Oxygen in silicon has been an important scientific and technological subject for over 30 years. It derives its technological importance from the fact that silicon crystals pulled from silica crucibles have been the standard substrates for fabrication of integrated circuits. The scientific interest originates both from support of this technology and the wide variety of phenomena that occur when cooled Czochralski (CZ) Si crystals become supersaturated with oxygen. Furthermore, the semiconducting nature of Si permits the use of very sensitive electrical and optical probes to complement the structural and chemical characterization of these varied solid state phenomena.

In spite of 30 years of study the most basic questions about the oxygen donor in silicon remain unanswered. There are not accepted models for the structure or formation kinetics. There is not even agreement on what the donor's constituents are. Nonetheless, substantial progress has been made in this field in recent years [1] that narrows model ideas and helps to focus continuing research.

The study of hydrogen in crystalline Si has enjoyed a resurgence of interest in recent times, principally because hydrogen is a common constituent of many semiconductor reagents, and also because of its ability to passivate the electrically activity of shallow acceptor dopants, and many deep contaminating centers in Si. This enables its use as a sensitive probe of the defect chemistry occurring in Si at relatively low temperatures – the presence of hydrogen can be detected electrically through the passivation of impurities and defects, and isotopic substitution with deuterium enables chemical analysis by secondary ion mass spectrometry.

This is the first symposium to explicitly include nitrogen in discussions on nondopant light impurities in Si. Ordering of the impurity listing (oxygen, carbon, hydrogen and nitrogen) deserves examination. It is obvious that the impurities are not listed alphabetically nor are they listed in the order they appear in the periodic table. Presumably the ordering does not simply reflect a bias of the organizing chairmen. The impurities are, in fact, listed in an order of recognition on a chronological scale of their importance in Si technology. Equilibrium solid solubility for the different impurities, as well as concentrations of the impurities in the processing environment, probably determined the chronology. Based upon published literature, the importance of oxygen was recognized in the mid 50's, carbon in the mid 60's and hydrogen in the mid 70's. Now in the mid 80's nitrogen is receiving increased attention as an impurity in Si.

The diffusivity and solubility are two key parameters required for understanding and modeling the behavior of oxygen in silicon. This paper gives an up to date review of experimental determinations of these parameters, including some recent unpublished data. There is very good agreement within the long-range diffusivity results determined by secondary ion mass spectrometry (SIMS), charged particle analysis (CPA), and x-ray diffraction. The oxygen diffusivity is independent of [O], orientation, ambient, or crystal doping. The data also extrapolate well to the diffusivities obtained by the intrinsic oxygen atomic hop frequency at low temperature to give a combined expression of D = 0.13 exp(−2.53eV/kT) cm2s−1. There is somewhat poorer agreement on the solubility measurements, in part due to inconsistent calibration factors and the observation of a processing-dependent extrinsic oxygen solubility. The intrinsic solubility derived from SIMS, CPA, and infrared absorption is described by [O] = 9E22 exp (−1.52 eV/kT) cm−3. Finally, the above diffusivity and solubility parameters are compared to modeling of oxygen related phenomena in silicon, such as thermal donor and precipitate formation kinetics, and interaction with point defects during the relaxation of stress-aligned dichroism.

The diffusivities of oxygen in Czochralski Si (CZ-Si) and float-zone Si (FZ-Si) have been measured by using secondary ion mass spectrometry. The diffusivity at 700–1160°C deduced from the outdiffused profiles of oxygen incorporated in CZ-Si shows little or no dependence on processing conditions and can be expressed as D = 0.14 exp(−2.53 eV/kT) cm2/s. Diffusivity at 700–1100°C of oxygen implanted in FZ-Si is insensitive to doses and follows D = 0.13 exp(−2.50 eV/kT) cm2/s, which agrees remarkably well with CZ-Si data. Since large variations in point-defect concentrations existed under the conditions studied, the excellent agreement among the diffusivities leads to the conclusion that point defects in Si have little effect on oxygen diffusion. This demonstrates that oxygen diffuses primarily via an interstitial mechanism in the temperature range studied.

A discussion of the isotope shift of the low temperature spectrum of the stretching mode of interstitial oxygen (Oi) in silicon introduces IR results showing the interaction of Oi with the silicon lattice. Evidence is given that the temperature dependence of a combination band observed at liquid helium temperature (LHT) at 1205.7 cm−1 is responsible for the room temperature (RT) band at 1227 cm−1 and that a weak band near 1013 cm−1 is an overtone of the 518 cm−1 band.

In this review we focus on oxygen-related defects created by electron irradiation (vacancy-oxygen defects) and subsequent thermal treatments. The annealing of the vacancy-oxygen pair (VO-center) at 300–350 °C is discussed as well as results from the formation and annealing of the successors to VO, the VO2 and VO3-centers. It -s found that VO2 is formed by diffusion of a vacancy-oxygen pair to an interstitial oxygen atom. It is suggested that VO3 is formed by the diffusion of interstitial oxygen to a VO2-center (in the temperature range 450–485 °C). At continued annealing at these temperatures VO3 is transferred to a new defect VO4 by attaching one more oxygen. Simultaneously thermal donors are developing in a normal way i.e. as in an unirradiated sample. It is therefore concluded that VO2 is not an important core for thermal donors but a possible nucleus for oxygen precipitation.

Measurements of optical bands in irradiated Si are combined with numerical modelling of the radiation induced reactions. No evidence is found for appreciable interaction of self-interstitials with O atoms for irradiations carried out at temperatures between 25 and 500°C. The reduction during electron irradiation of stress-induced dichroism in the 9μm oxygen band is shown to occur by sequential capture of a vacancy and a self-interstitial at the oxygen for irradiations carried out between 25 and 280°C. At higher temperatures repetitive capture and release of vacancies at oxygen atoms appears to dominate in the oxygen migration process.

Fourier Transform Infra-Red spectroscopy (FTIR) is commonly used to measure interstitial oxygen concentrations in silicon. The absorption coefficient αOX derived from the IR transmittance curve is converted to a concentration value through a multiplicative calibration factor γOX. Presently used calibration factors include 2.45 × 1017 cm−2 (ASTM F121−80), 3.03 × 1017cm−2 (JEIDA), and 4.81 × 1017 cm−2 (ASTM F121−76). These have been obtained from different analytical methods, including charged particle activation, helium carrier gas fusion-gas chromatography, vacuum fusion gas analysis, and photon activation combined with inert gas fusion. An alternative analytical method by secondary ion mass spectrometry (SIMS) is reported here. Oxygen ions of a known fluence are implanted into silicon samples of which oxygen content has been measured by FTIR. SIMS profiles of oxygen are performed using a cesium primary ion beam on a CAMECA IMS-3f. The result is a profile of the superposition of the implanted oxygen and the constant, bulk CZ oxygen. Integration of the implant signal followed by subtraction of the bulk signal can be used to calibrate the implant concentration from the implanted oxygen fluence, so that, in the same sample and during the same measurement, the SIMS calibration and FTIR calibration at the CZ bulk level can be directly compared. The γOX obtained from this method is 3.45 × 1017cm−2 and is closest to the JEIDA value of 3. 03 × 1017 cm−2.

Oxygen precipitation in CZ silicon is known to provide beneficial yield improvements in integrated circuit processing if the location and amount of precipitation can be properly controlled. The concentration of oxygen in the unprocessed silicon substrate is one of the most important variables to control for achieving these improvements. Fourier Transform Infrared Spectroscopy (FTIR) has successfully been used to measure [0] in silicon when the silicon resistivity is greater than about 0.1 Ω-cm. At lower resistivities typical of p+ and n+ substrates used for epi-wafers as free carrier absorption interferes with the FTIR measurement of bulk [0].

This work will focus on how to quantitatively measure oxygen in heavily-doped silicon by Secondary Ion Mass Spectrometry (SIMS) with a high sample thruput, low background signal, and tight σ/x distribution. SIMS calibration is performed against FTIR-calibrated substrates with resistivity higher than 0.1Ωcm. Typical background signals as measured in FZ are a factor of 20 below signals in CZ, and the 160− signal in CZ is over 105 count/sec. resulting in an excellent signal-to-noise ratio for each single measurement. Typical thruput is 18 samples per day where each sample is analyzed four to five times to obtain a σ/x of 3% for an oxygen level of 15 ppma (ASTM F121−80).

During aggregation, an assembly of oxygen atoms in silicon produces an electrically active site. The center is an effective mass, helium-like center (double donor) with a wave function of C2v symmetry. The formation reactions of the assembly reveal details of the invisible early stages of aggregation. The donor character of the center controls the aggregation process in heavily doped material. The atomic structure and the source of the electrical activity of the center remain unresolved.

The infrared spectrum of oxygen donor complexes in silicon under uniaxial stress has been examined for the neutral and singly ionized charge states. Our results are consistent with an effective mass-like ground state wave function that is constructed from a single pair of conduction band valleys for both charge states. A thermal ionization experiment in which the stress split components of the ground state are monitored by the absorption of polarized light confirm this interpretation and provide correlation with DLTS and EPR results. Additional small splittings, due to deviations from effective mass theory, show that the electronic wave function of the oxygen donor is distorted by an extended “central cell” with C2v symmetry. Previously observed splittings of 1s → np± transitions for the singly ionized charge state at zero stress are interpreted in terms of the effect of the anisotropic oxygen donor structure upon excited state wave functions constructed from the single pair of conduction band valleys

The electron spin resonance (ESR) of a number of vacancy-oxygen complexes created in electron irradiated Si containing oxygen is briefly reviewed. In these centers the unpaired spin density is highly localised (50–70%) on two Si sites within the complex. The ESR spectra of several paramagnetic thermal donors (TD+) formed by annealing oxygen-rich Si at 450 °C for one to several hundred hours are reviewed as well as recent experiments under uniaxial stress from which for the ground state a 2 valley effective mass-like wavefunction was derived. Results of electron nuclear double resonance (ENDOR) experiments on the TD+ ESR line “NL8” are presented for the first time. They reveal that all TD's identified in IR-spectroscopy are superimposed in the NL8 ESR line and that upon growth of the TD the structure of the core and the symmetry of the TD's is not changed. The hyperfine interactions with up to 7 shells of Si nuclei of 5 TD+'s were determined. Their size and tensor orientations are consistent with the two-valley effective mass-like wavefunction. 4 TD's identified in ENDOR could be correlated with IR-bands from growth kinetics. Although the ENDOR experiments so far do not yet lead to a detailed TD model, the model possibilities are narrowed down considerably by these results. The core must consist of nuclei with very low magnetic abundancy. It is tentatively suggested, that the core contains an O2 molecule.

Infrared absorption investigations of oxygen containing silicon reveal that the thermal donors generated by annealing around 450 °C consist of a series of shallow double donors with slightly differing ground state energies. The kinetics of formation and decay of these donors in samples with varying oxygen content annealed at temperatures between 460 and 600 °C support the relation of the thermal donors to aggregates of oxygen atoms. A preferential orientation of thermal donors induced by uniaxial stress during their growth demonstrates their low symmetry and extended defect-like nature. Optical cross sections evaluated by comparing IR-absorption spectra with resistivity measurements allow to estimate the concentrations of the different donors.

It is well known that Czochralski-grown silicon contains up to 2×1018 cm-3 oxygen atoms. When this silicon is annealed at 450°C a series of “thermal” donors are formed. In this work we present the results of a study of annealed Czcochralski-silicon samples by Photothermal Ionisation Spectroscopy and IR transmission measurements. All previously reported thermal donors, which occur in the spectral region above 330cm-1, and hove ionisation energies 53–69.3 meV, are observed by us. Also, our experimental results show the presence of at least three previously unreported thermally formed donor transition series below 330cm-1. Our results indicate that the ionisation energies of these new shallow thermal donors would be 36.3, 37.0 and 37.4 meV. Assignments are made by comparison of the energy differences between excited states to those predicted by the Effective Mass Theory.

We detect new luminescence lines in Czochralski-grown silicon after prolonged heat treatments at 450°C. The properties of the new optical transition are well explained by a thermal donor-free hole recombination. From the IR-absorption spectra measured on the same samples, we have evidence for an inhomogeneous distribution of the thermal donors. The spatial fluctuations of the thermal donors appear to be necessary for the new luminescence spectrum to emerge.

The electrical properties of thermal donors centers, generated after annealing CZ-silicon at 450° C are carefully evaluated. The energy levels are corrected for the Poole-Frenkel lowering are compared to the ionization energy of the dominant centers revealed by infrared absorption experiments. The capture cross sections are found to be ≈ 2 × 10-12 cm2 for TDI and ≈ 2 × 10-13 cm2 for TD2. Detailed examination of their temperature dependence reveals that they follow a common law σ α T-5.

The hierarchies of thermal donor binding energies produced by annealing oxygen-containing silicon or germanium at ca. 450°C are explained by using a generalized perturbation model which involves a standard repulsion parameter for the interaction between agglomerating oxygen atoms and the shallow donor electrons. This model is capable of fitting the ground state ladders for both charge states of the thermal donors in both Si and Ge, since differences between the two ladders can–ee explained entirely by the change in the electron-effective-mass and dielectric constant of the host.

Oxygen thermal donor formation under stress and following various pre-heat treatments was investigated by resistivity measurements. Thermal donor formation at 450°C, with and without bending stress, is monitored in p-type Cz-silicon following various heat-treatments. As has been shown by others, the thermal donor formation rate depends upon the pre-treatment of the samples. We find that the rate also depends on the stress, being faster, or slower, on the tensile side than on the compressed side depending on the pre-treatment.