Page images
PDF
EPUB

NBS Special Publication 400-23, ARPA/NBS Workshop IV, Surface Analysis for Silicon Devices, held at NBS, Gaithersburg, Maryland, April 23-24, 1975 (Issued March 1976)

SOME EFFECTS LIMITING SIMS DEPTH PROFILE ANALYSIS

AND METHODS FOR IMPROVEMENT

Robert K. Lewis

Cameca Instruments, Inc.
Elmsford, New York 10523

INTRODUCTION

Secondary ion mass spectrometry (SIMS) has proven to be a very attractive method for obtaining surface and bulk concentrations of materials used in semiconductor device technology. The popularity of SIMS has resulted primarily from the fact that the method has the sensitivity necessary to detect doping level concentrations, namely 10+15 to 10+17 atoms/cm3, and the ability to produce depth profiles at these concentration levels. 1,2 The disadvantage of being a destructive method is readily offset by the inherent depth analysis capability which results from the sputtering action of the primary beam. It is this depth profiling capability, in fact, that has provided the principal interest in SIMS for analyzing semiconductor material and the technique is now widely used for characterizing these materials. However, there are problems with SIMS depth profiling which must be taken into consideration. This paper discusses some of the effects which have been encountered and tend to limit the usefulness of the technique and describes methods to minimize them. These effects have been summarized in Table I.

DEPTH RESOLUTION

To achieve good depth resolution it is obvious that the surface of the crater bottom must be maintained as flat as possible. With normal or non-normal incidence of the primary beam a high degree of flatness can be easily maintained by rastering the beam. Schematic diagrams of craters produced by non-rastered and rastered beams are shown in Figure 1 and Figure 2. Figure 1 shows that with a non-rastered beam, 250μm in diameter, incident at 45° (this is approximately the bombarding angle in the Cameca IMS 300 instrument used for this study), and a gaussian distribution, the flatness variation over a typical selected area (50μm) is approximately 200 Ă. This is adequate for many depth profiles. However, if the beam is rastered as shown in Figure 2, the flatness over the area selected for analysis can be greatly improved. We have established that the variation in flatness can be less than 40 Å over an area of 100μm to depths of about 1,000 Å by bombarding amorphous Ta205 and

using the technique of observing the variation in color of fringes produced in white light.1 This would give a value of less than 5% for the variation in depth divided by the total depth sputtered. This value is consistent with the R value found by Werner3.

It is clear that ions sputtered from the crater wall must be rejected to have good depth resolution. This is easily accomplished in the Cameca ion microscope by mechanically aperturing in an image plane or in an ion probe by electronic gating (turning off the secondary detector when the primary beam is on the crater wall). Mechanical aperturing is shown schematically in Figures 1, 2, and 16. With electronic gating the primary beam must be kept smaller than that required with mechanical aperturing. This tends to limit the maximum average beam density obtainable and consequently the maximum sputtering rate available with ion probes.

If the sample surface does not sputter away evenly because of a lateral variation in sputtering rate, the depth resolution will suffer accordingly. Lateral variation in sputtering rate results from different lattice orientations encountered by the primary beam from grain to grain in polycrystalline material or from different lattice structures when different phases are encountered. For this reason the depth resolution attainable in such material is usually only ten to twenty percent of the total depth sputtered. Continuously rotating the sample while sputtering could reduce the lattice orientation effect; however, no commercial instruments presently provide this capability. When lateral differences in orientation or in phase are encountered the use of an oxygen leak can sometimes minimize the variation in sputtering rate.

Channeling (lattice effects) in SIMS have been described by Slodzian and Bernheim4. This is a variation in ion yield that occurs when the incident beam traverses transparent and opaque directions in the lattice. A plot of secondary ion intensity versus the angle of rotation is shown in Figure 3. Fortunately, this effect is only a second order effect so it is generally not a problem in depth profiling. However, it is important to be

aware of the fact that it can cause errors especially when comparing samples. A method of minimizing the effect is to bombard the surface with a stream of oxygen (oxygen leak). This produces a less severe variation in ion intensity with orientation as shown by the dotted line in Figure 3. Slodzian has pointed out that the effectiveness of the oxygen leak is determined by whether the surface oxide formed is crystalline or amorphous.

"Knock on" is the redistribution that occurs from the physical driving in and sub atomic mixing of the atoms being analyzed. The effect has been shown by McHugh5 analyzing Ta205 which had an approximately 50 Ă thick phosphorous layer located 230 Å below the surface. His results are shown in Figure 4. The effect is significant only at high primary bombardment voltages (above about five kilovolts). The effect would be less, of course, at a higher angle of incidence. angle of incidence can be changed for nonnormal incident primary beams by changing the polarity of the primary beam. The combinations positive primary-positive secondary and negative primary-negative secondary give the highest angles of incidence.

DETECTION SENSITIVITY

The

One of the most serious problems encountered in SIMS is the mass interference from the complex polyatomic species produced by the secondary ion process. These species are formed from the clusters of the matrix atoms formed by the sputtering action of the beam (e.g., M, M2, M3,...) and combinations with the primary ion when a reactive gas such as oxygen is used (e.g., MO, MO2,...M20, M202, ...etc.).

Typical spectra produced when bombarding silicon with a reactive gas (oxygen) is shown in Figure 5 due to Evans6. It can be seen that the problem of mass interference is more severe at the higher masses. This is just the opposite of the ion production process in spark source mass spectrometry (SSMS) where the interferences are more concentrated in the lower mass region of the mass spectrum. This is because in the spark source the energy available is much higher which breaks up the clusters and creates many multiply charged ion species.

The simplest method for minimizing mass interference from the polyatomic species is to take advantage of the large differences in energy distribution between the polyatomic and monatomic species. These distributions are shown in Figure 6 which is from Satiewicz7. It can be seen that the complex

polyatomic species can be rejected perferentially over the monatomic species by rejecting the low energy ions.

The low energy ions may be rejected in the Cameca instrument by using the low energy discriminator (L.E.D.) shown in Figure 7 or by appropriate adjustment of the electrostatic analyzer when the double focusing arrangement shown in Figure 8 is used. Rejection of the low energy ions with the ESA is possible because the double focusing geometry, used has a first order direction focus position such that slits can be used to precisely define the energy limits accepted. The SIMS instruments designed by 8 Herzog et. al. (GCA) and Tamura et. al. 9 (Hitachi) have this feature while the ion probe SIMS instruments described by Liebl (ARL) 10 and Banner et. al. (AEL)11 do not.

The attenuation of the silicon polyatomics in the spectra shown in Figure 5 with increasing attenuation of the parent monatomic peak using the L.E.D. on the Cameca instrument is shown in Figure 9. A typical application of the L.E.D. is shown in Figure 10. Here the detection of arsenic using the As0 species has been improved by one order of magnitude. An equally important use of the L.E.D. is the removal of tails on the low energy side of a mass peak. The presence of these tails can cause a serious loss in abundance sensitivity (and therefore detection sensitivity) on the low mass side of intense peaks. This is shown for the detection of copper and zinc in gallium arsenide in Figure 11. These tails result from the post ionization of the neutral species leaving the sample. This "kinetic" emission process has been described by Blaise and Slodzian12. It is not necessary to attenuate the intensity of the peak to eliminate the low energy tails with the L.E.D.

A second and more elegant method for eliminating interference from polyatomic ion species is the use of high mass resolution. The method of obtaining high mass resolution on the Cameca instrument is shown in Figure 8. Here the normal imaging mode path, that is prism-mirror prism, is interrupted and the beam is allowed to pass through a hole in the mirror (the mirror being displaced) into a sperical electrostatic analyzer (ESA). This combination of an ESA with the first magnetic deflection is an inverted NierJohnson double focusing configuration. this analyzer mass résolutions over 5000 have been achieved. It is possible with high mass resolution to completely separate the inter

With

fering polyatomic mass peak 27A12 from the 54Fe peak as shown in Figure 12 because of

the mass differences that occur.

Some

[blocks in formation]

A second effect limiting detection sensitivity is the contamination of the surface from the ambient gas molecules in the vacuum system. At pressures around 10-7 to 10-8 torr, which is the typical vacuum level in commercial SIMS instruments, it can be readily demonstrated that the low current densities present in the beam periphery produce an artificially high background signal. This is because the influx of ambient gas molecules now competes with the primary beam and react with the surface where they are sputtered away producing various interfering ion species. Low beam densities will always be present at the periphery of an ion probe. In addition there is a significant flux of neutral atoms (neutralized ions) over a large area outside the bombarded region that also sputter the target surface and produce ions 13 in from the ambient gas as shown by McHugh Figure 14. A cold plate near the sample surface will minimize the hydrocarbon contamination, but it will not eliminate it and it will have little or no effect on oxygen, nitrogen and CO2. The effect is shown dramatically using the Cameca instrument by bombarding aluminum with a beam of nonreactive ions (Art). By keeping the beam diameter less than the field of view (225μm) it is possible to see the effects of low primary beam density at the periphery of the beam on the fluorescent screen as shown in Figure 15. At the periphery there is a marked increase in the secondary ion intensity due to an ion yield enhancement ical effect) from the reactive ambient gas. By mechanically aperturing as shown in Figure 16 and selecting areas approximately 50μm in diameter at the center (Area A) and the periphery (Area B) the spectra shown in Figure 17 were produced. It is obviously impossible to achieve a low background signal without rejecting this ambient contribution. The only methods that have been demonstrated to be effective in eliminating the ambient contamination are lowering the vacuum below 10-9 Torr as Benninghoven14 has shown which unfortunately is impractical in commercial SIMS instruments or the ions produced outside the area of high beam density can be eliminated by the mechanical aperturing technique. However, there is no way to avoid this effect with the electronic gating technique used with ion probes.

(chem

[blocks in formation]

The most serious effects limiting the accuracy of a depth profile are those limiting the uniformity of the ion yield of the element being measured or a reference element (e.g. a matrix element) used for normalization. Any marked change in concentration of a reactive element such as boron, carbon or oxygen can change the ion yield of elements associated with it. This effect is shown in Figure 18 where phosphorous enhances the yield of the tantalum ion. It is important to determine if such species are present by in-depth mass spectral analysis whenever possible, and, if present, to measure the depth profile of the reactive species simultaneously with the element of interest. This also applies to any element being measured for normalization purposes. Sometimes it is useful to check for the presence of a reactive element by using an inert gas as the primary beam (e.g. argon) which does not enhance the matrix greatly increasing the contrast.

Whenever a reactive primary beam is used there can be a marked change in the ion yield over the first few hundred angstroms of sputtered material (15kV bombardment) due to the implantation effect of the primary beam as described by Lewis et. al.15. The effect is shown in Figure 19 where pure silicon was bombarded with 0- primary ions and positive secondary ions of oxygen and silicon measured together. This ion yield variation results from the fact that the implanted primary oxygen ion is first concentrated at a depth below the surface as shown schematically in Figure 20. Until the sputtering front reaches this depth there will be a variation in ion yield of all elements which are enhanced by the primary ion. The effect is minimized by keeping the bombardment voltage low (below 5kV) and can be essentially eliminated by the use of an oxygen leak as shown in Figure 21.

Variations in sputtering rate with depth can cause errors in depth measurements if the variations are not taken into consideration in the sputtering rate calibrations used to establish the depth scale. Often measurement of a matrix species and normalization against this signal will take care of small sputtering rate variations due to beam density changes. However, variations due to phase changes in the sample must be determined by depth measurements outside the instrument. If sufficient thicknesses of the material in the individual phases are available (a few hundred angstroms) it is possible to make their depth calibrations with a Talystep. larger thicknesses (a few thousand angstroms) are available it is more convenient to make the measurements with a light microscope equipped with a Michelson interferometer.

If

Fortunately this ion mobilization phenomena appears to be a problem only for the case of sodium in silicon dioxide. No other documented instances are known to this author.

References

It is not unusual to encounter a situation where what appears to be a variation in depth (shown for aluminum in the thin film depth profiles shown in Figure 22) is actually a variation in lateral distribution as shown in the ion image of Figure 23. This variation may be real or a result of variations in the lateral sputtering rate. The lateral distribution effect is observed most often at metal - metal oxide interfaces or thin film metallizations. It is difficult to minimize this effect. However, its presence can usually be taken into account by observing a series of ion images with depth.

1.

૧.

3.

4.

5.

6.

One of the major concerns in all depth profiling work is whether the element of interest is being redistributed by the bombardment process. This is the case of the "knock on" phenomena described above, which was described as a second order process. The only first order redistribution effect observed to date has been for the analysis of sodium in silicon dioxide layers. This redistribution occurs whenever a potential is applied to the surface. The sodium migrates (within milliseconds) to the Si02/Si interface or to the sample surface depending on whether the potential applied at the sample surface is positive or negative as shown in Figure 24. It is questionable whether an accurate profile of sodium in Si02 can be obtained at all since even neutral atom bombardment will release secondary electrons and charge the surface. The surface to substrate potential must not exceed a few tenths of a volt while the profile is being measured.

[ocr errors][merged small]

Morabito, J. M., and Lewis, R. K., Anal. Chem. 45, 869 (1973).

9.

Kofkner, W. K., Werner, H. W., Oosthoek, D. E., and de Grefte, H. A. M., Rad. Effects 17, 88 (1973).

Werner, H. W., Acta Electronica 18, 51 (1975).

10.

Slodzian, G., and Bernheim, G., Int. J. Mass Spectrom. Ion Physics (1975).

McHugh, J. A., Rad. Effects (1973).

Blattner, R. J., Baker, J. E., and Evans, Jr., C. A., Anal. Chem. 46, 2171 (1974).

Satkiewicz, R. G., Air Force Technical Report, AFAL TR-69-322 (1970).

Herzog, R. F. K., Poshchenrieder, W. P., Ruedenauer, F. G. and Satkiewicz, F. G., Proc. Fifteenth Annual Conf. Mass Spectrometry and Allied Topics, Denver, Colorado, May 1967, p. 301.

Tamura, H., Kondo, T. and Doi, H.,
Advances in Mass Spectrometry, Quayle,
Ed., (Inst. of Petroleum, London,
1971), Vol. V., p. 441.

Liebl, H., J. Appl. Phys. 38, 5277 (1967).

Banner, A. E., and Stimpson, B. P., Vacuum 24, 511 (1975).

[blocks in formation]

11.

15.

McHugh, J. A., to be published.

Lewis, R. K., Morabito, J. M., and Tsai, J. C. C., Appl. Phys. Lett. 23, 260 (1973).

« PreviousContinue »