Radiation-Induced Defects in 8T-CMOS Global Shutter Image Sensor for Space Applications

an author's https://oatao.univ-toulouse.fr/22962

https://doi.org/10.1109/TNS.2018.2820385

Le Roch, Alexandre and Virmontois, Cédric and Goiffon, Vincent and Tauziède, Laurie and Belloir, Jean-Marc and Durnez, Clémentine and Magnan, Pierre Radiation-Induced Defects in 8T-CMOS Global Shutter Image Sensor for Space Applications. (2018) IEEE Transactions on Nuclear Science, 65 (8). 1645-1653. ISSN 0018-9499

Radiation-Induced Defects in 8T-CMOS Global

Shutter Image Sens or for Space Applications

Alexandre Le Roch�,

Student Member, IEEE,

Cédric Yirmontois,

Member, IEEE,

Vincent GoiffonO_,

_{Member, IEEE,}

_{Laurie Tauziède, Jean-Marc BeUoir,} Clémentine DurnezO_,

_{Member, IEEE,}

_{and Pierre Magnan,}

_{Member, IEEE}

Abstract-We propose to identify the displacement damage defects induced by proton and carbon irradiations in a com­ mercial off-the-shelf pinned photodiode (PPD) 8T-CMOS image sensors (CISs) dedicated to space application operating in global shutter mode. This paper aims to provide a better understanding of defects creation in a specific space image sensor. Therefore, it leads to comparable results to those we could find during the mission. The study focuses on bulk defects located in the PPD depleted region which represents the main dark current contribution in PPD CIS. Four sensors have been irradiated with carbon ions and protons at different energies and fluencies. Using both the dark current spectroscopy and the random telegraph signal (RTS) analysis, we investigate defects behavior for different

isochronal a,mealing temperatures. By combi1ùng these results, we make the connection between two complementary phenomena and bring out the prevalence of divacancies-based defects in terin of dark current contribution.

Index Terms-CMOS image sensor (CIS), dark current spec­ troscopy (DCS), pinned photodiode (PPD), random telegraph signal.

1. INTRODUCTION

I

NSPIRED by microelectronic technologies, CMOS image sensors (CISs) are, nowadays, widely used in many appli­ cations ranging from medical imaging and nuclear industry to space observation. Over the past decade, numerous studies have led to technological enhancements and allowed CIS to achieve performances sometimes better than charge-coupled devices [I]. CIS are predominant in commercial market for embedded cameras and are now considered as the perennial path for imaging in the future space missions as many state­ of-the-art commercial off-the-shelf imagers can be certified and adapted to space applications shortening their development period. They are becoming a must for a growing number of space applications such as Earth and space observations as well as stellar sensors, planetology, and scientific missions requiring landers and rovers. This is particularly true for pinned photodiode (PPD) CIS operating with intrapixel charge

A. Le Roch, V. Goiffon, C. Dumez, and P. Magnan are with the Institut Supérieur de l' Aéronautique et de l'Espace, Université de Toulouse, 31400 Toulouse, France (e-mail: alexandre.le-roch@isae.fr).

C. Virmontois, L. Tauziède, and J.-M. Beiloir are with the Centre National d'Etudes Spatiales, 31400 Toulouse, France.

Digital Object Identifier JO. 1109/TNS.2018.2820385

transfer [2]. This technology off ers extremely low levels of noise and dark current compared to traditional CMOS imagers, leading to a considerable increase in sensitivity. However, in space environment, CIS are exposed to different sources of radiation which damage sensor performances espe­ cially in terms of dark current and random telegraph signal (RTS) [3]-[5]. Recently, a commercial 8T-CMOS PPD imager operating in Global Shutter mode for a high-resolution camera has been selected by CNES for the future NASA rover of the Mars Science Laboratory in the Mars 2020 project. This sensor will be a part of the SuperCam instrument [6], [7], an improved version of the ChemCam instrument currently embedded on the Martian rover Curiosity. Developed by CNES, SuperCam will be launched as a part of the rover in 2020. Unlike Earth, Mars has no major global magnetic field preventing radiations and charged particles from space reaching the planet surface. Payloads are, therefore, exposed to radiative environment throughout the mission [8]. In order to truly predict radiation degradations in this specific space imager and ensure that its performances will meet the mis­ sion requirements despite the radiation-induced degradations, displacement damages are studied using the dark current spectroscopy (DCS) technique [9]. This paper aims to pre­ dict the sensor degradation during the mission as well as providing technical enhancements for the next generation of space imager. Moreover, the on-space gathered data will allow verifying the correlation with this paper and will provide useful information to build a tailored degradation mode!.

Space radiations cause two types of damage to electronics named ionizing damage and displacement damage which are, respectively, quantified by the total ionizing dose (TID) and the displacement damage dose (DDD) [10], [11]. The ionizing damage occurs in the silicon dioxide (Si02) and leads to a buildup of positive charges in silica as well as interface states at the Si/Si02 interfaces. Contrary to the ionizing damage, the displacement damage affects the crystal regularity within the volume. It happens when the energy transferred to the silicon atom is sufficient to remove it from its crystal site. This atom is named primary knock-on atom (PKA) and can remove

other atoms from their crystal site giving rise to a cascade of displacement. Interactions produce vacancy-interstitial (V-1) defects labeled Frenkel pairs. Depending on the imping­ ing particle and its energy, Frenkel pair's cascades can be more or Jess significant and create after diffusion of some

point defects or aggregate of defects named clusters. The final radiation damage comes from the thermally stable defects formed by the reaction between several primary Frenkel pairs or by the reaction with atomic impurities such as phosphorus, boron, and oxygen atoms as well as metallic impurities. Defects resulting from the DDD can be seen as stable bulk traps with energy level within the silicon band gap [12]. These defects severely degrade sensor performances. Among them, we observed a huge increase of the dark current as well as RTS fluctuations. Although the ionizing damage in imagers has been extensively studied for many years, fewer studies only focus on degradation induced by the DDD. However, it is currently one of the most important phenomenon involved in state-of-the-art imagers due to the

reduction of Si/SiO2 interfaces using the pinned photodiode

technology. In order to truly avoid those degradations with specific designed detectors or to improve the sensor degrada-tion predicdegrada-tion, radiadegrada-tion effects need to be further studied to provide a better understanding as far as displacement damage is concerned. Radiations also have a strong impact on the RTS fluctuations. This degradation gives birth to blinking pixels and began to be observed recently because of the rise of the sensitivity of PPD CIS. Blinking pixels result from metastable defects which have, therefore, several generation rates. As a result, this discrete dark current variation makes the offset correction impossible. Nevertheless, RTS sources need to be differentiated. Whereas, the source follower RTS (SF-RTS) is well known, the mechanisms of the dark current RTS (DC-RTS) sources located in the PPD remain unclear. SF-RTS is no longer observed in state-of-the-art CIS image sensors thanks to their high intersample time [13]. Hence, only DC-RTS are expected to be found in our sensors.

II. EXPERIMENT

The sensor under test is a commercial 8T-pixel CIS designed

in the 0.18-μm CMOS technology using PPD’s pixels with

a 5.5-μm pitch. Studied frames have 512 × 512 pixels. Imagers operate in pipelined global shutter mode and use 10-bits resolution. The dark current measurements are based on several integration times which are optimized for each temperature ranging from 10 °C to 40 °C with a 10 °C step. The perpixel dark current is computed from the slope of the output signal of the pixel versus integration time. We use a linear fit algorithm which successively deletes measurements in the saturation regime from the highest integration time while

the correlation factor is below R2 = 0.90. In addition to the

correlation factor verification, a minimum of three integration times is required. Whereas, the dark current of the cold pixels are usually computed with eight integration times and the dark current of the hot pixels are mostly computed with five integration times. This technique allows an accurate estimation of the dark current by covering more than the half of the dynamic range both for the unaffected pixels and the so-called hot pixels presenting a short dynamic range. RTS analysis is performed on 28 800 images with a 1-s sample during 8 h at 20 °C. Both dark current and RTS analysis are performed after six isochronal annealing temperatures ranging from 80 °C

TABLE I IRRADIATION PARAMETERS

to 280 °C with a 40 °C step during 30 min. A last 30-min annealing is performed at 300 °C.

The experimentations are based on three kinds of irradiation made at the Centro Nacional de Aceleradores in Seville, Spain. The expected error on the indicated fluence values is± 10% and all the irradiations have been performed at room temperature. Irradiation parameters are summarized in Table I. It aims to study the induced defects for different DDDs as well as the different contributions which lead to it. The deposited DDD depends on the impinging particle as well as its energy and can be caused by two kind of interactions. Coulombic interaction refers to the electrostatic repulsion between the particle and the silicon atom. Energy transfer from the particle to the PKA is typically a hundreds of electronvolts and gives birth to localized point defects. Contrary to the coulombic interaction, nuclear interaction is based on the collision between the particle and the host nucleus. Therefore, the energy transfer to the PKA is much more important (few MeV) in comparison to the coulombic interaction and leads to a cascade of defect along the particle track. The two high energy protons radiations, referred to as A and B, lead to both ionizing and displacement damages. The DDD is caused both by nuclear and coulombic interactions. Low energy or end-of-range (EOR) proton irradiation, referred to as C, aims to produce a maximum of coulombic interactions by stopping the particle into the depleted microvolume of the PPD. The huge increase of the non-ionizing energy loss (NIEL) at the end of the particle track lead to a huge deposited DDD in the photodiode depleted volume, and despite the DDD cannot be estimated without specific simulations, a maximum of point defects can be legitimately expected [14]. Finally, the last sam-ple referred to as D, corresponds to a carbon ion irradiation. Since the NIEL is unknown in the literature, the DDD cannot be estimated neither. However, as a heavy ion (six protons and six neutrons), due to its large cross section as well as its charge and mass, we expect to have a huge DDD resulting from both nuclear and coulombic interactions [15]. Characterizations have been performed one week after irradiation.

III. DARKCURRENTGENERATIONUSING THESHOCKLEY

READHALLKINETIC

The computational approaches of DCS are absolutely cru-cial to obtain accurate defect identification. Hence, in order to find out precisely which generating centers are involved in the dark current, it is essential to build an accurate model from Shockley–Read–Hall (SRH) kinetic [12] using appropriate

simplifications. Although the SRH expression has some limita-tions especially with the thermodynamic correspondence [16], it is still the only satisfying tool to suggest a generation dark current expression. In the case of this paper, we only con-sider defects localized in the PPD depleted region. Therefore, the generation rate can be expressed as

G= σn· σp· νth· ni· Nt

σn· e Et −Ei

kB ·T + σ_p· eEi −EtkB ·T

(1)

with Nt is the traps concentration,σn andσp are the carrier

cross sections of electrons and holes, respectively, ni is the

intrinsic carrier concentration, νth is the thermal velocity,

Et is the trap energy level, Ei is the mid-gap energy level,

kB is the Boltzmann’s constant, and T is the temperature.

We assume temperature independence for the carrier cross sections. By introducing the temperature dependence of the intrinsic carrier concentration and the carrier thermal velocity

in (1) niα T3/2; νth(n,p)α T1/2, we determine the dark current

temperature dependence as Idcα T2· e E−Eg(T )/2 kB ·T e2·EkB ·T − 1 (2) with E = |Ei − Et|. Finally, if we consider e(2·E)/(2·kB·T ) 1 for E = 0, the expression (2) becomes

Idcα T2· e−

Eg(T )/2−E

kB ·T . ₍₃₎

Nevertheless, we have to keep in mind that the observed dark current results both from the generation and the diffusion. The physical origin of the so-called diffusion dark current is still under discussion. However, one of its origins could be the thermally generated charges out of the PPD depleted region. These charges are then diffused into the pixel and some of them are collected by the PPD electric field leading to an increase of the dark current. Even though the diffusion dark current relies on both the thermally generated charges and thermally assisted diffusion, the last one can be considered as the limiting mechanism for the charge collection. Hence, the diffusion dark current can be expressed as

Idcα T3· e−

Eg(T )

kB ·T . ₍₄₎

Due to these two contributions, the nature of the observed dark current strongly depends on the temperature. For high temperatures, the observed dark current mainly comes from the diffusion mechanism while at low temperature, the generation mechanism is predominant. Most of the time, for practical reasons, the experimental temperature range cannot sensitively measure one dark current contribution by lowering the second. Both contributions are observed simultaneously leading to a hybrid dark current.

IV. RESULTS

A. Characterization Before Irradiation

As mentioned earlier, we can see in Fig. 1 that the dark current distribution strongly depends on the operating tem-perature. In addition to the first peak containing the defect-free depleted volumes which corresponds to the diffusion dark

Fig. 1. Dark current histograms of a healthy imager for different

temperatures.

current, a second peak is visible. This peak contains fewer pixels with a higher dark current and is identified as generation dark current. It probably comes from defects induced during the CMOS processing such as ion implantation or impurities such as metal or oxygen coming from diffusion processes. According to the expressions (3) and (4), only the diffusion dark current can be observed for high temperatures. Indeed, for every 10 °C, the diffusion dark current is quadrupled whereas the generation dark current is only doubled. Hence, starting from the dark current distribution at 10 °C in Fig. 1, we can expect having the generation overwhelmed by the diffusion at 60 °C. Nevertheless, for experimental reasons, the imager and its board cannot operate at such temperature and Fig. 1 only depicts the regime where generation is higher than diffusion. In order to find energy levels into the silicon bandgap, we introduce the Arrhenius law where the activation

energy of the dark current is labeled Ea(T ) and A is an

pre-exponential constant factor

Idc= A · e−

Ea(T)

kB ·T . ₍₅₎

Hence, from (3), we can show that

Ea(T ) = 2 · kB· T + Eg(T ) 2 − T 2 · ∂ Eg(T ) ∂T − E. (6)

Moreover, we know that the temperature dependence of the silicon bandgap can be expressed as [17]

Eg(T ) = 1.166 −

0.000473 · T2

T + 636 . (7)

Therefore, we can express the activation energy at room temperature as a function of the energy level into the silicon bandgap as [15]

Ea(T ) = 0.65 + E. (8)

The activation energy expressed in (8) states that the

low-est measurable activation energy is 0.65 eV when E =

|Ei− Et| = 0 eV. It means that the dark current generation of

a defect presenting this activation energy is maximized. This defect is usually named the mid-gap defect. Fig. 2 depicts the 20 °C–30 °C temperature range activation energy as a function of the dark current at 30 °C as well as the previously shown

10' 10' 101 Phcl Count 1 10 100 1000 __________________________________ 0.9cV __ _ 0.8cV -�...:__ ._{. , . ___} 0_:7cV _{__ _} io"_o'---..__�_,00---₂�₀₀----'LJL..₃₀₀'--'-_u-₄₀₀�---_s.J. ooll.ll..Ll...lLJ D�ark Currcnt (c'.s"1) 1.2 08 � 0.6 .� _� OA 0.2

10

6

fi

=�����;;==

r::,

-;-;

�=::::'.::=::::::=::::'.:=::::'.:=::::-1

-Before l.1'1111diatioo (Ref') J.:<f'

-Higb cnergy Proton ltndîtirion (A)

1 cf --ffjgb c:nergy Proton tmdiation (B) Low encra>- Proton hraidia.tioo (C) -Caibon Ion Irradiation (D)

to'

10'

2000 3000 4000 6000 7000 8000 9000 J 0000 Dark Cun-ent (e·.s')

Fig. 2. Dark current histogram and its equivalent activation energy at 30 °C. Fig. 4. Dark current histograms after irradiations at 20 °C.

1.6 1_Difli-_{. -us-ion_P}_cak-_[I_O_'_

C_- -20'C-,-]---- --�---:::OOO_I,:--;:---, Diffusion Peak [20'C -30'C] 1.4

.g

0.8 !:

·=

<o.6 0.4

=

:M, .. ,�

,

._,.._w,,•,..,.__ ______ Generation Pcaks l [30'C - 40'C)

�,..::

· ,,

:�

·

1

·

::·.

. .. -�

-t

--�·

.

. ...

.

.

='='""'".7.t· •... ' _{.. . . .} î

Gen��:ion Peiok ;20'C -_30'C] ' . · Gcncration Peak [ 10°C -20°C) 100 10 0_·2_{0_'.---:5�00---I...J.000 _____ ���1500} Dark Currcot(c·.,1₎

Fig. 3. Activation energy evolution with the operating temperature. dark current distribution at 30 °C in Fig. 1. If we focus on the generation coming from the CMOS processing, we can note that its activation energy is about 0.7 eV and is close to the above-mentioned mid-gap defect. Those pixels have also a dark current higher than the healthy ones in the diffusion peak. This confirms that the closer to the mid-gap the energy level is, the higher the generation dark current will be. This is due to the equal probability to promote an electron from the trap energy level to the conduction band and a hole to the valence band. Moreover, Fig. 2 reveals another generation peak also disclosed on the dark current distribution. This generation has a higher activation energy (0.8 eV) and has, therefore, a lower dark current. Those pixels present another kind of defect leading to generation mechanism. Considering Fig. 3, the activation energies can also be computed using other temperature ranges. As this computation only takes into account two temperatures, Ea is not well established but it illustrates the rise of the activation energy with the operating temperature. This evolution points out the different dark current contributions along the temperature ranges.

B. lrradiated lmagers

Our imagers have been exposed to radiations whose para­ meters are summarized in Table 1. The expected mean dark current increase, labeled ll/uoF, uses the universal dam­

age factor (UDF) Kdark = 0.19 µm-3 • s-1 . TeV-1 . g [3].

For the high energy proton irradiations referred to as A and B, the observed mean dark current increase are ll/A = 424 e- · s-1 and !lis

=

515 e- • s-1 and are in fairly good agreement with the UDF estimation in Table 1. The mean dark current deviation from the UDF model can be explained by the variability of the empirical model due to the use of different proton energies. The considered depleted volume of 0.5 µm3 _{has been estimated from [18] which}

use a similar CIS. As the NIEL cannot be estimated for the other irradiations, the UDF factor cannot be certified. However, it fits the more conventional irradiations such as high energy proton and confirms the use of this model for this specific CIS. The dark current histograms available in Fig. 4 depict the impact of the irradiations in terms of dark current. A considerable quantity of defects has been created leading to multiple generation centers forrning a continuous dark current tail. As observed in several papers [19], [20], the dark current tail highly depends on radiation type. Carbon ion irradiation (blue) and low energy proton (green) show a larger dark current tail than high proton energy irradiation (orange and yellow). These results are in agreement with the expected trend regarding the deposited DDD estimated in Table I. The DDD is directly related to the size of the dark current tail and we clearly see that carbon ion irradiation has the higher DDD. We can also notice the similarity of the dark current tail slopes for all irradiations but one. Indeed, the dark current tail of the low energy proton is significantly different because of the prevalence of the coulombic interactions compared to the nuclear scattering [15]. Moreover, we clearly observe the TID contribution resulting in a translation of the histogram toward positive dark currents. This contribution is explained by an additive dark current coming from oxide interfaces and similar in all the pixels. Once again, the expected trend is observed. The TID translation is maximal for low energy proton and the iso-TID visible in Table I for high energy proton is proven. According to the results, carbon ion irradiation DDD prevails on the TID.

C.Annealing Ejf ects

/) High Energy Proton: Annealing has a strong influ­ ence on the dark current distribution as previously observed

10' 103 -Aft.,-lrrad -aO"C -120"C -160"'C -200°C -240°C -280°C 50 25 _{+o;m.,;o,.p,,,. /} � 20 �::::::::: ::: � •_�•15 _{_-..Ql.:11CltHilJII f>(.�l,;}}

�

·�

��---.

i :>

+--�-

---

.

0 � 0 so 100 150 200 250 Annealîng Tcmpenure (� 100 150 200 Dari< Currcnt (c.s1) 250 300

Fig. 5. Annealing impacts on dark current histogram after high energy proton irradiation at 20 °C.

in [4], [15]. First of ail, Fig. 5 points out the influence of the annealing on the TID-induced dark current with the translation of the histograrn toward lower dark currents. TID annealing occurs ail along the annealing temperature range. Fig. 5 also points out that the dark current tail after high energy proton irradiation decreases with annealing temperature and starts showing DCS peaks. Moreover, we clearly observe the diffusion peak increase with annealing. These changes result from the annealing of defects in a pixel which dominant dark current source switches from generation into the dark current tail to diffusion dark current. If we now consider the generation

peaks, we can name !!. I, the dark current increase from the

diffusion contribution. The dark current increase related to the generation corresponds to the measured dark current cutoff

from the diffusion contribution. Regarding !!. I, we notice that

al! peaks are multiples of the first one. Those different peaks correspond to the same defect present once or several times in the same pixel. The generation dark current adds up and takes a multiplying value of the single defect generation dark current. Finally, it is important to note that after 280 °C anneal­ ing, the previous generation peaks collapse. Thermal energy anneals the defects that were responsible for the generation peaks. To summarize, we notice three distinct regimes also outlined in [15], [21].

1) For low annealing temperatures (80 °C-120 °C), it only affects the defects induced by TID located at the Si/SiO2 interfaces.

2) For higher annealing temperatures (160 °C-240 °C), DDD-induced clustered defects rearrange, reducing the dark current tail and revealing preexisting generation peaks linked to point defects.

3) The final scheme is an annealing phase that drops the number of defects in the pixels induced either by the irradiation or by the lower temperature annealing.

This last annealing temperature occurs between 240 °C and 280 °C. Despite the slight differences in terms of activation energies of each generation peak visible in Fig. 6, the most appropriate peak is the one where generation contribution is the highest. Therefore, the one presenting the maximum of defects is considered to estimate the activation energy. The computed defect activation energy after a 240 °C annealing

1.2 1.1 �1.05 � 1

J

0.9S

.g

_{� 0.9} "fi < O.&S 0.8 50 1.2 1.1 0.8 100 150 200

Oark Current (e".s"1)

1000

100

600 800 1000

10

250 300

Fig. 6. Activation energy plotted at 20 °C after high energy proton irradiation and annealing at 240 °C. 1.2 _{r-:;r.:-:-�--�-,1co2.�===�==:::?::;=:::::;-7} 1.15 1.1 >'1.05

ti

t ,li 0.95

,g

_{� 0.9} "il < 0.85 0.8 0.75 1.1 _{Gencnition Peak} 200 400 600 800 1 000

Oark Cuaent (ti.s"1)

0.7 l---"o.---':...C-'-'=-..:...� --"-''-'-'-=-''-=-'--�'-""'"---""---L-=-''->---"'--' 0 50 100 150 200 250 300 Dark Currcnt (c".s"1) 1000 100 10

Fig. 7. Activation energy plotted at 20 °Cafter high energy proton irradiation and annealing at 280 °C.

visible in Fig. 6 is about 0.83 eV. After a 280 °C annealing, Fig. 7 points out the reduction of the defect concentration due to the annealing transition already observed in Fig. 5. From these observations and based on defect features available in the literature, those defects can be identified as divacancies labeled V2 with !!.Ev₂

=

0.17 eV [22], [23]. From (8), we find an expected activation energy of Eav₂

=

0.82 eV which is pretty close to our measurement at Ea

=

0.83 eV. The existence of such defects have also been pointed out in [15]. Due to the high contribution of the diffusion as well as the difference in the considered temperature range, a difference in the computed energy level exists between this paper and [15]. However, this result suggests that divacancies are also created by DDD in this specific 8T PPD CIS technology and consolidates the universality of the defect creation mechanism in CIS. Clusters induced by irradiation give birth to a continuous dark current tail in which each cluster has its own generation rate. Pre­ viously agglomerated into aggregates, clusters are rearranged and annealed to form stable divacancies V2. They partially disappear when reaching their annealing temperature which is about 250 °C [22], [23]. Both on Figs. 6 and 7, we identify another generation peak with a dark current higher than the earlier identified

V2

. Its dark current is about 500 e- • s-1 and its energy level is close to the rnid-gap. The origin of such high dark current generation source will be discussed in Section IV-C3.

1.1

111

:::::�

=�

":

.

7

•

.

--:-�-:-'"

.

i":i1.1[j.

�:;;:;::;:i:::::;=======,7

Di..flusion Peak _{. . . l.} �0.95 .:Ï 0.9

:i

0 85 0.8 0.75 0.9 1500 50 100 150 200 250 300 350 400 450 500

Dark Current (e-.s·')

100

10

Fig. 8. Activation energy plotted at 20 °C after carbon ions irradiation and annealing at 240 °C.

0.8 0.75

50 100 150 200 250 300 350 400 450 500

Dark CUrrent (e-.s1₎

100

10

Fig. 9. Activation energy plotted at 20 °C after carbon ions irradiation and annealing at 280 °C.

2) Carbon Irradiation: The carbon irradiation followed by annealing leads to the formation of the same defects as the ones induced by high energy proton irradiation. Indeed, both the dark current and the activation energy are similar

(/:,,/ =

30 e- • s-1; Ea

=

0.83 eV). Moreover, we observe the

same annealing phase between 240 °C and 280 °C annealing depicted in Figs. 8 and 9. Finally, with regard to both the defect behavior toward the annealing temperature and computed activation energy, this experiment suggests that divacancies are also involved after carbon irradiation. Moreover, as previously outlined after high energy proton, annealing at 280 °C reveals another peak with activation energy close to 0.65 eV as shown in Fig. 9. As the generation peak amplitude rises with the annealing, it suggests either that defects are formed during the annealing or that the dark current tail decrease reveals the pre-existing generation source.

3)Low Energy Proton: Contrary to the high energy proton

irradiation, the low energy proton irradiation only reveals one peak after annealing whose dark current is about !:,, l

=

500 e- • s-1 and visible in Fig. 10. Once again, it seems to be the same generation source that bas been found both after high energy proton irradiation and carbon ion irradiation. This generation source has been observed after irradiation and annealing at 260 °C in [15] after EOR particles. Both the dark

1o'r-;::::===================:::;-7

1-so•c -120'C -140'C -160-C-200'C-240'C-280'C!

102_.

10'

0 200 400 600 800 1000 1200 1400 1600 1800 2000 Dark CUrrent (e-.s1₎

Fig. 10. Annealing impacts on dark current histogram after low energy

proton irradiation at 20 °C.

current at 20 °C and the energy level are similar with this paper and the identity of the corresponding generation center bas not been clearly identified yet. Such generation sources could be identified as remaining clusters. Indeed, they are often observed as a constitutive part of the dark current tail after irradiation and could be disclosed after a sufficient annealing treatment. In this case, the high disparity in term of dark current could be seen as a consequence of the cluster structure complexity that could involve several energy levels as well as a disparity between the electron and holes cross sections leading to the complex generation mechanism, and thus to a dark current continuum. However, we can also bring out the existence of another point defect or chemical complex formed by a divacancy and an oxygen atom labeled V2O [22], [23]. Following Figs. 7, 9, and 10; V2O complex could explain these generation peaks with a high dark current because its energy level is close to the mid-gap. Regarding the high disparity in terms of dark current for this generation source, we can assume that it cornes from the existing dark current disparity among the diffusion and the generation at lower dark current. lndeed, the existing disparity refers to the dark current disparity with­ out any additive generation source. If an additive generation source cornes out in several pixels, the whole existing disparity will be shifted to higher dark current. Hence, the dark current disparity for those pixels is unchanged. As a result, the highest the disparity is at low dark current, the higher the disparity will be for the high generation dark current. Defects induced by irradiation could react during the annealing. In addition to give birth to divacancies, annealing reactions could also form the chemical complex divacancy-oxygen which results from the chemical reaction between a divacancy and an oxygen impurity. These complexes could be formed during the annealing when divacancies mobility is enhanced by the temperature which increases its probability to be trapped by an oxygen atom. Moreover, the annealing temperature for V2O is about 330 °C [22], [23]. As this chemical complex is still observed after an annealing at 250 °C, we can suggest that the chemical complex V2O is involved.

4) Last Annealing at 300°_{C and Comparison Between}

Radiation-lnduced Defects: An additional 30-min annealing at 300 °C has been performed for ail irradiated sensors. Results

1.1 10 100 100 200 300 400 soo 600 700 800 900 1000 100 200 300 400 soo 600 700 800 900 1000 .. . ... '� ·'::. 100 200 300 400 500 600 700 800 900 1000

Dark Current (e".s"1)

Fig. 11. Activation energy plotted at 20 °Cafter (top-to-bottom): high energy proton irradiation, carbon ion irradiation, and Iow energy proton irradiation after annealing at 300 °C.

are presented in Fig. 11. For the two earlier irradiations (high energy proton and carbon ion irradiations), divacancy peaks keep decreasing. On the low energy proton irradiation,

it finally reveals the first divacancy peak at /dc

=

65 e- · s-1.

These divacancies were previously invisible because of the prevalence of cluster in term of dark current. Due to the large DDD of the low energy proton irradiation, a larger amount of thermal energy is needed to anneal cluster and reveal point defects as divacancies. The reason why we still have some divacancies after a 300 °C annealing, which is above the annealing temperature, is probably because of the huge concentration of point defect. Formed divacancies already started to be annealed at lower temperature. However, previous annealing treatment was not long enough to totally anneal the defect population. Another explanation could be the formation of such defect during the annealing treatment. Moreover, the last annealing treatment leads to a relevant increase of the highest dark current generation source and thus for all irradiations. This last annealing is still below the V2O annealing temperature (330 °C) and does not exempt the hypothetical existence of V2O. The presence of this dark current source in ail the studied irradiations means that no matter is the irradiation, both the mechanisms and the defect creations are identical and only the amount of defects differs from the irradiation parameters. Indeed, the carbon ion irradiation and the low energy protons irradiation expected to have the larger DDD have also the higher high generation source concentration in comparison to the high energy proton. Hence, the formation of this generation source, corning from either clusters or point defects such as V2O, is linked to the deposited DDD. It is also interesting to notice a new peak

close to 0.75 eV at fctc

=

150 e- • s-1 after the last annealing

treatment in Fig. 11. Regarding its energy level, it could be

the chernical complex made from the reaction between a V2O complex and an oxygen atom giving rise to a V 202 chernical complex [22], [23]. This structure has an energy level

10' r-.::�----,---...---;:::===:i -t-2Lcvels ..,._3Lcvcls ... 41..evels -Â-5Levcls -61..evels 10' r-:---,---'-"' ... =lll..l..5..._..__...--__ --;:::===:-i -+--2 Levels Dark Cum:nt (c-.s') -+-31..evels ... 4Levels -Â-5Lcvels -6Lcvcls 15000

Fig. 12. Dark cUJTent histograms of RTS pixels after irradiation (top) and after annealing at 120 °C (bottom) .

HIGHER (Ea V2O2

=

0.81 eV) than the above-mentioned

V2O complex, and thus, could explain the fact that the generation rate is lower. As previously highlighted, the carbon ion and the low energy proton irradiations have the highest concentration of hypothetical V 20 complex. Tuen, we can log­ ically assume that they are also the ones which have the larger amount of V2O2 complex as its creation is directly related to the V2O concentration. Hence, we can also assume that this defect would be involved in the high energy proton irradiation but in a smaller amount. Regarding Fig. 11, we confirm the expected trend having more hypothetical V 202 in low energy proton and carbon ion irradiation compared to high energy proton irradiation.

D. Random Telegraph Signal Analysis

RTS analysis used a rising edge algorithm explained in [24]. First of ail, we notice an increase of RTS pixel with the DDD. As ail the irradiated sensors present the same RTS behavior toward annealing, for the sake of clarity, this section will only focus on the high energy proton irradiated sensor referred to as B in Table I at 20 °C. First, it is interesting to see in Fig. 12 that the higher the number of RTS level, the higher the average dark current. Moreover, the distribution remains unchanged no matter the annealing. Regarding the dark current associated with RTS pixels in the same figure, it is clear that RTS pixels are contained in the continuous dark current tail and mostly at very high dark current. However, there is no correspondence between the above-mentioned generation peaks and RTS pixels. In accordance with other works [25], we presume that RTS pixel relies on cluster of defect with a high global dark current. These complex structures of defect present several metastable configurations and orientations into the silicon crystal leading to different energy states within the bandgap, various generation rates, and thus to a discreet dark current. In Fig. 13, we observe that the annealing have decreased the number of RTS pixels. Hence, all RTS clusters are annealed simultaneously. Regarding the ratio loss, the higher the number of RTS level, the more sensitive is the annealing. Indeed, between annealing at 120 °C and 160 °C, the amount of four-level RTS pixels drop by 98% whereas the

104 io> -Aftff hrildi•tion --80"C --'k-12o·c --Â-16-0•c X 280°_C �102 .!S 101 10• X

Ê

2 4

Number ofRTS Level

Fig. 13. RTS levels histogram evolution with annealing.

Timt($)

Fig. 14. Dark current trace evolution with annealing.

two-level RTS pixels drop by 75%. Cluster sizes are reduced due to annealing leading to smaller RTS clusters. These new RTS clusters lower both their total dark current as well as their number of RTS levels as observed in [26] with the same temperatures. Severa! RTS behaviors are identified as follows. 1) One cluster in a pixel leading to RTS characterized by

its global dark current and its number of RTS level. 2) Severa! similar clusters in the pixel causing a multiple

RTS contribution (typically a lot of RTS levels), which can be reduced with the annealing of one of the RTS sources.

3) A single cluster presenting RTS and one continuous dark current source which can be annealed implying a reduction of the global dark current of the pixel without any change in terms of number of RTS level and the maximum amplitude transition.

The dark current evolution shown in Fig. 14 clearly depicts the Joss of the highest dark current level after annealing at 160 °C leading to a reduction of the average dark current from 2500 to 1500 e- • s-1• The pixel switches from a three-level RTS to a two-level RTS without any change in terms of RTS amplitude between the second and the third levels implying that the remaining two configurations are unchanged. The first annealing at 120 °C is not sufficient to anneal a specific RTS level but regarding the RTS time constant, the dark current rate seems more stable. Fig. 15 depicts the maximum

1o'_r---.--�---�--�----;!:::=======::1 - - k/1200 . .,,,1>(-�/1200) ] 10' ,,_ l(r 101 DOD 1 À =1-200 ---k/110, exp( -x/110) ... After Irndlation ---SOOC Annealin,a -Jr-120°_{c AnnoaaU.ng} � 160°_{C Anuealing} �280°_CAnncaling

Maximwn Tran.,;ition Amplitude (e"'

.s·1)

Fig. 15. Maximum amplitude transition histogram evolution with annealings.

amplitude RTS trans1t1on, a key parameter to study RTS fluctuations [24] and defined as the amplitude of the highest transition. We note that the maximum amplitude transition also decreases with annealing. As mentioned earlier, this section focus on the high energy proton involving both TID and DDD. After irradiation, the typical slopes related to TID and DDD previously identified in [27] are in fairly good agreement with the computed ones in the same work. This further proves the universality of this distribution. Along the annealing, the DDD slope remains unchanged if we consider the uncertainty related to the poor statistics. Hence, it proves this trend is still correct regardless the annealing. The slope is only related to the remaining DDD that has not been annealed yet. Moreover, we clearly see the annealing of the TID contribution between 80 °C and 120 °C as well as the continuous decrease of the DDD. This phenomenon is in accordance with the statement conceming the dark current histogram presenting the same trend.

V. CONCLUSION

Throughout several analysis, this paper gives one more evidence that sirnilar defects are involved in ail OS regardless the particle and its energy used during the irradiation. Among them, we found the divacancies as radiation-induced point defects and complex structures like clusters. We also pointed out an unusual generation source after annealing and proposed the hypothetical role of V20 chemical complexes rather than remaining clusters. It raises a question regarding the hypo­ thetical impact of oxygen impurities on the dark current rise. 0xygen concentration could have a leading role regarding CIS performances and particularly for those dedicated to space missions as they are constantly subjected to radiations. However, oxygen impurity concentration in silicon is about

1 x 1018 _{• cm-3. In comparison with the depleted volume of}

0.5 µm3_{, it represents approximately 5 x 10}5 _{oxygen atoms.}

Knowing that a single oxygen atom could form a genera­ tion center, improving the purity does not represent a major technological goal to reach. Expected

RTS

behavior has been observed, thus, confirming the universality of radiation effects on CIS. Performed studies have brought more information regarding the future radiation-induced degradations and further insights have been suggested for the next generation of CIS

such as RTS and dark current mitigation algorithm as well as on-board annealing recovery.

ACKNOWLEDGMENT

The authors would like to thank E. Locatelli from CNES for having made the annealings possible.

REFERENCES

[1] J. Janesick, J. T. Andrews, and T. Elliott, “Fundamental performance differences between CMOS and CCD imagers: Part 1,” Proc. SPIE, vol. 6276, p. 62760M, Jun. 2006.

[2] E. R. Fossum and D. B. Hondongwa, “A review of the pinned photodiode for CCD and CMOS image sensors,” IEEE J. Electron Devices Soc., vol. 2, no. 3, pp. 33–43, May 2014.

[3] J. R. Srour and D. H. Lo, “Universal damage factor for radiation-induced dark current in silicon devices,” IEEE Trans. Nucl. Sci., vol. 47, no. 6, pp. 2451–2459, Dec. 2000.

[4] G. R. Hopkinson, “Radiation effects in a CMOS active pixel sen-sor,” IEEE Trans. Nucl. Sci., vol. 47, no. 6, pp. 2480–2484, Dec. 2000.

[5] G. R. Hopkinson, V. Goiffon, and A. Mohammadzadeh, “Random telegraph signals in proton irradiated CCDs and APS,” IEEE Trans.

Nucl. Sci., vol. 55, no. 4, pp. 2197–2204, Aug. 2008.

[6] O. Gasnault et al., “The superCam instrument for the Mars2020 rover,” in Proc. Eur. Planetary Sci. Congr., 2015, p. 185.

[7] O. Gasnault et al., “SuperCam remote micro-imager on Mars 2020,” in

Proc. 46th Lunar Planetary Sci. Conf., 2015, p. 2990.

[8] G. De Angelis, M. S. Clowdsley, R. C. Singleterry, and J. W. Wilson, “A new Mars radiation environment model with visualization,” Adv.

Space Res., vol. 34, no. 6, pp. 1328–1332, 2004.

[9] R. D. McGrath, J. Doty, G. Lupino, G. Ricker, and J. Vallerga, “Counting of deep-level traps using a charge-coupled device,” IEEE Trans. Electron

Devices, vol. ED-34, no. 12, pp. 2555–2557, Dec. 1987.

[10] V. Goiffon et al., “Identification of radiation induced dark current sources in pinned photodiode CMOS image sensors,” IEEE Trans. Nucl.

Sci., vol. 59, no. 4, pp. 918–926, Aug. 2012.

[11] C. Virmontois et al., “Similarities between proton and neutron induced dark current distribution in CMOS image sensors,” IEEE Trans. Nucl.

Sci., vol. 59, no. 4, pp. 927–936, Aug. 2012.

[12] C.-T. Sah, R. N. Noyce, and W. Shockley, “Carrier generation and recombination in p-n junctions and p-n junction characteristics,” Proc.

IRE, vol. 45, no. 9, pp. 1228–1243, Sep. 1957.

[13] C. Virmontois et al., “Total ionizing dose versus displacement damage dose induced dark current random telegraph signals in CMOS image sensors,” IEEE Trans. Nucl. Sci., vol. 58, no. 6, pp. 3085–3094, Dec. 2011.

[14] H. J. Whitlow and S. T. Nakagawa, “Low-energy primary knock on atom damage distributions near MeV proton beams focused to nanometre dimensions,” Nucl. Instrum. Methods Phys. Res. B, Beam Interactions

Mater. Atoms, vol. 260, no. 1, pp. 468–473, 2007.

[15] J. M. Belloir et al., “Dark current spectroscopy in neutron, proton and ion irradiated CMOS image sensors: From point defects to clusters,”

IEEE Trans. Nucl. Sci., vol. 64, no. 1, pp. 27–37, Jan. 2017.

[16] D. K. Schroder, Semiconductor Material and Device Characterization. Hoboken, NJ, USA: Wiley, 2006.

[17] S. M. Sze, Semiconductor Devices: Physics and Technology. Hoboken, NJ, USA: Wiley, 1985.

[18] X. Wang et al., “A 2.2M CMOS image sensor for high-speed machine vision applications,” Proc. SPIE, vol. 7536, p. 75360M, Jan. 2010. [19] J. Bogaerts, B. Dierickx, G. Meynants, and D. Uwaerts, “Total dose

and displacement damage effects in a radiation-hardened CMOS APS,”

IEEE Trans. Electron Devices, vol. 50, no. 1, pp. 84–90, Jan. 2003.

[20] M. Beaumel, D. Herve, and D. Van Aken, “Cobalt-60, proton and electron irradiation of a radiation-hardened active pixel sensor,” IEEE

Trans. Nucl. Sci., vol. 57, no. 4, pp. 2056–2065, Aug. 2010.

[21] J.-M. Belloir et al., “Dark current spectroscopy on alpha irradiated pinned photodiode CMOS image sensors,” IEEE Trans. Nucl. Sci., vol. 63, no. 4, pp. 2183–2192, Aug. 2016.

[22] M. Moll, “Radiation damage in silicon particle detectors: Microscopic defects and macroscopic properties,” Ph.D. thesis, Hamburg Univ., Hamburg, Germany, 1999.

[23] J. Bourgoin and M. Lannoo, Point Defects in Semiconductors II:

Experimental Aspects. New York, NY, USA: Springer-Verlag, 1983.

[24] V. Goiffon, G. R. Hopkinson, P. Magnan, F. Bernard, G. Rolland, and O. Saint-Pe, “Multilevel RTS in proton irradiated CMOS image sensors manufactured in a deep submicron technology,” IEEE Trans. Nucl. Sci., vol. 56, no. 4, pp. 2132–2141, Aug. 2009.

[25] C. Durnez, V. Goiffon, C. Virmontois, J.-M. Belloir, P. Magnan, and L. Rubaldo, “In-depth analysis on radiation induced multi-level dark current random telegraph signal in silicon solid state image sensors,”

IEEE Trans. Nucl. Sci., vol. 64, no. 1, pp. 19–26, Jan. 2016.

[26] T. Nuns, G. Quadri, J. P. David, and O. Gilard, “Annealing of proton-induced random telegraph signal in CCDs,” IEEE Trans. Nucl. Sci., vol. 54, no. 4, pp. 1120–1128, Aug. 2007.

[27] C. Virmontois et al., “Dark current random telegraph signals in solid-state image sensors,” IEEE Trans. Nucl. Sci., vol. 60, no. 6, pp. 4323–4331, Dec. 2013.