The diffusion, solubility, complexing, and gettering of transition metals in silicon are reviewed with special emphasis on the observed chemical trends. 3d metals appear dominantly on interstitial sites in thermal equilibrium. Their diffusivity and solubility increases markedly with atomic number. Going from the 3d metals vertically down in the periodic table, a clear trend towards substitutional solution is obvious. Observation of deep level defects, complexing and gettering is strongly dependend on the respective mobility of the metal atoms. Understanding of the thermodynamics of complexing and gettering allows to define heat treatment procedures favourable to minimize detrimental effects from transition metals in silicon.

Neutron activation, atomic absorption, and voltammetric techniques are presented as appropriate analytical tools to monitor trace amounts of metallic impurities on large diameter silicon wafers. Detection limits and results are compared for standard wafers implanted with Fe, Cu, and Ni. DLTS measurements are included for implanted and as-grown doped wafers.

A new method is presented which is suitable to check gettering processes and to evaluate gettering efficiencies. This method is based on an intentional contamination of the respective silicon wafer by means of palladium or any other haze forming metal. The palladium is diffused and haze is revealed by preferential etching. In order to determine gettering efficiency the sizes of the haze areas on gettered and non-gettered wafers are compared. The quantity of the gettered palladium can be deduced from lateral palladium profiles.

In contrast to procedures published in the literature this method is very simple and fast. Therefore it is suitable for routine application to check back side damaged wafers or internal gettering.

We report the results of a study of the microstructural changes brought about by the gettering of metallic impurities in silicon. Phosphorus gettering proceeds by the formation of metal disilicides at the silicon/phosphosilicate glass interface. By showing that iron is strongly localized at the interface by phosphorus gettering, we present the first evidence that phosphorus gettering can proceed independently of Fermi level effects. Our first results on the intrinsic gettering of nickel lend support to the notion that the emission of silicon interstitials brought about by phosphorus diffusion (and SiP formation), or intrinsic gettering (SiOx, formation), plays an important role in the gettering process.

The annealing kinetics of the gold-iron complex have been studied over the temperature range from room temperature to 400°C. From a consideration of the charge states of the gold and iron centres at room temperature it was concluded that the pairing was by Coulombic attraction. The binding energy of the centre was found to be 1.22 eV.

During the annealing processes the interstitial iron was found to be precipitating and this was shown to modify the details of the growth and dissociation rates of the gold-iron complex.

Deep level defects in p/p+ epitaxial silicon were characterized by deep level transient spectroscopy (DLTS). Two dominant deep level defects were found in all samples which have been identified with Fe and CrB pairs. A third deep level defect was found in most of the samples which has tentatively been identified with Ti. The concentrations of these traps were established in a large number of samples as a function of epitaxial growth condition and substrate oxygen level.

Molybdenum contamination has been detected in silicon epitaxial layers and substrate wafers after processing in any one of several epitaxial silicon reactors. Greatly reduced minority carrier diffusion lengths and lifetimes are consistent with Mo concentrations measured by DLTS in the 1012 and 1013 cm−3 ranges. Depth profiling of diffusion length and the Mo deep level show much greater penetration than expected from previous reports of Mo as a slow diffuser. The data indicate a lower limit of 10−8 cm2/sec for the diffusion coefficient of Mo in silicon at 1200°C, consistent with high diffusivities measured for other transition metals.

To understand gettering action we have modeled the diffusion of metal to a backside getter. A first order model describing the diffusion of metals to an infinite backside sink is found not to fit experimental data. To fit the data for phosphorus gettering at 1000°C and argon implantation gettering at 800°C it is necessary to also hypothesize the injection of silicon interstitials. Assuming that the diffusion of the silicon interstitials is the rate limiting step in gettering, allows one to extract the diffusion coefficient and equilibrium concentration of the interstitial. The hypothesis of silicon interstitial injection agrees with recent work on phosphorus diffusion by Fahey et. al. To test this hypothesis for heavily damaged regions we have looked at the effect of an argon implanted region on the diffusion of a phosphorus buried layer. At both 800°C and 700°C, during nitrogen anneals we see significant enhancement of the phosphorus diffusion.

This paper presents strong evidence for carbon and oxygen gettering during high dose implantation of Cr and Al into Si by implanted ions or Si atoms. Rutherford Backscattering (RBS) and Auger Electron Spectroscopy (AES) experiments of Cr and Al implanted Si samples in a diffusion pumped vacuum (DPV) system (≃ 10−6 Torr) clearly indicate that the incorporation of carbon or oxygen occurs preferentially and depends on the surface composition which is controlled by the dose of implantation and by sputteripg.

The implantation of Cr and Al in a DPV system (≃ 10−6 Torr) results in a large increase in the concentration and retention of implanted atoms compared to implantation in a UHV system (≃ 10−8 Torr). For implantation in a DPV system, the Cr distribution broadens up to a factor of two. For Al a bimodal or broadened distribution could be observed by nuclear resonance profiling (NRP).

New results are presented on Ge doped Si epitaxial layers which contain interfacial misfit dislocations. Microscopic and chemical analyses showed the preferential gettering of several metallic species (Au, Cu, Ni, and Fe) at the misfit dislocations with semiquantitative correlation between dislocation density and the captured impurity concentration. Wafer curvature was measured and shown to be less than that for typical Si3N4 and SiO2 layers used in IC fabrication. The reduction of Schottky diode leakage current has been clearly demonstrated and attributed to gettering of residual impurities, as well as signifying that, the active surface device region is not deleteriously affected by spurious defect reactions at the buried epitaxial interface.

The diffusion of B and As in mechanically strnsed silicon has been investigated for initial implant doses of 1013, 1014, and 1015 cm-2, over a range of annealing temperatures. At stresses near the silicon yield point, no significant enhancement or retardation was observed. This was true even in plastically deformed samples with dislocation densities >1×107 cm-2. The results are consistent with the multiple charge state vacancy model of impurity diffusion in silicon. The B diffusivity appears to agree with the accepted activation energy of 3.59 eV and pre-exponential of 3.17 cm2/sec for intrinsic B diffusion.

The diffusivity of implanted As in crystalline Si has been measured using Rutherford backscattering and channeling for specimens annealed at temperatures between 850 and 1000°C under pressures up to 30 kilobars. The diffusivity, at a given temperature, was found to increase with pressure with a maximum increase of a factor of ten. This diffusivity enhancement can be described by an average activation volume of -5.7±.8 cm3/mole. The activation enthalpy ranges from an ambient value of 4.5 eV to 3.6 eV at 30 kilobars

The diffusion of various dopants( B, P, As, and Sb) in (111) silicon during high temperature( 1100°C ) heat treatment in nitrogen depends on the capping layers, i.e., SiO2, Si3 N4, and SiO2 + Si3 N4, layers. The junction motion of all dopants were retar ed in the bare silicon compared with that in the SiO2 + Si3 N4 capped silicon. The apparently retarded junction motion of As and sb diffused specimens in the bare silicon is due to the severe out-diffusion loss of dopant atoms and As and Sb diffusion are in fact enhanced during thermal nitridation of silicon in N2.

Enhanced diffusion of carbon Is observed to be produced during anneal ing of silicon at 900°C under conditions of surface oxidation and phosphorus in-diffusion. Silicon containing high concentrations of carbon (∼9 × 1017/cm3 substitutional) and varying levels of interstitial oxygen and differing defect concentrations has been studied. Diffusion coefficient enhancement over the value found for an anneal in an inert ambient is by a factor of three during oxidation and a factor of forty with phosphorus in-diffusion. Carbon accumulation takes place in a region 0.3–0.5 microns from the sample surface only under conditions of phosphorus in-diffusion and is attributed to carbon precipitation. A model that assumes Interactions between silicon self-interstitials and the carbon can explain both the enhanced diffusion and the carbon accumulation.

Boron diffusion profiles in single crystal silicon from highly doped polysilicon sources have been measured using SIMS after diffusion at 950°C for various times. These data have been analyzed to determine D(c) of boron in the single crystal. It is ncrmally assumed that at high boron concentrations D increases linearly with concentration. However, the shape of these profiles indicates that for a polysilicon source, this behavior does not appear to hold. Using Bolzmann-Matano analysis, D(c) was found to be insensitive to boron concentrations above 3.0E19 atoms/cm3. The results of this analysis were confirmed by using them as input to a diffusion simulation computer program and excellent agreement with the experimental profiles was obtained. The value of D was found to be unusually high at all concentrations in the single crystal and increased almost linearly with the doping level of the polysilicon. The effect of the furnace ramp-down cycle on the profiles near the crystal surface have also been investigated.

The effect of pressure on self diffusion in crystalline silicon is being studied using 30Si as a tracer. Diffusion experiments have been carried out in the pressure range of 1 to 35000 atmospheres at 1000°C. The 30Si is observed to diffuse faster at high pressures, indicating a negative activation volume.

In a first part, we deal with the influence of intrinsic point defects (vacancies and self-interstitials) on self- and impurity diffusion in silicon. Estimates of the diffusivities and thermal equilibrium concentrations of vacancies and self-interstitials are given. In a second partwe discuss the influence of point defects on the diffusion and precipitation of different types of metallic impurities in various gettering schemes as well as on the nucleation and growth of SiO2 precipitates.

The problem of the respective contributions of vacancies and self interstitials to diffusion mechanims in silicon is clarified by using a consistent model for the diffuion of P, As and B which involves the resolution of the coupled continuity equations for the impurities, the vacancies and the self-interstitials. This model allows also the understanding of basic features concerning the kinetics of the point defects in presence of impurities.

Our investigation of impurity diffusion in Si has clarified diffusion mechanisms and the point defect structure for the material. Thermal equilibrium between interstitials and vacancies in high temperature Si was clarified through observation of oxidation-retarded diffusion (ORD) of Sb in Si. Boron, Phosphorus, Aluminum and Galium were found to diffuse mainly due to an interstitialcy mechanism, while Sb did by a vacancy mechanism. Moreover, the atomic number in Si crystal compared to that at lattice sites was seen to play a dominant role in determining point defect concentration. Oxidation of Si and nitridation of SiO2 on Si lead to an increase in the atomic number of Si crystals and cause an increase in interstitial concentration as well as a decrease in vacancy concentration. On the other hand, nitridation of Si leads to a decrease in atomic number and causes vacancy super-saturation as well as interstitial under-saturation.

An interstitial model is developed to quantitatively account for the anomalous effects which accompany the diffusion of phosphorus in silicon from high surface concentrations. Phosphorus is assumed to diffuse as both a positively charged (intrinsic diffusion) and an uncharged interstitial (high concentration case). Silicon is assumed to diffuse as both an uncharged (instrinsic) and a negatively charged interstitial (high phosphorus concentrations). The effect of internal fields on charged species is included. An excellent fit is found with published experimental data if a surface enhancement factor, analogous to that for the case of oxidation enhancement, is used. Flat-top diffusion is modelled by assuming the formation of a dislocation array, which alters the point defect equilibria.