GB/T 24578-2024Test method for measuring surface metal contamination on semiconductor wafers - Total reflection X-Ray fluorescence spectroscopy (English PDF)

Relationship to previous standards

This document replaces GB/T 24578-2015 and GB/T 34504-2017.

3 Terms and definitions

The terms defined in GB/T 14264 apply, together with four terms defined here.

3.1 Total reflection: the phenomenon in which light travelling from an optically denser medium towards an optically thinner medium loses its refracted ray entirely and only the reflected ray remains, once the angle of incidence exceeds the critical angle. A note observes that for X-rays most solids are optically thinner than air.

3.2 Glancing angle: the angle between the plane of the sample surface and the virtual plane containing the X-rays incident on that surface. A note explains that when the X-rays strike the wafer surface at a small glancing angle they undergo total reflection and the angle of reflection equals the glancing angle.

3.3 Critical angle: the angle of incidence at which X-rays can undergo total reflection; below that glancing angle the surface under test totally reflects the incident X-rays. A note adds that if the angle of incidence is small enough the X-rays are reflected without passing through the sample, and the angle of incidence at the boundary between the refracted and reflected rays is called the critical angle.

3.4 Angle scan: the measurement of the emitted fluorescence signal as a function of the glancing angle.

4 Principle of the method

4.1 Monochromatic X-rays from an X-ray source strike the mirror surface of the wafer at a glancing angle below the critical angle and undergo total reflection. The evanescent wave passes into the wafer surface, excites the atomic levels to fluorescence levels and produces characteristic X-ray fluorescence spectra corresponding to the atomic numbers present; this energy dispersive spectrum is collected by a solid state detector such as a silicon-lithium detector. The evanescent wave decays exponentially in the process and the decay strength depends on the total electron density of the wafer surface or of its native surface oxide. A note states that for silicon wafers of all resistivity ranges the exponential decay length is about 5 nm. The principle is shown in Figure 1, whose key identifies the monochromator, the detector, the sample stage with free movement in X, Y and Z, and the X-ray source.

4.2 A calibration sample with a certified areal density of a given element above 10 to the eleventh atoms per square centimetre is used, the integrated count rate under the fluorescence peak being linear with the certified areal density. The calibration sample carries at least one element of known areal density within the measurement area; the total reflection X-ray fluorescence spectrometer analyses it and obtains the integrated fluorescence count rate corresponding to the known areal density. One or more samples are then measured under the same conditions, and the relative sensitivity factor (RSF) associated with each certified element allows the integrated fluorescence count rate of the elements in the sample under test to be determined. If the X-ray source is changed, a different set of RSF values shall be used.

5 Interfering factors

5.1 Factors of the TXRF method. The choice of glancing angle should take account of the main type of metal contamination on the sample surface, that is whether the contamination lies in the oxide layer, including the native oxide, or as particles on the surface; strictly the two types call for different glancing angles, and for particle contamination too low a glancing angle introduces a large error. Deviations in the RSF of a fluorescence line introduce a bias in the result. If the X-ray beam is diffracted by the wafer under test and the diffracted beam enters the detector, it excites metal in the detector window or in the detector itself and produces instrument peaks that affect the result. Where the fluorescence signal is not linear with the areal density of the impurity, detector dead time can arise at high total count rates. The degree of smoothing of the fluorescence curve affects the accuracy of the measured value. For elements below atomic number 16, such as sodium, magnesium and aluminium, the detection limit is higher, usually more than 10 to the eleventh atoms per square centimetre and sometimes higher still.

5.2 Factors of the TXRF equipment. The interferences known in X-ray fluorescence spectrometry all apply, including but not limited to overlapping fluorescence lines, overlap of escape peaks and sum peaks, calibration drift of the energy gain, stability of the X-ray source and instrument background peaks; the equipment is not required to correct for secondary fluorescence or matrix absorption. Common interferences from the software and the calculation can be estimated by comparing data systems, as described in A.2. The detection limit depends on atomic number, excitation energy, photon flux of the exciting X-rays, instrument background, integration time and blank value; a note adds that for constant instrument parameters the interference-free detection limit is a function of atomic number and varies by more than two orders of magnitude, and refers to Annex A for the relationship between repeatability and detection limit. If the calibration of the glancing angle is not repeatable, variability is introduced; if it is incorrect, a bias is introduced. Mechanical vibration lowers the energy resolution of the detector and may affect the detection limit. Different targets suit different elements, and whether the target is fixed or rotating also affects the detection limit. The vacuum level in the X-ray target chamber affects the accuracy. To prevent particle contamination and its accumulation, the nitrogen used to purge the main chamber shall meet the requirement for high purity nitrogen in GB/T 8979-2008, otherwise the chamber may be contaminated and the result affected.

5.3 Factors of the sample surface. Different glancing angles shall be set according to the type of metal contamination, since particles standing proud of the surface and particles inside the oxide layer call for slightly different angles; where the angle scan of the known element on the calibration sample differs from that of the element on the sample under test, for instance particle contamination on the sample against metal in the native oxide on the calibration sample, a bias in the measured value is introduced. Differences in surface roughness and waviness can also interfere: a surface that has not been chemically and mechanically polished lowers the detection capability, shifts the value and increases the variability, and differences in roughness or waviness produced by different cleaning processes can also interfere. A note states that when polished sapphire wafers are measured with an atomic force microscope the surface roughness Ra within a 5 µm x 5 µm area is not more than 3 nm, and that the effect of the surface roughness of other materials on the semi-quantitative TXRF result has not yet been determined. A bias in the certified areal density of the element on the calibration sample leads to a bias in the areal density measured by TXRF. Surface contamination introduced during handling or measurement biases the result where the contaminating element is one of the elements measured. Non-uniform surface contamination can give different results at different positions, which matters in particular when the method is compared with others; a note states that the method is non-destructive and complements other methods, and refers to B.1 for the comparison.

6 Test conditions

Testing shall be carried out at a temperature of (23 +/- 5) °C, with a temperature fluctuation during the test of not more than 2 °C; at a relative humidity of not more than 60 percent; in air of a cleanliness class not lower than class 5 of GB/T 25915.1-2021; and with the instrument placed where there is no appreciable vibration.

7 Apparatus and data system

7.1 The TXRF spectrometer shall provide a monochromatic X-ray source, a handling device for the sample under test, an energy dispersive X-ray detector, software for background subtraction, peak integration, RSF calculation and analysis, developed by the instrument manufacturer and stored in the instrument computer program, and an argon-free analysis environment such as a vacuum of 1.33 Pa or helium. It shall provide a method for calibrating the glancing angle and an attenuation routine for subtracting escape peaks so that the escape peak signal can be removed.

7.2 The data system for the calibration samples shall use statistical tools to study and confirm the repeatability of the instrument. Repeatability and detection limit are covered in A.1.

8 Samples

The sample surface shall be flat and clean. The surface to be measured shall have been chemically and mechanically polished to a mirror finish, or shall be a mirror finish carrying an oxide layer.

9 Calibration

9.1 The method used to certify the calibration elements and their areal densities shall be agreed between supplier and purchaser, as described in B.2. Certification measurements for the other elements detected by the TXRF spectrometer shall be made through relative sensitivity factors determined beforehand and stored in the instrument program. The RSF is a function of the X-ray source, the atomic number of the fluorescing element and the fluorescence energy level, so a different set of RSF values shall be used if the X-ray source is changed.

9.2 The calibration procedure is as follows. The calibration sample is loaded into the TXRF spectrometer. The same X-ray source voltage, current and glancing angle as for the sample under test are selected and the TXRF spectrum of the certified element on the calibration sample is measured; the measuring time and the environment of the analysis chamber, such as vacuum, gas, nitrogen or helium, may differ between calibration sample and test sample.

9.2.3 The glancing angle is set and calibrated either from Formula (1) or from the angle scan curve of the calibration sample. The symbols of Formula (1), which gives an approximate value of the critical angle for total reflection, are: phi c, the critical angle for total reflection, in degrees; rho, the density of the material of the sample surface, in grams per cubic centimetre; E, the excitation energy of the X-rays, in kiloelectronvolts; and pi, the ratio of the circumference of a circle to its diameter. Alternatively the angle scan curve obtained from the sample may be compared for similarity with the curves of Figure 2, and a glancing angle equal to 25 percent to 80 percent of the critical angle set according to the main type of metal contamination in the sample. A note gives silicon as an example: if the angle scan curve resembles curve a of Figure 2, for residues, the contamination is of the particle type and a glancing angle below 85 percent of the critical angle is chosen; if it resembles curve b, for contamination within a film, a glancing angle of 70 percent to 80 percent of the critical angle is chosen; and where both types are present in practice, the crossing point of curves a and b is chosen. The key to Figure 2 identifies curve a as residues, curve b as contamination in a film, curve c as the substrate, phi as the critical angle, and the plotted points as an electroplated or sputtered nickel sub-monolayer, an evaporated nickel salt solution, and the silicon substrate.

9.2.4 Wafers on which no element contamination is measured serve as blank samples, each wafer free of a particular element of interest forming part of a series of element-specific blanks. For all elements of interest, three TXRF measurements are made under the same test conditions to verify that there is no instrument background signal, which then forms their common blank. Figure 3 gives an example of a blank for all elements except sulfur.

9.2.5 The fluorescence signal of the certified element measured by the TXRF spectrometer is integrated and the background subtracted to give the net integrated count rate and thus the areal density of the corresponding element. The background may be subtracted by a conventional deconvolution routine or by a suitable linear routine.

10 Test procedure

The TXRF equipment is switched on. The analysis conditions for the sample under test are selected and recorded: X-ray source voltage; X-ray source current; type of target; glancing angle; integration time; laboratory environment; and the measuring positions on the sample. The sample is loaded, its TXRF spectrum is recorded, and the net integrated count rate is calculated for the element peaks detected.

10.6 Using the calibration sample data for the particular calibration element s and the RSF values of the other elements m detected, the areal density of each element detected on the test sample u is calculated from Formula (2). The symbols are: Du,m, the areal density of element m on the surface of test sample u, in atoms per square centimetre; Fs,m, the RSF of calibration element s relative to element m; CPSu,m, the integrated count rate measured for element m on the surface of test sample u; CPSs, the integrated count rate of calibration element s on the surface of the calibration sample; and As, the areal density of calibration element s, in atoms per square centimetre.

11 Precision

A single calibration sample with a nickel areal density of 2.80 x 10 to the twelfth atoms per square centimetre was measured five times in the same laboratory, using a tungsten target X-ray fixed anode, a glancing angle of 0.05° and an integration time of 300 s. The within-laboratory mean obtained was 2.87 x 10 to the twelfth atoms per square centimetre, the standard deviation 4.23 x 10 to the tenth atoms per square centimetre and the relative standard deviation 1.48 percent. Repeated measurement of the nine elements potassium, calcium, titanium, chromium, manganese, iron, nickel, copper and zinc under the same conditions gave standard deviations from 0.47 x 10 to the tenth to 3.21 x 10 to the tenth atoms per square centimetre, with a maximum relative standard deviation of 44.6 percent. There is no absolute standard for the accuracy of the method, so the bias cannot be evaluated. Reference information on precision and bias is given in B.3.

12 Test report

The test report shall contain information on the test sample; information on the calibration sample; the type of equipment, including model and manufacturer; the anode material; the X-ray source voltage; the X-ray source current; the type of target; the glancing angle; the integration time; the laboratory environment; the test results; the operator and the time of analysis; the number of this document; and any other items.

A Annex A (informative) Repeatability and detection limit

A.1 The consensus method for determining the photon spectral detection limit without subtraction of instrument peaks is given by Formula (A.1), whose symbols are: CL, the spectral detection limit, in atoms per square centimetre; sb, the standard deviation of the blank measurement, in atoms per square centimetre; and S, the sensitivity. The factor 3 chosen in that formula suits a strict one-sided Gaussian distribution and corresponds to a confidence level of 99.6 percent, although experience shows that at low concentrations a non-Gaussian distribution is more likely. For short-term measurements the standard deviation of the blank is usually assumed to be given by the Poisson statistics of the photons, which leads to the detection limit equation commonly reported in the technical literature, Formula (A.2); its symbols are: CL, the spectral detection limit, in atoms per square centimetre; As, the atomic areal density of the reference material, in atoms per square centimetre; B, the background counts, in atoms per square centimetre; and Cm, the net signal, in atoms per square centimetre. This detection limit rests on the key assumption that the standard deviation of the blank comes only from the Poisson statistics of the X-ray photons, with no other meaningful contribution to variability, an assumption valid only for a short-term estimate. For a long-term estimate the standard deviation of the blank comes not only from Poisson statistics but also from other sources of variability, which may include but are not limited to the calibration of the glancing angle and the divergence of the X-ray beam, so a long-term estimate of the detection limit is larger than a short-term one.

A.2 When the operating conditions of a measuring system are being established, the values of a reference sample may be compared with the values measured on the equipment under test; for instance, measurement data for several elements can reveal interferences that may come from the instrument software and the calculation. The comparison uses a standard wafer with a referee data set (RDS), each point of which is the mean of several repeated measurements. The standard wafer is measured on the equipment being evaluated to give a sample data set (SDS), and the difference between the two is calculated with Formula (A.3), whose symbols are: RDS, the referee data set of the standard wafer; SDS, the sample data set obtained on the equipment being evaluated; and DDS, the difference between the measured data and the referee data. The DDS contains a good deal of information; the simplest acceptance criterion is the maximum difference, that is the largest absolute value in the DDS, which represents the worst disagreement between the machine under test and the referee data. If that maximum difference is smaller than the value agreed between supplier and purchaser, the equipment is accepted for use. More elaborate calculations may also be used, for example element-by-element histograms of the DDS values together with statistics such as the mean and standard deviation, which can be compared with limits specific to the application and used to gain insight into the nature and origin of the differences. The parameters compared between the test wafer and the system under test, and the acceptable size of their difference, are agreed between the users.

B Annex B (informative) Comparison of TXRF with other surface metal test methods, and calibration of calibration samples

B.1 TXRF complements five other methods. Electron spectroscopy for chemical analysis has a surface areal density detection limit of 10 to the thirteenth atoms per square centimetre. Auger electron spectroscopy is listed with a surface areal density detection limit whose exponent is corrupted in the source text and is therefore not reproduced here. Nitrogen beam Rutherford backscattering has a detection limit of 10 to the tenth atoms per square centimetre for some elements but cannot separate heavy elements of adjacent atomic number. Secondary ion mass spectrometry can detect elements of lower atomic number over a range from 10 to the eighth to 10 to the twelfth atoms per square centimetre, but does not give an adequate detection limit for the transition elements between titanium and zinc, of atomic number 22 to 30, and is destructive. Vapour phase decomposition atomic absorption spectrometry has a range for surface metals from 10 to the eighth to 10 to the eleventh atoms per square centimetre, but gives no useful spatial information, takes longer than TXRF and is also destructive.

B.2 The calibration sample is a chemically and mechanically polished silicon substrate of current process quality with a surface metal areal density from 10 to the twelfth to 10 to the fourteenth atoms per square centimetre. The K-alpha fluorescence signal obtained from the certification standard is free of interference, that is free of escape peaks, sum peaks and the fluorescence peaks of other contamination, and there is no external source of contamination. Elements that easily accumulate as contamination, such as iron, or that diffuse into the depth over time, such as gold and copper, should not be the first choice; nickel or vanadium may be preferred. The certification uses a set of silicon wafers carrying known areal densities of different specified elements.

B.2.2 Four suitable certification methods are described. Nitrogen beam Rutherford backscattering spectrometry (N-RBS), with which no other metal within a range of +/- 5 atomic mass units may exceed 1 percent of the areal density of the certified metal; the N-RBS measurement is absolute and shall be made within the TXRF area, and other backscattering methods such as forward scattering Rutherford backscattering (F-RBS) and heavy ion backscattering spectrometry (HIBS) may also be used to certify the areal density. Vapour phase decomposition with atomic absorption spectrometry (VPD/AAS), which can certify a spin-coated contamination and is destructive; the contamination is shown by TXRF or SIMS mapping to be uniformly distributed over the whole wafer, the certified value rests on the atomic absorption spectrometry standards, and the accuracy of the method comes mainly from the VPD element recovery. Implantation, in which the reference element is implanted into a pre-existing amorphous silicon surface, which is then converted to single crystal silicon by solid phase epitaxial growth; if the reference element is more soluble in amorphous than in single crystal silicon, the process sweeps the implanted element to the sample surface, and the areal density is quantified from the ion dose determined by the implantation current. Diluted atomic absorption standard solutions, in which some diluted metal atomic absorption standard solution is deposited on a polished silicon substrate to form a local spot smaller than the TXRF analysis area; the method assumes that no certified metal at all is lost while the solution dries, TXRF must be able to find the deposited spot readily so that the whole dried deposit is analysed, and the certified value is determined by AAS. The deposited solution should produce one of two TXRF angle scans: either a metal fluorescence count rate that, as a function of angle, is independent of the glancing angle below 80 percent of the critical angle, as in curve a of Figure 2, or one that shows the character of metal contamination lying within 3 nm of the surface, as in curve b of Figure 2.

B.3 The precision and bias data of ASTM F1526-2000 are reproduced. Precision was evaluated with monochromatic TXRF equipment in two round robins. The within-laboratory precision was evaluated on 2 reference samples and 6 unknown samples with surface contamination of iron, nickel, copper and zinc from 10 to the eleventh to 10 to the twelfth atoms per square centimetre, with 13 laboratories taking part; each laboratory measured the samples once a day for 4 days. The conditions were a rotating anode X-ray source with a tungsten target, a LiF200 monochromator selecting the 9.67 keV line, 30 kV, 200 mA, a glancing angle of 0.05°, a vacuum of 1.33 Pa, an analysis area 10 mm in diameter and an integration time of 1000 s. Each laboratory reported the mean of 4 readings and their standard deviation; at 95 percent confidence the within-laboratory relative precision was 28 percent, that is 2.8 times a within-laboratory relative standard deviation of 10 percent. The between-laboratory precision was evaluated with 1 set of reference samples and 3 unknown samples, one of them a blank, the reference and unknown surfaces carrying nickel contamination from 10 to the eleventh to 10 to the twelfth atoms per square centimetre, with 17 organisations taking part and each laboratory measuring the samples over several days. The conditions were a rotating anode X-ray source with a tungsten target, a LiF200 monochromator selecting the 9.67 keV line or higher, 30 kV, 200 mA or more, a glancing angle of 0.1°, an analysis area 10 mm in diameter and an integration time of 1000 s. At 95 percent confidence the between-laboratory precision was +/- 8 x 10 to the tenth atoms per square centimetre at a mean of 15 x 10 to the tenth, and +/- 20 x 10 to the tenth atoms per square centimetre at a mean of 45 x 10 to the tenth. Other analytical conditions may be used, but their precision has not been evaluated. There is no absolute standard for the accuracy of the method, so the bias cannot be evaluated.

B.4 The method uses X-rays and protection shall be provided for persons exposed to them. It is particularly important to protect the hands and fingers from direct irradiation and the eyes from secondary scattered radiation. Film badges or dosimeters are recommended, together with periodic checks of the dose to the hands and other parts of the body using a Geiger-Muller counter calibrated against a standard nuclear source. For persons whose whole body is irregularly exposed to external X-rays of quantum energy not exceeding 3 MeV, the current maximum permissible dose is 1.25 R per quarter, that is 3.22 x 10 to the minus fourth C/kg, equivalent to 0.6 mR/h or 1.5 x 10 to the minus seventh C/(kg.h). Under the same conditions the maximum permissible dose for exposure of the hands and forearms is 18.75 R per quarter, that is 4.85 x 10 to the minus third C/kg, equivalent to 9.3 mR/h or 2.4 x 10 to the minus sixth C/(kg.h). Other government and regulatory bodies also have their own safety requirements.