J O U R N A L O F M AT E R I A L S S C I E N C E : M AT E R I A L S I N E L E C T RO N I C S 1 5 ( 2 0 0 4 ) 4 5 5 ± 4 6 1

Redistribution of P atoms in oxidized P-implanted silicon during annealing KATSUHIRO YOKOTA*, MAKOTO AOKI, KAZUHIRO NAKAMURA Faculty of Engineering, Kansai University, Suita, Osaka 564-8680, Japan E-mail: [email protected] MASAYASU TANNJOU, SHIGEKI SAKAI Nissin Electric Co., Kujo-Tonoshiro, Minami-ku, Kyoto 601-8205, Japan KOUHEI SEKINE, MASANORI WATANABE Ion Engineering Research Institute Co., Hirakata, Osaka 573-01 Japan Phosphorus ions were implanted into silicon at energies of 20*200 keV and then a region ranging from the surface to about 0.26 nm was oxidized in wet-oxygen atmosphere at 1000  C for 90 min. Two groups, samples (capped samples) for which the SiO2 was not removed, and samples (removed samples) for which the SiO2 was removed, were prepared from the oxidized samples. Subsequently, the samples were annealed in inert gas at 1000  C for 30*180 min. The annealed capped samples prepared from the 20 keV ion-implanted silicon had approximately the same P atom distribution pro®les as the as-oxidized sample regardless of annealing time: the implanted P atoms had not diffused into silicon during the annealing. The annealed removed samples had P atom distribution pro®les approximated by a solution of the diffusion equation with constant-total-dopant. When the samples prepared from the 200 keV ionimplanted silicon were annealed, the distribution pro®les of carriers were approximated the same regardless of whether or not the SiO2 ®lm was removed from the sample. # 2004 Kluwer Academic Publishers

1. Introduction

Thermally grown silicon dioxide (SiO2 ) ®lms on silicon are important in the fabrication of silicon discrete devices and integrated circuits. In the silicon device fabrication process, impurity atoms are implanted in silicon either before or after the silicon is thermally oxidized. During oxidation and annealing in order to recover the silicon crystalline lattice destroyed by ionimplantation, implanted impurity atoms in silicon are redistributed between silicon and SiO2 depending on their diffusion coef®cients and their segregation coef®cients, which are the ratios of the equilibrium concentration of the respective impurities in the silicon to that in the SiO2 [1]. Impurity atoms such as phosphorus (P) atoms with a small diffusion coef®cient in SiO2 compared to that in silicon and a segregation coef®cient greater than unity pile up near the silicon surface [1]. Then, the oxidation process of silicon accompanies the generation of silicon interstitials at and near the Si SiO2 interface. The silicon interstitials at and near the Si SiO2 interface can extend extrinsic defects produced by ion-implantation to large dislocations and stacking faults [2]. The dislocations and stacking faults give rise to large stresses in regions surrounding them [3]. The impurity atoms in the silicon are redistributed under the in¯uence of the stresses due to the dislocations and stacking faults in annealing.

Extrinsic defects are produced by ion-implantation in an extended region depending on the dose and energy of implanted ions. The silicon interstitials are produced at and near the Si SiO2 interface by oxidation and serve to extend small extrinsic defects to larger dislocations and stacking faults. In this paper, we studied the in¯uence of oxidation on the redistribution of implanted P atoms by measuring the redistribution pro®les of P atoms implanted in silicon under various conditions and planview transmission electron microphotographs on thermally oxidized silicon and post-oxidation annealed silicon.

2. Experimental procedure

Phosphorus ions were implanted in B-doped 10±20 O cm (1 0 0) CZ-silicon wafers with ion-implantation conditions as shown in Table I. The doses of P ions implanted in silicon were adjusted to be the same as the concentrations of the P atoms at the projected ranges, varying with the ion energies. The implanted silicon was oxidized in wet-oxygen atmosphere at a temperature of 1000  C for 90 min. Two groups, consisting of samples (referred to as capped P-xx samples) for which the SiO2 was not removed, and samples (referred to as removed P-xx samples) for which the SiO2 was removed, were prepared from the oxidized samples. Here, P-xx are the

*Author to whom all correspondence should be addressed.

0957±4522

# 2004 Kluwer Academic Publishers

455

T A B L E I Ion-implantation conditions Sample name

Ion-energy (keV)

Implanted dose61015 (ions/cm 2 )

P-21 P-22 P-71 P-72 P-202

20 20 70 70 200

0.3 4.8 0.83 18.0 40.0

sample names in Table I. The samples belonging to the two groups were annealed in argon atmosphere at a temperature of 1000  C for 30±180 min. Ohmic contacts were secured on the surfaces of the samples by vacuum evaporation of aluminum and subsequently alloying in argon atmosphere. Carrier distribution pro®les were obtained through differential Hall and resistance measurements after the removal of subsequent silicon layers by anodic oxidation and oxidestripping. The silicon surfaces were anodically oxidized in an n-methylacetamide-based solution, and the thin anodic oxide layers grown on the silicon surfaces were removed by dipping in HF diluted with water. The distribution pro®les of P atoms were analyzed in a 60 mm diameter area using a secondary ion mass analyzer (SISM) system with magnetic prism. The primary ion beams were 8 keV O2 ‡ with a beam current of 300 nA. A 100 kV transmission electron microscope was used for studying defects produced in annealed samples.

3. Results and discussion 3.1. TEM observations

Fig. 1 shows a relationship between the distribution pro®les of P atoms implanted into silicon, which were calculated by the TRIM Monte Carlo code [4], and the calculated thickness …dc ˆ 0:26 mm† of silicon consumed to grow an SiO2 ®lm with a thickness of about 0.61 mm oxide ®lms. The distribution pro®les of implanted P atoms in silicon can be approximated by Gaussian distribution functions and the maximum concentration of the implanted impurity atoms is at the projected range. The regions corresponding to about four times the projected range of the implanted P ions in the P-21 and P-22 samples were consumed to grow the SiO2 ®lm having a thickness of 0.61 mm, the regions corresponding to about twice the projected range of the implanted P ions in the P-71 and P-72 samples were oxidized, and in the P202 sample, a region corresponding to the projected range of the implanted P ions only was consumed to grow the 0.61-mm-thick SiO2 ®lm. Fig. 2 shows plan-view transmission electron microphotographs (TEM) of as-oxidized P-implanted silicon (referred to as as-oxidized P-xx samples). The asoxidized P-21 sample had stacking faults with a very high concentration. On the other hand, the P-21 sample, which was not oxidized but was annealed in inert gas at a temperature of 1000  C for 90 min, had small precipitates with a low concentration as shown in Fig. 2(a). Thus, it emerges that the stacking faults in the silicon were produced because P-implanted silicon was oxidized. Ionimplantation provided defect nuclei, which could grow to large stacking faults when a large number of silicon 456

Figure 1 A relationship between the implanted P atom pro®les calculated by the TRIM Monte Carlo code [4] and the calculated thickness of silicon consumed to grow a SiO2 ®lm.

interstitials were fed to the defect nuclei. Oxidation created an excess concentration of silicon interstitials at and near the Si SiO2 interface [2]. Thus, in the asoxidized P-21 sample, defect nuclei produced by ionimplantation could grow into large stacking faults under oxidation. However, the stacking faults were not measured in the as-oxidized P-22 sample, in which many more P ions were implanted than in the P-21 sample. The as-oxidized P-22 sample had small precipitates with high densities. The P-22 sample was prepared by implanting P ions at a dose much greater than the minimum dose (661014 cm 2 for P ions) necessary to form amorphous silicon [6]. Continuous amorphous regions were easier to grow again into damage-free layers by solid-phase epitaxy, and the amorphous±crystalline interface could move toward the silicon surface at a velocity of about 40 nm min 1 even at 600  C [6]. In the P-22 sample, crystalline lattices destroyed by ion-implantation were recovered for brief periods as the oxidation temperature was increasing to 1000  C. Most defect nuclei as seeds of stacking faults are annealed out at the beginning of the oxidation, and as a result, silicon interstitials produced by the oxidation were precipitated in silicon. When a P-71 sample with the same peak concentration of implanted P atoms as the P-21 sample but at a position deeper than the P-21 sample was oxidized, the sizes of the stacking faults were signi®cantly reduced in comparison with the as-oxidized P-21 sample. The P-71 sample was prepared by implanting P ions with an implant dose greater than the dose of 1:361014 cm 2 necessary to form amorphous silicon while the implant dose of P ions in the P-21 sample was slightly greater than the threshold dose [6]. The P-71 sample was prepared by implanting P ions with an implant dose

Figure 2 Plan-view transmission electron microphotographs of the as-oxidized samples, the capped samples, and the removed samples.

greater than a dose of 1:361014 cm 2 necessary to form amorphous silicon while the implant dose of P ions in the P-21 sample was about equal to the threshold dose. Defect nuclei remaining in the implanted region after recovery of destroyed crystalline lattices on the P-71 sample are less than that on the P-21 sample. The defect nuclei grew into small stacking faults by combining with excess silicon interstitials produced during oxidation. The distribution pro®les of the silicon interstitials and

vacancies produced by P-ion-implantation can be approximated by a Gaussian distribution function with a projected range of 0:63 Rp and 0:92DRp [4], where Rp and DRp are the projected range and the projected straggle of P ions implanted in the silicon, respectively. Their concentrations are greater than the ion concentrations. The large initial interstitial and vacancy pro®les are nearly equal. Subtracting them reveals that the surface is vacancy-rich while interstitials are kicked 457

deeper into the silicon. Most silicon interstitials produced by the oxidation can react easily with excess silicon vacancies introduced by ion-implantation near the surfaces and the number of silicon interstitials decreases. Thus, the defect nuclei were restricted to grow into stacking faults. However, on the as-oxidized P-21 sample with many stacking faults, the silicon vacancies produced by the ion-implantation were not suf®cient to reduce signi®cantly the number of silicon interstitials produced by the oxidation because the dose and energy of P ions implanted into the silicon were lower than those of the P-71 sample: the numbers of vacancies produced by implantation of 20 and 70 keV P ‡ ions into silicon are simulated to be 271/ions and 768/ions, respectively [4], and the dose of P ‡ ions implanted into the silicon on the P-71 sample was about 2.7 times greater than that on the P-21 sample. The P-72 and P-202 samples, which were prepared by implanting P ions at doses much greater than the minimum dose necessary to form amorphous silicon [6], had typical dislocation networks. The separation between the dislocations increased on increasing the implant dose and energy of P ions. The P-72 and P-202 samples had silicon interstitials and vacancies produced by the ion-implantation in amounts higher than those on the P-22 sample because the implant doses of P ions in the P-72 and P-202 samples were four and eight times greater than that on the P-22 sample, respectively. The excess silicon interstitials at and near the Si SiO2 interface created by the oxidation in any samples were of the same concentrations regardless of the dose of P ions implanted in silicon because the samples were prepared by oxidizing silicon layers of the same thickness at the same rate [2]. The number of the silicon interstitials produced by the oxidation of the P-22 sample was more than that of the silicon vacancies produced by the ion-implantation and as a result, many precipitates were left in the silicon as described above. However, the number of silicon vacancies produced by the ionimplantation in the P-72 and P-202 samples was much more than that on the P-22 sample and most of the silicon interstitials produced by the oxidation disappeared in silicon by combining with the excess silicon vacancies produced by the ion-implantation. Silicon vacancies still remaining in silicon served to form dislocations at and near the interface of Si SiO2 . Plan-view transmission electron microphotographs (TEM) on the capped and removed samples annealed at 1000  C for 30 min are shown in Fig. 2. The capped P-21 and P-71 samples had stacking faults at densities smaller than these as-oxidized samples. The densities of stacking faults in the removed P-21 and P-71 samples became smaller than those in the capped P-21 and P-71 samples, respectively. In the capped and removed P-22 samples, the precipitates on the as-oxidized P-22 sample varied to stacking faults. The density of the stacking faults on the removed P-22 sample with free surface was smaller on the capped P-22 sample, on which a silicon surface was linked with a thick SiO2 ®lm. The separation between dislocations on the removed P-72 samples became larger than that on the capped P-71 samples. On the annealed P-202 samples, the dislocation networks varied to small precipitates as on the annealed P-21 samples: when 458

P-implanted silicon (P-21 sample) was annealed in argon gas at a temperature of 1000  C for 120 min equal to the time adding the annealing time to the oxidizing time, plan-view transmission electron microphotographs were approximately the same regardless of annealing time as shown in Fig. 2(a) and (b). Thus, on all of the capped samples, the size of the stacking faults and the separation between the dislocations were not extended in annealing in argon gas because new silicon interstitials were not supplied near the silicon surface during annealing. Any capped samples have an intermediate layer between silicon and SiO2 [8]. The development of the intermediate layer restricts signi®cantly the ability of the silicon surface as sinks of defects. Since the removed samples did not have the intermediate layer, defects in the removed samples can disappear at the silicon surface during annealing. Thus, the annealed removed samples had the defects with low density.

3.2. Redistribution of implanted P atoms and carrier distribution

Fig. 3 shows redistribution pro®les of implanted P atoms and carrier distribution pro®les on the capped and removed P-21 and P-22 samples annealed for 30 and 180 min. The redistribution pro®les of the P atoms on the capped P-21 sample were scarcely diffused into a deep region in silicon regardless of annealing time. The P atoms in the capped P-22 sample were diffused slightly in comparison with those in the capped P-21 sample: most of the P atoms in the capped P-21 sample were distributed in a shallow region ranging from the silicon surface to about 0.4 mm while those in the capped P-22 sample annealed for 30 min were distributed in a region ranging from the silicon surface to about 0.7 mm. On the other hand, the P atoms in the removed P-21 and P-22

Figure 3 Redistribution pro®les of implanted P atoms and carrier distribution pro®les in the annealed P-21 and P-22 samples.

samples were diffused remarkably into a deeper region in silicon. It may be considered that the SiO2 ®lms on the silicon surfaces served to retard the diffusion of the P atoms in silicon on annealing. Thermal oxidation of silicon shows SiO2 to be in a state of compression on the surface and silicon to be in a state of tension because the stress is attributed from the difference in thermal expansion between silicon and SiO2 [5]. The tensile stress in silicon results in an expansion of the lattice spacing and as a result, the diffusion of impurity atoms in silicon is enhanced [8]. This is the very opposite of the results in this experiment: on the capped P-21 and P-22 samples having stacking faults with a high density and precipitates with a high density, respectively, the diffusions of P atoms were signi®cantly retarded. The dislocations, precipitates, and stacking faults give rise to large stress in a region surrounding them [3, 9]. On the capped P-21 sample, the implanted P atoms were scarcely diffused in the annealing because the density of the stacking faults was very much higher than those on other samples. On the capped P-22 sample, the implanted P atoms can be diffused during annealing into a deeper region in silicon than on the capped P-21 sample because the density of the stacking faults was lower than that on the capped P-21 sample. On the annealed capped and removed samples, plots for carriers fell well on the distribution pro®le. Fig. 4 shows redistribution pro®les of implanted P atoms and carrier distribution pro®les on the capped and removed P-71 and P-72 samples annealed for 30 and 180 min. The capped P-71 sample annealed for 30 min had a redistribution pro®le spread slightly into silicon in comparison with the as-oxidized P-71 sample and when the capped P-71 sample was annealed for 180 min, P atoms were diffused into a deep region in silicon in comparison with the capped P-71 sample annealed for

30 min. However, the retardation of the P atom diffusion was small in comparison with that on the capped P-21 sample annealed for 180 min as shown in Table II. The SiO2 ®lms on the capped P-71 samples did not serve to retard the diffusion of the P atoms in silicon as described on the capped P-21 samples. The density of the stacking faults increased in order of the removed P-71, the capped P-71, and the removed and capped P-21 samples. This order corresponds the order of the retardation of the P atom diffusion in a series of samples. The as-oxidized P-72 samples had only the dislocation network as shown in Fig. 2. The dislocation network was measured even on the removed P-72 samples although the separation between the dislocations became larger in the removed P-72 sample in comparison with the as-oxidized and capped P-72 sample. The extreme retardation of the P atom diffusion on the capped P-21 samples was not measured on the capped P-72 samples. That is, we ®nd that the redistribution of the implanted P atoms in the capped samples became signi®cant on increasing the implant dose of P ions. Fig. 5 shows the redistribution pro®les of implanted P atoms and carrier distribution pro®les on the capped and removed P-202 samples annealed for 30 and 180 min. On the capped and removed P-202 samples annealed only for 30 min, the dislocation networks disappear and small precipitates only were measured with low densities. The samples had nothing to hinder the diffusion of the P atoms in silicon. Thus, the capped and removed P-202 samples had the same carrier distribution pro®les although the carrier distribution pro®le varied with annealing time. The samples in this experiment had no kink as obtained in high P-doped silicon in their redistribution pro®les. The kink occurs at a P atom concentration of 361019 cm 2 and P atoms is followed by a rapid diffusion in the deeper region in silicon because a large

Figure 4 Redistribution pro®les of implanted P atoms and carrier distribution pro®les in the annealed P-71 and P-72 samples.

Figure 5 Carrier distribution pro®les in the annealed P-202 samples.

459

number of single-charged acceptor-type vacancies V is generated by dissociation of the coupled P-vacancy pair (P ‡ V2 ) [10]. In order to inhibit the occurrence of the kink on high P-doped silicon, it is required to suppress the generation of the P ‡ V2 pair. On the samples in this experiment, vacancies produced by the ion-implantation and the substitutional P atoms have combined rapidly with silicon interstitials produced by the oxidation and disappeared in silicon as described above. Thus, the kink seems to occur in the distribution pro®le on the capped and removed samples and on the samples subjected subsequently in annealing. The distribution pro®les of P atoms on the as-oxidized samples were slightly different from a Gaussian distribution function and those on other annealed samples could be represented roughly by Gaussian distribution functions. In order to simplify, we approximated the distribution pro®le of P atoms on the asoxidized samples as the Gaussian distribution function. In the case that the initial implanted ions are given by a Gaussian distribution function and the solution of a limited-source diffusion is also Gaussian, the redistribution pro®le can be approximated by replacing the projected range DRp in a Gaussian distribution function 1=2 for P ions implanted into silicon by …DR2p ‡ 2Dt† , where D is the diffusion coef®cient of P atoms: C…x† ˆ fQ=…2pDR2p ‡ pDt†1=2 g expf …x Rp †2 =2=…DR2p ‡ 2Dt†g, where Q is the implant dose of P ions [11]. A characteristic depth at which the concentration of the implanted P atoms becomes 1=e ( ˆ 0.368) times the maximum concentration at the projected range is given 1=2 as 21=2 …DR2p ‡ 2Dt† . The diffusion coef®cients of P atoms during annealing can be calculated from an equation substituted the …DR2p ‡ 2Dox t† in the Gaussian distribution function in the oxidized samples by …DR2p ‡ 2Dox t ‡ 2Dan t†, where Dox t and 2Dan are the diffusion coef®cients of P atoms in oxidized and annealed samples, respectively, because the initial impurity atom distribution pro®le is approximated with a Gaussian distribution function and the solution of a limited-source diffusion is also Gaussian. Table II shows diffusion coef®cients of P atoms in the as-oxidized, capped, and removed samples. The diffusion coef®cient of P atoms on the as-oxidized P-21 samples was approximately the same value of 2610 14 cm2 s 1 in the literature [12]. However, on the as-oxidized, capped, and removed samples, the diffusion coef®cients of P atoms became larger on increasing either the implant dose or the energy of P ions. The samples having stacking faults with a high density have small diffusion coef®cients. The samples with dislocation networks had diffusion coef®cients larger than the samples with defects such as stacking faults and precipitates. A noteworthy result in Table II is the fact that a remarkable enhanced diffusion of the P atoms was observed on the removed samples annealed for a short time of 30 min but not on the capped samples. An enhancement of the diffusion of impurity atoms in silicon has occurred under the existence of excess silicon interstitials as seen at the beginning of annealing of boron-implanted silicon [13]. The enhanced diffusion of impurity atoms is associated to an interaction of silicon interstitials with impurity atoms through the kick-out impurity atoms. The silicon 460

T A B L E I I Diffusion coef®cients D of P atoms on the as-oxidized samples and the capped and removed samples Sample name

D for as-oxi (cm2 s 1 )

As-oxidized P-21 Capped P-21 Removed P-21

2:6610 Ð Ð

14

As-oxidized P-22 Capped P-22 Removed P-22

3:4610 Ð Ð

14

As-oxidized P-71 Capped P-71 Removed P-71

4:7610 Ð Ð

14

As-oxidized P-72 Capped P-72 Removed P-72

5:1610 Ð Ð

14

D for 30 min (cm2 s 1 )

D for 180 min (cm2 s 1 )

Ð 1:4610 4:6610

Ð 1:4610 2:7610

14

Ð 2:6610 4:7610

14

Ð 1:3610 3:1610

14

Ð 4:4610 1:1610

14

Ð 3:6610 8:7610 Ð 1:9610 8:1610 Ð 4:6610 1:5610

14 14

14 14

14 14

14 13

14

14

14

13

interstitials remain still in silicon at high densities immediately after the oxidation was stopped. On the removed samples, the silicon interstitials remaining in the silicon can diffuse to the surface and into a deeper region in silicon where the enhanced diffusion of the implanted P atoms was measured. On the other hand, the capped samples have still many defects such stacking faults. The stacking faults can extinguish silicon interstitials remaining in silicon immediately after the oxidation. Thus, the enhanced diffusion of P atoms in silicon subsequently annealed after oxidation was restricted on the capped samples and occurs only on the removed samples.

4. Conclusion

Silicon, into which P ions were implanted at energies of 20*200 keV, was oxidized in wet-oxygen atmosphere at 1000  C for 90 min and subsequently annealed in argon gas at 1000  C. Plan-view transmission electron microphotographs and P-atom and carrier-distribution-pro®les were measured on the samples. Stacking faults were formed in silicon implanted with small doses of P ions because silicon interstitials produced during oxidation were more than vacancies produced by ion-implantation and served to grow the defect nuclei. When P ions were implanted in silicon with a dose much more than the amount of silicon interstitials produced during oxidation, dislocation networks were measured on the silicon because the silicon interstitials produced by the oxidation disappear by reacting with the silicon vacancies produced by the ion implantation. The density of the stacking faults and the size of the dislocation networks made an in¯uence on the redistribution of the implanted P atoms during subsequent annealing: the diffusion coef®cients of P atoms in silicon became greater with decreasing the density of the stacking faults and increasing the size of the dislocation networks.

Acknowledgment

This research was ®nancially supported in part by the Kansai University Grant-in-Aid for Progress of Advanced Research in Graduate Course, 2003.

References 1.

2. 3. 4. 5. 6. 7. 8.

F. S H I M U R A ,

``Semiconductor Silicon Crystal Technology'' (Academic Publishers, Boston, 1989) Chapter 4. S . M . S Z E , ``VLSI Technology'' (Mcgraw, Auckland, 1983) Chapter 6. H . A L E X A N D E R and H . T E I C H L E R , in ``Handbook of Semiconductor Technology'', vol. 1, edited by K. A. Jackson and W. Schroter (Wiley-VCH, Weinheim, 2000). J . F. Z I E G L E R and J . P. B I E R S AC H , ``TRIM-91 User Manual, IBM-Research, 28-0'' (Yorktown, New York, 1991) p. 10598. S . M . S Z E , ``VLSI Technology'' (Mcgraw, Auckland, 1983) Chapter 4. B . E L - K A R E H , ``Fundamentals of Semiconductor Processing Technologies'' (Kluwer, Boston, 1995) Chapter 6. F. H E R M A N , I . P. B AT R A and R . V. K A S O W S K I , in Proceedings of the International Topical Conference on Physics of SiO2 and its Interface, Pergamon, Oxford, 1978, p. 333. D . S H AW , in ``Atomic Diffusion in Semiconductors'', edited by D. Shaw (Plenum, London, 1973).

9. 10.

S. M. SZE,

Chapter 1.

``VLSI Technology'' (Mcgraw, Auckland, 1983)

A . F. W. W I L LO U G H B Y , in ``Impurity Doping Processes in Silicon'', edited by F. F. Y. Wang (North-Holland, Amsterdam, 1981). 11. H . G L AW I S C H N I G , in ``Handbook of Ion Implantation Technology'', edited by J. F. Ziegler (North-Holland, Amsterdam, 1992). 12. Y. I C H I K AWA , Y. S A K I N A , H . TA N A K A , S . M AT S U M OT O and T. N I M I , J. Electrochem. Soc. 129 (1982) 644. 13. E . R I M I N I , ``Ion Implantation Basics to Device Fabrication'' (Kluwer, Boston, 1995) Chapter 7.

Received 8 September and accepted 18 November 2003

461