by sweeping the magnetic field. Ion images of a preselected mass are obtained by rastering the primary beam on the sample and using the photomultiplier current output to modulate the brightness of a CRT by synchronizing the raster with the primary beam.

Figure 8 is a photograph of the ARL-IMMA instrument. This particular one is the instrument in our laboratory at Texas Instruments. The duoplasmation and primary magnet are to the right in the main console and the secondary ion analyzer system is to the left. The data collection system is interfaced to a Texas Instruments 960A computer (installed in the top rack on the left console). The 960A can drive a digital plotter which has been used to generate most of the data figures in this presentation.

This introduction to the instruments, which is by no means complete, is only intended to cover the general principles; let us go on to their application to surface analysis. No review of secondary ion mass spectroscopy would be complete without showing Anderson's famous figure 9 depicting the ion yield comparison between the kinetic and chemical ionization processes. The curve on the left is the ion yield of Al as a function of sputter time (or depth). Note that the ion yield falls off dramatically as the surface oxide is sputtered away. The curve on the right is a repeat of the same experiment + using 02 as the sputter beam.

Here the

surface yield is high; then the yield takes a small dip, and then returns to the high surface point after about 40 seconds of sputtering. The conclusion is that in both cases the surface yield is primarily the result of chemical ionization (oxides) and in the Art case the mechanism quickly converts to kinetic ionization. For the 02 beam the dip is related to a combination of kinetic and chemical ionization which becomes completely chemical when the sputtered depth reaches the original primary ion implant depth. This experiment has lead to the widespread use of oxygen as the primary ion sputter beam while maintaining a relative high partial pressure of oxygen near the sample surface.

Figure 10, also from Anderson's work, shows the relative intensities of secondary ions obtained from pure metals using an 0 sputter beam. From this table Al appears to be the most intense, and noble metals represented by Au the least intense. This chart, even though taken from the pure metals, does correlate moderately well with the sensitivities determined for these elements in the silicon or silicon oxide matrix. However, for bulk type analysis in silicon, the ordinate is

from 1 ppb to 10 ppm going from top to bott This illustrates that secondary ion mass spectroscopy is a very sensitive technique even when material quantities are small, as is the case in surface analysis.

Figure 11 is a series of spectra taken from i silicon surface with identical primary 02+ beam density, with the analyzed surface depti controlled by either rastering the primary beam or moving the sample. The calculated sputter rate for these spectra was about 50 A/sec. As can be seen in these spectra, by far the most sensitive method in surface analysis is to move the sample and thereby constantly expose a fresh area to the primar beam. In this spectrum the carbon, hydrocarbon fragments, sodium, potassium and calciu peaks are quite pronounced. The 400 micron raster spectrum does show the same peaks but not nearly so intense. Also, note that the overall intensity is reduced, which tells us that the whole spectrum was recorded in the region between the surface and primary ion implant depth. The 50 micron and stationary spectra both suggest the surface is quite clean with only traces of Na and K appearing. I have not labeled the peaks on these spectra since the point I wish to make is the number of peaks and their relative intensity. However, if you care to do some quick mental exercises, I will remind you Si has 3 isotopes at mass 28, 29, and 30, and Si related peaks will always appear as groups of three to five peaks. Also, you may note that the Si02 group at mass 60 (spectra shifted to the right) are prominent in the 50 micron and static beam spectra. This, along with the higher intensity, tells us that the spectra were recorded after reaching the implant depth. This molecular spectra is characteristic of structural chemical bonding and has been investigated by several of my European colleagues for surface fingerprint spectra. Other techniques such as decreased beam intensity or faster recording could have been used to insure the surface wasn't sputtered away before the entire spectra was recorded, but it must be remembered that sensitivity is directly related to ions/sec delivered to the detector, and that a little integration goes a long way at the low end.

beam raster, the count acceptance area was restricted to the central 1/4 of the crater bottom. The depth assignments were made by measuring the total crater depth with optical interferometry and assuming a uniform sputter rate. This family of curves all show the same trend, a very intense surface peak rapidly falling off before the 50 angstroms depth, then increasing slightly as the primary ion implant depth is approached, and then taking a second sharp decrease to a constant level well above background. These shapes show that much of the surface Ag is knocked into the silicon by the primary ion beam, hence, the second rapid decline. This is bad if we were interested in a concentration depth profile, but actually beneficial in a surface analysis since it does contribute to detectability time for quantitative analysis. However, if we are struggling for quantitative analysis, which is every mass spectroscopist's impossible goal, we would like to minimize this knock-on effect even though it is only 1 part per hundred. Figure 13 is an attempt to reduce the knock on, or at least the depth, by lowering the primary ion energy. The general trend does show that the second sharp drop is indeed decreased as the energy is reduced. Only the 12 kV curve violates this trend and is probably in error. The actual depths for these curves were extremely difficult to measure and could easily be as much as 100% in error. One final point to note on these curves is that only the 20 kV curve shows a positive slope after the first surface drop. This results from the decreased sputter rate, with the lower energy beams allowing the ambient oxygen to keep a constant surface oxide throughout the analysis. Figure 14 is a closer look at the surface region as a function of primary ion energy. The set illustrates that even in the worst case at 20 kV, we will have collected about 90% of the integrated intensity before 50 angstroms have been sputtered away. Consequently, this illustrates that for quantitative surface analysis by secondary ion mass spectrometry, it will probably be for only one mass per area sputtered. If the area is limited such as on a small signal diode, this can be a severe limitation. However, on pilot slices, which encompasses most of my work, the problem is not limiting as it can be overcome by continual movement of the sample. Figure 15 is one last look at the curves as a function of primary ion energy. These are the same curves as Figure 13, but now the abscissa is plotted as a function of time. This set shows that for qualitative analysis, as much as 10 minutes was available to detect the Ag which was present at about 5% of a monolayer.

Figure 16 shows how quantitative the results shown back in Figure 12 were. The line is a least squares fit to the five points indicated. Each point represents the total integrated count under the respective curve, with the base line taken where the curve leveled off after the second sharp drop. This empirical calibration curve is very encouraging even though two of the low points are off by as much as 100%. No correction was made for ion pick-up efficiency by means of a matrixion and this is known to vary by as much as 50% from sample to sample. This study does show that given a set of standards, and I believe that bulk standards can be used, secondary ion mass spectroscopy surface analysis can be empirically quantitated, provided the surface impurity concentrations are not high enough to drastically alter the effective matrix which controls the ion yield efficiency. Figure 17 turns the topic over to the analysis of SiO2 surfaces, whereas up to now I have concentrated on silicon surfaces. This is a series of spectra taken under various conditions on a 10,000 angstrom thick oxide film. The oxide was not coated with a conductive film, as would be the normal practice for bulk or in depth analysis, since this would mask the surface of interest. As can be seen from this set of spectra the use of a positive beam on a naked sample gave only a few peaks, none of which are normal patterns expected from a SiO2 matrix. The absence of a spectra is the result of a high positive charge buildup at the point of impact by both the positive ion beam and loss of secondary electrons. Also, without a conductive surface coating there is insufficient bias on the sample surface to accelerate and direct the secondary ions to the pickup electrode. The negative beam on the naked sample does show the silicon peaks but they are very weak. This does show that the secondary electron loss is compensated by the negative primary ion which allows the point of impact to come to an equilibrium state. The absence of the remainder of the spectrum here is entirely due to the absence of the accelerating sample bias. The two spectra on the right were taken through the open areas on a 300 mesh electron microscope copper grid which was electrically connected to the 1500 volt sample bias. The primary beam was rastered rapidly over both the copper bars and the openings. The theory is that the primary beam bombarding the copper emitted sufficient low energy secondary electrons, which neutralized the positive charge on the Si02 surface and provided the necessary secondary ion accelerating potential as well. These spectra do show that the mechanism works; both positive and negative beams give

satisfactory spectra. However, these electroplate formed copper grids are not noted for their purity, so a much more pure grid must be employed to render meaningful results. Grids formed by e-beam evaporation of high purity Al or Au should be excellent. This discussion only applies to oxide films greater than 1500 angstroms thick. Thinner oxides behave the same as bare silicon. Therefore, when searching for the source of contaminants, such as furnace tubes, the oxide thickness should be kept below about 1200°C which just happens to be the gate oxide thickness of most MOS devices.

No presentation concerning oxide analysis would be complete without illustrating the mobile ion problem shown by Ron Baxter in Figure 18. These profiles were taken on the same sample using both a positive and negative beam. The conclusion is that using a positive beam, the Na all moves to the interface and with a negative beam, it is all drawn to the surface. The grid technique may help this problem as well, but to date the grid itself has been the dominant source of sodium.

To summarize this presentation, a look at secondary ion mass spectroscopy with respect to five criteria for surface analysis is in order.

1.

2.

Anderson, C. A., International Journal of Mass Spectrometry and Ion Physics 2, 61 (1969).

Fralick, R. D., and Roden, H. J., ASMS Committee VII Workshop, 22nd Annual ASMS Conference, Philadelphia, (May 1974).

4.

The sensitivity has been shown to be as good as a few ppm.

Hughes, H. L., Baxter, R. D., and Phillips, B. IEEE Trans. Nucl. Sci. NS-19, 256 (1972).