SURFACE CLEANLINESS

Meeting the challenge of submicron defect characterization on final-polished wafers

James J. Shen and Lee M. Cook, Rodel

As ULSI devices continue to shrink and device layers become thinner and thinner, the critical size of particles and wafer defects also decreases. For devices with a design rule of 0.25 µm, for example, the critical size of "killer" particles or defects on silicon substrates is believed to be approximately 0.08 µm. Interestingly, such submicron-size particles are known as light point defects (LPDs), while defects developed from bulk crystals are called crystal-originated particles (COPs). This misterminology is probably partly due to the historical difficulty of distinguishing a particle from a defect when characterizing wafer surfaces using light-scattering instruments. Although scatterometer capabilities have also been evolving, this difficulty still exists and is even worsening as the to-be-detected particles and defects become tinier and tinier. The light-scattering behavior of intrusion defects such as pits is not fully understood, and the sources that scatter laser light have not been directly confirmed on actual wafers. The research reported here revealed some limitations of the systems used for the identification and characterization of submicron particles and defects. After discussing those findings, the article also suggests some areas where additional research could lead to new technologies that would better satisfy the needs of the semiconductor industry.

Characterization of Particulate LPDs

Particulate LPDs on final-polished wafers can come from various sources, including the abrasives used in slurries (e.g., SiO^₂ particles); polishing debris; airborne particles in cleanroom fabs; and contaminants transferred from such process and handling equipment as wafer carriers, cleaning tanks, and inspection instruments. Theoretically, one should be able to observe LPDs much smaller than 0.08 µm by scanning electron microscopy (SEM) and atomic force microscopy (AFM). In practice, however, SEM is a poor candidate for production applications because of its small scanning area at high magnifications, the high reflectivity of silicon, and the high probability of contamination and electric-charge buildup on the silicon surface during the inspection process. The capabilities of AFM for characterizing and studying wafer defects have been developing rapidly. Unfortunately, the technique's usefulness for particulate LPD inspections in manufacturing is still limited by its low throughput, the difficulty of accurately aligning wafers on the microscope's x-y coordinating stage, and the possibility of mechanical and electrical interactions between AFM tips and mobile particles. Consequently, laser scatterometers are the current instrument of choice for particle sizing and counting in both production and research activities. The sensitivity of scatterometers such as Tencor Instruments' (Mountain View, CA) Surfscan 6200/6220 and ADE's (Charlotte, NC) WIS series is claimed by their manufacturers to be within the 0.08—0.10-µm range, depending on the inspected wafer's surface roughness.^1,2 Most current production specifications cite only particles >0.12 µm, and measurements above this threshold are believed reliable from machines with this nominal sensitivity.

The particle-size capability claimed for scatterometers is calibrated using polystyrene latex spheres deposited on fine-polished silicon wafers. However, because the nature of real particles varies and their differing composition and geometric shapes affect their scattering behaviors, the accuracy of the measurements reported by scatterometers is open to question. In addition, the roughness of final-polished silicon wafers also varies, and they may sometimes be far rougher than the wafers used for calibration. (An Ra range of 1—5 Å is typical for final-polished and cleaned wafers, measured by AFM in the scale of 15 x 15 µm.) The rougher a wafer surface is, the higher the background noise level will be, which may "add" particles to a scatterometer measurement if the instrument's gain level and threshold are not selected properly. In order to minimize the influence of surface roughness on particle detection, the Surfscan 6420 scatterometer (Tencor Instruments) uses an oblique incident angle, selectable laser polarization, and side detectors.³ This scanner is sensitive on rough surfaces and has a nominal sensitivity of 0.12 µm.⁴

A comparison of LPD maps and size distributions of the same wafers from the two Tencor scatterometer models revealed that the 6420 reported smaller particle sizes than did the 6200 (see Table I). This difference in sizing could be attributable to several factors, including the weaker scattering intensity that results from the 6420's oblique incident angle and side detectors, and the orientation dependence of the scattering cross section of an irregularly shaped particle. The significant sizing difference for the LPDs designated numbers 7 and 8, in particular, indicates that reported LPD size is strongly dependent on the shape and orientation of the real particles or defects.

Instrument	LPD ID No.
	1	2	3	4	5	6	7	8
6200	0.18	0.25	0.35	0.38	0.4	1.1	2.1	.9
6420	0.16	0.11	0.13	0.25	0.35	0.8	0.42	0.25

Even though it is less affected by surface roughness, the 6420 has a lower sensitivity specification than the 6200 does.^1,4 In comparative testing, we observed that the minimum particle size detectable by the 6420 on standard polished wafers was 0.13—0.15 µm, with a signal-to-noise ratio of approximately 2:3. However, using an improved polishing process, we were able to produce very smooth wafers (Ra = 1.51 Å, measured using a Zygo with an 80-µm filter) on which particles as small as 0.08 µm were detected clearly by the 6420, with a signal-to-noise ratio of approximately 5:1. In another experiment, a 6220 was used to characterize a wafer with a high level of roughness. As seen in Figure 1, the wafer's haze map pattern and LPD pattern matched each other, which suggests that the LPD map was a "false" reading of the haze pattern. The instrument may have picked up the background haze signal because the gain level and particle-size threshold were inappropriately set for this wafer. The above findings indicate the importance of wafer surface roughness for reliable measurement of LPDs by either the 6420 or the 6200.

Figure 1: Haze pattern map of a 6-in. wafer measured by a 6220 (a), and the LPD pattern map of the same wafer (b).

Characterization of Crystal Defects

As mentioned above, it is difficult for measurement instruments to distinguish between a particle and a bulk crystal defect. The LPDs reported as particles by scatterometers may actually be COPs, which can form during prepolishing processes, such as crystal growth, cutting, lapping, and etching, or result from stock and final polishing. COPs can be classified into three categories: point defects such as vacancies, interstitials, and impurity substitutes; line defects such as dislocations; and three-dimensional defects such as point-defect clusters, oxide/silicate precipitates, etch pits, and oxidation-induced stacking faults (OISFs). Some of these defect types can coexist or be cocreated, such as interstitials and dislocation loops with OISFs. The size of most point and line defects is in the angstrom range, while three-dimensional defects may measure in the nanometer, submicron, or micron ranges.

The most common defects seen on final-polished wafers are subsurface damage, pits, and protrusions (nodules or bumps). Subsurface damage presents itself as lattice distortion or residual stress. This damage may be attributable to nonuniform plastic deformation resulting from nonuniform interactions between the wafer surface and slurry solids (or pad surfaces), especially in the presence of precipitates or OISFs. Pits and protrusions on final-polished and cleaned wafers may result from uneven surface removal during polishing or from preferential etching by slurry and cleaning chemicals at defective sites—for instance, the exposure of vacancies (pits), pulling out of precipitates (pits), and protrusion of precipitates (bumps). Pits on wafer surfaces in the submicron and micron size ranges have been reported elsewhere.^5—7 Depending on defect sizes and instrument resolutions, optical Nomarski microscopes, SEMs, optical profilometers, AFMs, or laser scatterometers can be used to characterize pits and protrusions, while subsurface damage can be evaluated by x-ray diffraction technology, and possibly by gate-oxide integrity measurement.⁸

Figure 2: AFM image of etch pits on a final-polished wafer. These pits are approximately 3—8 µm in size.

Figure 3: Nomarski micrograph of etch pits on a final-polished wafer. These pits are approximately 5—15 µm in size.

The 6200, which has a normal incident angle, displays both protrusions and intrusions as LPDs on defect maps, and because of this it is impossible to distinguish one defect type from another, but the 6420 is only sensitive to protrusions because of its laser's grazing angle. By using both instruments to characterize the same wafers and then comparing the LPD maps, we determined that approximately 50% of the LPDs reported by the 6200 were actually intrusions.⁷ As seen in Figures 2 and 3, AFM and Nomarski measurements revealed that these were pits that ranged in size from 1 to 15 µm. These observations reinforced our suspicions regarding the accuracy of the LPD sizes given by scatterometers. Clearly, the scattering action of an incident laser beam at a pit differs from that at a particle, but almost all scatterometers are calibrated solely with polystyrene latex spheres. In the experiments discussed below, the hypothesis that scatterometers may underestimate pit size was verified by AFM measurements.

Confirmation of LPDs Using AFM

The presence of submicron LPDs identified by laser scatterometers on final-polished silicon wafers is nearly impossible to confirm by optical microscopy or scanning electroscopy because of the smooth wafer surfaces, the potential for in situ contamination, and the lack of reliable accuracy in the instruments' x-y coordinating systems. Therefore, an AFM (Dimension 5000, Digital Instruments, Santa Barbara, CA) was used for defect characterization in this study. The AFM has an x-y coordinating stage capable of navigating LPD maps from the 6220, so that identified defects can be searched and located. A certified standard silicon wafer (VLSI Standards, San Jose) was used for the comparative 6220 and AFM measurements, which were conducted in a Class 10 cleanroom. The standard wafer was a (100) P-type, 150 mm diam.

Figure 4: Defect map of a standard silicon wafer measured by the 6220.

Figure 5: AFM images of the four types of scattering site on the standard wafer: (a) a single pit, (b) a triangular 4-pit pattern, (c) a circular 10-pit pattern, and (d) a circular 20-pit pattern.

As shown in Figure 4, the standard contained 12 squares with 9 light-scattering sites in each quadrant of each square (some sites are missing in the figure). Using AFM, it can be seen that those sites contain a single pit or cluster of pits. There are four types of sites, shown in Figure 5, which scatter approximately the same light intensity as do 0.168-, 0.208-, 0.269-, and 0.343-µm polystyrene latex spheres, respectively. Figure 6 depicts the profile of the pits, which all measure 2.7 µm, and Table II compares the 6220 and AFM measurements of various sites on the standard. (The sizes reported by the 6220 varied with gain setting; the results in the table were obtained at gain level 7, which is the most widely used setting for inspections of final-polished wafers.) These data clearly demonstrate that defect sizes were underestimated by the 6220 by more than an order of magnitude, confirming our previous observations.⁷ In the worst case, 20 2.7-µm etch pits, in a pattern measuring 1017.4 µm², were counted as an area defect of 14,000 µm² by the scatterometer. The underestimations of pit size may be attributable to the difference in scattering mechanisms between a pit and a particle, while the reporting of pit patterns as single defects may be attributable to the poor spatial resolution (50 µm) of the scanner. (The spatial resolution of a scatterometer is determined by its beam size and step length. In this case, the 6220 has a 80-µm-diam beam, and at a setting of "low throughput and sweep correlation 2," the wafer moves at a pace of 10 µm/scan.) Furthermore, a particle or defect will not be registered as an LPD by the scanner unless it is scanned twice and the intensity of the scattered light is above the detect threshold each time.

Figure 6: AFM image of a pit cluster showing the profile of the pits.

Site	6220	AFM
	LPD Size	Pit Size (µm)	Pit Pattern Area (µm2)
Type 1 (1 pit)	0.195 µm	2.7	N/A
Type 2 (4 pits)	0.332 µm	2.7	266.4
Type 3 (10 pits)	1.9 µm	2.7	543.6
Type 4 (20 pits)	14,000 µm2	2.7	1017.4

To confirm that similar undercounting and underestimation would occur with particles (rather than pits), the same standard wafer was intentionally contaminated and then cleaned by passage through a scrubber (OnTrak Systems, San Jose). As seen in Figure 7, additional defects were identified on the wafer by the 6220. Some of these new LPDs, labeled alphabetically in Figure 7, were then measured using AFM. Table III lists the comparative results, which indicate that particles on wafers are not well resolved by current light-scattering technology. For example, the patch of 0.1—1.26-µm particles shown in the AFM image in Figure 8 was identified as a single 0.172-µm LPD by the 6220. In addition, a scratch measuring approximately 5 x 100 µm, shown in Figure 9, was reported as a large-area defect by the 6220. It seems that, because the detection of scratches is dependent on scanning orientation, the scatterometer detected only the large particles contained within the scratch. Finally, particles smaller than the sensitivity of the scatterometer (0.1 µm) were also observed using AFM.

Figure 7: Defect map of the standard wafer after contamination and scrubber cleaning; added LPDs of interest are labeled A—H.

Figure 8: AFM image of an LPD site identified in the defect map shown in Figure 6.

Figure 9: AFM image of a scratch reported as a large-area defect by the 6220.

Instrument	LPD ID No.
	A	B	C	D	E	F	G	H
6220	4.59 µm	23,300 µm2	0.198 µm	0.203 µm	15,400 µm2	0.172 µm	No LPDs	0.16 µm
AFM	No particles	3.57 µm	No particles	A patch of particles, 0.2—0.5 µm	3.2 µm	A patch of particles, 0.1—1.26	A patch of particles, 0.3—1.1	A patch of particles, 0.2—0.8

Table III: Comparison of AFM and 6220 measurements of selected LPDs.

The x-y coordinating accuracy for these AFM measurements was controlled within ±1.0 µm by using the pit patterns on the standard wafer as reference points. The AFM was set in tapping mode to scan the wafers, and some particles were moved and rearranged by the AFM tip during scanning, but the presence of particles in the LPD areas of interest was indubitable. In order to eliminate the possibility that contamination from the AFM tip interfered with results, the wafers were rescanned after AFM measurements and no additional particles were observed. Therefore, we are confident of the accuracy of the correlations between the LPDs on the 6220 map and the defects measured by AFM.

Conclusion

Although the accuracy and sensitivity of defect and particle characterizations of silicon substrates have been improving thanks to the introduction of laser scatterometers and AFM, current technologies do not meet the changing requirements of the semiconductor industry. Improvements in the spatial resolution and sensitivity of laser light—scattering systems are particularly urgent. As the research described above indicated, an intrusion defect will be sized approximately one order of magnitude smaller than it really is by laser scatterometers. The number of particles is also underestimated by scatterometers, and wafers in the real world may be far dirtier than the industry thinks. Furthermore, the confusion in distinguishing particulates from crystal defects remains. There is a need to develop novel wafer inspection technologies. Areas of investigation might include determining differences in the light-scattering action of particles and various types of defects; improving spatial resolution by such methods as using finer laser beams; and integrating currently available systems to extend overall capability—for example, by combining SEM or AFM with accurate x-y coordinating systems.

References

1. Surfscan 6200 product brochure, Mountain View, CA, Tencor Instruments.

2. WIS CR-80 product brochure, Charlotte, NC, ADE Optical Systems.

3. Howland RS, "Detecting Killer Particles on Rough Surfaces," Semiconductor International, 17(8):164—170, 1994.

4. Surfscan 6420 product brochure, Mountain View, CA, Tencor Instruments.

5. Pirooz S, Shive LW, Malik IJ, et al., "Predicting Technology Advances for Wafer Surface Inspection Systems," Microcontamination, 11(10):21—25, 1993.

6. Morita E, Okuda H, and Inoue F, "Distinguishing COPs from Real Particles," Semiconductor International, 17(8):156—160, 1994.

7. Shen JJ, Cook LM, Pierce KG, et al., "Non-Particulate Origins of Light Point Defects (LPDs) on Polished Silicon Wafers," Journal of the Electrochemical Society, 143:2068—2074, 1996.

8. Sethuraman A, private communication, Newark, DE, Rodel, November 1996.

9. Resnick PJ, Adkins CLJ, Clews PJ, et al., "A Study of Cleaning Performance and Mechanism in Dilute SC-1 Processing," in MRS Symposium Proceedings, vol 386, Pittsburgh, Materials Research Society, pp 21—26, 1995.

James J. Shen, PhD, is a scientist at Rodel, Newark, DE. Before joining the company in 1995, he worked in the fields of semiconductor crystal growth and defect analysis at the New York State Center for Advanced Materials Processing. A member of the Materials Research Society and the Electrochemical Society, Shen received a BS from the Jiangxi Institute of Metallurgy in China, an MS in the same field from the Beijing University of Science and Technology, and a PhD in engineering science from Clarkson University (New York). (Shen can be reached at 302/366-0500, ext. 6397; e-mail, jshen@rodel.com)

Lee M. Cook is a vice president of research and technology at Rodel, which he joined in 1992. Previously he managed an R&D team at Galileo Electro-Optics. A winner of three R&D 100 awards and author of numerous publications, he is a member of the American Ceramic Society and the Society for Precision Engineering. Cook has a BA in chemistry from Grinnell College (Grinnell, IA).

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit