Journal of ELECTRONIC MATERIALS, Vol. 39, No. 6, 2010
DOI: 10.1007/s11664-010-1212-6 Ó 2010 TMS
Study of Metal Contamination in CMOS Image Sensors by Dark-Current and Deep-Level Transient Spectroscopies F. DOMENGIE,1,2,3 J.L. REGOLINI,1 and D. BAUZA2 1.—STMicroelectronics, 850 rue Jean Monnet, 38926 Crolles Cedex, France. 2.—IMEP-LAHC, Minatec, 3 rue Parvis Louis Ne´el, BP 257, 38016 Grenoble Cedex 1, France. 3.—e-mail: florian.
[email protected]
Pixels in complementary metal–oxide–semiconductor (CMOS) image sensors (CISs) are being scaled downward toward 1.0 lm. In this context, improvements in crucial parameters such as dark current per pixel, which suffers from defects incorporated during processing, need to be achieved. Indeed, accidental metallic contamination is a critical issue that induces dark current and reduces yield. In this paper, detection and characterization of gold and tungsten implanted in CISs using dark-current and deep-level transient spectroscopies are reported. Deep levels responsible for dark current are identified, and tungsten is studied for the first time with dark current spectroscopy. Key words: Dark current, deep level, contamination, image sensors, gold, tungsten
INTRODUCTION Complementary metal–oxide–semiconductor (CMOS) image sensors (CISs) are receiving much attention for large-volume electronic applications such as mobile phones, digital cameras, webcams, and automobiles. Although this technology presents advantages over charge-coupled devices (CCDs) due to its lower power consumption and manufacturing cost along with its high-speed imaging and CMOS standard technology compatibility,1 the dark current in CISs is higher than that in CCDs.2 As pixels are being scaled downward toward 1.0 lm, to keep acceptable pixel performance, a serious challenge is to improve sensor sensitivity, quantum efficiency, etc. while keeping the dark current as low as possible. Dark current in image sensors is a parasitic current created by carriers not generated by photons in the photodiodes. It is enhanced by individual defects such as metallic contamination dissolved in silicon, by interface states, or by structural defects. Its main cause is remaining defects incorporated during the whole process. Plasma processing that affects gate oxide integrity and increases the dark (Received September 12, 2009; accepted March 18, 2010; published online April 20, 2010)
current has been presented.3 Si/SiO2 interfaces from transfer gate or shallow trench isolation regions are known to be a dark-current source that can be minimized by hydrogen passivation.4 Metal contamination introduced by ion implantation has been reported several times5,6 and is a potential source of contamination on the production line. Usual in-line detection techniques are not sensitive enough to detect contamination levels that impact the production yield. In the 1990s, dark-current spectroscopy (DCS) was developed to study such levels of pixel contamination in CCD sensors.7,8 The quantization of dark-current generation was first observed in 1987.9 More recently, this technique allowed the characterization of radiation-induced defects in CCDs,10 and unknown contaminations have been investigated by DCS and deep-level transient spectroscopy (DLTS) in CISs.11 In this work, detection and characterization of gold and tungsten contamination by ion implantation in CISs have been studied using both DCS and DLTS. EXPERIMENTAL PROCEDURES A complete description of the pixel technology used in this work has already been published.12 The pixel architecture is a 4T-type pinned photodiode 625
626
Domengie, Regolini, and Bauza
Fig. 2. Number of white pixels compared with the reference wafer versus implanted dose for gold and tungsten.
Fig. 1. Zoomed image of pixels contaminated with gold taken at 45°C and integrated for 400 ms in the dark, showing white pixels. Here the pixel size is 2.2 lm length.
with 1.75 equivalent shared transistors per pixel. Several doses of metallic contamination were introduced using ion implantation through the sacrificial oxide grown before the gate stack was processed. Au and W were implanted at energies corresponding to a projected range between 70 nm and 160 nm into silicon, where the photodiode potential is at its maximum. They underwent the whole process thermal budget. These elements were chosen for their expected dangerousness for image sensors. Their diffusivities at 1100°C13 differ by more than eight orders of magnitude: Au (8.9 9 106 cm2/s) is expected to diffuse over the whole wafer thickness during process thermal treatment, whereas W (7.6 9 1014 cm2/s) is expected to stay in the first micron below the surface of the photodiode active area. The dark current was measured for several dies per wafer at different temperatures, and histograms of the number of pixels versus dark current were plotted at 30°C, 45°C, and 60°C. A pixel is called ‘‘white’’ if, without external illumination, its current is large enough that it appears white in the image taken by the sensor. Figure 1 shows such a picture in the case of an image sensor contaminated with Au. The white pixel threshold was 25 dark current code at 45°C. DLTS measurements were carried out directly on an N+/P diode that was specially included into the CISs wafers. Doping concentrations were 5 9 1017 cm3 for the N+ region and 5 9 1016 cm3 for the P region. For DLTS measurements, the reverse voltage used was 1.5 V and the filling pulse voltage was 0.1 V.
white pixels by a factor of about 1000 for the 100 a.u. implanted dose, and by a factor close to 10,000 for the 1000 a.u. dose. The number of white pixels compared with the reference is increased by a factor of 2000 for the 100 a.u. W implanted dose. Different metal contaminants form deep levels in the silicon bandgap. Graff published an interesting synthesis of these levels.13 The main parameters determining their generation rate are their activation energy Et and their capture cross-sections for electrons rn and holes rp. To model accurately the generation current produced by these deep-level centers, the Shockley– Read–Hall (SRH) formalism is used, where en and ep are the emission rates of the trap for electrons and holes, and sn and sp represent the associated emission time constant14: en ¼
ðEt Ei Þ 1 ¼ rn mth;n ni e kT ; sn
(1)
ep ¼
ðEi Et Þ 1 ¼ rp mth;p ni e kT ; sp
(2)
where ni is the intrinsic carrier concentration, Ei is the intrinsic level, vth,n, and vth,p are the thermal velocities for electrons and holes, k is the Boltzmann constant, and T is temperature. The generation process requires combined emission of one electron and one hole by the deep level. It is limited by the slowest of these emission rates. g, the generation rate, is given by15 pffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rn rp mth;n mth;p ni en ep : g¼ ¼ (3) en þ ep 2 cosh Et Ei þ 1 ln rn mth;n 2 kT rp mth;p Assuming rn rp and vth,n vth,p, r being the mean capture cross-section, g becomes
RESULTS AND DISCUSSION In Fig. 2, the number of white pixels is plotted as a function of the implanted dose of Au and W. The number of white pixels for a wafer not contaminated is used as the reference. Au increases the number of
g¼
rvth;n ni E : i 2 cosh EtkT
(4)
Even if it is acceptable in some studies to use Eq. 4 and consider rn rp,10,11,15 in this work this may be
Study of Metal Contamination in CMOS Image Sensors by Dark-Current and Deep-Level Transient Spectroscopies
too great an approximation,16 since Au contamination produces a deep level with rn/rp ratio of about 0.02 and one goal of this study is to model generation currents accurately. The distribution of metal atoms over the pixel matrix due to contamination is considered random and therefore will follow a Poisson distribution, in agreement with previous observations.7,17 Some pixels do not contain any impurity atoms and have a dark current due to intrinsic sources common to all pixels. A fraction of these contains one atom and will exhibit an additional dark current due to thermal generation at this center in the pixel volume. Other pixels have two metal atoms and therefore twice the dark current with regard to one impurity center is generated, etc. As a result, a histogram of the number of pixels versus the dark current provides equally spaced peaks which shift with temperature in accordance with the activation energy of the deep level.7 In addition, each peak can be approximated by a Gaussian curve whose amplitude is related to the mean concentration of the contaminant in the pixel volume, i.e., to the average number of atoms per pixel. Finally, the generation rate is known to be enhanced by the combination of two major phenomena: energy barrier lowering due to the Poole– Frenkel effect, and the phonon-assisted tunneling mechanism. The natural electric field in the photodiode is quite large and can be estimated as a few 1 9 104 V/cm. These two effects tend to increase the generation rate so that, for a midgap defect at room temperature, the enhancement factor k(E) for the generation rate is believed to be around 2.18,19 The generation current for the Nth peak, IN, is calculated from Eq. (3). The bandwidth of the peak, rN, increases with temperature and with the number of contaminant atoms: rN = Ö(r02 + NrT2), where r0 is the intrinsic and rT is the contaminant atom contribution (see Fig. 4).7 The Nth peak is then simulated using the function fN(I): ! ðI I N Þ2 fN ðIÞ ¼ APN ðXÞ exp ; (5) 2r2N with IN ¼ qNkðEÞg ¼ qNkðEÞ
pffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rn rp mth;n mth;p ni : r m Ei 2 cosh EtkT þ 12 ln rnp mth;n th;p (6)
In Eq. (5), PN(X) is the Poisson probability of having N atoms in one pixel, considering X to be the average number of atoms per pixel. A is a normalization factor, and q is the electronic charge. Figure 3 shows clear evidence of this quantized dark current present in CISs contaminated with Au implanted at 100 a.u. dose. The bandwidth of the peak, rN, increases with peak order and
627
Fig. 3. Histogram of number of pixels versus dark current for different temperatures on an imager contaminated with 100 a.u. gold dose.
Fig. 4. Histogram of number of pixels versus dark current for gold contamination at 100 a.u. dose at 45°C and corresponding simulation of DCS peaks. The peak width parameters used are r0 = 0.15 dark current code and rT = 0.4 dark current code.
temperature, so that the peaks become wider and the amplitude decreases, but the number of pixels from integration of the same peak remains constant at all temperatures. The generation current of the peaks increases with temperature. The first peak is called the intrinsic peak and is the sole peak visible for pixels that do not have any electrically active Au atom. The second peak, associated with pixels that contain one Au atom, shows a dark current of 1.65 code at 30°C, 4.95 code at 45°C, and 14.75 code at 60°C. From Eq. (6), the generation current of Au acceptor deep level at 45°C can be calculated using the parameters reported by Graff13: Et = 0.55 eV below the conduction band, rn = 1.4 9 1016 cm2, and rp = 7.6 9 1015 cm2. This leads to a generation current of 2.70 dark current code per atom. An enhancement factor k(E) of 1.7 means that it can generate 4.59 dark current code, which is actually the difference observed between two peaks at that temperature. In Fig. 4, the histogram extracted for the 100 a.u. Au implanted dose at 45°C and the simulation based on Eqs. (5) and (6) of quantized dark-current peaks are compared. Up to the fourth peak, the simulation follows the experimental results quite closely. DLTS measurements prove (Fig. 5) that the Au donor level is present in pixels at a concentration
628
Domengie, Regolini, and Bauza
Fig. 5. DLTS peak due to the 0.33 eV gold donor level in silicon for the three different implanted doses.
Fig. 6. Histogram of number of pixels versus dark current for tungsten contamination at 10 a.u. dose at 60°C and corresponding simulation of the DCS peaks. Peak parameters used are r0 = 0.2 dark current code and rT = 0.4 dark current code.
40 to 120 times larger than that of the acceptor level deduced from DCS peaks. The total Au acceptor and donor concentrations found are 70 times greater than expected assuming diffusion of Au over the whole wafer thickness during processing. It is common to observe an increase of Au concentration in the first micron below the surface after thermal diffusion, resulting from preferential interaction of Au with vacancies.20,21 A Au activation energy of 0.33 eV above the valence band is extracted from an Arrhenius plot. This is slightly different from the commonly admitted activation energy for this level of 0.34 eV,13 which could be a sign of enhancement of emission rate due to high electric field in the photodiode. Au acceptor and donor levels are known to be not related to the same defect,22 i.e., Au is present in both donor and acceptor states.
Due to the high supposed concentration of W, a slow diffuser, in the active area of the pixel, DCS peaks are expected to be observed at larger dark current code, and first quantized dark-current peaks should not appear clearly. However, first individual peaks are observed in Fig. 6 for the W 10 a.u. dose, although they appear clearly only at 60°C. At 30°C and at 45°C, the dark current produced by individual W atoms is too low to produce visible peaks. Peaks were simulated with the donor level of W in silicon, with parameters from the literature and confirmed by our DLTS measurements: Et = 0.71 eV below the conduction band, rp = 5.0 9 1016 cm2, and rn = 4.8 9 1015 cm2.13,23 An enhancement factor k(E) of 1.8 was used to fit the peaks to obtain a peak spacing of 1.73 dark current code. DLTS measurement indicates that the W donor level has a maximum concentration of about 9 9 1012 cm3. The concentration deduced from the fit to the DCS peaks is 2.7 9 1012 cm3, of the same order of magnitude but lower. Indeed, since two different devices are tested here and because a much larger volume is probed by DLTS, 16,000 lm3 versus less than 2 lm3 for the pixel volume studied by DCS, the difference in concentrations is not surprising. It is interesting to note that, for higher doses of implanted W, the information regarding W atom electrical activity and its fingerprint in the individual peaks are lost (not shown). This may be due to the fact that, for the resulting concentrations, W is no longer in dilute form as required to induce observation of quantized dark-current peaks. The sensitivity of the DCS technique was estimated by using the smallest peak still visible above the reference histogram. For Au we find 2 9 108 cm3 and for W, 3 9 108 cm3. Table I and Fig. 7 summarize the deep-level characteristics and concentrations of contaminants found in the photodiodes studied. From Fig. 7, the W donor level concentration is increased by only a factor of three when the implanted dose is increased by ten, indicating that some of the W is not diluted, which may explain the loss of quantized dark-current peaks. CONCLUSION Inductively coupled plasma-mass spectrometry (ICPMS), secondary ion mass spectrometry (SIMS), surface photovoltage (SPV), or DLTS techniques have been exploited in previous works to monitor
Table I. Deep-level characteristics obtained from DCS and DLTS measurements on gold- and tungstenimplanted CISs DCS Deep Level
Et (eV)
Au acceptor Au donor W donor
0.55 0.71
rn (cm2)
DLTS rp (cm2)
1.4 9 1016 7.5 9 1015 Level not detected 4.8 9 1015 5.0 9 1016
k(E)
Et (eV)
1.7 1.8
0.33 0.71
rn (cm2)
rp (cm2)
Level not detected – 4.0 9 1014 – 3.0 9 1016
Study of Metal Contamination in CMOS Image Sensors by Dark-Current and Deep-Level Transient Spectroscopies
629
REFERENCES
Fig. 7. Au and W deep-level concentrations from DCS and DLTS measurements for the three doses used.
metals introduced accidentally by implantation in silicon.5,6 However, using these methods, the contaminant is either not identified or their detection limits are too high. In this study, it has been confirmed that DCS is an efficient method to monitor contamination of image sensors due to implantation. It uses pixels as electrical microsensors to detect dilute defects at concentrations as low as a few 108 cm3 for gold and tungsten. The DCS technique has been developed for metals monitoring on CISs, leading to the first observation of quantized dark-current peaks from the W slowdiffuser contaminant. DLTS is less sensitive than DCS but is shown to be a complementary technique for the detection of deep levels not close enough to the midgap to produce large dark current for visible DCS peaks. An accurate SRH model including electric field enhancement considerations in combination with Poisson statistics permitted accurate simulation of observed dark-current peaks. W contamination study with DCS may not result in clear dark-current peaks for concentrations >5 9 1012 cm3, but at this concentration other traditional methods are able to detect this element. ACKNOWLEDGEMENTS The authors thank C. Augier, D. Herault, and S. Hulot from STMicroelectronics, Crolles, for their help in dark-current measurements and fruitful discussions, and M. Gri from IMEP, Grenoble, for her help in DLTS sample preparation.
1. M. Bigas, E. Cabruja, J. Forest, and J. Salvi, Microelectron. J. 37, 433 (2006). 2. A. El Gamal and H. Eltoukhy, IEEE Circuit. Dev. 21, 6 (2005). 3. K. Tokashiki, K. Bai, K. Baek, Y. Kim, G. Min, C. Kang, H. Cho, and J. Moon, Thin Solid Films 515, 4864 (2007). 4. Y. Kunimi and B. Pain, Proceedings of the 2007 International Image Sensor Workshop, Ogunquit Maine, USA, June 7–10, 2007, p. 66. 5. E. Colelli, A. Galbiati, D. Caputo, M.L. Polignano, V. Soncini, and G. Salva, Proceedings of the 8th International Symposium on Plasma and Process Induced Damage, Corbeil-Essonnes, France, April 24–25, 2003, p. 81. 6. R.B. Liebert, G.C. Angel, and M. Kase, Proceedings of the 11th International Conference on Ion Implantation Technology, Austin Texas, USA, June 16–21, 1996, p. 135. 7. W.C. McColgin, J.P. Lavine, and C.V. Stancampiano, Mater. Res. Soc. Symp. Proc. 378, 713 (1995). 8. W.C. McColgin, J.P. Lavine, C.V. Stancampiano, and J.B. Russell, Mater. Res. Soc. Symp. Proc. 510, 475 (1998). 9. R.D. McGrath, J. Dory, G. Lupino, G. Ricker, and J. Vallerga, IEEE Trans. Electron. Dev. 34, 2555 (1987). 10. C. Tivarus and W.C. McColgin, IEEE Trans. Nucl. Sci. 55, 1719 (2008). 11. H.I. Kwon, I.M. Kang, B.-G. Park, J.D. Lee, and S.S. Park, IEEE Trans. Electron. Dev. 51, 178 (2004). 12. M. Cohen, F. Roy, D. Herault, Y. Cazaux, A. Gandolfi, J.P. Reynard, C. Cowache, E. Bruno, T. Girault, J. Vaillant, F. Barbier, Y. Sanchez, N. Hotellier, O. LeBorgne, C. Augier, A. Inard, T. Jagueneau, C. Zinck, J. Michailos, and E. Mazaleyrat, IEDM Technical Digest, 2006, pp. 127–130. 13. K. Graff, Metal Impurities in Silicon-Device Fabrication (Berlin: Springer, 2000). 14. D.K. Schroder, Semiconductor Material and Device Characterization (New York: Wiley, 1998). 15. D.K. Schroder, IEEE Trans. Electron. Dev. 29, 1336 (1982). 16. E. Simoen and C. Claeys, IEEE Trans. Electron. Dev. 46, 1487 (1999). 17. W.J. Toren (PhD thesis, Twente University, 1997). 18. J. Bogaerts, B. Dierickx, and R. Mertens, IEEE Trans. Nucl. Sci. 49, 1513 (2002). 19. A.F. Tasch Jr. and C.T. Sah, Phys. Rev. B 1, 800 (1970). 20. R. Kalyanaraman, T.E. Haynes, V.C. Venezia, D.C. Jacobson, H.-J. Gossmann, and C.S. Rafferty, Appl. Phys. Lett. 76, 3379 (2000). 21. M. Schulz, A. Goetzberger, I. Fra¨nz, and W. Langheinrich, Appl. Phys. A Mater. 3, 275 (1974). 22. S.J. Pearton and A.J. Tavendale, Phys. Rev. B 26, 7105 (1982). 23. S. Diez, S. Rein, and S.W. Glunz, Proceedings of the 20th European Photovoltaic Solar Energy Conference, Barcelona, Spain, June 6–10, 2005, p. 1216.