Powerful CO2 lasers operating at a wavelength of10.6 �m are at present widely used in science and technol-ogy. The transmissive optical elements used for forming andtransporting the powerful laser radiation are ordinarily fabri-cated from polycrystalline zinc selenide obtained by chemi-cal deposition from the vapor phase. A necessary conditionfor the trouble-free operation of such optical elements is thatno defects must be formed on their surfaces during abrasiveprocessing.1
The process of polishing zinc selenide has special fea-tures that must be studied for a deeper understanding of themechanisms by which the polishing compounds interact withthe crystal surface, as well as for enhancing the quality of theresulting items. Our earlier papers2–4 discussed how the pol-ishing materials and the conditions under which the polish-ing process is carried out affect the surface quality of poly-crystalline zinc selenide. However, these papers did notdescribe the optimum conditions for polishing the zinc chal-cogenides.
The process regime of polishing optical elements basedon zinc chalcogenides assumes that there is a great deal ofvisual monitoring of the surfaces being polished. Surface-quality evaluation reduces to the investigator’s subjectivejudgement; it largely depends on his experience and qualifi-cation and is therefore rather approximate.
New possibilities in solving these problems are openedup by modern digital technologies and computer methods ofpattern recognition. Current developments give the generalprinciples and some methods of this technology. For a spe-cific image-processing task, one usually has to solve theproblem de novo. The solution of this problem makes it pos-sible not only to simplify and shorten the work, but also toincrease the uniformity of the criteria used to evaluate thesurface-processing quality.
The goal of this paper was to investigate how the ther-mophysical characteristics of the polishing resins, the pres-
67 J. Opt. Technol. 77 �1�, January 2010 1070-9762/2010/0
sure, the grain size of the abrasive being used, the rotationalrate, and the oscillation frequency of the processing tool af-fect the resulting surface quality of zinc selenide and sulfide,using the technique developed for computer recognition ofdefects on micrographs of the polished surfaces.
II. EXPERIMENTAL TECHNIQUE
To achieve this goal, studies were carried out of how thethermophysical characteristics of the polishing resins and thegrain size of the polishing suspension affect the processingand the surface quality of polycrystalline zinc selenide andsulfide. The dependences of the removal rate and the surfacequality on the hold-down pressure, the rotation rate, and theoscillation rate of the hold-down tool during the chemome-chanical polishing of the zinc chalcogenides were studied.Polycrystalline zinc selenide and sulfide obtained by chemi-cal deposition from the vapor phase5 were used in the experi-ments. The samples were disks 20 mm in diameter and5 mm thick. These samples were cut from a single-synthesiswafer and were preprocessed identically. The processing ofthe samples in the experiment was carried out unit-by-unit.The unit was a duralumin faceplate on which eight zinc se-lenide samples were cemented. Rosin-based cementing res-ins were used on the faceplate to combine the samples intounits. The experimental conditions were chosen on the basisof the results obtained earlier and described in Ref. 2. Dis-tilled water was used as a lubricant-coolant. Small amountsof a 1-M solution of nitric acid were used to carry out ch-emomechanical polishing �CMP� of the zinc selenide. Thepolishing was done using ASM 2 /1 and 1 /0 diamond mi-cropowder with grain size 2 and 1 �m, respectively. Rosin-based polishing resins were used as the material of the pol-ishing wheel. A computer-vision technique with softwaredeveloped for this purpose was used to evaluate the qualityof the zinc chalcogenide surfaces obtained in the process ofpolishing.
The criteria used in the program for evaluating the clean-liness of the optical surfaces are based on GOST �State Stan-dard� 11141-84, currently active in Russia. The initial photo-graphic images of the test surfaces were obtained as follows:From ten to fifty digital photographs were made of eachsample of polished zinc selenide or sulfide in the form of acircular optical element 20 mm in diameter, using the Ax-ioplan 2 optical microscope �Carl Zeiss, Germany� equippedwith a digital camera. The digital microphotographs �Fig. 1�were arrays of about 1000�1000 image elements—pixels—
a
b
c
FIG. 1. Successive stages of the computer processing of an image of apolished surface of zinc selenide. �a� Original micrograph, �b� photographequalized in background density, �c� discriminated high-contrast sectionscorresponding to defects.
68 J. Opt. Technol. 77 �1�, January 2010
containing information on the brightness of the correspond-ing section. The size of the photographed section of thesample surface depended on the microscope’s chosen mag-nification. The minimum size was about 100�100 �m, andthe maximum was about 1 mm across. Because of the pres-ence of photometric aberrations, the edges of the frame wereappreciably darker than its central part �Fig. 1a�. To eliminatethis effect, the image was corrected by computer �Fig. 1b�.The most contrasty sections, corresponding to the defects,were then distinguished �Fig. 1c�.
The images of the limits of the defects �Fig. 1b� are notusually ideal: they are blurred and have numerous bumps andnotches. The defective sections shown in Fig. 1c were dis-tinguished by determining the critical pixel brightness Kcrit.When its brightness is lower than the critical level, a sectionis considered defective, and brighter sections are regarded aspolished background. To determine the critical brightnesslevel of the entire image, the distribution function of thepixels in brightness was constructed. A typical form of thisfunction is shown in Fig. 2.
The value of Kcrit was determined from Eq. �1�, startingfrom certain characteristics of the distribution function of thepixel brightness on the initial photographic image,
Kcrit = N + A�B − �i�N
�ini�� �j�255
nj� , �1�
�i�N
ni� �j�255
nj � C , �2�
where A, B, and C are certain constants chosen as a functionof the process conditions, i and j are the pixel brightnesses�whole numbers from the interval 0-255�, and ni and nj arethe numbers of pixels with the corresponding brightness val-ues. The value of N is determined from inequality �2� as thesmallest number for which this condition is satisfied.
Only a few percent of the pixels usually have the criticallevel. The results strongly depend on the accuracy withwhich this value is determined. Figure 3 shows what canresult from the displacement of the critical level. The firstphotograph �Fig. 3a� shows the surface of a high-quality zinc
0 100
Number of pixels
Pixe
lbri
ghtn
ess
200 300K crit
5000
10000
15000
20000
0
FIG. 2. Form of the pixel-distribution function over brightness. The verticalline indicates the position of the critical brightness.
68Gavrishchuk et al.
selenide sample. Figure 3b illustrates the results for the nor-
a
20 µm
b
c
FIG. 3. Photographs of a polished zinc selenide surface. �a� Original micro-graph of a section with high polishing quality, �b� image of defects for thenormal position of the critical level, �c� image of the defects for a smalldisplacement of the critical level.
69 J. Opt. Technol. 77 �1�, January 2010
mal placement of the critical level. However, a large numberof defects that are virtually indistinguishable by eye are de-tected when the critical level is displaced �Fig. 3c�. Evenscratches corresponding to a width of one pixel are visible inthis picture, and this is less than 100 nm.
The next stage is to classify the defects. The most effi-cient algorithm was based on covering the distinguished de-fective sections of the image with circles having the maxi-mum possible size. The algorithm is implemented in thefollowing sequence: The longest chain of circles that can beconstructed while approximately maintaining the same direc-tion is first located, and the length and width of the defect isdetermined. According to the normative documentation con-sidered here, a defect is regarded as a scratch if the ratio ofthe length to the breadth exceeds 3:1, and it is regarded as apoint if it is smaller than this ratio. The program then deter-mined the size and number of defects from each photo-graphic image.
The program combined a set of several tens of frames ofvarious sections of a single sample with the correspondingweighting functions into a single set of data, and distributionfunctions of the defects �points and scratches� according tosize were constructed from them. Based on these data, theprogram at once classified the test surface in cleanliness inaccordance with GOST 11141-84 according to certain pa-rameters.
The error of the resulting data was determined by a num-ber of components, the basis on which was the following:
Representativeness of the sample. The micrographscovered lass than 1% of the surface. It is important to includeall the typical surface elements and to assign the correctweighting function, which is used when combining all thephotographs of a certain sample into a single block of data.
How correctly the program works for defect recogni-tion. The program might not cover the entire variety of situ-ations and in a number of cases used a simplified approach.However, as shown by an expert comparison of the results ofvisual observation of the defects on the photographs and theresults of their computer processing, more than 90% of thedefects �with respect to their area� were correctly recognized,and the contribution from this part of the errors is not large.At the same time, the normative documents �state standard,adopted in 1984� does not provide for such a detailed studyof the surface using a computational technique and does notprovide sufficiently distinct criteria to be used in such studiesfor determining the defective sections.
Statistical error caused by random scatter of thedata. Since there were several tens of micrographs for eachsample, this form of the errors can be evaluated using stan-dard techniques. The method of calculating them can be var-ied, depending on the parameters to be studied.
The described technique thus objectively gives the quan-titative characteristics of the surface defects, and these quan-tities make it possible to study in more detail the mechanismof the processes that occur during polishing.
III. RESULTS AND DISCUSSION
The dependences of the removal rate and the surfacequality of zinc chalcogenides for polishing resins that have
69Gavrishchuk et al.
various thermophysical characteristics were obtained experi-mentally. Figure 4 shows the distribution functions of thescratches and points on the zinc selenide surfaces as theydepend on the softening temperature of the polishing resin. Itcan be seen that the size and number of defects on the samplesurface increase as the softening temperature of the resinincreases. However, there is a certain optimum: thus, on aresin with a softening temperature of 64 °C, the surface hadthe minimum number of point defects, the size of whichdepended only on the grain size of the abrasive being used.At the same time, the minimum number of scratches wasobtained on resin with a softening temperature of 70 °C.With respect to the combined defect concentration, the bestsurface was obtained on a resin with softening temperature64 °C. The use of the computer-vision technique thus madeit possible to optimize the thermophysical characteristics ofthe polishing resins for the chosen conditions for polishingzinc chalcogenides.
In the course of the studies on how the grain size of thepolishing suspension affects the processing and the surfacequality of zinc selenide, it was determined how the variation
0
2
4
6
8
Width, µm
Len
gth,
nm
0 2 4
12
3
4
5
1000
2000
3000
0
Num
ber
ofpo
ints
Diameter, µm0.5 2.5 3.5
12
3
4
5
a
b
FIG. 4. Distribution functions of points �a� and scratches �b� on the surfacesof zinc selenide samples with an area of 1 mm2 for various softening tem-peratures of the polishing resin. 1—Tp=56 °C, 2—Tp=59 °C, 3—Tp
=70 °C, 4—Tp=73.5 °C, 5—Tp=64 °C.
70 J. Opt. Technol. 77 �1�, January 2010
of the removal rate depends on the polishing time. It wasshown that the removal rate was significantly increased byincreasing the grain size of the abrasive while keeping theother polishing conditions constant. The greatest removalwas observed when ASM 5 /3 micropowder was used, andthe defects left on the polished surface by this abrasive arealso the largest. Figure 5 shows the distribution functions ofthe points and scratches on the sample surfaces. It should bepointed out that the resulting surface quality increases as thegrain size of the polishing abrasive decreases; i.e., the sizeand number of chips decrease, along with the width of thescratches.
As a result of experiments to study how the hold-downpressure affects the CMP process of zinc chalcogenides, thedependences of the sample thickness on the polishing timeare obtained. Based on these results, the dependences of theremoval rate on the hold-down pressure during CMP of thezinc chalcogenides are constructed. It is shown that the de-pendences have an identical character for both zinc selenideand zinc sulfide. The only difference is observed in the re-moval rates. Thus, the removal rate under identical experi-mental conditions is higher for zinc selenide than for zincsulfide. However, higher-quality surfaces were obtained onthe zinc sulfide samples. This result is associated with thesuperior mechanical characteristics of zinc sulfide by com-parison with zinc selenide. The surface cleanliness and ge-ometry also substantially depend on the pressure applied tothe surface during polishing �Fig. 6�. The range of pressuresused here is as follows: P=19.6–66.7 kPa for a surface areaof 24.96 cm2, and P=7.8–34.3 kPa for a surface area of85.97 cm2m. Experiments were also carried out on both me-chanical polishing and CMP. When zinc selenide and sulfideare mechanically polished, the best results are obtained whena pressure range from 15.7 to 29.4 kPa was used. For CMPof zinc selenide, the best results in surface quality are ob-tained when a pressure range from 39.2 to 49.0 kPa is used.A comparison of the experimental data from mechanical pol-ishing and CMP of zinc selenide and sulfide thus made itpossible to distinguish the optimum pressure range.
Studies of how the rotation rate and the oscillation rateof the hold-down tool affect the CMP process of zinc se-lenide allowed us to obtain the dependences of the removalrate and the surface quality on these process parameters. Therange of variation of the rotation rate is 5.5–23 rpm, and thatof the oscillation rate is 22–140 cycle /min. Figure 7 showsthe defect-distribution function on the surface of the zincselenide as it varies with the oscillation frequency of thespindle. It is shown that the best results in surface quality areobtained at the minimum rotation rate of the polishing wheel.Increasing the oscillation rate of the unit to 100 cycle /minappreciably reduced the size and number of surface defects.A further increase of the oscillation rate to 140 cycle /mindegraded the quality of the surface being processed. It isshown that increasing the processing rate of zinc selenide
70Gavrishchuk et al.
increases the amount of the material being removed from thesample surface and consequently increases the removal rate
0
40
80
120N
umbe
rof
poin
ts
Diameter, µm0 2 4
1
2
3
a
0
0.01
0.02
0.03
0.04
Width, µm
Len
gth,
mm
0 6 12
1
2
3
c
FIG. 5. Distribution functions of points �a and b� and scratches �c and dabrasive-grain sizes. 1—ASM 0.5 /0.2, 2—1 �m Al2O3, 3—ASM 1 /0, 4—
100
200
300
0
Num
ber
ofpo
ints
Diameter, µm0 4 62
1
2
3
4
5
a
FIG. 6. Distribution functions of points �a� and scratches �b� on the surfac2—17.1 kPa, 3—22.5 kPa, 4—27.5 kPa, 5—33.3 kPa.
71 J. Opt. Technol. 77 �1�, January 2010
and reduces the processing time. However, the surface qual-ity and its geometry are degraded in this case.
0
100
200
Num
ber
ofpo
ints
Diameter, µm0 2 4 6
6
4
5
60
0.01
0.02
Width, µm
Len
gth,
mm
0 63 9
4
5
d
b
the surfaces of zinc selenide samples with an area of 1 mm2 for various2 /1.5, 5—ASM 3 /2, 6—ASM 5 /3.
0
0.04
0.08
0.12
Width, µm0 2 4
1
5b
f zinc sulide samples for various pressures during polishing. 1—11.4 kPa,
� onASM
Len
gth,
mm
es o
71Gavrishchuk et al.
Surfaces of zinc sulfide samples of various qualitieswere obtained as a result of the experiments carried out bymechanical polishing.6 The cleanliness class of the surfaceswas determined according to GOST 11141-84, and theirquality corresponds to cleanliness classes 7, 5, and 3. Micro-graphs of these surfaces were then processed according tothe computer-vision technique described above. As a result,complete agreement between the results of computer and vi-sual monitoring was obtained. Figure 8 shows the distribu-tion functions of the scratches and points on the surfaces ofthe zinc sulfide samples for the three surface-cleanlinessclasses. Curve 1 corresponds to cleanliness class 7, curve 2to cleanliness class 5, and curve 3 to cleanliness class 3.
IV. CONCLUSION
Our studies have shown how the thermophysical charac-teristics of the polishing resin, the pressure, the grain size ofthe abrasive being used, the rotation rate, and the oscillationfrequency of the processing tool affect the processes of me-
0
0.02
0.04
0.06
0.08
Width, µm
Len
gth,
mm
0 2 4 6
1
2
3
4
5
0
20
40
60
Num
ber
ofpo
ints
Diameter, µm0 4 8
1
2
3
4
5a
b
FIG. 7. Distribution functions of points �a� and scratches �b� on the surfacesof zinc selenide samples for various oscillation rates of the spindle �rotationrate of polishing wheel 5.5 rpm� during chemomechanical polishing.1—22 cycle /min, 2—44 cycle /min, 3—69 cycle /min, 4—100 cycle /min,5—140 cycle /min.
72 J. Opt. Technol. 77 �1�, January 2010
chanical and chemomechanical polishing of zinc selenideand sulfide and the quality of the resulting surfaces.
The main principles have been determined, and a tech-nique has been developed for recognizing the form of theresulting defects on images of the polished surface of zincchalcogenides. Algorithms and software have been proposedfor obtaining detailed quantitative characteristics of the sur-face defects, making it possible to evaluate the cleanliness
0
200
600
400
Width, µm
Len
gth,
mm
0 4 8 12
1
2
3
0
0.4
0.8
1.2
Num
ber
ofpo
ints
Diameter, µm0 4 8
1
2
3
a
b
FIG. 8. Distribution functions of points �a� and scratches �b� on the surfacesof zinc sulfide samples for various surface-cleanliness classes.
72Gavrishchuk et al.
class of the polished surface in accordance with the currentstate standards.
1M. A. Okatov, Handbook for Optical Technicians �Politekhnika, St. Pe-tersburg, 2004�.
2E. M. Gavrishchuk, O. V. Timofeev, A. A. Pogorelko, and A. I. Suchkov,“The effect of the polishing conditions on the optical properties of thesurface of zinc selenide,” Neorg. Mater. 40, 267 �2004�.
3E. M. Gavrishchuk, V. V. Potelov, B. N. Senik, and O. V. Timofeev, “The
effect of the polishing conditions on the processing quality of the optical
73 J. Opt. Technol. 77 �1�, January 2010
surfaces of zinc selenide elements for articles that operate in the IR re-gion,” Prikl. Fiz. No. 5, 107 �2005�.
4O. V. Timofeev, S. R. Kushnir, E. M. Gavrishchuk, and B. A. Radbil’,“The development of polishing and cementing resins for fabricating opti-cal elements from CVD ZnSe,” in Abstracts of Reports of the TwelfthConference on High-Purity Substances and Materials �Production, Analy-sis, Application�, Nizhni� Novgorod, 2004, p. 299.
5G. G. Devyatykh, I. A. Korshunov, E. M. Gavrishchuk, G. L. Murski�, S.V. Ignat’ev, and D. V. Nikonenko, “Study of the volume inhomogeneitiesin polycrystalline zinc selenide obtained by chemical deposition from thevapor phase,” Vysokochist. Veshchestva 3, 16 �1993�.
6E. M. Gavrishchuk, E. Yu. Vilkova, O. V. Timofeev, B. A. Radbil’, and S.R. Kushnir, “Study of the process of polishing polycrystalline zinc se-lenide using rosin-based polishing resin,” Opt. Zh. 759, 83 �2008� �J. Opt.
