Reticle programmed defect size measurement using low voltage SEM andpattern recognition techniques

Larry Zurbrick, KLA-Tencor Corporation, San Jose, CA, USASteve Khanna, KLA-Tencor Corporation, San Jose, CA, USAJay Lee, KLA-Tencor Corporation, San Jose, CA, USAJim Greed, VLSI Standards Incorporated, San Jose, CA, USAEllen Laird, VLSI Standards Incorporated, San Jose, CA, USARene Blanquies,' VLSI Standards Incorporated, San Jose, CA, USA

Abstract

The use of programmed defect test reticles to characterize automatic defect inspection equipment has long been anestablished practice in the maskmaking industry. Measurement of the defect sizes on these programmed defect test masks isnot necessary if one only desires to qualitatively investigate differences in system performance. However, more meaningfulcomparisons in inspection system performance require a calibrated programmed defect test mask. Historically, commerciallyavailable programmed defect test reticles have not had traceable or well-documented defect sizing methods nor wasinformation regarding the precision of these measurements provided. This paper describes the methods used and resultsobtained from the work performed to address these issues. Using a low voltage scanning electron microscope as an imageacquisition system, defect sizing is accomplished using automated pattern recognition software. The software reports defectsize metrics such as maximum inscribed circle diameter and area. Measurement precision better than 30 nm has beendemonstrated for the maximum inscribed circle method. The conelation of SEM based measurements to historical opticalmetrology measurements is also discussed.

1 Introduction

Programmed defect test masks such as DuPontPhotomasks' VerimaskTM and VerithoroTM test masks areutilized to test defect detection sensitivity of automateddefect inspection equipment. The defect sizemeasurements provided with these test masks are stated asthe one-dimensional height of the defect. Thesemeasurements are performed using a manually operated,visually based optical microscope utilizing an imageshearing measurement method. This microscope useswhite light supplied by a tungsten halogen lamp, a 100X0.9 numerical aperture objective operating at a 0.7 sigma,and lOX eyepieces. For programmed defect sizes lessthan 0.5 .tm and linewidths less than 1 im, therepeatability of this measurement method is inadequatefor ensuring inspection system performancespecifications. Figure 1 shows a histogram of the typicalrepeatability of this measurement method and the SEMmeasurement method discussed herein for a VerithoroTM69OEXS test mask. The histogram was constructed byperforming a same defect pair-wise comparison betweentwo sets of measurements, calculating the absolutedifference between the defect measurement pairs, andbinning these absolute differences.

Fig. 1 Comparison of Optical and SEM basedmeasurement repeatability based upon pairedmeasurement differences

i_ii•o o 0 0 0 0 0 0 0C\J (0 (0 0 C'J (C) CO

0 - ' '('4 - (0 co 0 ('4 (0

The optical measurement repeatability appears to be nobetter than 180 nm. The source of this lack of

64In 1 6 European Conference on Mask Technology for Integrated Circuits and Microcomponents,

Uwe F. W. Behringer, Editor, Proceedings of SPIE Vol. 3996 (2000) • 0277-786X100/$1 5.00

90

80

Cl) 70C)aC)

50aa 400

30:$$: 20

100

•Optical

I.''0 00 0('4 ('4

A

CO

Bin Range (nm)

repeatability is related to the small image scale(magnification) presented to the microscope operator andrelatively large point spread function of the optics ascompared to the defect sizes of interest (resolution).Figure 2 illustrates the image scale that the microscopeoperator observes for a VT69OEXS type test mask (thechrome edge defect is circled).

Fig. 2 Typical Optical measurement image scale

In order to improve the measurement repeatability, higherresolution and magnification are required. With theintroduction of CD measurement scanning electronmicroscopes (SEM's) specifically designed for masks andreticles such as the KLA-Tencor SIOOXP-R, issues ofsubstrate handling and sample charging have been largelyeliminated. Figure 3 shows the same defect as Figure 2imaged at 50K)( with the KLA-Tencor 8100XP-R CDSEM. With the improve resolution, magnification andstored digital image, several different measurementmethods (e.g. area, one dimensional height, XY houndingbox) can he employed. However, a measurement methodthat closely correlated to historical optical measurementswas desired.

Fig. 3 SEM image of defect shown in Fig. 2

Defect imaging was performed with a KLA-Tencor8IOOXP-R CD SEM equipped with charge reductionhardware and utilizing secondary electron detection. Thelanding energies utilized for imaging were in the range ofI to 2 kV.

Image analysis software was written to perform edgeextraction, reference image alignment, defect extraction.and defect measurement on stored SEM digital imagesgathered with a KI.A-Tencor flOOXP-R CD SEM. Theimage analysis operation utilizes a reference image(without a programmed defect) in order to perform thedefect measurement. A reference image was utilized inorder to compensate for intrinsic corner rounding on thetest mask patterns. This was primarily done for themeasurement of corner type defects. The defectmeasurements reported by the software include area,maximum inscribe circle diameter, and hounding box XY1dimensions.

To determine the correlation between historical opticalmeasurements and SEM based measurements. SEMimages from multiple VerlthoroTM masks were gatheredand defect size measurements and comparisons to theoptical defect size measurements performed.

65

2 Experimental

66

3 Results and Discussion

Initially, one dimensional defect height measurementswere performed with the SEM mimicking the opticalmeasurement method. It was soon discovered thatprogrammed defect shape was not well controlled.Figure 4 shows two chrome extension defects ofapproximately the same height from two different testmasks. It can be seen that the defect shape variesconsiderably between the masks. It was decided that aSEM based one dimensional height measurement wouldnot be adequate since it only provided limited informationregarding the defect size. This also implied that ahounding box XY size measurement would have similarlimitations (since the Y dimension of the bounding box is

essentially the one dimensional defect height).Additionally, the X dimension of the bounding box could

vary considerably depending upon line edge roughnessand the gradual slope that exists at the base of somedefects (referred to as "tails").

Fig. 4 Defects of similar height but differentshape

Defect square root of area was investigated as a defectsize measure. The square root of area was of interestbecause it states defect size in terms of an equivalent"square defect" and that the measurement unit, micron(tim), is much more familiar than that of square microns(ftm2). It was observed that defect size based upon thismeasurement method could be influenced by sample lineedge roughness and straightness on the defect andreference images. Figure 5 illustrates this point. Infigure 5, defect C3 (a chrome extension) from twodifferent masks are compared. The reference edge issuperimposed as a white line on the defect image alongwith the maximum inscribed circle. It can he seen thatsignificant "tails" in terms of area can develop as in thecase of the defect shown on the left (snlOl9) of the figureas compared to the image on the right (snlOSO).Although the defect on the left is smaller in height, itmeasures greater in square root of area, It was observedthat the area in the "tails" can he greatly influenced by theedge roughness/straightness in the reference image. This

type of change in defct area (i.e. low aspect ratio "tails)and its effect on defect printability is not entirely clear.Further investigations into the relationship of defect shapeupon printability will he necessary. Additionally, use ofsquare root of defect area for defects such as edgemisplacements (CD error and misplaced contacts del'ectsand pinholes/pindots conflicts with the accepted definitionfor these dctct types. It was desired to have a singlemeasurement method that could he used for a wide rangeof defect types.

Fig. 5 VT69OEXS defect C3, snlOl9 (left) andsnlOSO (right)

0.26 MinMax.0.22 pm

A compromise between square root of area and onedimensional dekct size is the maximum inscribed circlediameter method of measurement. This measurementmethod determines the maximum diameter circle that canhe fit into the identified defect. This method has theadvantage of working with edge misplacement, edge.corner, pinhole and pindot defects. See Figure 6 forexamples.

Sq. Root Area = 0.292 JimInscribed Circle = 0.I tim

Fig. 6 Examples of maximum inscribed circledefect sizing method. Defect sizes range from 0.094to 0.250 mm.

One of the largest issues in determining an averagecorrelation between historical optical measurements andSEM based measurements was the variability in thehistorical optical measurements. Figure 1 shows therelationship between optical and SEM based InscribedCircle Diameter (lCD) measurement repeatability for thesame VerithoroTM 69OEXS. The histogram was generated

by performing a pair-wise comparison of measurementsof the same defect on the same mask from twoindependent optical measurement data sets. As seen inFigure 1, 95% of the optical repeatability data extendsover a range of 0 to 160 nm. Using the same test mask.three independent sets of SEM images were captured at5OKX magnification over the period of two weeks and themaximum lCD defect sizes determined using the imageanalysis software. The absolute size range wasdetermined on a defect by defect basis and plotted on thesame graph for comparison purposes. For the SEM basedlCD measurement repeatability, 95% of the data is in the0 to 20 nm range with a maximum observed difference of

3() nm. The defects included in this repeatability studyinclude edge defects (rows A through I)), corner defects(rows E through H), pinhole defects (row S). and pindotdefects (row T). Figure 7 shows defect C3 (chromeextension) troni tour different Verithoro IM 69OEXSmasks. All four defects appear to he similar in size andshape, with the SEM Id) si/cs in the range of 0. I 8 to0.22 him. However, the optical one dimensional defi.ctheight measurements differ considerably ranging in sizefrom 0. 13 to 0.27 J.tm and do not appear to correlate to theSEM lCD sizes. The smallest optical measurementcoincides with the largest ICI) measurement (Sn 1050).Examination of the defect sizes and shapes reaffirms thebelief that the optical measurement data is not repeatable.

XY plots of the Optical minus SliM ICE) measurementdifference versus SEM LCD defect size were generated bydefect type. See Figures 8 and 9. The different symbolsin each plot represent a difkrent serial number test mask.Analysis of this data showed that clear intrusions, clearnotches on chrome corners, clear extended corners, andchrome notches on clear corners had an average

Fig. 7 VT69OEXS, detectOptical measurements showdefect physical size

C3. multiple test masks.poor correlation to actual

SEM lCDOptical sizeSerial #

0.18 Jim0.25 Jim 0.20 JimsnlOl9 snlO4()

SEM LCDOptical sizeSerial #

0.18 Jim(1.27 Jimsn1047

0.22 m1)13 Jimsn 1050

67

68

difference less than 0.05 tm over the defect size rangestudied with a slope close to zero. As these defect sizesapproached 0.70 rim, the variability of the data decreasedand approached a difference value within 0.05 .tm ofzero. It should be noted that the point spread function ofthe optical measurement instrument is approximately0.75 pm in diameter. For extended chrome corners,values less than 0.60 im averaged approximately 0.05 .tmand did not appear to converge to zero for larger defectsizes. Chrome extension defects exhibited a morecomplex correlation behavior. Regression analysis of thechrome extension defects showed a positive slope of0.267 I.tn1/tm with an intercept of —0.0 19 over the defectsize range studied. It is not entirely clear why chromeextension defects did not appear to converge to a fix valueat the larger sizes, but may be due in part to the defectsnot reaching a large enough size. It also should be notedthat the regression R2 value was only 0.35 which indicatesa very low degree of correlation. Pinhole defectsexhibited two interesting behaviors. First, it appears that

the data is bimodal in that two distinct groupings of thedata occur by serial number of test mask. The exact causeis not know, but may be related to an operator bias. Thisgrouping of data does not occur for the other defect types.Second, the data appears to indicate that pinhole opticalmeasurements change slope when the pinhole size is lessthan the optical measurement system's point spreadfunction diameter, approximately 0.75 .tm. Based uponapriori knowledge of the measurement characteristics ofthe optical tool, a 0.15 to 0.20 xm bias was expectedbetween the two measurement techniques where theoptical tool would measure pinhole sizes smaller (and theopposite being true for pindots). This is true for the uppergroup of the pinhole data larger than 0.75 jim. Pindotdefect data exhibits a more random distribution with anaverage difference of 0.178 jim. The pindot defect opticalmeasurements are larger than the SEM measurementswhich agrees with the apriori expectation.

Fig. 8 Defect size correlation for edge and corner defects

Defect Type D . Chrome Extension

0.25 . .0.20 U0.15 1 axx ;U. A0.10 • ) Ax0.05 • A

: AU x

0.000 0.200 0.400 0.600 0.800 1.000

SEM CD Size

Defect Type F - Extended Chrome Corner

0.25 -————————- S

0.20 —U.

0.15 • X•X *X. •U0.10 . )c xA. A •0.05 . Ax A0.00 . AA A-0.05 A A

x .-0.10 .-0.15

-0.20

-0.25 I t-

0.000 0.800 1.000

Defect Type A - Clear Intrusion Defect Type B . Clear Intrusion

0.25

0.20

0.15—0.10 --

XAl AX A0.05 x < X0X •X U00.00

U —-0.05 • U U

-0.10 • • . U •—-0.15 • • . .—a -0.200

-0.25 —0.000

E

00C)

UiU)

00.200 0.400 0.600

SEM CD Size

0.25

0.20

0.15 X x0.10 AX A0.05

A AX xA X U0.00 AX • •xx x U U-0.05 • U• U • •-0.10 • • U •.0.15 •-0.20

—0.25

0.000 0.200 0.400 0.600 0.800 1.000

SEM lCD Size

0.800 1.000

E

8

0C.)

LUU)

0

Defect Type C - Chrome Extension

0.25 —————--——-——-——————————————-—- ——

:i: U I,'A : IA —0.0 X*

005 X C

I. • A .x0.00 A S —

-0.05 * • —

-0.10 •. • —

-0.15 —

-0.20 —

—0.25 t , I 4 I —

0.000 0.200 0.400 0.600 0.800 1.000

SEM lCD Size

E

8

0C)

ILlU)

:. -0.20

-0.25

Defect Type E . Clear Notch on Chrome Corner

I8

U)ia0

0.20

0.15 X A -X X A0.10 X A* -x A0.05 *A)k A A U •0.00 A • • •,x) X • U

-0.05 • U • • • -0.10 —

-0.15 —-0.20 —-0.25 —

0.000

6

0C)

LUU)

00.200 0.400 0.600 0.800 1.000

SEM lcD Size

Defect Type G . Clear Extended Corner

0.200 0.400 0.600

SEM cD Size

0.25

- 0.20e 0.15

0.10

0.05

0 0.00C); -0.05LUU) 0.10: -0.15- -0.20

-0.250.000

Defect Type H - Chrome Notch on Clear Corner

X X X • AA A A • —

A AX • * —: • •U

U

U U

: U • —. U —S .

0.25

0.200.15

0.10

0.05

0.00

-0.05

-0.10

-0.15

-0.20

E

8

150C)

UiU)

0

0.200 0.400 0.600

X AX X A AxA XA AX XXAx AXX x U U— . SU •I. .• ••• U U

U U

SEM CD Size0.800 1 .000 0.000 0.200 0.400 0.600 0.800 1.000

SEM CD Size

69

Fig. 9 Defect size correlation for pinhole and pindot defects

70

4 Conclusions

Use of the KLA-Tencor 8 100XP-R CD SEM and customdeveloped image processing software has improved defectsizing repeatability by a factor of 9X. The maximuminscribed circle diameter defect sizing method correlateswith average historical optical measurements.

5 Acknowledgements

VerimaskTM and VerithoroTM are trademarks of DuPontPhotomasks Incorporated.

The authors would like to thank Lantian Wang of KLA-Tencor and Paul Konicek of VLSI Standards for

performing SEM imaging work. The authors also thankDuPont Photomasks, Santa Clara, for use of their KLA-Tencor 8100XP-R CD SEM and providing opticalmeasurement repeatability data.

6 Literature

[1] Semiconductor Equipment and MaterialsInternational (SEMI®) Draft Document 2780B,13 October 1998, Revision to SEMI P22, Guidelinefor Photomask Defect Classification and SizeDefinition

Defect Type S - Pinhole

0.10

0.05

0.00

-0.05

-0.10ao -0.15C.)j -0.20

-0.25

-0.30

-0.35

-0.400.000

0.-,,

Defect Type T . Pindot

AA AX

• - A.*X A • ,-

X •• X•

U

E

0

a0LU(I)

-0.05

0.30 •U U • •

0.25X • •

0.20 £ AX X* •0.15 A A A0.10 • • A. **0.05 .0.00

—u.Ju

0.200 0.400 0.600 0.800

SEM lCD Size

1.000 1.200 0.000 0.200 0.400

SEM lCD Size

0.600 0.800