shot noise of the signal, dark and the bias voltage. In the following subsections
current and thermal-background radiation, and these dependencies will be quanti®ed.
noise of the ampli®cation electronics. Detector
noise depends on material parameters like doping
concentration, material composition, quantum ef- 3.1. Thermal background
®ciency, and on instrument settings such as inte-
gration time, bias voltage, detector temperature, The temperature of the instrument is cooled to
and instrument temperature. In this section a 255 K in order to reduce thermal-background ra-
theoretical model is developed describing these diation, ®rst identi®ed by Goede and coworkers
noise processes and their relation with material [10,11] as a signi®cant source of dark signal and
parameters and instrument settings. Some of the noise for SCIAMACHY channels 7 and 8. On the
material parameters are unknown and are ob- cylindrical lens of the channel 7 DM a low-pass
tained by analysis. In particular, the dependence of wavelength ®lter is placed to block the background
the detector dark current and noise on the bias radiation with wavelengths longer than the spec-
voltage and detector temperature yields insight in tral band. On the lens of the channels 6 and 8 de-
the underlying physical processes. tector only an anti-re¯ection coating is applied,
In the absence of illumination, the detectors still because here the quantum-eciency curve drops o€
produce a signal as a result of thermal-background sharply at the end of the spectral band providing
6 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16

an e€ective blocking function. The calculated ther- temperature T by Paul et al. [17] with the ®tting
mal-background current, Ith-bg , follows from inte- parameters measured by Shen and by Fang [18]:
gration of the Planck curve, Pk , for black radiators
multiplied by the detector pixel area, Apixel , the 
aGaAs T 2 aInAs T 2
®eld of view (FOV) to the 255K spectrometer Eg ˆ Eg GaAs ÿ ‡ Eg InAs ÿ
T ‡ bGaAs T ‡ bInAs
(about 1.2p sr) and the quantum eciency of the 
2
diodes, QE…k†: aGaAs T
ÿ Eg GaAs ‡ x ÿ 0:475x…1 ÿ x†
Z kmax  ÿ1 T ‡ bGaAs
hc dPk
Ith-bg ˆ Apixel q FOV QE…k† dk …1†
kmin k dk
Z kmax
ˆ 2pcApixel q FOV kÿ4 QE…k† where x is the ratio of In to As (for numerical
kmin values see Table 1), Eg GaAs ˆ 1:512 eV, the energy
ÿ1
 ‰ exp…hc=kkB Tinst † ÿ 1Š dk: band gap of GaAs at T ˆ 0 K, Eg InAs ˆ 0:42 eV,
the energy band gap of InAs at T ˆ 0 K,
Here q is the charge of an electron, kmin and kmax aGaAs ˆ 5:1  10ÿ4 eV/K, a ®tting parameter for
are the minimum and maximum wavelength over GaAs, aInAs ˆ 2:7  10ÿ4 eV/K, a ®tting parameter
which it is integrated (determined e.g. by the op- for InAs, bGaAs ˆ 190 K, a ®tting parameter for
tical coating on the DM lens), kB is BoltzmannÕs GaAs and bInAs ˆ 83 K, a ®tting parameter for
constant, c the speed of light, h PlanckÕs constant InAs. The cut-o€ wavelength kc can be found
and Tinst the temperature of the instrument (255 making use of the equality kc ˆ hc=Eg .
K). The quantum eciency, de®ned as the ratio of Based on Eq. (1), the wavelength shift of the
detected electrons relative to the number of inci- responsivity curve versus detector temperature is
dent photons, has been measured by Epitaxx at calculated (typically 2 nm/K), assuming that the
operating temperature and at room temperature temperature dependence of the band gap also
[12]. An example of the responsivity of channel 8 is holds for wavelengths shorter than kc . Fig. 3 shows
shown in Fig. 3. that the quantum eciency at room temperature
The temperature dependence of the quantum calculated from the measured quantum eciency
eciency can be modelled using the empirical re- at 150 K deviates less than 5% from the measured
lation between energy band gap Eg and detector quantum eciency at room temperature, which is

Fig. 3. Measured responsivity of channel 8 at 150 and 298 K respectively. The drawn line represents the theoretical responsivity curve
for 298 K, derived from the measured responsivity at 150 K. The maximum quantum eciency is approximately 55%.
 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16 7

Table 1
Calculated physical parameters of the SCIAMACHY near-infrared channels at operating temperatures
Channel
6 6‡ 7 8
Base material In0:53 Ga0:47 As In0:60 Ga0:40 As In0:81 Ga0:19 As In0:83 Ga0:17 As
Energy band gap (eV) 0.770 0.699 0.527 0.511
E€ective mass for holes (mh =m0 )a 0.47 0.43 0.42 0.41
E€ective mass for electrons (me =m0 )a 0.041 0.039 0.035 0.034
Permittivity (F/cm) 1:230  10ÿ12 1:240  10ÿ12 1:267  10ÿ12 1:270  10ÿ12
a
The e€ective mass is calculated by interpolating the values for InAs and GaAs [18]. m0 is the free electron mass.

within the error bar of the measured values. From lattice mismatch between layers. These trapping
the relation between quantum eciency and de- states generate electrons, which are thermally ex-
tector temperature the thermal-background signal cited to the conduction band, causing the interface
can be calculated as a function of detector tem- dark current. A theoretical expression for the in-
perature. The thermal-background signal is inde- terface current density is given by [20]:
pendent of the bias voltage. q p
Jif ˆ ni rn rp vth Nis pkB T …A cmÿ2 † …3†
2
3.2. Dark current p
where vth ˆ …3kB T =mh † is the mean free carrier
Dark current is a spurious output signal of the thermal velocity and q is the elementary charge.
diode array, caused by thermally generated elec- The variables rn and rp represent the electron and
tron±hole pairs. Four components can be identi- hole capture cross-section with values depending
®ed in the dark-current density Jd : interface on the detector material and doping type. The
current Jif , di€usion current Jdiff , generation± density of interface states Nis over the energy band
recombination current Jg±r , and tunnelling current gap is a measure for the quality of the diode array.
Jtun . The strong temperature dependence occurring Note that the interface current is bias-voltage in-
in dark current is primarily due to the thermally dependent and exponentially dependent on tem-
generated intrinsic carrier concentration ni which perature through the charge carrier concentration
depends exponentially on the inverted temperature ni .
T [19]:
 3=2   3.2.2. Di€usion current
2pkB T 3=4 ÿEg Outside the depletion region charge carriers are
ni ˆ 2 M c …m m
e h † exp …cmÿ3 †
h2 2kB T thermally generated and can subsequently di€use
…2† into the depletion-layer where the drift induced by
the electric ®eld will separate the electrons and
where me is the e€ective mass for electrons, mh is holes. These extra charge carriers cause a di€usion
the e€ective mass for holes, Mc ˆ 1 is the number dark current per unit area which, for one-sided
of equivalent minima in the conduction band, kB is abrupt junctions (np‡ diode) with a reverse bias
BoltzmannÕs constant and h is PlanckÕs constant. voltage Vbias , can be expressed as [19]:
Eg is the energy band gap calculated according to s   
Eq. (1). Numerical values are given in Table 1. qn2i Dp ÿqVbias
Jdiff ˆ exp ÿ 1 …A cmÿ2 †
N D sp kB T
3.2.1. Interface current …4†
At the interface between the active and the cap
layer of the detector, trapping states exist due to where Dp is the di€usion constant for holes, sp
contamination or due to dislocations caused by is the hole lifetime and ND is the donor
8 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16

concentration. Note that the di€usion component s  

has a strong temperature dependence due to the q3 Fm Vbias 2me ÿ8p q3
Jtun ˆ exp 2me Eg …A cmÿ2 †
quadratic carrier concentration ni dependence. It is h2 Eg 3qFm h
also dependent on the bias voltage. …7†

where Fm is the electric ®eld at the junction, which

3.2.3. Generation±recombination current depends directly on the doping concentration
In the depletion region (the region of the diode (about 1  1017 cmÿ3 ) of the active layers. Note
under in¯uence of the electric ®eld) the main cause that the tunnelling current is proportional with the
of the dark current is the thermal electron±hole bias current and that it is only weakly dependent
pair generation and recombination at doping, on temperature, through the temperature depen-
dislocation and contamination sites. It can be ex- dence of Eg .
pressed as [19]:
   
ni W ÿqVbias 4. Measurement and modelling of thermal back-
Jg±r ˆ q exp ÿ 1 …A cmÿ2 † …5†
seff kB T ground and dark current

An example of measured dark current of the

where W is the depletion width. The depletion
1024 pixels of the channel 8 (2265±2380 nm) de-
width W depends on the donor concentration of
tector array is shown in Fig. 4. Local variations in
the active layer according to W ˆ …2es Vbias =qND †0:5 ,
doping concentration, dislocations and contami-
where es is semiconductor permittivity (see Table
nation densities result in pixel-to-pixel variations
1). Based on this relationship the depletion width
in the dark signal, although most pixels display a
is estimated to be approximately 0.1 lm. The time
very similar behaviour. In modelling of thermal
constant seff is the dominating lifetime of the
background and dark current all pixels are treated
charge carriers due to trapping states, and can be
individually. For the purpose of our model only
expressed as [21]:
pixels are considered of dark current within certain
  de®ned bounds. Not considered are so-called hot
1 Et ÿ Ei pixels and dead pixels de®ned as follows:
seff ˆ exp …s† …6†
rt vth Nt kB T
1. Hot pixels: the dark current cannot be mea-
where rt is the dominating capture cross-section, sured due to saturation at all temperatures
Nt is the trapping state density and Et represents and bias voltage settings.
the dominant energy level (referenced to the in- 2. Dead pixels: the dark current is always near
trinsic Fermi energy level Ei ) of the trapping sites zero (non-connected pixels).
of the doping, dislocations or contamination. Note
that the generation±recombination current com- Only a few percent of the pixels belong to one of
ponent contains an exponential temperature de- these two categories.
pendence through carrier concentration ni , and The di€usion and generation±recombination
that it is also dependent on the bias voltage. dark current depends exponentially on the bias
voltage: exp …ÿqVbias =kB T † ÿ 1 (Eqs. (4) and (5)).
At low bias voltages (Vbias < kB T =q ˆ 13 mV at
3.2.4. Tunnelling current 150 K) this dependence may be replaced by a ®rst
When a diode consists of multiple layers of order approximation qVbias =kB T. A linear relation
di€erent material, the di€erence in band gap in- with bias voltage is also expected for the tunnelling
troduces an energy barrier. Charge can tunnel current (Eq. (7)), while the interface current (Eq.
through this barrier under certain circumstances (3)) contains no bias-voltage dependence. From
resulting in tunnelling dark current [19,7]: the linear relationship between dark current and
 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16 9

Fig. 4. Measured total current (thermal background and dark current) of the 1024 pixels of SCIAMACHY channel 8 at a detector
temperature of 150 K. The large variation in signal is ascribed to pixel-to-pixel variations in tunnelling current (see text).

bias voltage the diode resistance R0 and the zero- The diode resistance is composed of three
bias current J0 can be derived at various detector components: the generation±recombination cur-
temperatures. Temperatures have been varied rent, the tunnelling current and the di€usion cur-
from 140 to 180 K for channels 7 and 8 and from rent. All are linearly dependent on the bias
200 to 240 K for channel 6. From the zero-bias voltage. Very strong temperature dependence is
current as a function of temperature the bias- predicted for the di€usion current (Eq. (4)). This
voltage independent quantities can be derived: the was not reproduced in the measured dark current
thermal-background current Jth , the bias-voltage as function of temperature. Based on this obser-
o€set Voffset and the interface current Jif . The bias- vation and also backed by theoretical estimates, it
voltage o€set is an unknown deviation of the real is concluded that the di€usion current can be ne-
bias voltage over a diode and the pre-set bias glected at the temperature range considered in this
voltage due to the fact that the individual ampli- paper. The detectors tested show di€erent tem-
®ers of each diode shows a slight o€set. The o€set perature dependencies for the dark current for
is assumed to be independent of detector temper- each individual pixel. By ®tting the model pa-
ature. The interface current contains a strong rameters Fm and seff for each individual pixel, the
temperature dependence, and the thermal-back- tunnelling current and generation±recombination
ground signal only has a weak temperature de- current were derived. Table 2 shows the model
pendence through the quantum eciency of the parameters obtained, which are ®tted employing
detector (Eq. (1)). Using the measured tempera- the Levenberg±Marquardt method [22]. The re-
ture dependence of J0 , the thermal background, sults for the tunnelling current and the generation±
the interface current and the bias-voltage o€set recombination current are shown in Table 3.
correction can be determined independently. The The uncertainties given in Table 2 re¯ect the
results are shown in Table 3. variation from pixel to pixel rather than an error

Table 2
Median values for theoretical model parameters derived from measurement
Channel
6‡ at 200 K 7 at 150 K 8 at 150 K
7 7
Electric ®eld Fm (V/m) 2:38  0:08  10 1:45  0:05  10 1:39  0:04  107
Charge carrier lifetime seff (ls) 3:1  2 0:85  0:5 0:82  0:3
10 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16

Table 3
Calculated median values of detector dark-current constituents at ÿ2 mV bias voltage
Channel
6 at 200 K 6‡ at 200 K 7 at 150 K 8 at 150 K
Resistance R0 A (MX cm2 ) 8:3  104 10 6.9 5
Thermal-background signal (fA) 0:6  0:03 05 10  1 200  20
Interface current (fA) 0:2  0:1 11 2  0:5 01
Bias-voltage o€set (mV) ± 01 01 02
Gen.±rec. current (fA) ± 4 25 36
Tunnelling current (fA) ± 21 10 14
Dark current Jd (fA) 0.003 25 36 50

on the value. The values for the ®tted electric ®eld generation±recombination current is smaller than
of channels 7 and 8 correspond to the calculated the tunnelling current. Because of the weak tem-
values …Fm ˆ qND W =es †. For channel 6‡ the cor- perature dependence of this tunnelling current the
respondence is better assuming a donor concen- dark current cannot be further reduced by lower-
tration that is 4 times higher, resulting in a ing the temperature. This situation is already
depletion width of 0.05 lm (instead of the previ- reached at 200 K for channel 6‡ where the higher
ously assumed 0.1 lm). This adjusted depletion electric ®eld and charge-carrier lifetime result in a
width for channel 6‡ has been used in Eq. (5) to lower generation±recombination current and a
derive the charge-carrier lifetime seff given in Table higher tunnelling current as compared with chan-
2. Note that the charge-carrier lifetime is temper- nels 7 and 8.
ature dependent (see Eq. (6)). Results for median dark-current values for a
The uncertainties given in Table 3 for thermal single pixel of channels 7 and 8 are shown in the
background, interface current and bias-voltage left-hand side plots of Fig. 5. Plotted are the mea-
o€set re¯ect the variation over the pixels rather sured values of the dark current, as well as the
than an error on the value. The values without an modelled value and its two main components. The
error bar show a large pixel-to-pixel variation. The two right-hand side plots of Fig. 5 show the dark
pixel-to-pixel variation of the dark current is current for pixels with a much higher dark-current
caused by the generation±recombination current level, the so-called ``bad'' pixels. Note that the
as well as the tunnelling current, the latter causing di€erence between a good and a bad pixel is mainly
the largest variations. The interface current is of caused by the di€erence in tunnelling current. It is
negligible value, while the bias-voltage o€set shows our conjecture that large dislocations in the active
a variation of less than a few mV over all channels. layer give rise to a distorted di€usion of the p
Because the dark current of channel 6‡ is domi- doping, resulting in a locally higher electric ®eld
nant, the small thermal-background signal can not Fm and thus in a larger tunnelling current. The
be accurately derived. The temperature depen- pixel-to-pixel variation of the phenomenon is ex-
dence of the dark current of channel 6 at 200 K is plained by the random distribution of the dislo-
too small as to allow determination of the dark- cations. Based on this conjecture, a few general
current components and bias-voltage o€set accu- rules for the production of good extended-wave-
rately. For channel 8 the thermal background length InGaAs detector material may be formu-
dominates the dark signal, whilst generation± lated in addition to those given earlier by Linga
recombination current is the next contribut- et al. [7] and Joshi et al. [23]: Doping levels should
ing factor. For channel 7 the thermal-background be high such that the depletion width is reduced
current, the generation±recombination current and and thereby the generation±recombination cur-
the tunnelling current are more or less equal. At rent, but not so high as to cause the tunnelling
the lowest operating temperature of 140 K, the becoming dominant. For InGaAs the optimum
 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16 11

Fig. 5. Modelled and measured dark-current components as function of temperature for a typical good pixel (left-hand side plots) of
channel 8 (top) and channel 7 (bottom) and a typical poor pixel (right-hand side plots) of channel 8 (top) and channel 7 (bottom).
Below 140 K the tunnelling current takes over from the generation±recombination dark-current component, yielding relatively little
further reduction in dark current.

doping concentration turns out to be just under 5.1. Electronic noise

1017 cmÿ3 which is lower than the values reported
elsewhere [7,23]. The exact value for the doping The pixel read-out circuit of channels 6±8 MUX
concentration is determined by material parame- of the SCIAMACHY DM is based on a CTIA and
ters and the operational temperature. In order a sample and hold circuit to perform CDS. The
to reduce the number of ``typically'' bad pixels a electronic noise is modelled including the transfer
better control of wafer growth is required such as functions for the CTIA and CDS circuit. The most
to minimise the number of dislocations. Also, a important electronic-noise sources turn out to be
better control of p doping is required during wafer the preampli®er noise, ADC noise, power-supply
processing. noise and the thermal noise of the reset FET and
the sample and hold circuit. The preampli®er and
thermal noise are induced by thermal movements
of electrons inside the resistors of the electronic
5. Noise theoretical model circuit. Therefore this noise is temperature de-
pendent. The noise of the 16-bit ADC used in
Noise consists of the following contributions: SCIAMACHY consists of three components:
electronic noise, Johnson noise of the diode and quantisation noise, di€erential non-linearity and
shot noise of the signal, dark current and thermal- broadband thermal noise. Since the ADC is kept
background radiation. on a constant temperature outside the detector
12 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16

package, this noise component is independent of by surface e€ects, of which the numerical value
the detector temperature. depends on the fabrication process and therefore
Noise on the power supplies, caused by inter- has to be ascertained by experiment [24]. The 1=f
ference signals, also adds to the read-out noise. A noise is not included in the theoretical calculation.
worst-case estimate, based on crosstalk from the For frequencies relevant to SCIAMACHY the
digital lines in the MUX itself, has been made for noise spectrum Sd …f † on the feedback capacitor of
the resulting power-supply noise added to the this circuit, due to Johnson noise of the diode, is:
read-out noise. It turned out that the power-supply s
noise was higher than expected during the devel- 1 4kT 1 p
opment phase of the MUX. Better shielding the Sd …f † ˆ …electron= Hz† …8†
q Rd 2pf
on-chip current lines of the MUX and better on-
chip ®ltering could have reduced this noise-source,
Considering the CDS transfer function the result-
but this option would have required a second
ing rms Johnson noise equals:
production run, which was not further pursued
because of cost and schedule reasons. The present s
 2
Z 1
MUX ful®lls the instrument requirements. 2 sin …pf d†
rj ˆ Sd …f † 2 sin …pftint † df
The preampli®er and thermal noise of the 0 pf d
electronic circuit are integration-time dependent s
1 2kTtint
and scale with the square root of the temperature.  …electron rms† …9†
The ADC and power-supply noise are independent q Rd
of detector temperature and integration time. For
all components of the electronics noise there is no where d is the sample time length and tint is the
bias-voltage dependence. integration time …d  tint †.
For a typical integration time of 1.0 s and a On both the dark current and the thermal-
minimal integration time of 31.25 ms, noise ®gures background signal shot noise appears, which
have been calculated in Table 4. equals the square root of the total detected signal.
Table 5 gives the resulting noise for the shortest
integration time of 31.25 ms and Table 6 for the
5.2. Diode noise typical integration time of 1.0 s. For channel 6‡ an
upper limit is given because the thermal-back-
The main noise sources of the diode are John- ground signal is not exactly known. Because the
son noise and shot noise of the dark current, the dark current shows pixel-to-pixel variations, the
thermal background and the optical signal itself. median value of the dark-current noise has been
Random thermal movements of the charge carriers shown.
in the diode generate Johnson noise (also thermal Apart from the dominating noise sources dis-
noise). There is also a 1=f -noise component caused cussed above (Johnson diode noise, electronic

Table 4
Electronic noise (electron rms) derived from theoretical model at 31.25 ms and 1.0 s integration times
Channel
6/6‡ 6/6‡ 7/8 7/8
Integration time 31.25 ms 1.0 s 31.25 ms 1.0 s
Electronic thermal noise 341 564 118 321
Preampli®er noise 79 77 203 205
ADC noise 269 269 90 90
Power-supply noise 1300 1300 435 435

Electronic noise total 1373 1444 502 585

 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16 13

Table 5
Theoretical diode noise (electron rms) at bias voltage ÿ2 mV and integration time 31.25 ms
Channel
6 at 200 K 6‡ at 200 K 7 at 150 K 8 at 150 K
Johnson diode noise 3 290 301 355
Thermal-background noise 11 31 44 198
Dark-current noise 1 70 84 99
Total diode noise 11 300 316 418

Table 6
Theoretical diode noise (electron rms) at bias voltage ÿ2 mV and integration time 1.0 s
Channel
6 at 200 K 6‡ at 200 K 7 at 150 K 8 at 150 K
Johnson diode noise 18 1:6  103 1:7  103 2:0  103
Thermal-background noise 61 177 249 1:12  103
Dark-current noise 4 484 581 560
Total diode noise 64 1:7  103 1:8  103 2:4  103

noise and shot noise on thermal-background ra- tion, it is independent of the settings and quality of
diation and dark current), the following noise the detector. With sucient light, photon noise
sources turned out to be insigni®cant: will become dominant and one obtains the ideal
situation where observations are photon shot-
· current noise of the ampli®er (eliminated by noise limited.
CDS),
· reset interference signals (eliminated by CDS),
· electromagnetic coupling (reduced by electro- 6. Noise measurement and interpretation
magnetic shielding of the housing),
· mechanical vibration (suppressed by mechani- In Table 7 the measured and modelled total-
cal design). noise ®gures are summarised for operating tem-
peratures and a bias voltage of ÿ2 mV, both for
Note that under illumination condition the the minimum integration time of 31.25 ms and the
photon shot noise is introduced by the Poisson nominal integration time of 1.0 s.
distribution in time of the arrival of photons at the For most pixels the total noise for the shortest
detector. Photon noise contributes to the total integration time is determined by the electronics,
noise and depends on the detected signal only. although for some low resistance pixels diode
Being a fundamental property of photon absorp- Johnson noise dominates. For channel 6=6‡ the

Table 7
Comparison of theoretical and measured total noise (electron rms) of the SCIAMACHY near-infrared detectors (bias voltage ÿ2 mV)
Channel
6 6‡ 7 8
Modelled value for integration time 31.25 ms 1373 1404 593 653
Measured value for integration time 31.25 ms 1132 1488 636 668
Modelled value for integration time 1.0 ms 1450 2230 1900 2500
Measured value for integration time 1.0 ms 1175 1950 1500 2100
14 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16

power-supply noise of the MUXes is the domi- and bias-voltage combinations. For channel 8
nating noise contribution. At the nominal inte- operated at 150 K temperature and ÿ2 mV bias
gration time of 1.0 s, diode Johnson noise is voltage a comparison between theoretical and ex-
dominating, except for channel 6 where the power- perimental noise values is given in Fig. 6 (1024
supply noise is largest. Comparison with the pixels, 31.25 ms integrating time scale). The mea-
measured noise values shows good agreement. For surement error on the noise is 30%. The ®gure
the noise at 1.0 s, the measured values are slightly shows that the measured noise is in good agree-
below the modelled values, which is probably due ment with the model. A few pixels produce a sig-
to the rather pessimistic choice of the median dark ni®cantly higher noise than the theoretical value.
current, yielding too large values for the Johnson This may be caused by 1=f noise [24] or popcorn
noise of the diode. noise.
The noise model presented allows predictions of The noise is temperature dependent up to a
the noise values obtained at di€erent temperature point where it cannot be further reduced by low-

Fig. 6. Comparison between theoretical and experimental noise of all pixels of SCIAMACHY channel 8 (integration time 31.25 ms).

Fig. 7. Temperature dependence of the noise for SCIAMACHY channels 6‡, 7 and 8. Theoretical model compared with measure-
ments (integration time 31.25 ms).
 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16 15

ering the temperature. At lower temperatures and Johnson noise of the diode. Especially at longer
short integration times, the noise becomes limited integration times (1 s) the Johnson noise is the
by electronic noise and tunnelling current noise. dominant noise source. Based on the model out-
This e€ect is shown in Fig. 7 for a typical pixel of puts, suggestions have been made to further reduce
channels 6‡, 7 and 8. For longer integration times the dark current and noise.
Johnson noise of the diode becomes dominant
which can be reduced by cooling. The noise is only
little a€ected by the bias voltage.
Acknowledgements

SCIAMACHY is a joint German/Dutch

7. Conclusion
project funded by the National Space Agencies
DLR/NIVR with a Belgian contribution funded
High resistivity InGaAs detector diodes sensi-
by BUSOC. The system engineering of the
tive in the near-infrared 1.0±2.4 lm spectral
SCIAMACHY DMs was carried out at SRON
range have been developed for the space-based
under responsibility of P.J. de Groene, with ther-
SCIAMACHY atmospheric chemistry instrument
mo-mechanical design by P. Schreinemakers, de-
on the ESA ENVISAT. Dedicated read-out elec-
tail electronic design by J.F. Zwart and digital
tronics have been developed allowing low bias-
electronics by M. van der Linden.
voltage operation and low-noise performance.
Dark-current levels achieved are: R0 A of 5:0 
106 X cm2 for channel 8 (2265±2380 nm) and
6:9  106 X cm2 for channel 7 (1940±2040 nm). References
The overall performance of the detectors is such
that the SCIAMACHY instrument is predicted to [1] J.P. Burrows, K.V. Chance, P.J. Crutzen, H. van Dop, J.C.
Geary, T.J. Johnson, G.W. Harris, I.S.A. Isaksen, G.K.
detect total CO in the earth atmosphere with 10%
Moortgat, C. Muller, D. Perner, U. Platt, J.-P. Pomme-
accuracy and total CH4 with 1% accuracy. reau, H. Rodhe, E. Roeckner, W. Schneider, P. Simon, H.
A theoretical model has been developed for the Sundqvist, J. Vercheval, SCIAMACHY ± A European
analysis of the dark current and noise of the Proposal for Atmospheric Remote Sensing from the ESA
InGaAs detectors of SCIAMACHY. Based on this Polar Platform, Max-Planck Institut f ur Chemie, 55122
Mainz, Germany, 1988.
model, the bias voltage, detector temperature and
[2] H. Bovensmann, J.P. Burrows, M. Buchwitz, J. Frerick, S.
integration time settings have been optimised. An Noel, V.V. Rozanov, K.V. Chance, A.P.H. Goede,
operational detector temperature around 150 K SCIAMACHY: mission objectives and measurement
for channels 7 and 8 and around 200 K for channel modes, J. Atmos. Sci. 56 (2) (1999) 127±150.
6=6‡ yields lowest optimal dark current and [3] A.P.H. Goede, SCIAMACHY instrument concept,
ESAMS'99 ± European Symposium on Atmospheric
Johnson noise. Lowering the temperature further
Measurements from Space, ISSN 1022-6656, ESA-ESTEC
will decrease the dark current, but will also shift publication WPP-161, vol. 1, 1999, p. 147.
the quantum-eciency curve towards shorter [4] C. Smorenburg, H. Visser, The SCIAMACHY optical
wavelengths, thus reducing the signal and yielding system, Proc. Int. Symp. Opt. Syst. Des. SPIE 1780 (1992)
no net gain in performance. At lower temperatures 612±622.
[5] G.H. Olsen, A.M. Joshi, V.S. Ban, Current status of
dark current is shown to be dominated by tun-
InGaAs detector arrays for 1 to 3 lm, SPIE 1540 (1991)
nelling current which is characterised by a weak 596.
temperature dependence. As a result little is gained [6] K.R. Linga, G.H. Olsen, V.S. Ban, A.M. Joshi, W.F.
by further lowering of operation temperature. An Kosonocky, Dark current analysis and characterization of
operating bias voltage of 0 mV is selected. InGaAs/InAsP graded photodiodes with x > 0:53 for
response to longer wavelengths (>1.7 lm), J. Light Wave
The theoretical model allowed to identify the
Technol. 10 (8) (1992) 1050.
contributing noise and dark signal sources. The [7] K.R. Linga, V.S. Ban, A.M. Joshi, S. Mason, 1024 element
dominant noise sources are shown to be power- linear InGaAs/InAsP detector arrays for environmental
supply noise, thermal-background noise and sensing from 1 to 2.6 lm, SPIE 2021 (1993) 90.
16 R.W.M. Hoogeveen et al. / Infrared Physics & Technology 42 (2001) 1±16

[8] K.R. Linga, C.S. Wang, Y. Dzialowsky, M. Louri, S. ing of the Atmosphere and Clouds, Beijing, ISBN 0-8194-
Mason, High performance 1024 element InGaAs hybrid 2960-0; Proc. SPIE 3501 (1998) 429.
focal plane arrays from 1 to 2.6 lm, Opto-94 Conf., Paris, [16] H. Schrijver, Retrieval of carbon monoxide, methane and
1994. nitrous oxide from SCIAMACHY measurements,
[9] W.F. Kosonocky, B. Kleinhans, Chang Lau, Design and ESAMSÕ99 ± Eur. Symp. Atmos. Meas. from Space, ISSN
lay-out of SCIAMACHY InGaAs multiplexer, Dornier/ 1022-6656, March 1999; ESA-ESTEC publ WPP-161 1
Epitaxx contract 05 20008 AO 5220, 1993. (1999) 285.
[10] A.P.H. Goede, P. de Groene, R.W.M. Hoogeveen, J. de [17] S. Paul, J.B. Roy, P.K. Basu, Empirical expressions for the
Vries, R.J. van der A, C. Smorenburg, H. Visser, alloy composition and temperature dependence of the band
SCIAMACHY instrument development for POEM-1, gap and intrinsic carrier density in Gax In1ÿx As, J. Appl.
29th COSPAR Conference, Washington, 1992; Adv. Space Phys. 69 (2) (1991) 827±829.
Res. 14 (1) (1994) 17±20. [18] P. Bhattacharya (Ed.), Properties of Lattice-matched and
[11] A.P.H. Goede, R.W.M. Hoogeveen, R.J. van der A, J. de Strained Indium Gallium Arsenide, INSPEC, London,
Vries, Performance calculations and test of SCIAMACHY 1993.
detector modules, Optical Remote Sensing of the Atmo- [19] S.M. Sze, Physics of Semiconductor Devices, Wiley, New
sphere, Technical Digest, vol. 5, Optical Society of York, 1981.
America, Salt Lake City, Washington, DC, 1993, p. 436. [20] B.E. Burke, S.A. Gajar, Dynamic suppression of interface-
[12] A.P.H. Goede, R. Hoogeveen, S. Slijkhuis, R.J. van der A, state dark current in buried-channel CCDs, IEEE Trans.
J. de Vries, SCIAMACHY instrument performance and Electron. Dev. 38 (2) (1991) 285±290.
test, Atmospheric Radiation, in: K. Stamnes (Ed.), Proc. [21] G. Vanstraelen, K. de Backker, I. Debusschere, C. Claeys,
SPIE/EOS, vol. 2049, Troms o, 1993, p. 174±182. G. Declerck, Generation lifetime monitoring, Nucl. Instr.
[13] R.W.M. Hoogeveen, H. Spruijt, B. Broers, A. Selig, Near- Meth. A 288 (1990) 48±53.
infrared focal-plane arrays for SCIAMACHY, advanced [22] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flan-
and next-generation satellites, in: H. Fujisada, M. Sweeting nery, Numerical Recipes in C: the Art of Scienti®c Com-
(Ed.), Proc. Europto, Proc. SPIE/EOS, vol. 2583, Paris, puting, Cambridge University Press, Cambridge, 1992.
1995, p. 459±470. [23] A.M. Joshi, V.S. Ban, S. Mason, M.J. Lange, W.F.
[14] R.W.M. Hoogeveen, A.P.H. Goede, S. Slijkhuis, A. Selig, Kosonocky, 512 and 1024 element linear InGaAs detector
J.P. Burrows, SCIAMACHY development, Int. Symp. arrays for near-infrared environmental sensing, Proc. SPIE
Space Optics: Earth Observation and Astronomy, Gar- 1735 (1992) 287±295.
misch-Partenkirchen; Proc SPIE 2209 (1994) 78±85. [24] A.M. Joshi, G.H. Olsen, V.S. Ban, E. Mykietyn, M.J.
[15] H. Schrijver, A.P.H. Goede, M.R. Dobber, M. Buchwitz, Lange, D.R. Mohr, Reduction of 1=f noise in multiplexed
Retrieval of carbon monoxide, methane and nitrous oxide linear In0:53 Ga0:47 As detector arrays via epitaxial doping,
from SCIAMACHY measurements, Optical Remote Sens- IEEE Trans. Electron. Dev. 40 (1993) 303.