Uniformity mapping using X-ray diffraction

XRD Mapping

Uniformity mapping using X-ray diffraction Chris Moore, PhilipsAnalytical Today's advanced semiconductor industry increasingly requires process control tools that are robust, fast and easy to use. This article, part of lll-Vs Review's continuing series on characterization, describes the benefits that can gained from using X-ray diffraction techniques in process control.

he c o m p o u n d semiconductor industry, as part of its maturation process, has developed a n eed for several process control tools. Modern c o m p o u n d semiconductor p r oduc t i on lines are more and more heavily staffed by line operators, with fewer and fewer scientists involved in the actual pr oduct i on process. Instruments and techniques that were suitable for scientists have to be significantly modified and adapted to allow operation with minimal user intervention. In fact, one can see the convergence of tool design from the technology-oriented models of the past to the silicon industry operator-based model. Process control tools need to satisfy a basic set of needs:

Reproducibility Average =6.68 ( + / - 0 . 0 2 ) "

T

6.73

N=

6.68 6.63 0

20

Figure 2. A reproducibility measurement for the FWHM of the rocking curve of a 167point spatial map of a Si substrate showing the variation in the average value.

• The technology must measure a useful process control parameter (and not just produce pretty data). • The tool must provide reproducible measurements w h e n run by an operator with minimum intervention.

• Mapping measurements must be sufficiently fast to provide timely uniformity data for the process.

-lEO0

-1450

-1400

Peak Angle FWHM Separation Substrate 943202 -1441.0 121

Figure 1. A double crystal X-ray diffractometry rocking curve obtained from a commercial GaAs substrate (angle measured in arc seconds) (S = substrate).

30

Ill-Vs Review• Vol.11 No. 6 1998

60

Run Number

• The tool should analyse the measurement to produce relevant data without the need of 'expert interpretation'.

-t600 -1550 Rock Angle (")

40

This article will concentrate on c o m p o u n d semiconductor process control uses of X-ray diffraction (XRD). Subsequent articles will discuss the use of room temperature photoluminescence, optical reflectometry and combinations of these techniques. All of the data present-

ed in this article are from typical production samples.

Substrate screening For all epitaxy processes, the starting material for building the structure is the substrate. Imperfections in the substrate or its surface are invariably reflected in the quality of the epitaxy. In general, the p o o r e r the quality of the epitaxy, the p o o r e r the device performance. Although substrate manufacturers have significantly improved their production processes, there can still be large differences from boule to boule of the same material from the same supplier.This problem of substrate variability is made considerably more difficult as one tries to buy substrates from a n u m b e r of different wafer suppliers. In the past, epitaxy growth and substrate cleaning procedures

0961-1290/98/$19.00© 1998 Elsevier Science Ltd. All rights reserved.

XRD Mapping

(mm)

B i n Size ~ 1.00 "

-I'% 0

10

20

~ltidth - A v e r a g e - Std Oev.

30

40

50

80

70

0

5

10

15

20

25

: 12.85 "

: 2109 "

Intensity - U n d e r F - ] 2 4 . 7 %

Figure 3. The spatial variation of the FWHM of the X-ray rocking curve obtained from a commercial 3" diameter InP substrate,

Substrate Correlation 18 -r-

17

u-~

16

= o,. uJ

15 14 11

12

13

S u b s t r a t e F W H M (sec) Figure 4. The observed correlation between substrate rocking curve FWHM and epitaxy rocking curve FWHM.

were modified for each change of bottle material. The push to larger and larger production facilities has led to multiple wafer epitaxial reactors which no longer allow the adaptation of the growth process for each boule change. This increase in throughput has led to a need for robust, reliable substrate screening procedures. After a significant number of comparative measurements, we have found that X-ray screening is a much more reliable indication of substrate quality than secondary probes such as photoluminescence. In this way substrates can be screened for structural quality and for precise crystal orientation. Although X-ray topography has been used for many years to screen boules, it is a very qualitative process and difficult to quantify as a numerical screen.This article will concentrate on the use of X-ray dif-

fraction rocking curves and maps for substrate screening. The basic technology of XRD has been described earlier in this series of articles on characterization [1].All X-ray data presented in this article are from a Philips Analytical RD100 or a Philips Analytical DCDII high speed diffractometer.These systems are double crystal systems using blocked

20001

,oooo

,,oot\ I000~ k 600 4

11055" (CU) M5648 CU)

Offcut = 82686" (0.2300 Fit Error= 3.80 [0.23%]

C)

alpha collimators. This type of system blocks the K alpha2 beam emitted from the X-ray source. Although this lowers the overall beam intensity only one X-ray sampling beam probes a defined sample location. Typical probe beams were 0.5 mm wide by 1 to 3 mm long. This use of a single probe beam is unlike most double crystal systems, which allow both sampling beams onto the sample and blur the resulting data both spatially (because of the two sampling beams) and spectraUy (due to the two different wavelengths of the probe beam). The data plotted in Figure 1 is a typical XRD rocking curve from a GaAs substrate. These data were measured in 0.25 arc second steps and it took 1.5 seconds to measure the entire curve. The width of the rocking curve is a direct measure of the amount of crystalline imperfection (dislocations) inside the volume sampled by the X-ray beam.Typically the Cu X-rays penetrate 15 lam in GaAs, so this is a near-surface volume and not a complete bulk measurement. Simply put, the more damage (imperfections) within the sampling volume, the wider the rocking curve. Normally the FWHM (width of the curve at 50% peak intensity) is used as a sensitive screen for substrate variation from boule to boule or from wafer to wafer within the boule. One requirement for this type of measurement is that the tool

/ \

/

/

\

(oo ~)

1

1155,5" (1558o CL

S

400001 /

(/T~'.., ,q i ", '

',~

3o°1111 ,,,

f

100004

/ ,'/

',,

\,,

~',,.,~_..d" 0 J

"-150"-100

-50 ' I~ " 513 ' 1130'150' Rotation (")

.

,

11 o 113o 11 o 1140"1150 Rock Angle(")

Figure 5. An off-axis 'offcut' measurement from a 0.2 ° misoriented substrate (see text).

IIl-Vs Review • Vol.11 No. 6 1998 31

XRD Mapping

s1~5"

2865 5 "

(cu) (~5o cu)

~.s co)

8800 71300 6800.

t

5000. 4000. 3800• 2000. 1000. J

~o

~

~o

~o

31oo

315o

Rock A n g l e (") Rock A n g l e Step Rotation Tilt Detector Type Crystal

1.0 " 180 ° -200 " Silicon GaAsNew

Peak Substrate 8 2 3 6 5 E p i l a y e r l 5599.4

Angle 3067.2 2946.7

FWHM 1t.2 15.6

Separation -120,5

004 full wafer [Peek: Mid 90 %]

Figure 6. X-ray diffraction rocking curve from an AIGaAs layer grown on a GaAs substrate measured in four seconds using a solid-state detector. S indicates the substrate peak; I is the epi peak.

p r o d u c e s reproducible data w h e n run b y a non-expert. In an X-ray m e a s u r e m e n t this translates to a n e e d for automatic peak finding, automatic peak optimization and robust data analysis, w h i c h does not involve simulation of the data. Figure 2 shows the results of a reproducibility test of a Philips DCD II diffractometer on a Si substrate. The only o p e r a t o r intervention was to load and unload the wafer. The data plotted in this figure is the average value of a 167 point spatial map. The outlying measurements w e r e related to changes in the ambient t e m p e r a t u r e caused b y the heating/cooling system. Reproducibility m e a s u r e m e n t s like this are necessary to give users

(ram)

Separation

i - Average

: .1"18.92 AreSec

- Std. Bey. :

1.58 Ar¢Seo

(rnm) 80"t

'Jr . . . . . . . . . . . . . . . . | i N

[ t | ~ l i ] t l

c o n f i d e n c e in the tool's capability to accurately measure small changes in the FWHM. Figure 3 contains a map o f rocking curve FWHM m a p p e d over an InP substrate. The spatial variation of the FWHM is s h o w n in the colour map o n the left of the figure w i t h the histogram o n the right showing the relation o f the colour to FWHM. Note the large range in values from 12 to 23 arc seconds (a factor o f two in range). The basic quantitative screen for substrate quality consists of t w o parts; a m a x i m u m allowable value for the average FWHM and a m a x i m u m allowable value for the standard deviation. The actual numerical value used for the c u t o f f values d e p e n d s o n the nature of the devices being p r o d u c e d . If the value is set too large, yields go down; if set too small, the material reject rate bec o m e s too high and average substrate costs go up (the standard U curve). This particular s c r e e n i n g param e t e r is sensitive to defects in the n e a r surface and the surface o f the material. If the screening criteria is c h a n g e d f r o m 50% intensity (FWHM) to the w i d t h at 10% o f p e a k intensity, the sensitivity is increased to surface effects such as polishing damage or to inappropriate e t c h back after lapping and polishing. This variation in FWHM has a direct effect o n the quality o f the epitaxy g r o w n o n such a substrate.

t [ai~

t|i]al41|:it

~eparation - Average : - 1 1 8 . 4 7 A f e S e o - S t d . Oev, : 1 . 0 7 AroSeo

1o 2o

Figure 7. The spatial variation of the diffraction peak separation for two sample rotations: (above) 0 ° and (below) 180°. Their agreement demonstrates that tilt is very important in this system.

III-Vs Review • Vol.11 No. 6 1998 32

Data V=lue - Avetlge • Std Oev. • Median • 10 16 - O0 16

30

4o

: 41.01% : 0,54 % : 41.1o 16 : 40,23 16 : 41,61 16

oo',~-¢o

"e~ Ir4endty

,Under •over Thfe~,hold .BeJoul -,&bowe

r-13*3,4% [ ] o.o *4 • 0,0 % • 0.0 16

Figure 8. The spatial variation orAl composition for the sample of Figure 7 as determined by X-ray diffraction.

XRD Mapping

The effects of substrate damage o n epitaxial g r o w t h can be clearly seen in Figure 4. This is a plot of the FWHM of the diffraction peak from the epitaxial layer (plotted o n the y axis) versus that from the substrate peak (plotted as the x axis).The data shows that as the substrate FWHM increases, the epitaxy FWHM also increases. In fact, the substrate FWHM sets a l o w e r limit for the epitaxy FWHM and, in effect, determines the best case scenario for the quality o f the epitaxy. Imagine the spread in this data set w h e n the substrate FWHM varies by a factor of two (like the substrate in Figure 3)! The s e c o n d type of XRD screen for substrates measures the orientation of the lattice planes with respect to the physical surface of the sample. Substrate orientation is det e r m i n e d w h e n the boule is cut and is not related to the lapping and polishing processes used to make the substrate surface optically flat for lithography and growth. A n u m b e r o f studies have b e e n d o n e w h i c h relate the misorientation of the substrate to the dislocation density of epitaxial alloys g r o w n o n the substrate. Because of this sensitivity, typical substrate misorientations used in p r o d u c t i o n processes can vary from 0 degrees to as m u c h as 15 ° off-axis, determined by the structure and type of devices being p r o d u c e d . Substrate manufacturers normally measure this misorientation by measuring crystal orientation (using simple diffraction) at two different rotation angles. The crystal misorientation determines the angle at w h i c h Bragg's law is satisfied. If the lattice planes are tilted with respect to the physical surface o f the sample, the angle at w h i c h Bragg's law is satisfied (that is the angle w h e r e the substrate peak is found) describes a sinusoidal variation with the rotation angle of the substrate. The amplitude of this variation determines the misorientation and the average value ( w h e n related to

-4039 0 " (CU) (-7 6 CU,~

(mm)

0

10

20

Rock Angle Step Rotation Tilt Detecter Type Crystal

30

40

(26, 26 ram)

-3439 0 " F4.1 CU)

-4000 -3900 -3800 -3700 -3600 -3500 Rock Angle (")

50

05 " 0 ° -8000 Silicon GaAsnew 004

Peak Angle FWHM Substrate 55413 -3633,0 164 Epilayer 1 310I 7 -3774.1 20,0

Separation -141.1

[Peak: Mid 90 %]

Figure 9. A typical rocking curve of an InGaAIP epitaxial layer grown on GaAs. The two peaks show a small lattice mismatch.

a calibrated sample) can be used to d e t e r m i n e the lattice constant of the substrate. However, this technique only works if the sample tilt is optimized at each m e a s u r e m e n t rotation.A misorientation measurem e n t for a 0.2 ° (nearly on-axis) crystal is s h o w n in Figure 5.The actual measured offcut (the angle that the substrate was cut off axis) was 823 arc seconds or 0.23 ° . The data o n the right side of Figure 5 is the actual rocking curve measured o n this sample. The data on the left plots the measured peak angle (after tilt optimization) as a function of rotation angle of the substrate. The solid line o n the left is the sinusoidal fit to the data using a 0.23 ° offcut angle.

Ternary epitaxy screening Bragg's law states that the angle at w h i c h diffraction occurs is directly related to the lattice p a r a m e t e r of the layer being sampled. In practice, this relationship can be used to d e t e r m i n e the lattice constant of an epitaxial layer assuming that the lattice constant of the substrate is well known. Figure 6 is a plot of a rocking curve for an AIGaAs on GaAs structure. Because AIGaAs is a ternary material, the composition of the alloy can be d e t e r m i n e d from the peak separation, assum-

ing that the epitaxy is not tilted with respect of the substrate (this can be verified by measuring the peak separation as a function of rotation of the sample) and the substrate lattice constant is known. This calculation leads to a technique of high precision (the ability to see small changes) but low accuracy (due to assumptions about the substrate lattice constant). Figure 7a shows the spatial variation of the peak separation of the 3')~;0 s (CU) 16 q h~

(48, 26 mm)

-3300 3 " (23 4 CL

'°1i

.oo,,

!/

t°°°I

J

';J

,j<

-°3~ -3~o .3~oo -3~o° -'~o -3X® ' Neck Angle (")

-4074 'fl " (CU) d -7, 70OO

(16, 10 mm)

-34749 (-0 5 CU

60OO 5OOO 4000

S

!?

30oo

O

i::l .

,

,

,

.

,

.

,

.

r

,

-4000 -3900 -3800 -3700 -3600 -3500 Rock Angle (")

Figure 10. Two rocking curves from different parts of the sample in Figure 9 showing local spatial variation in diffraction conditions.

III-Vs Review • Vol.11 No. 6 1998

33

XRD Mapping

(ram) ~° ~

(Arc:3e¢) , -20:3.o, I

Bin Size is9.20 Af0:3¢0.

I

40:

II =

i

.102 3. I

ill

I:.1too.

2o

?

1:

:

[

!; I

0

t0

20

:30

40

~i0

•

0

5

10

15

20

25

30

Separation-Average

: .1tq.:3 Ato:3eo • $td. Dev. : 17.5 A r ¢ : 3 e o • Median : -110.0 Aro:3e¢

Inten.~ity

• 10 % • 90 %

Threshold - B e l c ~ • -Above •

: -138.2 Ar¢:3eo : -00.2 Ar¢:3e¢

-Under [ ] 2 7 . 1 % -Over [] 0.0 % t.5 0.5 %

[Peal(: Mid gO %]

Figure 11. The diffraction peak separation map for the sample in Figure 9.

same AIGaAs structure as in Figure 6. Figure 7b is the same sample measured at 180 ° sample rotation. If the epitaxy lattice planes are tilted with respect to the substrate lattice planes then the measured peak separation will change with sample rotation. If epitaxy tilt is a problem, one would normally measure the two rotations shown and average the result to remove this effect. In the data shown in Figure 7, a change in average peak separation from -118.92 arc seconds at 0 ° changed slightly to - 118.47 arc seconds at 180 ° . In this case there is very little epitaxy tilt evident. A spatial map of Al composition of an MBE grown sample is contained in Figure 8. Note the circular symmetry evident in the spatial map (to the left of the figure). This pattern is caused during production by the combination of sample rotation, non-uniform substrate temperature and spatial variation of the source flux. This particular composition map used a previously published calibration [2]. Unfortunately, there are no traceable standards for composition measurements of this type and a number of 'peak separation to composition' correlation curves exist. Fortunately for process control, the reproducibility of the measurement (precision) is of more importance than the accuracy.

III-Vs Review = Vol.11 No. 6 1998 34

A statistically significant spatial map for uniformity requires 150 or more spatial measurements. TypicaUy this represents a 5-10 minute measurement using a solid state detector which has limited dynamic range but high speed.This detector works well for thicker epitaxial layers but does not have the dynamic range to measure thin epitaxial layers. For thinner layers, a Xe proportional counting detector has been satisfactory. This system has a dynamic range in excess of six orders of magnitude. The automated analysis of thin layer structures such as HEMTs and laser structures will be discussed in a later article in this series.

Quaternary epitaxy screening A typical rocking curve for a quaternary system is shown in Figure 9, measured at the location shown by the cursor on the left-hand spatial map. The sample is an InGaAIP epitaxial layer on GaAs. Figure 10 shows two additional rocking curves taken at selected locations on the sample. Note the significant variation in the data sets. It is this variation that reinforces the need for a minimum of 150 point spatial measurements. Figure 11 shows the spatial variation of the peak separation of the InGaAIP layer to

the substrate. The circular symmetry is the result of wafer rotation during MOCVD growth. Unfortunately, for a quaternary alloy a single measurement by either photoluminescence, to give the bandgap energy, or lattice constant is insufficient to determine the alloy composition. However, the combination of photoluminescence peak wavelength (to determine bandgap) and X-ray diffraction peak separation (used to determine lattice constant) can be used to uniquely determine alloy composition. The generation and use of these alloy maps will be discussed in a later article.

Conclusions This article has concentrated on describing a few robust, simple quantitative screens based on X-ray diffraction that can be used for compound semiconductor process control. Although the basic measurement technology can be very complicated, and the resultant data sets difficult to interpret, there is significant benefit to be gained by these approaches.The bottom line for building an epitaxial structure is similar to that for building a house: the poorer the foundation (substrate), the poorer the wall (epitaxy).

References [1] B.Tanner, III-Vs Review, vol 11, No. 4 (1998) pp. 24-29. [2] I.Bassignana et al., J. AppL Crystallogr 22 (1989) pp. 269-276. Contact:

Chris Moore Philips Analytical 101 Randall Drive Waterloo

Ontario Canada N2V 1C5. Tel: + 1-519-746-6260 Fax: + 1-519-746-8270; E-mail: [email protected] URL: www.analytical.philips.com