Oxygen Precipitation in Highly Doped Silicon Substrates

This paper reviews oxygen precipitation in heavily arsenic-doped silicon. A wide arsenic concentration range is explored, from 3 × 1018 cm−3 to 4 × 1019 cm−3. A precipitation retardation effect is observed in the arsenic doped samples when the dopant concentration is higher than 1.7 × 1019 cm−3 compared with lightly doped samples having the same initial oxygen content and grown under similar conditions. The oxygen precipitate density in the heavily arsenic-doped samples is uniform along the wafer radius, with no rings or cores, contrary to what is commonly observed in lightly doped samples grown with similar V/G values. This finding is discussed by considering the role played by vacancies in the formation of oxygen precipitates and the impact of the arsenic concentration on the equilibrium concentration of point defects in silicon, deduced from the experimentally observed voids revealed as light-scattering surface defects in polished wafers. © The Author(s) 2019. Published by ECS. This is an open access article distributed under the terms of the Creative Commons Attribution 4.0 License (CC BY, http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse of the work in any medium, provided the original work is properly cited. [DOI: 10.1149/2.0081901jss]

Heavily doped silicon substrates are widely used for the fabrication of discrete devices for power applications. The basic wafer template is generally composed of an epitaxial layer over a heavily doped silicon substrate whose dopant type and resistivity vary depending on the final device characteristics. In either case of p-and n-type substrates, the demand is more and more oriented toward very low resistivity ranges, such as 0.5-5 m · cm, in order to minimize the substrate contribution to the transistor drain-source resistance. Furthermore, the continuous effort to reduce manufacturing costs is moving discrete device fabrication from 100-150 mm in wafer diameter to 200 mm and even to 300 mm. Often a vertical device structure is chosen, where the electric current flows through the entire wafer thickness, therefore the properties of the wafer bulk affect the device performance and the bulk defectivity level has to be understood and controlled carefully. In this respect, oxygen precipitation is of utmost importance. The aim of this paper is to present a study of the behavior of oxygen precipitation in heavily doped silicon substrates, with particular focus on heavy arsenic doping.

Oxygen Concentration and Incorporation
Oxygen incorporation is of course the first factor to be considered when dealing with oxygen precipitation in silicon. Crystal growers have a number of process parameters that they can tune in order to achieve the desired oxygen concentration, such as crucible rotation rate, magnetic field intensity and so on. However, when the melt is heavily doped with volatile elements, such as arsenic, antimony or phosphorus, the oxygen concentration shows a strong dependence on the dopant concentration itself and it becomes very difficult to decouple these two parameters and achieve any desired combination of oxygen and dopant concentration. Specifically, a decrease in oxygen concentration with increasing dopant concentration is typically observed. This phenomenon was first reported in the literature in the case of antimony doping, 1 and the responsible mechanism was concluded to be an enhanced oxygen evaporation from the melt. 2 The crucible dissolution rate was found to be enhanced, too. 1 The reasons of the increased oxygen evaporation are still under debate. A direct evaporation of antimony oxide species has been proposed by some authors 3 but excluded by others on the basis of thermodynamic calculations and of the analysis of evaporated deposits. 4,2 Some authors 4 have proposed an accelerated SiO evaporation caused by the simultaneous evaporation of elemental antimony, similar to the steam distillation of a substance of low volatility and to diffusion pumping. More recently, the enhanced SiO evaporation was explained by the reaction of Sb with SiO in the gas phase above the melt: the Sb x O y formation lowers the SiO partial pressure in the gas boundary layer, favoring its evaporation from the melt. 5 A deeper discussion is outside the scope of this paper. The same phenomenon is observed also in the case of arsenic and phosphorus doping: the oxygen concentration decreases with increasing dopant concentration, if all other process conditions are kept the same. An example relative to arsenic doping is shown in Figure 1. The effect of the arsenic concentration on the oxygen content becomes appreciable below 7 m · cm (corresponding to a concentration of As of 8 × 10 18 at/cm 3 ).
The mechanism responsible for the reduced oxygen concentration in the case of phosphorus and arsenic doping is likely the same as that proposed for antimony-doping: an enhancement in oxygen evaporation. The evaporating species could be either SiO only, as suggested for the case of Sb, 2,4,5 or a mixture of SiO and volatile P and As oxides, respectively. Evidence in support of this mechanism is the formation of heavier oxide deposits on cold puller parts during the crystal growth process, as well as a faster crucible erosion, compared to undoped melts. In particular, in the case of arsenic doping, the crucible erosion rate at the melt line was measured and found to be twice as high compared to undoped melts: 22.4 μm/hour in the case of arsenic-doped melts vs 10.5 μm/hour in the case of undoped melts under the same process conditions. It is interesting to observe that in Figure 1. Oxygen concentration in arsenic-doped silicon samples measured by Gas Fusion Analysis as a function of the sample resistivity. All data points refer to crystals grown under the same conditions, in particular the melt quantity in the crucible is the same (same melt aspect ratio). For reference, the average oxygen concentration of lightly doped silicon grown in the same conditions is 17 ppma. The GFA was calibrated against FTIR using lightly doped samples. The FTIR calibration coefficient used is 2.45 × 10 17 atoms/cm 3 (ASTM F 121-83). the case of heavy boron doping (boron is non-volatile), no appreciable change in oxygen concentration is generally observed. 6,7 The strong coupling between dopant concentration and oxygen concentration in the case of volatile dopants poses a challenge when trying to assess the impact of the dopant concentration on the oxygen precipitation, because it is very difficult to obtain a good set of samples where only the dopant concentration varies. In this case, the use of undoped or lightly doped samples with the same oxygen content and grown under the same conditions is necessary.

Oxygen and Oxygen Precipitation Measurement in Heavily Doped Silicon
Oxygen measurement in silicon wafers down to 1 × 10 16 at/cm 3 is usually performed by Fourier Transform Infrared Spectroscopy (FTIR). 8 However, the method cannot be applied when the dopant concentration exceeds 1 × 10 15 -1 × 10 16 at/cm 3 due to excessive free carrier absorption. 8 For heavily doped silicon, the most used methods are Gas Fusion Analysis (GFA) and Secondary Ion Mass Spectroscopy (SIMS). GFA is based on the reaction of oxygen with the carbon of the graphite crucible after having melted the silicon sample under test. GFA is needs a significant amount of silicon for each analysis, a few mm 3 . The detection limit is approximately 1 × 10 16 at/cm 3 for a standard GFA. 9 For SIMS, the sample is bombarded by a primary ion beam under UHV conditions which are necessary to improve the detection limit; for oxygen this is around 1 × 10 17 at/cm 3 . SIMS can best measure the oxygen concentration on polished wafers and can also perform a depth profiling of the oxygen concentration. 10 Contrary to the FTIR technique which is fast and not destructive, GFA and SIMS are both destructive methods and require much longer measurement time, especially SIMS. GFA is often used in the industry for in-line product certification: the tool has affordable costs and reasonable measurement cycle time, of the order of one day including the sample preparation. SIMS is typically used when GFA is not possible, i.e. for the measurement of thin wafers, or for depth profiling. Both GFA and SIMS must be calibrated, this is usually accomplished using lightly doped silicon samples certified by FTIR. Finally, it must be reminded that both GFA and SIMS measure the total oxygen concentration of the sample, without the possibility to distinguish between the interstitial form and other oxygen species.
The most common method to determine the volume density of oxygen precipitates in heavily doped silicon is preferential etching and defect counting under a microscope. Usually, the sample is cleaved to expose a <111> plane and it is etched with a preferential etch mixture that creates a pit on the surface of the sample in correspondence to each precipitate. The etching depth is measured and the defect volume density is calculated by dividing the total number of defects by the total volume sampled. Another method used to characterize oxygen precipitates in heavily doped silicon is transmission electron microscopy (TEM), but this method is mainly useful to characterize the defect morphology rather then the density, due to the small volume measured. It is also difficult to find defects with TEM when their density is less than 10 10 cm −3 . Light scattering tomography (LST) can be also used when the investigated material is transparent to the wavelength of the laser. The scattered light is analyzed and the size, number and distribution of the light scattering defects can be calculated. Modern LST setups can detect defects larger than about 20 nm.

Oxygen Precipitation Behavior in Heavily Doped Silicon
Heavily doped N+ silicon wafers generally exhibit very low oxygen precipitation when analyzed after a typical nuclei stabilization followed by growth thermal cycle, such as the standard precipitation treatment consisting of 4 hours at 800 • C followed by 16 hour at 1000 • C. 3 An example of this behavior in the case of heavily arsenicdoped silicon is shown in Figure 2, where the oxygen precipitate density is plotted as a function of the sample resistivity. An abrupt From the plot in Figure 2 it cannot be immediately concluded that the drop in oxygen precipitation is due to the increased dopant concentration, since also the oxygen content decreases with increasing dopant concentration, as shown previously in Figure 1, although not so sharply. To some extent, the decrease in oxygen content can be corrected by tuning the crystal pulling process, but not completely.
In order to fully decouple the two factors, a comparison with lightly doped samples was done. For this purpose, a group of lightly boron doped samples grown under similar conditions was chosen. Attention was put to compare samples not only with the same oxygen content, but also with similar growth thermal histories, to reduce as much as possible the influence of other factors. More details on this study can be found in Ref. 11. The results are shown in Figure 3, where the oxygen precipitate density is plotted as a function of the oxygen content, comparing lightly boron doped and heavily arsenic doped samples. The reduced oxygen precipitation in the case of arsenic doped samples is confirmed. For completeness, it should be mentioned that the possibility of an incomplete detection of precipitates does exist, in principle, and one could also explain the reduced oxygen precipitation observed in the heavily arsenic-doped samples as a result of their size becoming smaller than the detection limit of the technique, rather than their volume density becoming lower. Although this possibility  cannot be ruled out completely, to the authors' knowledge no evidence is available that oxygen precipitates become smaller with increasing arsenic concentration, therefore this work will assume that the detected defects are well representative of the total number of defects for both sample groups.
Regarding the radial distribution of oxygen precipitates, the samples did not show any presence of rings or cores: defects were distributed quite uniformly on the whole sample radius, with the exclusion of few millimeters near the edge. A quantitative measurement is reported only for the center, where also the oxygen concentration was measured.

Impact of Heavy Donor Doping On Point Defects
The precipitation of oxygen is strongly affected by the concentration of vacancies. For this reason, before moving to the discussion of the oxygen precipitation in heavily doped silicon, we will review the impact of donors on point defects in silicon. [12][13][14][15] Experimentally it was found that increasing the concentration of donors at first shifts the silicon growth regime toward the vacancyrich regime, until a critical dopant concentration is reached, and then toward the interstitial-rich regime for even higher concentrations. 15 In the case of antimony, the impact on point defect concentration was observed already at a concentration of the order of 10 17 cm −3 , 16,17 while for arsenic and phosphorus a higher concentration is required, of the order of 10 18 cm −3 . 15 At these concentrations, a downward shift in the critical V/G (ratio between pull speed and axial thermal gradient) was measured. The maximum of the vacancy concentration (considering the number of voids counted by wafer surface laser inspection as indicator) is achieved at a critical concentration of 1.7 × 10 19 cm −3 (3.9 m · cm) in the case of arsenic, and at approximately 2.9 × 10 19 cm −3 (2.4 m · cm) for phosphorus. 15 The Figure 4 illustrates the impact of arsenic on the formation of voids detected as Crystal Originated Particles (COP) on the surface of 200 mm diameter polished silicon wafers by an automatic laser inspection tool. A similar trend was found also for phosphorus. 15 The sudden drop is explained by a quick increase in the concentration of interstitial arsenic As i and a corresponding sharp transition from the vacancy growth mode (with the dominant VAs point defects) to the interstitial mode (with the dominant As i point defects). According to Ref. 18, the concentration of VAs (at the interface) increases in proportion to the doping level [As], while the concentration of As i increases much faster, roughly as [As] 3 -due to the assumed charge difference between substitutional As (positive) and interstitial state As i (negative). For this reason, As i will dominate at sufficiently high doping level. A complementary view is shown in Figure 5, where the diameter of the vacancy-rich region is plotted as a function of the resistivity for several arsenic-doped crystals. The diameter of the vacancy-rich region is determined by visual inspection under collimated light of samples subjected to defect decoration (by copper) and preferential etching. At the highest resistivity explored of 25 m · cm (corresponding to an arsenic concentration [As] = 8.6 × 10 17 cm −3 ), the silicon crystal has a small vacancy core of approximately 50 mm diameter surrounded by an annulus of interstitial-type dislocation loops. This behavior is similar to that of undoped or lightly doped silicon grown under the same V/G conditions, confirming that at this low concentration arsenic does not have an important impact on point defects concentration. In the range between 4 and 12 m · cm ([As] = 1.6 × 10 19 to 3.3 × 10 18 cm −3 ) the crystal is fully vacancy-rich. From 4 to 2.5 m · cm ([As] = 2.7 × 10 19 cm −3 ) the density of the defects drops quite abruptly to zero while remaining substantially radially uniform, apart a few cases where the density at the crystal edge is already zero while some rare defects can still be observed in the central area. No rings or cores are observed.
Moving further down in resistivity, another feature becomes visible by copper decoration in samples with resistivity <1.8 m · cm ([As] > 3.9 × 10 19 cm −3 ), i.e. the presence of a pattern that resembles closely the dopant striations as far as the macroscopic pattern is concerned, but that is formed by tiny individual copper decorated points instead of being continuous lines, as shown in Figure 6. A possibility is that it consists of tiny arsenic clusters or precipitates. In the literature, the formation of arsenic clusters/precipitates has been studied mainly in silicon wafers implanted with arsenic at levels higher than the solubility limit, up to 4 × 10 21 cm −3 , where they are considered  responsible for a fraction of electrically inactive arsenic. [19][20][21][22] In the case of as-grown silicon, no direct evidence of arsenic clusters was found by examining samples with TEM, 23 however their presence can be considered possible on the basis of sectional X-ray topography results in silicon samples doped with arsenic to a concentration of 4 × 10 19 cm-3. 24 In our case, arsenic precipitates may be formed by agglomeration of minority interstitial arsenic atoms As i that coexists with the majority substitutional species As s . The precipitating species -As i -is incorporated at the interface where it is in equilibrium with As s . Upon cooling, the As i to As s ratio becomes strongly above the equilibrium value and accordingly the As i solid solution may become supersaturated (while the substitutional component is undersaturated).
The presence of arsenic precipitates at very high arsenic concentrations would support the hypothesis of a switch from vacancy-richto interstitial-rich growth mode of the silicon at such high donor concentration.
To verify that the disappearance of voids takes place without rings or bands, a longitudinal slab was cut from a crystal in the resistivity range of 2.7 to 6.2 m · cm ([As] from 2.5 × 10 19 cm −3 to 9.2 × 10 18 cm −3 .). The longitudinal sample was decorated with copper and submitted to preferential etching. The defect pattern is shown in Figure  7 and confirms that the defects disappear gradually with increasing arsenic concentration.

Reduced oxygen precipitation in heavily donor doped silicon.-
The results presented in the previous sections have shown a reduced oxygen precipitation in heavily arsenic doped silicon below 4 m · cm compared to lightly boron doped silicon with the same oxygen content and growth conditions. For resistivities higher than 4 m · cm, no appreciable difference is observed between heavily arsenic doped and lightly boron doped silicon. These results provide an experimental evidence to the general belief that arsenic doping reduces oxygen precipitation and clarify more precisely the concentration range where this effect is appreciable.
The same threshold in arsenic concentration above which a reduction in oxygen precipitation is observed -1.7 × 10 19 at/cm 3coincides with the arsenic concentration above which a rapid drop in COP formation takes place. A reduced vacancy availability, caused by the switch from vacancy-dominated to interstitial-dominated growth mode at high arsenic concentration, can explain both the reduction of oxygen precipitation and the reduction of COP formation.
Not much comparison can be done with previously published experiments, as the available literature is very limited in this field. Zhao et al. 25 have studied one arsenic-doped sample with a resistivity of 5 m · cm and an oxygen content of 1.6 × 10 18 cm −3 in comparison to one lightly boron doped sample with an oxygen content of 1.2 × 10 18 cm −3 (calibration factor for oxygen was 3.14 × 10 17 cm −2 ). They observed a retarded oxygen precipitation in the arsenic doped sample after a ramped anneal from 650 • C to 1050 • C with different ramping rates: 0.5, 1 and 2 K · min −1 in comparison to the boron doped sample subjected to the same anneal. Before the ramped anneal, the samples were pre-annealed at 1150 • C for 2h and then kept at 650 • C for 1 h. The effectiveness of the permanence at 650 • C in promoting oxygen precipitation in arsenic doped samples is in agreement with other published results. 11,26 A very systematic investigation was published by Sugimura et al., 26 where the impact of three different arsenic concentrations: 1.7 × 10 15 at/cm 3 (lightly doped, 3 · cm), 1.7 × 10 18 at/cm 3 (18 m · cm) and 1.6 × 10 19 at/cm 3 (3.9 m · cm), on oxygen precipitation was studied. In their work, an increased arsenic concentration was found to cause a retardation in oxygen precipitation and a reduced precipitate density after a nucleation anneal (temperature: 600 • C -750 • C, Time: 1 -96 hours) applied after an epi dissolution treatment and followed by a growth anneal of 16 hours at 1000 • C. In this work, the oxygen concentration of the arsenic-doped samples was not reported, therefore it remains uncertain whether the observed differences in oxygen precipitation behavior are solely due the different arsenic concentration or partially also to a different oxygen concentration between the samples. Interestingly, the oxygen diffusivity was also studied by SIMS oxygen depth profiling after an out-diffusion treatment in the temperature range of 700 • C-1050 • C without finding any impact of the arsenic concentration on the oxygen diffusion constant. 26 This result would rule out a reduction in oxygen diffusivity as a reason for the reduced oxygen precipitation in heavily arsenic-doped silicon.
The impact of vacancies on oxygen precipitation in heavily doped silicon was studied recently 27,28 by means of high temperature rapid thermal anneal (RTA)-induced vacancy injection. Arsenic doped samples with a resistivity of 3 m · cm ([As] = 2.2 × 10 19 cm −3 ) and antimony doped samples with a resistivity of 20 m · cm ([Sb] = 1.3 × 10 18 cm −3 ) were studied. Comparing the oxygen precipitation in samples with and without a prior RTA treatment at 1250 • C for 60 s, the authors observed an enhanced oxygen precipitation in samples with RTA compared to those without RTA, similar to what is normally observed in lightly doped silicon. For the arsenic doped samples, the enhancement was found only if the subsequent precipitation cycle is at 800 • C (32h) + 1000 • C (32 h), while no precipitates are observed for a cycle at 900 • C (32 h) or at 1000 • C (32 h). In the case of antimony doped samples, all thermal cycles showed an RTAinduced precipitation enhancement. According to the authors, in the antimony doped samples, less RTA generated vacancies are trapped by the dopant atoms and more vacancies remain available to reduce the critical size of the oxide precipitates compared to the arsenicdoped samples. 27 No comparison with lightly doped silicon was made. This result gives an additional evidence of the effectiveness of vacancies as oxygen precipitation enhancers also in heavily arsenic-doped silicon.
On the basis of these results, the reduced oxygen precipitation observed in heavily arsenic doped silicon above a threshold concentration of about 1.7 × 10 19 at/cm 3 can be reasonably attributed to the reduction in incorporated vacancies observed in this dopant concentration range, rather than to a reduced oxygen diffusivity. The fairly flat precipitate density around 10 10 cm −3 observed in arsenic doped samples above this resistivity is discussed in the next section.
Why the oxygen precipitate density is often of the same "magic" value.-In lightly doped silicon, the oxygen precipitation induced by a standard annealing cycle 800 + 1000 • C is controlled by the vacancies. The vacancies can be introduced into a wafer by Rapid Thermal Annealing (RTA) in a concentration C V that is higher at a higher RTA temperature. 29 The precipitate density N p is an increasing function of C V that seems to saturate 30 at a level of about 10 10 cm −3 . As-grown crystals are often composed of vacancy and self-interstitial regions. In the vacancy-dominated regions, the incorporated point defects are vacancies; they aggregate into voids that consume most of the vacancies, and yet some vacancies remain as residual point defects thought to be VO 2 31 -vacancies trapped by oxygen dimers. The precipitate density N p is low in the interstitial regions but high (typically 10 10 cm −3 ) in the vacancy regions, due to the residual vacancies. In our As-doped crystals the precipitate density was found to be the same as in lightly doped vacancy-type crystals: around 10 10 cm −3 up to some high As concentration; above this threshold, the density is however strongly reduced.
The existence of a "magic" density value, ≈ 10 10 cm −3 , is further supported by the data on oxygen precipitation in boron doped silicon. 32,33 The boron impurity is known 34 to induce, at fixed growth conditions, a transition from vacancy to interstitial growth mode; this effect is manifested, in particular as a shrinkage of the OSF ring 35 that is located within the vacancy region, close to the boundary between a vacancy and an interstitial region. In moderately doped crystals, when the OSF ring is still present (and thus the crystal still consists of an inner vacancy region and a peripheral interstitial region), the precipitate density was found to be 10 10 cm −3 -uniformly through a wafer including the interstitial region (see Fig. 4b in Ref. 32). In the interstitial region, the self-interstitials (I) are replaced, upon lowering T, with boron interstitials (B i ), by the kick-out reaction I + B s → B i . A high precipitate density within the I-region can be attributed either to prevention of vacancy recombination with I (due to a conversion of I into B i ) or to the mobile B i species themselves: small boron atoms can be incorporated into the oxide precipitate nuclei. In heavily boron doped crystals pre-annealed at 1120 • C (which eliminates grown-in vacancies if there are any) one needs a lower temperature for the first anneal: 700 • C instead of 800 • C, to induce appreciable nucleation of oxygen precipitates. 33 Remarkably, nucleation rate is faster at a higher boron concentration but the density tends to the same final value of about 10 10 cm −3 (see Fig. 2 in Ref. 33).
A simple explanation of the persistent precipitate density is that crystals always contain some heterogeneous sites for nucleation of oxide precipitates. These sites, of a density N s ≈ 10 10 cm −3 , could be small inclusions captured from the melt -or homogeneously formed small precipitates of some mobile impurity. The sites themselves do not give rise to the oxide precipitates. Yet they can be "activated" in the presence of some species that are incorporated, along with oxygen atoms, into the oxide nuclei, reducing the nucleation barrier. The enhancing (activating) species can be vacancies -or interstitial boron atoms in boron-doped material. At a lower concentration of activating species, only a part of sites is activated, for fixed annealing conditions. The observed precipitate density N p is then smaller than the site density N s . Above some threshold amount of the activating species, all the heterogeneous sites are activated, and N p become identical to N s , irrespective of the amount of activating species.
Within this general concept of vacancy-enhanced heterogeneous nucleation of oxide precipitates, the persistent value of N p in As doped silicon is explained in the following way. Slightly doped crystals already contain enough residual vacancies to activate all the heterogeneous sites by 800 + 1000 • C cycle. Doping with As -at not too high concentration -results in a strongly increased concentration of incorporated vacancies 18 due to the presence of VAs (vacancies trapped by As) along with the free vacancies V. This is manifested by an increased density of voids revealed as COP. We can expect that also the residual vacancy concentration is increased, and accordingly all the heterogeneous sites are activated, just as in lightly doped material. The precipitate density N p thus remains identical to the site density N s -close to 10 10 cm −3 -in spite of a strong variation in the vacancy concentration induced by As doping.
However, in As doped crystals, the As-related point defects are not only VAs but also interstitial As atoms (As i ), and their concentration, initially small, increases faster than that of VAs. 18 The total incorporated vacancy concentration (represented by V and VAs) begins to decrease above some As concentration, when [As i ] approaches [VAs] -due to the annihilation reaction V + As i → As s . This is manifested 18 as a sharp drop in the COP density at [As] > 1.7 × 10 19 cm −3 . The residual vacancy concentration, although reduced at this point, still seems to be enough for full activation of the sites: the observed density N p is still kept at the same value, 10 10 cm −3 . Only at a higher [As], the crystal becomes of interstitial type -populated mostly with As i instead of VAs, and thus without residual vacancies. Accordingly, the precipitate density is then strongly reduced.

Conclusions
The phenomenon of the reduced oxygen precipitation in heavily donor doped silicon has been reviewed, with special emphasis on heavily arsenic-doped silicon. An impact of arsenic on oxygen precipitation is experimentally confirmed in heavily arsenic doped silicon below 4 m · cm compared to lightly doped silicon with the same oxygen content and growth conditions. For resistivities higher than 4 m · cm, no appreciable difference is observed between heavily arsenic doped and lightly doped silicon. The drop in oxygen precipitation below 4 m · cm is quite rapid and can be explained by a strongly reduced vacancy availability in this resistivity range, as indicated also by the rapid disappearance of COPs. When the vacancy concentration is high enough, the density of oxygen precipitates is correspondingly high and tends to saturate at approximately 10 10 cm −3 also in the case of arsenic doped silicon. This can be explained by a heterogeneous precipitation mechanism controlled by the activation of specific sites with density of the order of 10 10 cm −3 .