Appl. Phys. A 66, S415–S419 (1998) Applied Physics AMaterialsScience & Processing Springer-Verlag 1998

Recombination dynamics of traps inSiO2 layer onSi by scanningcapacitance microscopyC.J. Kang1, C.K. Kim 1, Y. Kuk 1, C.C. Williams2

1Department of Physics, Seoul National University, Seoul 151-742, Korea(E-mail: ykuk@phya.snu.ac.kr)2Department of Physics, University of Utah, Salt Lake City, UT 84112, USA

Received: 25 July 1997/Accepted: 1 October 1997

Abstract. The depth-dependent carrier density and trappedcharges in an arsenic-implanted silicon sample were meas-ured by using scanning capacitance microscopy (SCM). Theposition-dependent capacitance versus voltage scans weremeasured at various dc bias voltages. A strong dc bias de-pendence was observed at the interface of an abrupt junctionbetween n+ and p. The bias-dependent SCM images wereused to follow the recombination dynamics of traps. Theresults show good agreement with quasi-three-dimensionalsimulations, suggesting that they can be used to map devicestructure.

With the recent developments in silicon submicrometer tech-nology, the characterization of the three-dimensional carrierdensity profile in a device has become one of the mostimportant issues. In a sub-micron device, it is very diffi-cult to directly measure the carrier density profile aroundthe gate because many tools measure only the profile av-eraged over a large area. Secondary ion mass spectrometry(SIMS) and spreading resistance profilometry (SRP) havewidely been used to measure dopant profiles. The tomo-graphic approach [1] and the angle lapping technique [2] havebeen developed to measure two-dimensional profiles withimproved resolution. Recently, nano-SRP, utilizing scanningprobe microscopy, has further improved spatial resolution [3–5]. Other important problems in submicron MOS devices aredevice failure and noise from hot electron effects and trappedcharges. If the gate oxide in a MOS device becomes thinnerthan10 nmand5 V is still applied to the gate, problems re-lated to the hot electron effect and trapped charges becomemore serious. At present there is no tool to measure the lo-cal defects such as trapped charges and the defects createdby hot electrons. When data is taken by macroscopic measur-ing tools and then compared with the simulated results pooragreement is found.

Since the advent of the scanning tunneling microscope(STM) and the atomic force microscope (AFM), geometri-cal and electrical measurements of a semiconductor surface

have become possible with atomic resolution. As a probe isscanned over a specific area, the carrier density profile [6], thepotential across abrupt junctions [7], the impurity distribu-tion [8], and the carrier scattering near defect centers [9] canbe measured at the same time. The carrier density of a semi-conductor can be mapped by the capacitance-voltage (C-V)dependence, and the geometrical structure can be measuredby STM or AFM. In order to have good spatial resolution, thesensitivity of the detector has to be better than10−19 F [10].This can be achieved with the scanning capacitance micro-scope (SCM) operating under ambient conditions by usinga sample with only a capping oxide and without any furtherextensive sample preparation. However, it only measures thecarrier density rather than the dopant concentration of thesemiconductor surface. If the dopant density does not varyabruptly within a few Debye lengths, it can be estimatedfrom the measuredC-V dependence [11]. In an earlier SCM,the capacitance signal was fed back to control the gap be-tween the sample and the probe, achieving a25 nm lateralresolution [10]. The SCM was able to detect the small vari-ations in capacitance even at highly doped regions by sens-ing the depletion modulation [12]. The carrier density profileat one point and the two-dimensional (2D) map at a fixedbias were measured, showing good agreement with SIMS re-sults within the dopant range of1017–1020 cm−3 [13–17].The dopant profile can be estimated by comparing the meas-uredC-V dependence with the SIMS result. In this work, bymeasuring theC-V dependence at every point, it has beenpossible to obtain dc-voltage-dependent capacitance images,which are similar to a quasi-3D carrier density profile. Themeasured dc-bias-dependent images were compared with thesimulated result.

The SCM measures not only the carrier density in equi-librium but also the trapped charges at the interface or ina dielectric material. Once there are trapped charges underthe probe, the electric field is altered around the traps, thusobeying the Poisson equation. The trapped charges then shiftthe C-V dependence toward the negative or positive direc-tion depending on whether they are hole or electron traps. Theamount of the shift would depend on the amount of trapped

S416

charges. In this study we were able to measure the amountand the recombination dynamics of the trapped charges bySCM.

1 Experimental

The SCM used in this study is similar to the one describedpreviously [12–17]. Briefly, a contact-mode AFM, with thebeam deflection method, was used. An RCA video disc sensorthat operates at915 MHz was used to detect the capacitancesignal. This high frequency enables us to detect a relative ca-pacitance variation less than10−18 F [18]. The tip used wasfirst coated with50 nm thick Ti (or Cr ) by e-beam evapo-ration and annealed shortly before being attached to the ca-pacitance sensor. Figure 1 shows the schematic diagram ofthe SCM used in this study. A dc bias (−5 V to 5 V) withan ac modulation (mainly20 mV peak to peak ,50 kHz) wasapplied to the sample to obtain theC-V dependence. It wasfound either by experiment or by circuit simulation that withan ac modulation, more than90% of the dc bias can be ap-plied to the sample. Since the resistance of a semiconductorsample is inversely proportional to the carrier density, the ap-plied dc bias voltage should be rescaled by considering thecarrier density (substrate resistance) and the oxide thickness(oxide resistance). An electrical stress can also be appliedfrom the sample side. A dc stress can be applied from theprobe side through an opto-couple to the probe, as shown inFig. 1. The capacitance signal measured by the RCA sensorwas fed into a lock-in amplifier and recorded by a computer.The measured signal sensitivity (5µV) which gave the ca-pacitance variations in sub-aF was large enough to scan atthe rate of1.25 s/line. Since the force between the probeand sample was regulated by the AFM, the topographic andcapacitance images can be obtained simultaneously. The sam-ple used was a p-typeSi substrate (boron doped,1015 cm−3).With an oxide mask, a dose of2×1013 cm−2 of arsenic wasimplanted at40 keV energy as shown in Fig. 2. The samplewas then annealed at950◦C subsequently underN2 ambient.Finally, the oxide mask was removed and a25 nmgate oxidewas grown for the capacitance measurement.

2 Results and discussion

The oxide mask was imaged by using SCM before it was re-moved (see Fig. 3). The circular mask is visible and it is thickenough for implanted As not to reach the p-typeSi substrate.The distribution of the dopant profile can then be calculatedby solving the Poisson equation. At present, there is no propersimulation package by which a 3DC-V dependence undera nonequilibrium condition can be calculated. We have mod-ified the TSUPREM-4 and MEDICI packages to explain ourresults [19]. The depletion layers for various probe–samplebiases were calculated in a cross-sectional plane along theZ Z′ direction as shown in Fig. 4a. The edge of the dark areadenotes a line of equal arsenic and boron dopant concentra-tion. By scanning a50 nmwide simulated rectangular elec-trode along theZ Z′ direction, we were able to calculate thedepletion lines. The lines of depletion were drawn at the pos-ition where the density of the majority carrier (hole) is 1/2.

Fig. 1. Schematic diagram of the scanning capacitance microscope used inthis study. An opto-couple was used to give stress from the tip

Figure 4b shows a characteristic 1D (dC/dV)-V curve cal-culated at the marked points of Fig. 4a. The series resistanceeffect of the inversion layer [11] with a depleted capacitancewas considered.

Figure 5 shows the measured one pointC-V dependenceat various points. The measured (dC/dV)-V (curve B) isquite similar to that at low frequencies, even though the datawere measured at a high frequency. This is due to the accurrent at the abrupt junction between theAs-doped n andsubstrate p. The measured data agree well with the simulateddata in Fig. 4b. ThedC/dV monotonically increases withbias voltage (curve A) at an implanted area. The slight devi-ation from the ideal non-equilibrium (dC/dV)-V curve may

Fig. 2. The preparation process of the sample used in this study

S417

Fig. 3. The atomic force microscope image of the oxide masked sample (7×7µm2)

be caused by the excess carriers generated in the depletion re-gion from the scattered laser light in AFM. The carriers canmove out of that region before recombination caused by thefield produced by the built-in potential of thepn junction orby the externally applied bias [20]. Humidity, mobile ions onthe surface, immobile charges trapped in the oxide may makeshunt paths, or a local high electric field, resulting in a de-viation. As the probe moves away from the junction edge tothe unimplanted region, the measured curve shifts to the rightwith little change in its shape. The curve of the implanted areakeeps nearly the same shape irrespective of the tip position.

As an electrode moves from A to E in Fig. 4b, the curveshifts to the right but as it moves from F to I (implanted area),it is nearly unchanged. The contrast in the derivative capaci-tance images comes from the spatial variation of thedC/dVat a given dc bias. Thus the SCM image at the sample bias of−0.1 V will look like an inverted one at1.3 V but a brightedge ring appears at∼ 0.5 V. The derivative capacitance im-ages were calculated by the process simulation by assumingthat this system has a rotational symmetry about the centerof the oxide mask as shown in Fig. 6. The SCM images of7×7µm2 were observed at various bias voltages. Since an ar-ray of circular masks was used, arsenic was mainly implantedonly outside the circular patterns (lateral straggle should beconsidered in a strict sense). The topography shows a heightdifference less than3 nm between the circles, even after themask oxide is removed (inset in Fig. 7a). This could be dueto the different etching rates when the oxide mask was re-moved by hydrofluoric acid. A dark edge ring appeared onthe unimplanted area at a positive sample bias (Fig. 7a), thenthe ring became brighter with a decreasing sample bias (Figs.7b,c). The width was broadened toward the center of the cir-cle (Fig. 7d) and the ring disappeared at−0.15 V (Fig. 7e).The image contrast between the two areas was too low to im-age clearly at< −0.7 V (Fig. 7f). Since the active region of

b

a

Fig. 4a,b.Depletion layers formed by scanning with the50 nmwide rectan-gular electrode calculated by TSUPREM-4 in 2D (a), and one-dimensional(dC/dV)-V characteristic curve calculated at marked points ofa withMEDICI (b)

a device is often depleted even at zero bias, it is difficult tomeasure the impurity concentration within the depletion layerwith SCM. This can be overcome by varying the applied dc

Fig. 5. One-point, high-frequency (dC/dV)-V characteristic curves fromVs=3 V to −3 V at unimplanted area (B) to implanted area (A) (if the tipis considered as a the gate, the voltage is varied from accumulation to in-version as usual. Therefore the plus sign at the sample bias is the same asminus at the gate bias)

S418

Fig. 6a–f. Simulated capacitance im-ages after 2D calculation by MEDICIa Vs=+1.3 V, b +1.1 V, c +0.7 V,d +0.5 V, e +0.3 V, f −0.1 V

bias while controlling the depletion width. This method canshow the carrier distribution of a real device under the operat-ing voltage. Therefore the images in Fig. 7 are quite differentfrom the 2D SCM results on a cleaved sample surface. Thusthe present image inversion and edge effects are due to a true3D effect caused by varying the depletion width. The imagein Fig. 7 shows good agreement with the simulated resultswith only a shift of −0.2 V of the C-V dependence in theaccumulation region of the unimplanted area. This may becaused by the previously described deviations in (dC/dV)-Vfrom the ideal one. At highly doped implanted areas this ef-fect is negligible, but at unimplanted areas this effect givessimilar results to that of hole injection caused by applyinga positive bias to the sample. The image calculation can be

Fig. 7a–f. Capacitance images whichdepend on the sample biases withgrounded tipa Vs=+1.3 V and topog-raphy (inset), b +0.9 V, c +0.4 V,d +0.3 V, e −0.15 V, f −0.7 V (7×7µm2)

refined further by considering realistic probe-sample geome-tries [16].

In order to understand the recombination dynamics of thetrapped charges in theSiO2, we applied electrical stressesof −6 V to the tip for 60 s. After the stress, theC-V de-pendence was shifted toward the positive direction, resultingan inversion of thedC/dV image. Figure 8 shows the time-dependent SCM images after the electrical stress was applied.The amount of charge decreased with time, so did the shiftof the minimum voltage in (dC/dV)-V . If this trap is an in-terface trap, the recombination time should be less than1 s.However, the trap recombined after395 s. If the trap is a fixedone and is excited thermally and recombines with interfacestates, the retention time will be much longer than a few

S419

Fig. 8a–d. Time-dependent SCM images of trapped charges which wasstressed at−6 V for 60 s. The images with the scan rates ofa 60, b 112,c 218,d 395 s, respectively

hours. Therefore, other mechanisms such as recombinationwith mobile ionsNa+ or protons should play a major role inexplaining the present result.

3 Conclusion

Quasi-3D information of impurity profile was measured bythe position-dependent (dC/dV)-V. Inversion of the imagecontrast was observed with the applied sample bias. These re-sults agree well with a quasi-3D simulation. The applied biasis not only a function of the carrier density of the sample butalso of the circuit parameters of the sensor, which were stud-ied by simulation. Therefore one has to take this into accountwhen comparing the measured data with the calculated re-sults. The recombination of trapped charges was measured asa function of time. A theoretical model is needed to explainthe recombination mechanism.

Acknowledgements.This work was partially supported by Korean Scienceand Engineering Foundation through Science Research Center at YonseiUniversity and Korean Ministry of Science and Technology.

References

1. X. Liu, S. Goodwin-Johansson, J.D. Jacobson, M.A. Ray, G.E. McGuire:J. Vac. Sci. Technol. B12, 116 (1994)

2. W. Vandervorst, V. Privitera, V. Raineri, T. Clarysse, M. Pawlik: J. Vac.Sci. Technol. B12, 276 (1994)

3. P. De Wolf, J. Snauwaert, L. Hellemens, T. Clarysse, W. Vandervorst,M. D’Olieslaeger, D. Quaeyhaegens: J. Vac. Sci. Technol. A13, 1699(1995)

4. C. Shafai, D.J. Thomson, M. Simard-Normandin, G. Mattiussi, P.J.Scanlon: Appl. Phys. Lett.64, 342 (1994) ; P. De Wolf, J. Snauwaert,T. Clarysse, W. Vandervorst, L. Hellemens, Appl. Phys. Lett.66, 1530(1995)

5. P. De Wolf, T. Clarysse, W. Vandervorst, J. Snauwaert, L. Hellemens:J. Vac. Sci. Technol. B14, 380 (1996); J.N. Nxumalo, D.T. Shimizu,D.J. Thomson, J. Vac. Sci. Technol.14, 386 (1996)

6. M.B. Johnson, O. Albrektsen, R.M. Feenstra, H.W.M. Salemink: Appl.Phys. Lett.63, 2923 (1993)

7. P.M. Thibado, T.W. Mercer, S.Fu.T. Egami, N.J. Dinardo, D.A. Bon-nell: J. Vac. Sci. Technol. B14, 1607 (1996)

8. T. Teuschler, M. Hundhausen, R. Eckstein, L. Ley: J. Vac. Sci. Tech-nol. B 12, 2440 (1994)

9. Y. Kuk, R.S. Becker, P.J. Silverman, G.P. Kochanski: Phys. Rev. Lett.65, 456 (1990)

10. C.C. Williams, W.P. Hough, S.A. Rishton: Appl. Phys. Lett.55, 203(1989); C.C. Williams, J. Slinkman, W.P. Hough, H.K. Wickramas-inghe, Appl. Phys. Lett.55, 1662 (1989)

11. E.H. Nicollian, J.R. Brews: MOS Physics and Technology Chaps. 4,9(Wiley, New York 1982)

12. J.J. Kopanski, J.F. Marchiando, J.R. Lowney: J. Vac. Sci. Technol. B14, 242 (1996)

13. Y. Huang, C.C. Williams, M.A. Wendman: J. Vac. Sci. Technol. A14,1168 (1996)

14. A. Erickson, L. Sadwick, G. Neubauer, J. Kopanski, D. Adderton,M. Rogers: J. Electron. Mater.25, 301 (1996)

15. G. Neubauer, A. Erickson, C.C. Williams, J.J. Kopanski, M. Rogers,D. Adderton: J. Vac. Sci. Technol. B14, 426 (1996)

16. Y. Huang, C.C. Williams, H. Smith: J. Vac. Sci. Technol. B14, 433(1996)

17. Y. Huang, C.C. Williams, J. Slinkman: Appl. Phys. Lett.66, 344(1995)

18. R.C. Palmer, E.J. Denlinger, H. Kawamoto: RCA Rev.43, 194 (1982)19. Computer code TSUPREM-4 and MEDICI. Technology Modeling As-

sociates, Inc. 595, Lawrence Expressway, Sunnyvale, CA 94086-3922,version 6 and 2.1, respectively

20. See, for example, S.M. Sze: Physics of Semiconductor Devices,Chaps. 7, 13, 14, 2nd edn. (Wiley, New York 1981)