The results of calculations of absorbed dose as a function of electron-beam energy are shown in figure 24. The calculations were made for the case of the deposition of aluminum or chromium metallization on a 50-mm diameter silicon wafer covered with a 100-nm thick layer of silicon dioxide and located about 500 mm from the melt. The wafer subtends a solid angle of 0.01 sr. For each energy, the beam current was taken as the value necessary to maintain the beam power at 5 kW. A typical deposition rate of 3 nm/s was assumed and the process was continued for 333 s in order to deposit 1 μm of metal.

With this set of parameters, the absorbed radiation dose in the oxide due to both K and

α

and bremsstrahlung radiation was calculated as a function of beam energy. In the range 5 to 15 keV the absorbed dose was found to be essentially independent of energy; approximately 60 kGy (6 Mrads (SiO2)) was absorbed in the oxide film for aluminum evaporation and approximately 20 kGy (2 Mrads (SiO2)) for chromium evaporation. It should be noted that no significant dose reduction can be expected during aluminum evaporation by operating the electron gun at energies below 1.5 keV, where the aluminum K x-ray is not excited, because of the increase in the bremsstrahlung radiation at low energies.

α

To evaluate the shielding effect of the metal film, the x-ray dose in the oxide was evaluated under the assumption that the metal absorbed no energy. At a beam energy of 10 keV the aluminum film was found to reduce the absorbed dose in the oxide by a factor of 1.7 while the chromium film was found to reduce it by a factor of 3.9. The calculation of energy deposited per unit thickness of oxide indicated that the dose is essentially uniformly distributed throughout the oxide, being only approximately 10 percent higher at the metal-oxide interface than at the oxidesilicon interface.

The effect of absorbed dose on the metal deposition rate was calculated for several beam energies. The radiation dose for depositing 1 um of metal as a function of deposition rate is given for a beam energy of 10 keV and a beam current of 0.5 A in figure 25; the results of calculations at other beam energies were similar.

These calculations are based on the assumption of the deposition of equal thicknesses of aluminum and chromium for comparison. However, in actual practice, a chromium gate

layer of 100 to 300 nm would probably have about 1 μm of aluminum deposited over it. Assuming the depositions are carried out with a 10-keV electron beam at 0.5 A yielding a deposition rate of 2 nm/s, figure 26 illustrates both the dose absorbed in the oxide as a function of chromium thickness and the dose absorbed in the oxide with a 1-μm thick aluminum layer deposited over the chromium layer of thickness as shown on the abscissa. The shielding effect of the chromium is clearly illustrated.

The discrepancy between the results of these calculations and the preliminary experimental results is thought to arise from the sensitivity of the thermoluminescent dosimeters (TLDs) used to measure the absorbed dose to in-situ annealing effects caused by optical radiation and thermal effects due to the kinetic and condensation energies of the metal film being deposited. A temperature rise during irradiation causes a well known glow peak indicative of release of trapped electrons. The net reading after irradiation of TLDs in contact with a silicon wafer therefore corresponds to only a fraction of the actual dose deposited.

Therefore to test the accuracy of the calculations, a second experiment was performed in which polychlorostyrene and nylon film dosimeters, 50 and 60 μm thick respectively, were used to measure the x-radiation produced in the electron-beam evaporator. These films were exposed at the normal wafer location in the evaporator while a 10-keV electron beam impinged on aluminum. The beam intensity was reduced to 0.04 A while the crucible containing the metal was cooled to prevent any aluminum evaporation. Under these conditions, an x-ray exposure equivalent to that obtained in a typical metal deposition is obtained provided the total number of electrons (or total charge) incident on the metal is equivalent to that during metal deposition. Two groups of nine 1-cm2 polychlorostyrene and nylon films were exposed in pairs to increasing xray doses. Figure 27 shows the measured nylon film absorbance as a function of total electron charge incident on the metal. The measured absorbance must be corrected to account for the sensitivity of the film to optical radiation which is present in the evaporator chamber during metallization. The figure also shows the correction factor and the corrected results for nylon films. Similar results were obtained for the polychlorostyrene films which, however, required no correction for optical sensitivity.

The dose absorbed in the dyed films when a 10-keV electron beam deposits 100 C in aluminum was calculated. This charge incident on the metal at an appropriate current is typical of that necessary to deposit 1 μm of aluminum at a rate of about 3 nm/s. The calculalations yielded an absorbed dose of 3.9 kGy (0.39 Mrads (nylon)) and 4.5 kGy (0.45 Mrads (poly)) in the nylon and polystyrene films, respectively. This compares reasonably well with the experimentally measured values of 4.3 kGy (0.43 Mrads (nylon)) and 3.0 kGy (0.30 Mrads (poly)) and provides the basis for the deposited dose calculations in silicon dioxide films during the metallization process. Discrepancies can be explained in terms of errors introduced by the extrapolation of the aluminum bremsstrahlung below 2 keV.

The trends in x-ray dose due to the electronbeam metallization process illustrated by these calculations can be used in examining experimental results on device hardness. The deposition of chromium results in reduced dose when compared to aluminum deposited under equivalent evaporation conditions. However, since chromium sublimates, less electron beam injection power is necessary to deposit a chromium layer of the same thickness as an aluminum layer. This can result in a significant reduction in dose absorbed in the oxide. Several experimenters have noted that MOS devices fabricated with electron-beam evaporated chromium gates exhibit a higher degree of radiation resistance than devices fabricated with aluminum gates [65-68]. Gates formed by depositing aluminum over chromium result in a reduced x-ray dose in the oxide when compared to gates formed only of aluminum as illustrated in figure 26.

An attempt to reduce the electron-gun damage in the fabrication of aluminum gates by reducing the electron-gun beam energy to about 3 keV and reducing the deposition time to about 20 s has been reported [69]. The authors concluded that even under these conditions considerable damage to the oxide occurred. Figure 24 illustrates that reducing the electron-beam energy does not reduce the x-ray dose absorbed in the oxide; however, figure 25 indicates that reducing the evaporation time (increasing the deposition rate) can reduce the oxide dose.

(S. Mayo, K. F. Galloway, and T. F. Leedy)

5.3. Ion Implantation Parameters

This task is concerned with the development of methods for measuring critical parameters

associated with ion implantation. Study of Schottky barriers for use in measuring implanted profiles by the capacitance-voltage (C-V) technique was extended to the case of phosphorus-implanted silicon. As part of a comprehensive study of methodology for dose measurement, the characteristics of various current integrators were investigated.

Schottky-Barrier Capacitance-Voltage Technique The four metallization systems, aluminum, gold, titanium-gold, and molybdenumgold, previously studied on boron-implanted silicon (NBS Spec. Publ. 400-25, pp. 31-32), were studied on cleaned, natural (unetched) surfaces. Gold consistently yielded rectifying current-voltage (I-V) characteristics, stable C-V profiles, and barrier capacitance in the range 200 to 400 pF/mm2. Results with aluminum were mixed; some barriers yielded fairly rectifying I-V characteristics, satisfactory but slightly unstable C-V profiles, and barrier capacitance in the range 200 to 300 pF/mm2 while others yielded non-rectifying I-V characteristics, unstable C-V profiles, and barrier capacitance in the range 2000 to 3000 pF/mm2. Molybdenum-gold yielded poorly rectifying I-V characteristics, barrier capacitance of the order of 1600 pF/mm2, and C-V profiles which were only occasionally satisfactory. Titanium-gold yielded resistive I-V characteristics, barrier capacitances in the range 1000 to 5000 pF/mm2, and

no C-V curves.

The quality of the I-V characteristic and the value of the barrier capacitance are good indicators to determine whether the C-V curve is going to be good or bad. For satisfactory C-V profiles, the I-V characteristics should be at least fairly rectifying and the barrier capacitance should be in the range from 200 to 400 pF/mm2. It was found that the metal film thickness should be about 100 nm; films less than 50 nm or more than 200 nm thick were likely to yield poor or time-dependent C-V curves. In the case of silicon implanted with phosphorus, gold is the preferred metallization material.

Figure 28. Typical target configuration for characterizing the ion beam before ion implantation.

rors in current measurement which would result from the emission of secondary electrons for the target collector. The dimensions and spacing of the various electrodes can be changed to provide for differences in target

area.

In order to obtain a uniform ion current over large areas, the beam must be scanned in both vertical and horizontal directions or the target must be scanned mechanically. The ratio of scanning frequencies is usually such that a series of high frequency pulses is superimposed on a series of low frequency pulses. . Since the area of the ion beam itself is much smaller than the area of the circular defining aperture, this type of scanning produces a square pattern over the aperture as shown in the inset of figure 29. The ratio of the area of the circular defining aperture, md2/4, to the area covered by the beam scan, D2, which is known as the overscan efficiency, is a measure of the utilization of the ion beam. As shown in figure 29, this is about 80 percent for the case where the scanned area fills the entire circular aperture (D=d) and drops to less than 10 percent when the beam is overscanned one full target diameter on each side of the aperture (D=3d), a condition necessary if the beam shape is badly distorted.

During the set-up procedure, the Faraday cup current (see fig. 28) can be compared with the total target collector current to determine when the current density is equal in the two areas. Typical plots of the two current

Figure 29. Scanning geometry and overscan efficiency. (Overscan (%) = [(D/d)-1] x 100; overscan efficiency (%) (πd2/4D2) × 100.)

=

densities are shown in figure 30. With no scanning voltage applied, the steady-state ion beam is centered on the cup and shows a relatively high reading depending on how well the beam can be focused. As the scanner voltage is increased, the beam starts to move back and forth across the cup, resulting in a series of pulses which are averaged by the current measuring instrument (an electrometer, for example). The target collector current. measured by a current integrator remains essentially the same, increasing slightly as less current is collected by the cup, until

the scan voltage is high enough to deflect the beam past the aperture defining the target collector area. As the voltage increases beyond this point the target collector current drops proportionately. When the collector cup current density and the total collector current density are equal, a satisfactory beam uniformity is obtained over the entire target area. For the case shown in figure 30 the current densities are equal for current densities less than about one-fourth of the unscanned target current density. From figure 29, this is seen to be equivalent to an overscan of about 80 percent so that the beam deflection distance, D, is equal to 1.8 times the diameter of the defining aperture.

Measurement of target area is critical in determining the correct implantation dose. It is advisable to use a calculated area based on the physical measurements of the system being used. The distances between the scanner, target, and defining aperture must be known to calculate the ratio by which the target area is larger than the defining aperture. Other methods, such as measurement of visible implanted areas, are subject to a number of uncontrolled factors, such as beam size and instability, that cause shifting of the beam and uncertainty in the location of the implanted region boundaries.

As the beam passes over the center of the target in the typical 80-percent overscan case, the integrator measures a square-wave type pulse. The pulse width gradually decreases with distance from the center, becoming zero during the maximum levels of the slower sweep voltage. The integrator must therefore be capable of measuring this wide variety of pulse widths which ranges from zero to about 250 us.

In order to investigate the accuracy of the current measurements, two sources of current were used: 1) actual ion beam currents obtained from an implantation system and 2) simulated current pulses obtained from a pulse generator. The currents in both instances were measured and compared using five different commercially available current integrators as well as other current measuring instruments. The readings were taken from dc conditions to 10,000 pulses per second (pps). The ion beam scanner unit used was variable over a range from 20 to 2000 pps; the remaining frequencies were covered with the pulse generator. Average current readings were checked over the range from about 50 nA to 500 μА, usually with a 50-percent duty cycle.

The preliminary conclusions from these studies are that all the current integrators agree within a few percent for dc conditions, but only three of the five agreed within 10 percent for 10 to 2000 pps. Inconsistencies were observed between 1 and 10 pps; above 2000 pps other anomalies were observed. Therefore, both scan frequencies must be between 10 and 2000 pps; a ratio of 100 to 1 between the horizontal and vertical scan frequencies was found to be most desirable.

Errors related to measuring current pulses with peak amplitude greater than the full scale setting of certain of the current integrators were also encountered. All the current integrators performed well for peak ion beam currents less than 1 μA; four were satisfactory to 10 μA; three, to 100 μА; and two to 1 mA. Significant inaccuracies were observed for higher beam currents. Since the present trend in industry is toward the use of implantation currents of the order of 1 mA or more, these inaccuracies may cause difficulties in defining and comparing implantation conditions. (R. G. Wilsons and D. M. Jamba§)

5.4. Passivation Integrity

A corona discharge, reverse decoration technique was found to be a rapid, nondestructive method for decorating cracks and other defects in passivation overcoats [70]. This is a two-step method. In the first step, ions from a glow discharge at atmospheric pressure are deposited on insulating regions of the device. Then the device is immersed in a suspension of oppositely charged particles which are attracted to the charged insulating regions.

The charging process is illustrated schematically in figure 31. The source of ions is an array of seven parallel 40-um diameter nickel alloy wires, 1.8 cm apart, held in a plastic frame in a horizontal plane. The power supply is typically an rf-type, highvoltage, dc source capable of ±10 kV and a maximum current of 10 mA. A grounded plate, not shown in the figure, is placed about 3 cm above the wires to provide greater current uniformity and to protect the wires from mechanical damage. The specimen is placed on a

W

Figure 31. Schematic representation of corona charging with positive ions.

grounded plate about 2 cm below the wire array.

The charging process is carried out in a nitrogen atmosphere with about 25-percent relative humidity for 15 to 20 s. It was established that the charging of a surface containing closely spaced insulator and conductor regions is limited mainly by the conducting areas. As the ions collect on the insulator they generate an electric field in the gas ambient which causes subsequent ions to move toward the regions of ground potential before arriving at the wafer surface. Thus, for the geometries of interest, the oxide surface voltage is not determined by the oxide breakdown potential (except for very thin oxides or very large insulator areas).

The ambient relative humidity must be controlled to prevent lateral current flow on the oxide surface. The oxide surface voltage decay rate after charging was found to increase greatly above 30-percent relative humidity for thermal silicon dioxide, chemical vapor deposited (CVD) silicon dioxide, and CVD phosphosilicate glass (PSG). The surface voltage also decays rapidly if the bulk conductivity of the oxide is high due to the presence of moisture. Thus, for some wafers with CVD silicon dioxide or PSG passivation, it was found to be necessary to remove adsorbed or absorbed water before charging. This can be done by heating in room air at 200°C for 5 min. This has been made the initial step in the standard charging process to

ensure retention of surface charge on the specimen.

This procedure is satisfactory for devices or wafers in which the passivating layer has been etched open at the contact pads or at the grid lines. These etched regions furnish the closely spaced insulator-conductor geometry required so that the surface voltage does not depend on whether the passivating layer is over metal or over insulator. It is also possible to decorate wafers before the passivation layer is etched open by placing a grounded conducting grid over the wafer during the charging. A convenient grid consists of an aluminum sheet 0.8 mm thick with 3.5-cm diameter holes on 5.0-cm centers. When the grid is used, charge is deposited only at The entire wafer can openings in the grid. be charged if the grid is moved laterally during charging while it is spaced about 0.3 mm above the wafer.

It is necessary that the surface charge remain on the passivating layer for a time about equal to the delay between charging and deposition plus the deposition time. The dielectric relaxation time of the passivation layer must be longer than the total time which is about 20 s.

Thus, a bulk resistivity of about 1013 cm or greater is needed. The resistivity of an insulating layer may be increased by reducing its temperature; this can be accomplished by cooling the specimen during charging and by the use of cooled particle suspensions.

The reverse decoration process is illustrated schematically in figure 32. This process has been accomplished with various powders including phosphors, zinc oxide, and lead aluminosilicate glass, but the best results in terms of sensitivity and contrast were obtained using carbon black suspensions. Various carbon black suspensions were tested. These included commercially available electrophotographic toner concentrates and specially prepared formulations.