Figure 32. Schematic representation of the reverse decoration of defects with negative ions following corona charging with positive ions.

་

Best results were obtained with a carbon black suspension prepared in the following way. First, a concentrate is made from 17.0 g Raven 1255 carbon black [71], 100.0 ml toluene, 10.0 ml of a mixture of 50 g Lubrizol 894 [72] in 50 ml toluene, and 10.0 ml of a mixture of 50 g A-C Polyethylene, grade 430, [73] in 90 ml toluene. These materials are readily available. The carbon black has a mean particle diameter of 22.0 nm and a nitrogen absorption surface area of 130 m2/g. The surface of the carbon black as supplied is acidic by addition of volatiles [71]. Lubrizol 894 is an ashless dispersant additive for heavy duty engine oils; the weight percent of nitrogen ranges from 1.65 to 1.95 percent [72]. A-C Polyethylene, grade 430, is a polyethylene-vinyl acetate copolymer (30 percent vinyl acetate); its softening point is 60°C [73]. The concentrate is agitated for about 5 min with a 100-W ultrasonic probe to disperse the carbon black.

To make the dilute suspension, about 1 ml of the concentrate is added to 400 ml of Freon TF and stirred. This results in a suspension which contains about 0.1 g of carbon black. This amount has been found to be adequate for decorating at least ten 3-in. (76mm) diameter wafers. Both the concentrate and the dilute suspension are stable for long periods (months). In this suspension, the carbon black is negatively charged so that

positive corona potentials are used for reverse decoration.

Measurements were made to characterize the electrical state of the carbon particles in the dilute suspension. These measurements involved depositing the carbon black on metal foils while recording the current and then weighing the deposited carbon. This was done as a function of applied voltage and time. The weight of carbon black deposited and the charge transported varied approximately linearly with voltage and time. For electrode spacings of 1.7 cm on each side of a metal foil, an applied voltage of 10 V for 4 min resulted in a deposition of 3.2 × 10-4 g/cm2 of carbon black. The average charge per single particle of carbon black, with the assumption that all such particles were 22.0 nm in diameter and singly dispersed, was 0.05 times the electronic charge. Since the minimum charge per particle is one electronic charge, it is clear that the carbon black is aggregated to at least a factor of 20 times. This aggregation is essential for obtaining dense black on the charged insulator surface.

The detection of cracks and pinholes decorated in this way is quite convenient under microscopic observation. An example is shown in figure 33. It was demonstrated that every defect detected by etching of the underlying aluminum through the defects was detected by the carbon black reverse decoration procedure. This was done in two ways. Wafers and chips were decorated with carbon black and photomicrographs were taken of crack and pinhole patterns. The carbon black was then removed and the aluminum was etched (NBS Spec. Publ. 400-25, p. 33) at 50°C for 5, 7, 10 or 15 min. The specimens were then photomicrographed again and the results compared. Hundreds of micrographs from many different specimen types were compared, and, in every case, all defects detected by etching were also clearly outlined by the carbon black. In the second method, specimens were decorated with carbon black and then aluminum etched with the carbon black in place. Again, the correlation was excellent. An example of the latter test is shown in figure 34.

defect sites per unit area, using one of several microscopic techniques discussed previously [74]. The choice depends on the uniformity and density level of defects, the type of material used in decorating, the degree of accuracy required, and the specific purpose of analysis. Incident white-light, bright-field illumination is most suitable for observing specimens decorated with carbon black. The magnification used should be sufficient to allow resolution of individual demarcated defect sites, but not higher, so that the largest possible specimen area is included within the field of view. It is impossible to specify a fixed magnification that would hold for all cases, but a magnification in the range from 50 to 500 diameters has been found to be most useful.

The greatest difficulty in determining the density of defects by any method concerns the distribution uniformity. If the uniformity of distribution is "good", counting within a few sites may provide an adequate measure of the true statistical distribution. If it is "fair", which is the most common occurrence, examining a substantial number of areas (5 to 15) over the specimen may be necessary to arrive at a valid average density value. Specimens with "poor" uniformity are best treated by dividing it into "good", "fair", and "poor" regions and computing separate density values for each, rather than computing a meaningless average value.

The sampling areas for counting should be spread reasonably uniformly over the specimen surface. If the specimen is a device wafer

Figure 34. Photomicrograph of portion of reverse decorated device aluminum-etched at 50°C for 5 min without removing carbon black. (Magnification: ~325x; etched defects appear inside the regions without carbon black as denoted by the arrows.)

or is otherwise patterned, one may choose to examine every nth unit, and count the defects within that defined area. If the specimen is not patterned, then the microscope stage micrometer can be used to space the observation sites uniformly over its area. The circular area of the microscope field of view for the magnification may be used as a convenient unit. Photomicrography affords a useful record of the defect density and often facilitates counting of the defect sites, especially if the density is high; it also provides an accurately defined area of the specimen surface.

A major advantage of the carbon black reverse decoration method is the high optical contrast available in reflected-light microscopy. This high contrast permits a fairly simple form of automated quality assessment with a microscope-mounted photocell to measure the integrated light reflected from the defect areas. This technique is much simpler than the use of computer-implemented image analysis. Although the latter would provide a defect count, which the integrated reflectedlight method does not furnish, it is believed that the simple method explored here could be used in a production facility to rank wafer quality and set a pass, no-pass criterion.

Initially, photomicrographs were made of carbon black reverse decorated devices on integrated circuit wafers. Transparencies made from these micrographs were then mounted in an optical system so that a collimated light

beam was passed through the transparency and then focused on the faceplate of a vacuum photocell to form an image of the light source. An image of the transparency is not desired since the photocell response may not be uniform across the faceplate. The current output of the photocell varies linearly with incident light. It was found that the current output varied inversely with subjective judgments of device quality made by counting pinholes and estimating crack lengths.

Based on these preliminary encouraging results, a photocell-microscope system was assembled to take direct readings of reflected light from decorated wafers. The arrangement used is shown schematically in figure 35. By placing a plano-convex lens with a focal length of 57 mm at the end of the camera tube, an image of the light source field stop was obtained at the photocell faceplate about 6 cm from the end of the camera tube. The photocell employed an S-11 cathode and was operated at 100 V. Light intensity was adjusted to give convenient current levels and the

overall magnification was chosen so that a large portion of the circuit was visible, but bond pads and grid lines were excluded. Shielding prevented introduction of ambient light. Photocell current readings were taken on 20 circuits of each of two wafers. For the "good" wafer, the lowest individual current was 2.9 nA, the highest was 5.5 nA, and the average was 3.5 nA. For the "bad" wafer the lowest current was 6.7 nA, the highest was 15 nA, and the average was 10 nA. Pinhole counts on these wafers resulted in a difference of a factor of two in defect density. Thus, the photocell current measurement can easily detect a difference of a factor of two in pinhole density and is probably more sensitive than this.

been placed face down in the bottom of the ultrasonic tank and it is believed that the mechanical contact caused these defects. Subsequent tests with the ultrasonic tank used in this manner did induce defects.

A third test involved automatic probe testing of C-MOS wafers. These were chosen since it is expected that C-MOS devices are most susceptible to static electrical damage. Two 3-in. (76-mm) diameter C-MOS device wafers were probed, and a map was made of good and bad devices. Then the wafers were charged, decorated, and cleaned. The wafers were probed again. One wafer showed a net loss of four devices out of about 130 good devices, initially. The other wafer showed a net gain of five devices out of about 100 good devices, initially. These variations are within the normal tolerance of repeated wafer probe testing.

Transport

Boron Nitride Diffusion Sources studies were made of known amounts of zinc from simulated boron nitride diffusion sources. Two glasses, which were designed to simulate the oxide layers on activated boron nitride of two types (pure and mixed with silica), were prepared with admixtures of 0.75 mole percent zinc oxide [75]. One of the glasses was boron oxide (B203) and the second was a borosilicate glass composed of silica (SiO2) and boron oxide in a 3 to 7 weight ratio. The transfer experiments were conducted at 1000°C in a dry nitrogen atmosphere for times up to 12 h. The glass which formed on the surface of the silicon wafer was analyzed for zinc by means of flameless

atomic absorption. In the case of the B203 glass, no zinc transfer was observed within the detectability limit of one part per million by weight. The zinc content of the borosilicate glass is shown in table 10. The confidence limits for these data have not been established at this time, but it seems likely that the variations are not statistically significant. (J. Stacht.

T. A. Yager, and R. E. Tressler+)

Hydrogen Chloride Gas The SOLGAS program [76] was used to compare the equilibrium partial pressures at 1100°C of the chemical species in the hydrogen chloride (HCl) and trichloroethylene (TCE) oxidation systems. The most significant difference is that the partial pressure of water vapor in the TCE system is lower by an order of magnitude when compared with that in the HC1 system [77]. The calculations also indicate that the partial pressure of C10 is nearly equal in both systems whereas the partial pressures of HC1, HO, and HOC1 are approximately five times greater in the HC1 system. (R. Tressler+ W. H. Grubbs*, M. B. Das*, and J. Stach*)

6.1. Line-Width Measurement with Spatially Filtered Coherent Optical Radiation

where σ is the lower cutoff frequency, 02 is the upper cutoff frequency, 2a is the width of the line, Si[x] = [(sing)/5]d (the sine integral), and the pre-superscript zero on the inverse sine integral indicates that the lowest value (between 0 and T) must be taken. The upper cutoff frequency is determined from (NA)/X (10)

02 = where (NA) is the numerical aperture of the objective lens and is the wavelength of the incident illumination; this is the diffraction limit. Thus, for any given microscope system, the optimum values of σ1 and 02 are fully defined for a specific line width, 2a. Since lines of unknown test width cannot a priori be paired with their optimum filters, measurements must be made on images formed by non-optimum filters. To assess the ability to correct routine measurements made with non

The physical makeup of the microscope in the o-plane determines the frequency content of the image. The outer aperture of the system determines the spatial frequency limit of this plane; the origin of the o-plane is equivalent to optical "dc", and spatial frequency increases linearly outward to the aperture limit. A clear aperture therefore acts as a low-pass filter, passing without attenuation all frequencies below its upper cutoff, but truncating all those above. If an opaque line of width 201 is inserted in the center of the o-plane, frequencies lower than 01 will also be truncated. With such an occluding band-pass filter, it can be shown that the image of an opaque line in a clear background is identical with that of a clear line