In this paper we review the performances of a few techniques for the analysis of metal contamination in silicon. A few case studies are discussed to compare the ability of these techniques in detecting contaminants with different properties as silicon impurities. Common contaminants and elements recently introduced in the fabrication process are considered. TXRF (Total reflection X Ray Fluorescence), recombination lifetime measurement techniques, DLTS (Deep Level Transient Spectroscopy) and micro-photoluminescence analyses are compared. The results reported in this paper show that it is not possible to define a unique recipe that can be applied in all cases. Different approaches are required depending, on one hand, on the contaminant diffusivity and solubility, and, on the other hand, on the process step to be monitored. The techniques for monitoring metal contamination can schematically be divided into two main groups, one consisting of the techniques that measure the chemical concentration at the silicon surface and the other of the techniques that measure the electrical activity of the contaminant in the silicon volume. Mostly in case of metal contamination detection during wafer manufacturing, techniques which intentionally deposit thin films of polysilicon or thermal oxide and then perform chemical etching of them associated to ICP-MS analysis are also used, mostly addressed to detect fast diffusers like Cu and Ni [1], not always associated to clear electrical response. Indeed, these elements retain significant diffusivity even near room temperature, where the solid solubility is negligibly low, so they mostly segregate at the silicon surface or in getter sites, and for this reason they are hardly revealed in the silicon volume. However, these elements can be revealed by the recovery from deep depletion of MOS capacitors, because the segregated metals significantly increase the surface generation velocity. Some medium-fast diffusers like iron diffuse through several hundred microns during an ordinary thermal treatment, and mostly remain in the solid solution in silicon [1]. A comparison was carried out between TXRF and carrier recombination lifetime measurement for their ability to reveal iron contamination after a cleaning process. To prepare the samples for carrier lifetime measurements, the samples had to be thermally treated by RTP (Rapid Thermal Process). This study showed that TXRF and recombination lifetime measurements have essentially the same sensitivity to iron contamination. Vice versa, the detection of slow diffusers is found to be critical, because a very low concentration per unit area (≈107cm-2) may result in a non-negligible volume concentration in the device region (≈1010cm-3). As a consequence, the sensitivity per unit area required for these elements is difficult to reach with surface techniques such as TXRF. Molybdenum and tungsten [1] are examples of this sort. These elements do not diffuse deep enough to be efficiently revealed by recombination lifetime measurements, but they are easily revealed in the silicon volume by DLTS, which is probably the best approach for these elements, though it requires a considerable sample preparation. On the other hand, these elements are very common contaminants in ion implantation or epitaxial processes, and are very harmful for devices sensitive to carrier recombination, such as image sensors. Therefore, a technique able to detect near-surface contamination with no or limited sample preparation would be needed. Band-to-band micro-photoluminescence proved to be a good candidate, though DLTS still has the best sensitivity. Silicon contamination may also happen in the final part of the device process flow when the wafer frontside is protected by several dielectric layers and low-temperature thermal treatments only are allowed. In that case, contamination may come from the wafer backside only, but fast diffusers are still able to reach the wafer frontside and damage the device. In a study of palladium contamination from a contaminated chuck, TXRF proved not to be sensitive enough to palladium contamination. Palladium is best revealed by recombination lifetime measurements, such as the SPV (Surface PhotoVoltage) technique, but a high temperature treatment is required to activate it, so also in this case some sample preparation is required. Finally, we recall that the device itself can be a very sensitive monitor of metal contamination [2], though of course methods that prevent the impact of contamination of the device are generally preferred. The analysis of the dark current in image sensors not only reveals the presence of metal contamination (see the enclosed figure), but also provides some hints about the contaminant element. References [1] K. Graff, Metal impurities in silicon devices fabrication, Berlin, Springer, 2000 [2] C. Claeys and E. Simoen, physica status solidi (a) 216, (2019) Figure 1

We have analysed the effects of intentional metal contamination on the dark current of complementary metal-oxide-semiconductor image sensors. A few contaminants (molybdenum, tungsten, vanadium, titanium, copper) have been selected for this study, because they previously showed the most relevant impact on the dark current. The dependence of the dark current on temperature has been analysed in contaminated diodes from different regions of the dark current distribution. The generation current is always dominant in the diodes contaminated with the metals in this study, whether diodes in the peak of the distribution or in high current tails are considered. Diodes contaminated with slow diffusers never deviate from a diffusion-generation model of the current vs temperature. Copper contaminated diodes are the only example showing a limited but significant contribution approximately independent of temperature, which can be ascribed to a local tunnel current due to copper precipitates. From the point of view of dark current spectroscopy, this analysis shows that a correct identification of the dominant impurity is more easily obtained if diodes in the peak region of the dark current distribution are analysed. Indeed, in these diodes the generation current is characterized by an energy level which is a fingerprint of the dominant impurity. Vice versa, in high current diodes the dominant energy level can be different from this level, though the generation current is even more important in these diodes.

As Si is an indirect band gap material, the PL generated by phonon assisted band-to-band (B2B) radiative recombination (of energy equal to energy gap of Si) is very weak; about 10 orders of magnitude lower than the exciting photon flux [1] . If crystallographic defects are present then at room temperature an additional broad defect PL peak is generated (DPL) with energy smaller than the band gap of Si [1] , [2] , [3] . At room temperature, defect-band PL intensity is orders of magnitude lower than the B2B intensity [1] .

In this work, molybdenum and tungsten-implanted wafers are analyzed by an innovative technique based upon photoluminescence measurements, with the aim to assess the ability of this technique to detect metal contamination in the near-surface region. Surface Photovoltage (SPV) measurements of carrier diffusion length are compared to the results of photoluminescence measurements. It is shown that molybdenum and tungsten contamination are easily detected by photoluminescence intensity measurements down to about 10 10 cm − 2 contaminant dose. Vice versa, SPV has limited sensitivity to these elements ( ≥ 5 · 10 10 cm − 2 ), because of their low diffusivity. Therefore, photoluminescence intensity measurement can be a valid alternative to conventional carrier diffusion length measurements for monitoring slow diffuser contamination.

In this work, we review some techniques used for the characterization of ion implanted layers, with the aim to identify the best approach in various experimental conditions.With regard to dopant profiles, Secondary Ion Mass Spectrometry (SIMS) or Time-of-Flight (ToF)-SIMS are discussed. In the case of very thin layers (of the order of 10 nm) and for specific elements, ToF-SIMS is the best choice. For some devices, it is also necessary to analyse rather thick layers with low dopant concentration, and in this case dynamic SIMS gives better performances.Mass spectroscopy cannot provide information about the electrical activity of dopants. The profile of electrically active dopants can be obtained by the Differential Sheet Resistance (DSR) and Hall Resistance (HR) technique. This technique has the advantage that the carrier density and mobility are independently measured, thus providing information about the crystal quality in the layer, in addition to the dopant distribution. In the Spreading Resistance (SR) technique, a resistance profile is measured and turned into a carrier concentration profile with the aid of literature mobility data or by comparison with data from reference samples with known doping concentration. This approach may lead to wrong concentration data if the carrier mobility is degraded, for instance because of unrecovered damage.Various microscopy techniques are commonly used for the analysis of residual crystal defects, for instance the Transmission Electron Microscopy (TEM) and the Scanning Electron Microscopy (SEM) associated with selective etching. Recently, a new technique based upon micro-photoluminescence (μ-PL) measurements (so-called “EnVision”) proved to be a valid alternative to selective etching. A study about the residual damage after silicon implantation and annealing is reported. The conclusions reached by selective etching and conventional microscopy and by micro-photoluminescence are consistent with each other, but micro-photoluminescence has the advantage of providing a non-destructive analysis of large silicon areas.The best approach for monitoring metal contamination depends on the specific contaminant and hence of the contamination mechanism. Minority carrier lifetime measurements are suitable for fast diffusing contaminants (e.g. iron), whereas DLTS provides better sensitivity for slow diffusers such as molybdenum and tungsten.

In this work we present the results of experiments aimed at comparing the performances of various techniques for the detection of metal contamination in the silicon technology. Techniques for the measurement of surface contamination such as Total Reflection X-Ray Fluorescence (TXRF) and Time-of-Flight Secondary Ion Mass Spectrometry (ToF-SIMS) are compared with techniques for the measurement of contamination in the silicon volume, specifically the Deep Level Transient Spectroscopy and techniques for the measurement of carrier lifetime. Carrier lifetime measurements were obtained by photocurrent measurements and by Surface Photovoltage measurements. In a first experiment, Synchrotron-Radiation TXRF (SR-TXRF) was used as the reference technique to assess the sensitivity of a commercial TXRF instrument and of carrier lifetime measurements to detect iron contamination in wafers cleaned by ordinary cleaning processes. Samples intentionally contaminated by spinning with various elements were used for another experiment comparing TXRF and ToF-SIMS measurements of surface contamination. Then, a few case studies are discussed, specifically tungsten contamination by sputtering in ion-implanted samples and palladium contamination due to contact with a contaminated chuck. In all these experiments, advantages and disadvantages of the different techniques are discussed. The results of this study clearly show that it is not possible to define a unique recipe that can be applied in all cases. The maximum tolerated contaminant concentration per unit area depends on the contaminant diffusivity, and is much lower for slow diffusers. The contaminant diffusivity and solid solubility in silicon determine the in-depth distribution of the contaminant, and hence the most effective approach.

The aim of this work is to characterize the electrical properties of the defects induced by hafnium (Hf) contamination in silicon by means of different experimental techniques. Hf is introduced in silicon by shallow ion implantation at low doses in order to match experimental conditions close to an accidental contamination; moreover annealing is performed in RTP, that allows to maintain the metallic contaminant in solution. During the annealing the most part of Hf atoms segregate at the SiO2/Si interface, increasing the density of interface states. A detectable part of Hf atoms diffuse through the silicon bulk producing generation and recombination centers and traps for holes and electrons. Eight levels are detected, six in the upper and two in the lower half of the silicon bandgap, respectively. Hf acts differently in p-or n-type substrates: it is more effective as hole trap and as a recombination center in p-type and it de-activates above an implanted dose of 3 x 10(11) cm(-2) only in p-type silicon. The accurate determination of Hf-induced defects even at low concentration allows to find and pinpoint possible hafnium contamination affecting electronic devices. (c) 2018 The Electrochemical Society.

The objective of this paper is to study the phenomenology of the extended defects generated by phosphorus (P) implantations in silicon with a special attention to the interaction with a previous implantation of the same specie already recovered by a furnace annealing. For this purpose we decided to analyze two P implantations: the first is shallower with a projected range of 110 nm and a dose high enough to amorphize the silicon, the second one is implanted after the furnace annealing of the first implantation and has a projected range of 1.4μm. For the deeper implantation different doses (from 1×1013 to 1×1014cm-2) and two different schemes of annealing have been tested. More in depth there is a third P implantation with a projected range of 2.3μm and a dose in the order of some 1013cm-2. Selective etching of silicon, transmission electron microscopy (TEM) and micro-photoluminescence, combined with secondary ion mass spectroscopy (SIMS) were used for the analysis of the implantation defects and for the determination of P distribution, respectively. Analyzing the two implantations separately we note that the shallower and amorphizing implantation left the dislocation loops aligned at the projected range, as expected; instead the couple of the deeper implantations shows the presence of threading dislocations which generate in the implanted zone (between 1 and 3μm) and which increase for many μm reaching the surface and also the deepest zone of the wafer (6.5μm). The situation changes if the implantations are done in the same sample, first the shallow implantation followed by the double implantations in depth. The dislocations are more numerous, but shorter. The interesting fact is that the growth of the defects is inhibited mostly towards the surface, while the zone affected by the defectiveness in depth remains more or less the same.

We have investigated the effects of deliberate heavy metals contamination on dark current and image defects in CMOS Image Sensors (CIS). Analysis of dark current in these imager dice has revealed different behaviors among most important 3d metals present in the process line. We have implanted directly in 3 Mega array pixels the following metals: Cr, V, Cu, Ni, Fe, Ti, Mo, W, Al and Zn. Analyzing the dark current "spectrum" as obtained for fixed integration periods of time by means of standard image-testing equipment, these impurities can be identified and detected with a sensitivity of similar to 10(9) traps/cm(3) or higher. (C) 2017 The Electrochemical Society. All rights reserved.

In this paper the tungsten contamination in ion implantation processes is studied by DLTS analysis both in typical operating conditions and after contamination of the implanter by implantation of wafers with an exposed tungsten layer. Of course the contaminant concentration is orders of magnitude higher after contamination of the implanter, but in addition our data show that different mechanisms are active in a not contaminated and in a contaminated implanter. A moderate tungsten contamination is observed also in a not contaminated implanter, however in that case contamination is completely not energetic and can be effectively screened by a very thin oxide. On the contrary, the contamination due to an implantation in a previously contaminated implanter is reduced but not suppressed even by a relatively thick screen oxide. The comparison with SRIM calculations confirms that the observed deep penetration of the contaminant cannot be explained by a plain sputtering mechanism.

Any noble gas characterization that includes chemical ionization for the detection process is hard to manage because of the very high ionization potential: the range is from 10 to 25eV; electronic affinity is also close to zero. So ionization processes used in secondary ion mass spectroscopy (SIMS) analysis have very poor cross section. It is known that amorphous carbon mechanical properties are changed by implant; sp2/sp3 ratio and H content in the film are modified, allowing high density and good transparency needed for dry etching and lithography steps. In this paper, we propose to monitor secondary ion clusters Cx–Hy by time of flight‐SIMS to obtain information on amorphous carbon layer implanted with Ar and to study Cn/C5 ratio to understand sp2/sp3 ratio behavior. Further consideration on 13CH as hydrogen representative will be addressed. A binary collision code Stopping and Range of Ions in Matter (SRIM) was used to estimate the depth of the region damaged by Ar+ implantation. In implanted samples, the profiles of Cx–Hy clusters have an in‐depth gradient. This is an indication that the film was involved in an energy release process. Therefore, these profiles give information on the implanted region even if the Ar profile is not collected. Although the SIMS technique uses ion beams that produce damages similar to those due to the implant process, a 13CN depth profile behavior was found able to discriminate between damage effects produced by that or the other. Copyright © 2016 John Wiley & Sons, Ltd.

In this paper, we review the results of some studies addressing the problem of tungsten contamination in implantation processes. For some tests, the implanter was contaminated by implantation of wafers with an exposed tungsten layer, resulting in critical contamination conditions. First, DLTS (deep level transient spectroscopy) measurements were calibrated to measure tungsten contamination in ion-implanted samples. DLTS measurements of tungsten-implanted samples showed that the tungsten concentration increases linearly with the dose up to a rather low dose (5x10(10) cm(-2)). Tungsten deactivation was observed when the dose was further increased. Under these conditions, ToF-SIMS revealed tungsten at the wafer surface, showing that deactivation was due to surface segregation. DLTS calibration could therefore be obtained in the linear dose regime only. This calibration was used to evaluate the tungsten contamination in arsenic implantations. Ordinary operating conditions and critical contamination conditions of the equipment were compared. A moderate tungsten contamination was observed in samples implanted under ordinary operating conditions. This contamination was easily suppressed by a thin screen oxide. On the contrary, implantations in critical conditions of the equipment resulted in a relevant tungsten contamination, which could be reduced but not suppressed even by a relatively thick screen oxide (up to 150 angstrom). A decontamination process consisting of high dose implantations of dummy wafers was tested for its efficiency to remove tungsten and titanium contamination. This process was found to be much more effective for titanium than for tungsten. Finally, DLTS proved to be much more sensitive that TXRF (total reflection X-ray fluorescence) in detecting tungsten contamination. (C) 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

In this paper, the impact of molybdenum and tungsten contamination on the properties of MOS capacitors is analyzed, with the aim to investigate whether measurements of MOS capacitors can be used to detect this sort of contaminants. Surface Photovoltage (SPV) measurements of carrier diffusion length and of the oxide charge (obtained from high injection SPV) are compared to the results of MOS capacitor measurements. In agreement with previous DLTS data, both SPV and generation lifetime data show that molybdenum remains in the solid solution in silicon in the studied dose range. As a consequence, molybdenum contamination has limited impact on the interface state density at the oxide-silicon interface. Vice versa, tungsten was shown to deactivate starting from rather low doses (1011 cm(-2)). Tungsten deactivation is associated with the formation of new peaks in interface state density spectra. However, no tungsten segregation at the oxide-silicon interface could be detected by ToF-SIMS measurements. We suggest that in MOS capacitors tungsten deactivation is associated with segregation in the oxide layer. SPV has limited sensitivity to these impurities. Generation lifetime is a good monitor of tungsten contamination, whereas for molybdenum DLTS is the most sensitive contamination monitor. (C) 2016 The Electrochemical Society. All rights reserved.

Contaminant reduction is a key issue for SOI substrate which cannot make use of back-side gettering. H 2 annealing has been proven to be effective in Si reconstruction, influencing diffusion by breaking strained Si bonds and generating cavities for contaminant gettering. These properties could help in reducing contaminants in BCD-SOI process. Unfortunately, H 2 annealing integration can be highly critical and the process optimization has to take into account 3-D morphology evolution and contaminant reduction efficiency. Aim of this work is to understand the physical mechanisms behind Si surface reconstruction and metallic contaminants reduction.

A few key techniques for the analysis of contamination in silicon are compared for their ability to detect different impurities. Both metal and organic contamination is included in this study. In addition, common contaminants and elements recently introduced in the fabrication process are considered. For what concerns metal contamination, it is shown that different approaches are required depending on the in-depth distribution of the contaminant and hence on its diffusivity. Copper, iron, molybdenum, and tellurium are chosen as examples of contaminants with different diffusivity and solubility properties. Total reflection X-ray fluorescence (TXRF), recombination and generation lifetime measurement techniques, deep level transient spectroscopy (DLTS) and capacitance versus voltage measurements are compared. The detection of slow diffusers is found to be very critical, because a very low dose may result in a non-negligible concentration in the device region. As a consequence, the sensitivity per unit area required for these elements is difficult to reach with surface techniques such as TXRF. On the other hand, very fast diffusers such as copper can hardly be revealed in the solid solution in silicon. Copper in silicon can be revealed at the oxide-silicon interface by TOF-SIMS measurements, or by surface generation velocity measurements with the Zerbst method. For what concerns organic contamination, surface recombination velocity and gate oxide integrity tests were compared. The most relevant effects of organic contamination were observed by electrical stress of the oxide. Indeed, the fraction of capacitors with degraded breakdown voltage increased dramatically in wafers with intentional organic contamination. (C) 2015 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

In this work we report the results of a set of experiments carried out to assess the ability of recombination lifetime measurements for the detection of palladium contamination in silicon. Palladium is found to be a very effective recombination center, so recombination lifetime measurements are a very sensitive method to detect palladium in silicon. The surface segregation of palladium was monitored by the reduction of its recombination activity in the silicon volume. The palladium segregation at the wafer surface was checked by selective etching, and by Transmission Electron Microscopy (TEM) and Energy Dispersive X-ray (EDX) analysis.After validating recombination lifetime measurements for palladium detection, we use these measurements to define suitable approaches to the prevention of palladium contamination of silicon devices. The efficiency of a diffusion barrier layer (silicon nitride) and of decontamination by wet cleaning are tested.

Various measurement techniques are compared and the most suitable methods for contamination detection are identified. The results of this study show that it is not possible to define a unique recipe that can be applied in all cases. Concerning metal contaminants, the stratigraphic in-depth distribution and hence the diffusivity of contaminants determines the most effective approach. Iron and palladium are chosen as the examples of fast diffusers, molybdenum and tungsten as slow diffusers. Fast diffusers like iron and palladium diffuse through several hundred microns during an ordinary thermal treatment. Minority carrier lifetime measurements are probably the best choice to detect these contaminants. Molybdenum and tungsten do not diffuse deep enough to be efficiently revealed by recombination lifetime measurements, but are easily revealed in the silicon volume by DLTS. Because of their low diffusivity, a very small amount of these elements per unit surface may result in a significant concentration in the near-surface region where devices are built. Ion implantation is confirmed to be an important source of metal contamination. It is shown that ion implantation can be responsible both for iron contamination and for contamination by slow diffusers, such as molybdenum and tungsten. A procedure for monitoring molybdenum and tungsten contamination in ion implantation processes by DLTS is defined and calibrated. Finally, the efficiency of some gettering techniques in reducing iron, molybdenum and tungsten contamination is discussed. Gettering is found to be active at relatively high contaminant concentrations, but low contamination levels are not gettered under our experimental conditions. Carbon implantation showed partial efficiency in gettering molybdenum and tungsten, whereas gettering did not take place after silicon implantation. (C) 2015 The Electrochemical Society. All rights reserved.

In this work palladium is characterized as a silicon contaminant by recombination lifetime, DLTS, C-V and C-t measurements of palladium-implanted wafers. Palladium introduced by ion implantation is found to remain in the solid solution in silicon after rapid thermal treatments, and to be a very effective recombination center. For this reason recombination lifetime measurements are the most sensitive method to detect palladium in silicon.Two palladium-related levels were found by DLTS in p-type material. One of these levels corresponds to a level reported in the literature as the single donor level of substitutional palladium.For what concerns MOS capacitors, palladium is responsible for negative oxide charge and for degradation of the generation lifetime.In addition, palladium is confirmed to be a very fast diffuser, which segregates at the wafer surface even with low temperature treatments (250 degrees C). Microscopy inspections showed that palladium precipitates and surface defects were formed upon segregation. (C) 2015 Elsevier Ltd. All rights reserved.

A procedure to measure molybdenum and tungsten contamination in implantation processes by DLTS (Deep Level Transient Spectroscopy) is defined and calibrated for the evaluation of molybdenum and tungsten contaminant dose. The obtained calibrations are used to study molybdenum contamination in BF2 implantations and tungsten contamination by sputtering from a previously contaminated wafer holder.