• Aucun résultat trouvé

Investigation of dark current random telegraph signal in pinned photodiode CMOS image sensors

N/A
N/A
Protected

Academic year: 2021

Partager "Investigation of dark current random telegraph signal in pinned photodiode CMOS image sensors"

Copied!
5
0
0

Texte intégral

(1)

To cite this document

: GOIFFON Vincent, VIRMONTOIS Cédric, MAGNAN Pierre.

Investigation of dark current random telegraph signal in pinned photodiode CMOS image

sensors. In: IEEE International Electron Devices Meeting (IEDM 2011), 05-07 Dec 2011,

Washington, USA.

This is an author-deposited version published in:

http://oatao.univ-toulouse.fr/

Eprints ID: 5125

Any correspondence concerning this service should be sent to the repository administrator:

(2)

Investigation of Dark Current Random Telegraph Signal in

Pinned Photodiode CMOS Image Sensors

Vincent Goiffon, Cédric Virmontois, and Pierre Magnan

Université de Toulouse, ISAE, Image Sensor Research Team 10 ave. E. Belin, Toulouse, F-31055, France

Phone +33561338093, Fax : +33561338351, email : vincent.goiffon@isae.fr

Abstract

The characteristics of Dark Current Random Telegraph Signal (DC-RTS), observed in Pinned PhotoDiode (PPD) CMOS Image Sensors (CIS), are investigated thanks to a dedicated analysis tool. Our results demonstrate, for the first time in PPD CIS, that this DC-RTS is due to meta-stable oxide interface SRH generation centers located in the transfer gate depletion region.

Introduction

Because of the continuous reduction in average dark current and noises, blinking pixels are becoming one of the main limiting factors of state of the art CMOS Image Sensor (CIS) performances. Two phenomena can cause this discrete variation of dark signal in CIS: the Source Follower Random Telegraph Signal [1-3] noise (SF-RTS) and the Dark Current Random Telegraph Signal (DC-RTS). Whereas significant efforts have been made to understand and mitigate SF-RTS in Pinned Photodiode (PPD) CIS [2,3], very few results have been published [4] on dark current RTS in PPD. Therefore its origin and typical characteristics remain unclear.

In this paper the characteristics of dark current RTS in Pinned PhotoDiode (PPD) are investigated thanks to a dedicated original RTS analysis tool. Our results demonstrate, for the first time in pinned photodiode CIS, that this RTS is due to meta-stable oxide interface SRH generation centers located in the Transfer Gate (TG) depletion region. The RTS defects appear to behave similarly to the SRH centers responsible for Variable Retention Time (VRT) in DRAM [5-8]. The influences of temperature and transfer gate voltage on RTS characteristics are also reported for the first time in PPD CIS. A good understanding of the mechanisms at the origin of DC-RTS in PPD CIS is required for the further reduction of this phenomenon in CIS, but also in other devices (e.g. such as VRT in DRAM or Variable Junction Leakage (VJL) in MOSFET source/drain [9]).

Experimental details

The studied device is a 7 µm pitch 256x256 4T-PPD pixel array (see Fig.1 for a simplified 4T-PPD cross-section), manufactured in a commercial 0.18 µm CMOS CIS process and designed for scientific applications. The average conversion factor was around 65µV/e-. A dedicated RTS

detection and analysis tool was used for this study. The details of this software can be found in [10]. Dark current measurements were performed at 60°C (except for activation energy determination).

Fig.1: Simplified cross section of the studied four-transistor (4T) Pinned PhotoDiode (PPD) pixel. Insets show the transfer gate depletion region for

two bias cases: (a) VLoTG=0V: the TG depletion region merges with the PPD

depletion region. (b) VLoTG<0V: the gate is accumulated and the PPD

depletion region does not touch any oxide interface anymore. The P+ pinning layer and the spacers are not represented for the sake of clarity.

Fig.2: Measured dark current variations with time of a selection of pixels exhibiting a DC-RTS behavior. The studied RTS can exhibit very different shapes, time constants, duty cycles and amplitudes (X and Y axis scales differ from one plot to another). Each data point represents a single dark frame. Integration time = 265 ms, temperature = 60°C.

(3)

General observations

Selected dark current variations along time of pixels exhibiting RTS behaviors are presented in Fig.2. It is important to emphasize that each data point corresponds to a single dark frame. This figure illustrates the various forms that can take this DC-RTS, especially in terms of amplitudes, number of levels and time constants. Fig. 3 shows the output voltage variation of a typical RTS pixel in the dark for several integration times. It can clearly be seen that the observed discrete switching amplitude (RTS amplitude) is rising proportionally to integration time, clearly demonstrating that the observed RTS is a dark current RTS due to discrete leakage current variations and not a SF-RTS noise.

The mapping of dark current on a 60x60-pixel-area is presented in Fig. 4a to show the hot pixel distribution. The spatial distribution of detected RTS pixels in the same 60x60-pixel-area is presented in Fig. 4b (almost only 2-level RTS were seen in the device). As compared to the average dark current map (Fig. 4a), one can see that both distributions are pretty uniform spatially and that no obvious correlation

appears between hot pixels (i.e. pixels with high average dark current value, also called white or bright pixels) and DC-RTS pixels. This interpretation is confirmed in Fig.5 where no clear correlation arises between RTS amplitude and mean dark current value. It indicates that hot pixels are not necessarily RTS (in the time range of the experiment) and that RTS pixels do not necessarily exhibit the highest average current values. The dark current standard deviations of non-RTS pixels are also plotted in Fig. 5 (right-hand y-axis). One can notice that most RTS amplitudes are more than one order of magnitude larger than the dark current shot noise.

Fig.5 : RTS amplitude versus mean dark current of RTS pixels. The dark current standard deviation of non-RTS-pixels is also shown. No correlation can be seen between RTS amplitude and mean dark current. A significant number of RTS amplitudes are more than one order of magnitude larger than the Dark Current (DC) shot noise.

Fig.6: RTS amplitude versus RTS switching time constant. RTS time constants are almost uniformly distributed over the detection time range (10s to 2h). No correlation can be seen between RTS amplitudes and RTS time constants. The alignment of data points, by column, for time values above

103 s is due to the fact that only a few RTS transitions occurred in these

signals, leading to apparent quantified time constants (with values equals to

τ = Tobs/Ntrans, where Tobsis the total observation time and Ntransthe number of

observed RTS transitions during Tobs.).

Fig.3 : Typical dark output voltage variation with time of a DC-RTS pixel for several integration time values. The RTS amplitude is proportional to the integration time clearly demonstrating that the observed RTS is a dark current RTS (not a Source Follower MOSFET RTS noise). T = 60°C.

Fig.4 : (a) Dark current map compared to (b) detected DC-RTS pixel map on the same 60x60 pixel area. Dark current values and detected DC-RTS pixels appear uniformly distributed (no clustering). No obvious correlation can be seen between dark current extreme values (hot pixels) and detected DC-RTS pixels. Integration time tint = 265 ms, temperature T = 60°C.

DC shot noise (b) (a) RTS No t RTS

(4)

Similarly, no correlation can be seen between DC-RTS amplitudes and DC-RTS time constants in Fig. 6. It should be emphasized that the observed time constants do not appear restricted to a particular time range.

Influence of temperature

DC-RTS amplitudes and time constants appeared to vary exponentially with temperature. The activation energies

extracted for DC-RTS amplitude are presented in Fig. 7. Most RTS pixels that stayed in the detection range during the whole temperature sweep (44°C to 68°C) have dark current switching amplitude activation energies around or slightly above the mid-gap value [11] (with an average value of 0.69eV±0.1eV). Therefore, Shockley-Read-Hall (SRH) Recombination-Generation (R-G) centers located in the PPD depletion region appear to be responsible for this RTS. To generate discrete dark current values, these SRH centers are necessarily meta-stable centers. The average extracted time constant activation energy was 0.77eV±0.3eV.

Fig.7: Activation energy of RTS amplitudes as a function of RTS amplitude

at 60°C. Temperature range used: 44°C-68°C, tint = 265 ms. The average

values (0.69 eV) slightly above midgap (0.56 eV) strongly suggests a Shockley-Read-Hall generation mechanism.

Fig.8: Number of detected RTS pixels and sensor average dark current as a function of TG OFF voltage (VLoTG). When a negative TG voltage is used, there is a clear disappearance of RTS pixels showing that the oxide interface in the TG vicinity is responsible for the observed RTS.

Fig.9: DC-RTS amplitude distribution for several TG OFF voltages (VLoTG). The clear left-shift of the distribution with decreasing VLoTG confirms the disappearance of DC-RTS centers in the pinned photodiode when the transfer gate is accumulated.

Fig.10: Typical DC-RTS evolution with time for several TG voltages. The amplitude is not influenced by the TG voltage until the RTS centers is out the depletion region (for VLoTG=-0.3V).

(5)

Influence of transfer gate OFF voltage during integration

During integration, the TG OFF (or low) voltage VLoTG is known to have a strong impact on PPD dark current [12,13]. When VLoTG is 0V, the TG depletion region merges with the PPD depletion region (inset (a) of Fig.1) whereas the PPD depletion region does not touch any oxide interface (inset (b) of Fig.1) when the TG is in the accumulation regime (VLoTG = -0.6V in Fig. 8). It can clearly be seen in Fig. 8 that in the latter case, the number of detected RTS pixels is almost zero. Further insight into this phenomenon is provided by Fig. 9 which clearly shows that the overall DC-RTS distribution shifts to the left when VLoTG decreases. This indicates that the less the PPD depletion is in contact with an oxide interface, the less RTS pixels are detected. Therefore, the DC-RTS centers appear clearly located at the oxide interface, in (or near) the TG channel (gate oxide or top of the shallow trench isolation sidewalls, on the channel borders).

Fig. 10 illustrates the typical behavior of RTS centers when VLoTG is reduced: the amplitude stays roughly constant while it stays in the depletion region, and then disappears rapidly when the center reaches the depletion region boundary (for VLoTG=-0.3V in Fig. 10). Most of the studied RTS pixels exhibited the same behavior. This is clear evidence that the TG voltage (and thus, the induced electric field) does not have a strong influence on the DC-RTS amplitude, as far as the RTS center stays in the depletion region. Therefore, electric field enhancement processes [14], such as trap assisted tunneling, do not seem to be at the origin of the large dark current fluctuations observed. It agrees well with the average RTS amplitude activation energy determined around 0.6 eV.

Some rare pixels also exhibited a strong amplitude dependence on VLoTG suggesting some scarce electric field

enhancement processes but this behavior was clearly not representative of the whole RTS pixel population. Hence, reducing the electric field magnitude is not expected to reduce the amplitude of the observed DC-RTS.

Conclusion

Dark current RTS was detected and studied in 4T-PPD-CIS. Meta-stable interface states acting as SRH R-G centers in the TG channel, in the vicinity of the PPD, appeared to be responsible. The observed characteristics (amplitudes, time constants…) were similar to DC-RTS due to silicon bulk displacement damages in solid state imagers (CCD and CIS) exposed to high energy particles [15], and also to the DC-RTS recently observed in 3T-CIS-pixel conventional photodiode [16]. However, in PPD CIS, these centers are clearly located in the TG vicinity whereas they are located in the bulk or on the photodiode perimeter in the previous cases (irradiated imagers or 3T-pixel-CIS). The reported behaviors also strongly suggest that the meta-stable interface states responsible for this phenomenon are the same than the ones at the origin of VRT in DRAM [7,8] and of VJL in MOSFET source/drain [9]. Finally, accumulating the TG can completely remove DC-RTS but it may not be applicable in most applications where anti-blooming is necessary (VLoTG must be slightly greater than 0V for an efficient anti-blooming). Moreover, biasing VLoTG negatively can cause long term reliability issues due to negative-bias-temperature stress as recently disclosed [17].

Acknowledgment

The authors are grateful to Philippe Martin-Gonthier, Franck Corbière and Sébastien Rolando for their help on circuit design. We also would like to thank Paola Cervantes and Barbara Avon for their support on characterization.

References

[1] M.J. Kirton et al., Adv.Phys., vol. 38 (4), p. 367, 1989. [2] C. Leyris et al.,Proc. ESSCIRC, p 376, 2006 [3] X. Wang et al., IEDM Tech. Dig., p. 1, 2006. [4] K. Ackerson et al., ASMC, p. 255, 2008. [5] D.S. Yaney et al., IEDM Tech. Dig., p. 336, 1987 [6] P.J. Restle et al., IEDM Tech. Dig., p. 807, 1992. [7] Y. Mori et al., IEDM Tech. Dig., p. 1034, 2005. [8] T. Umeda et al., Appl. Phys. Lett., Vol. 88 (1), 2006. [9] K. Abe et al, Proc. IRPS, p. 683, 2010.

[10] V. Goiffon et al., IEEE TNS, vol. 56 (4), p. 2132, 2009. [11] A.S. Grove et al., Solid-State Electron., vol. 9 ( 8), p. 783, 1966. [12] H. Han et al., Proc. IISW, 12-6, 2007.

[13] T. Watanabe et al., IEEE TED, vol. 57 (7), p. 1512, 2010. [14] G. Vincent et al., J. Appl. Phys., vol. 50 (8), p. 5484, 1979. [15] G.R. Hopkinson, IEEE TNS, vol 47(6), p. 2480,2000. [16] V. Goiffon et al., IEEE EDL, vol. 32 (6), p. 773, 2011. [17] H. Yamashita, Proc. IISW, R29, 2011.

Fig.11: Rare example of DC-RTS evolution with time for several VLoTG values showing a clear amplitude dependence on TG voltage. Electric field enhancement could be the cause of this rare behavior dependence.

Figure

Fig. 10 illustrates the typical behavior of RTS centers when  VLoTG is reduced: the amplitude stays roughly constant  while it stays in the depletion region, and then disappears  rapidly when the center reaches the depletion region  boundary (for VLoTG=-0.

Références

Documents relatifs

On the suction side, URANS predicts the same transition process than on the pressure side: first a periodic transition induced by incoming wakes before to reach a

Again in the limit of large inertia ratios, LSA predicts existence of two unstable oscillating modes which are nothing but those associated with the linear wake instability past a

In this paper, we present globally adaptive control variate, a numerical integration method based on control variate, targeting arbitrary scalar or vector-valued integrands

chemical results of milk analysis from three dairy farms, located in the surrounding of Algiers area (Algiers, Blida and Tipaza) show a significant effect (p &lt; 0.05) of the

The split-root assay results described above suggested that AOI was mediated by a mobile signal migrating from the proximal to the distal root either via the plant

C’est dans ce contexte que nous avons choisi d’étudier spécifiquement les états de surface induits par la technique de projection laser, en considérant un

Comme la m´ ethode num´ erique exploite ces estimations, il suffira de coupler ` a ces es- timations les deux techniques propos´ ees pour produire op´ erationnellement la

L’objectif principal de ce chapitre est l’étude de l’impact de la taille de pores sur la colonisation microbienne des électrodes et la production de courant