Issue



Evaluating the effects of internal gettering in epi Si


11/01/2003







Overview

A rigorous study of the efficiency of internal gettering under different conditions, comparing it to external and p+ gettering, has shown it to be effective in p+ substrates with backside polysilicon. This is true when the density of oxygen precipitates is high and their size small. However, internal gettering is not effective for a low density of oxygen precipitates when using either p+ substrates or polysilicon.

Deep levels in the band gap associated with metallic impurities present in semiconductors strongly affect the lifetime of the minority carriers. For silicon, third transition-metal atoms (e.g., Mo, Fe, Cu, Ni) are inevitably introduced during the wafer processing following the crystal growth step. The high diffusivity of these metals in silicon further contributes to their harmful presence. The concentration level of metal impurities that can be tolerated is mainly determined by device dimensions and is now generally 1 x 1010cm-3 or below. To achieve this contamination level in the active device region, gettering procedures are used that allow removal of metal atoms from some regions of wafers to other predetermined regions where their presence is not harmful [1].

Gettering techniques include internal gettering, high-boron-concentration activated gettering, and external gettering [2] such as backside polysilicon gettering, Al-backside gettering (Al-Si alloying), phosphorus-diffusion gettering, and gettering at nanocavities introduced by helium or hydrogen implantation [3]. The various physical mechanisms associated with these procedures lead to classifications as relaxation, segregation, or injection-induced gettering [4].


Figure 1. TEM image of oxygen precipitate-dislocation complex in the p/p+, high-oxygen, polysilicon on backside sample. Bar is 200nm.
Click here to enlarge image

null

Internal or intrinsic gettering involves oxygen precipitates, often associated with dislocations, as sinks for the metallic impurities [5–7]. The oxygen precipitates are intentionally created in a wafer (e.g., by an appropriate high, low, high annealing process). With backside polysilicon external gettering, the impurities are gettered by the grain boundaries of a polycrystalline silicon layer deposited on the wafer backside.

High-boron-concentration activated gettering applied to Fe is due to the electronic interactions between Fe interstitials and the dopant B ions controlled by the temperature-dependent Fermi level position [8, 9]. It can be applied to p/p+ epitaxial structures with the epi layer having a p concentration smaller than the boron concentration of the p+ substrate, which becomes the sink of the metal atoms. The use of p/p+ epitaxial structures in silicon technology has the advantages of having the active region (built in the epi layer) unaffected by crystal originated particles (COP) because any possible COP remains localized at the substrate surface, and also of reducing device latch-up [10].

Recent work

In recent work, we studied the efficiency of internal and polysilicon-external gettering, and substrate p+ doping gettering. We looked at these methods on two types of epitaxial silicon structures — 2µm thick p/p- and 4µm thick p/p+ epi (both 10Wcm, 200mm) grown in a single-wafer epi reactor, without HCl baking during deposition, on CZ silicon wafers having different properties. We assessed gettering efficiency through the observation of the nonradiative recombination properties of oxygen precipitates done by the analysis of the slope of the electron beam-induced current (EBIC) contrast vs. temperatures.

For each type of substrate, we used 11.5–12.5ppma "low oxygen" and 14–15ppma "high oxygen" contents. For some samples, polysilicon was deposited on the substrate backside by CVD at 680°C. The boron-doped p- and p+ 200mm (100) CZ substrates had a resistivity of 10–20Ωcm and 5–10mΩcm.

We subjected the epi structures to a 64MB DRAM thermal process budget in a horizontal furnace with inert or oxidizing ambient for a total thermal cycle of 23 hrs. Temperatures ranged from 750–1100°C to simulate actual wafer-processing sequences for film deposition, gate oxidation, field oxidation, and dopant drive-in.

We then prepared Al-sputtered Schottky diodes on the epi surface, which were used to carry out EBIC measurements in a scanning electron microscope as a function of temperature (T) from 73–300K. The beam energy used was in the range 30–40keV with beam currents between 0.5 and 5pA. The contrast (%) at the nonradiative recombination centers detected by EBIC was evaluated as

C = (Io - Id)/Io

where
Io = the EBIC signal collected far from recombination center, and
Id = the EBIC signal at the defect.

We also used transmission electron microscopy (TEM) in the bright-field mode to determine the structure of oxygen precipitates in cross-sectional specimens.

Specific evaluations

By TEM we observed that the crystallographic defects present in substrates were oxygen precipitates and dislocation loops that punch out from them (Fig. 1). The precipitates were octahedron-type in agreement with the annealing temperature used [11–12]. The densities of oxygen precipitates as measured by decorative etching were 106 and 107cm-3 for the p/p- low- and high-oxygen content samples, and 5 ¥ 108 and 1010cm-3 for the p/p+ low- and high-oxygen content samples.

In p/p- high-oxygen content samples, we compared the influence of backside polysilicon on the internal gettering efficiency.


Figure 2. For the p/p-, high-oxygen with backside polysilicon sample, a) EBIC images at 73K and b) 160K, and c) the behavior of the EBIC contrast (%) vs. temperature between 73 and 300K.
Click here to enlarge image

null

We found that the EBIC contrast at the oxygen precipitates and associated dislocation at 160K is lower than at 73K (Fig. 2), and it vanished completely at 300K (image not shown); the EBIC contrast vs. temperature curve has a negative slope (Fig. 2c). When no polysilicon was present and all the other parameters (p/p-, high oxygen) were constant, EBIC contrast was also visible at 300K where it was greater than at lower temperatures (Fig. 3).

Our interpretation of EBIC contrast curves is based on published models [13–15]. The EBIC detectability of the electrical activity at all temperatures including room temperature, as seen in the no-polysilicon sample (Fig. 3), indicates that in this sample the oxygen precipitates-dislocation complexes are contaminated with impurities having some associated deep level in the silicon band gap; this is supported by other published models [16, 17]. In our case, such impurity is expected to be Fe since Fe has been seen by DLTS to be the main metallic impurity present in samples prepared in the same way as the samples used here, with a density of ~5 x 1011cm-3 [18]. One published model suggests that the energy level of the contaminating impurity has to be at an energy ≥0.35eV from the valence band. Interstitial Fe with the energy level 0.39eV above the valence band edge matches this condition.

On the other hand, the negative slope of the EBIC contrast vs. temperature curve — with vanishing EBIC contrast as temperature increases to room temperature for the sample with polysilicon — indicates that the observed electrical activity at low temperatures is due to minority carrier recombination at the shallow levels intrinsic to the crystal defects. Very likely the shallow levels are those of dislocations punched out from the oxygen precipitates. This type of EBIC contrast is characteristic of "clean" dislocations.

With polysilicon on the backside, therefore, metal impurities do not contaminate oxygen precipitates-dislocation complexes compared to when polysilicon is absent. Thus, external gettering seems to prevail over internal gettering for p- substrates when both are present.

We examined the influence of substrate resistivity when polysilicon is also present on the backside. Looking at EBIC images at 300K for p/p+ samples with high oxygen content and polysilicon, we detected EBIC contrast at the oxygen precipitates-dislocation complexes (Fig. 4a). Although with a smaller value, the contrast was clearly detected even at a lower temperature (Fig. 4b). The nonradiative recombination area at the defects is smaller in this sample because the oxygen precipitates in p+ silicon are smaller.

Using the same models, these results suggest that in the p/p+ samples with high oxygen content, Fe contamination of the oxygen precipitates-dislocation complexes takes place so that internal gettering of Fe seems to be effective, despite the presence of polysilicon on the backside and the high doping level of the p+ substrate that can both cause Fe gettering.

The effect of p+ doping on the gettering of Fe has been discussed by some authors using grown p+ substrates or boron-implanted float zone silicon, for which it seems to be the prevailing gettering mechanism for high p doping due to enhanced solubility of Fe [19]. The action of Fe gettering in a p/p+ structure takes place by two concomitant mechanisms:

  • The formation of Fei+-B pairs (Fei+ is the positive charged state of the interstitial Fe). In p-type silicon, Fei+ becomes positively charged when the Fermi level is below the deep level state of Fei+, which is at 0.39eV above the valence band edge. The Fermi-level controlled fraction of Fei+/Feio (Feio is Fe in the neutral state) is a function of p-doping (boron) and temperature, and it increases for increasing boron and decreasing temperature. A Coulomb attraction arises between Fei+ and B- with the formation of Fei+-B- pairs.
  • The redistribution of Fe between the p and p+ regions. The higher number of negative ions in the p+ region exerts a Coulomb attraction on the Fei+ ions that then migrate from the p region to the p+ substrate. The ratio of Fei+ in the highly boron-doped region to that in the less-doped region increases exponentially with decreasing temperature. Gettering of Fe in the p+ regions is due to the simultaneous action of both Fermi-level induced Fe redistribution and Fermi-level controlled Fe-B pairing.

The theory of the Fermi level dependence of Fe gettering by p+ regions predicts that such gettering is possible only for temperatures <600°C, since above this temperature the electronic effects discussed previously are too weak. It becomes extremely effective at temperatures <400°C.

Our results show that internal gettering is effective in p/p+ structures based on CZ substrates when the level of oxygen is high (14–15ppma). This can be explained by considering that precipitation of Fe, at oxygen precipitates, on cooling from the annealing temperature, can occur at temperatures >400°C. It has been reported that efficient internal gettering at precipitates can take place at T>520°C. On the other hand, at temperatures <400°C (for example, when p+ gettering is also operative) it is likely that a part of Fe atoms are trapped at oxygen precipitates instead of forming Fe-B pairs.

According to one report for p = 1 x 1019cm-3 (resistivity of 10mΩcm), the solubility for Fe is 5 x 1011cm-3 for temperatures between 650 and 700°C, so that Fe precipitation can start at temperatures lower than the expected onset of p+ gettering.


Figure 3. For the p/p-, high-oxygen without backside polysilicon sample, the behavior of EBIC contrast (%) vs. temperature between 73 and 300K.
Click here to enlarge image

null

The high density of oxygen precipitates in the p+ substrate may favor internal gettering by increasing the number of trapping sinks, as reported in other work [20]. Such competition between the two different sinks (precipitates and Fe-B pairs) does not exist in float zone silicon where oxygen precipitates are not expected and p+ gettering may be the only operative gettering mechanism.

Furthermore, with p/p+, high-oxygen content samples with backside polysilicon, internal gettering is operative even in the presence of the backside polysilicon that was shown to prevail over internal gettering for p- substrates. Internal gettering is believed to be competitive with external gettering in this case due to the higher density of oxygen precipitates in p+ substrates with respect to the p- substrates, which makes them much closer to each other than in the case of smaller density p- substrates. This property can increase the trapping probability of Fe atoms at precipitates before they reach the backside polysilicon.


Figure 4. a) EBIC image at 300K and b) EBIC contrast (%) vs. temperature behavior between 73 and 300K for p/p+ samples with high oxygen content and backside polysilicon.
Click here to enlarge image

null

The EBIC data for the p/p-, high-oxygen backside polysilicon samples (see Fig. 2) show that in this case internal gettering does not work. Very likely trapping of Fe at oxygen precipitates is negligible due to their low density. Since the substrate resistivity is high, gettering should only be due to the backside polysilicon.

For low oxygen content and backside polysilicon, both p/p+ and p/p- samples exhibit similar EBIC behavior — EBIC contrast curves with negative slope similar to Fig. 2c with no detectable EBIC contrast at high temperatures. This indicates negligible internal gettering and dominance of external gettering for the p/p- samples. For the p/p+ case, our results do not allow us to say whether p+ gettering or external gettering prevails. It is possible that external gettering at polysilicon prevails because diffusion of Fe to the back side may occur at T>400°C, the temperature at which p+ gettering starts to be really effective and therefore earlier than p+ gettering.

What works

We found that with p/p+ epi structures based on CZ substrates with a high density of oxygen precipitates, a condition obtained with high oxygen content using our thermal treatment, p+ gettering seems to be less important than internal and external gettering, because it is expected to effectively take place only at T<400°C, whereas the other two mechanisms may occur even at higher temperatures. We found evidence that internal gettering was achieved even in the presence of external gettering from backside polysilicon and is believed to be due to the high density of small oxygen precipitates with associated dislocations. It is not possible to say to what percentage the different gettering mechanisms contribute to trapping of the Fe atoms.

For p/p+ samples with low density of oxygen precipitates, a condition obtained with low oxygen content, internal gettering was not detected. In such a case, our techniques do not allow us to establish to what extent external gettering and p+ gettering are operative. Qualitative speculations based on the temperature dependence of the two mechanisms would suggest that external gettering prevails.

For p/p- samples, internal gettering has not been effective when backside polysilicon is applied. Thus, only external gettering should be active, as p+ gettering is not expected to work. Internal gettering worked, however, when no polysilicon was applied.

Cesare Frigeri, CNR-IMEM Institute, Parma, Italy
Gabriella Borionetti, Pierangelo Godio, MEMC Electronic Materials SpA, Novara, Italy

Acknowledgments

This work was supported by PN Microelettronica.

References

    S.M. Myers, et al., J. Appl. Phys. 88, 3795, 2000.
  1. N. Gay, S. Martinuzzi, Solid St. Phenom. 57–58, 115, 1997.
  2. J.S. Williams, et al., Appl. Phys. Lett. 75, 2424, 1999.
  3. W. Schröter, et al., Materials Science and Technology 4, 540, 1992.
  4. H. Hieslmair, et al., Appl. Phys. Lett. 72, 1460, 1998.
  5. Z. Laczcik, et al., Solid St. Phenom. 47–48, 177, 1996.
  6. D. Gilles, et al., Phys. Rev. Lett. 64, 196, 1990.
  7. J.L. Benton, et al., J. Appl. Phys. 80, 3275, 1996.
  8. P.A. Stolk, et al., Appl. Phys. Lett. 68, 51, 1996.
  9. S.A. McHugo, et al., Appl. Phys. Lett. 73, 1424, 1998.
  10. K. Sueoka, et al., Appl. Phys. Lett. 65, 1686, 1994.
  11. H. Fujimori, J. Electrochem. Soc. 144, 3180, 1997.
  12. M. Kittler, W. Seifert, Mater. Sci. Eng. B24, 78, 1994.
  13. M. Kittler, W. Seifert, Mater. Sci. Eng. B42, 8, 1996.
  14. V. Kveder, et al., Phys. Rev. B. 63, 115208, 2001.
  15. P.R. Wilshaw, et al., Point and Extended Defects in Semiconductors, Part 2, G. Benedek, et al., Eds. Plenum Pu. Co., New York, p. 243, 1989.
  16. P.R. Wilshaw, T.S. Fell, J. Electrochem. Soc. 142, 4298, 1995.
  17. G. Borionetti, P. Godio, unpublished results, 2000.
  18. M. Aoki, et al., Appl. Phys. Lett. 66, 2709, 1995.
  19. S.A. McHugo, Appl. Phys. Lett. 66, 2840, 1995.

Cesare Frigeri received his degree in solid state physics at the School of Specialization in Materials Science at Bologna U. He is senior scientist at CNR-IMEM Institute, Parco area delle Scienze 37/A, Fontanini, 43010 Parma, Italy; fax 39/0521-269206, e-mail [email protected].

Gabriella Borionetti received her degree in solid state physics at the School of Specialization in Materials Science at Milano U. She is an MEMC Associate Fellow and the manager of MEMC's R&D Lab in Europe.

Pierangelo Godio received his degree from ITIS, Turin. He is a senior engineer specialist at MEMC Europe R&D Labs.