Using photoluminescence to map uniformity

PL mapping

Using photoluminescence to map uniformity Chris Moore, Philips Analytical

I

In this latest article in III-Vs RevieWs characterization series, the spotlight turns to photoluminescence (PL). From qualitative assessment of substrates to quantitative composition mapping of quaternary materials, PL is proving a powerful tool in the control of c o m p o u n d semiconductor processes.

T

he maturation of c o m p o u n d semiconductor process technology and the need for high-volume high-yield processes has created a need for process control tools similar to the silicon industry's operator-based model. As discussed in an earlier article (see 'Uniformity mapping using X-ray diffraction', III-Vs Review, Vol 11, No 6, 1998, pp. 30-34) these tools must satisfy a set of basic needs, ineluding minimal operator intervention and sufficient measurement speed. This article will concentrate on the use of photoluminescence (PL) and PL mapping measurements for the control of c o m p o u n d semiconductor processes. Because PL provides information about the bandgap of the material, it closely matches the assessment requirements of many c o m p o u n d semiconductor processes. Measurements of the spectral response as a function of probe p o w e r are beyond the scope of the current article. For readers interested in these and similar aspects, the basics of the PL measurement and techniques can be found elsewh er e in many texts on luminescence.

Substrate screening The final device performance in many epitaxy processes relies heavily on the quality of the starting material.Thus, substrate screening to assess initial damage or impurity variations is an important III-Vs Review ° Vo1.12 No. 21999 40

areas of high PL intensity. Unfortunately, in a real substrate the intensity is also affected by changes in doping and by the carrier recombination velocity at the surface of the substrate. In many cases, therefore, the intensity of PL from a substrate is dominated by the last surface treatment that the substrate received and not the substrate defect density. Substrate screening by PL should thus be treated as a qualitative, inspection type of measurement rather than a quantitative measurement technique. For a quantitative determination of crys-

quality control tool for minority carrier devices. A number of groups have attempted to use integrated PL intensity as a quantitative measurement of substrate quality. Typically this measurement is done by filtering out the laser beam and collecting all other wavelengths emitted by the sample, which are passed by the detector bandpass filter to the detector. This type of screening is based on the assumption that the controlling factor in the change in PL intensity is the defect density of the substrate.Thus, areas of low PL intensity have more damage than

(a)

.C

PLM100 Reproducibility Average = 1184.88 (_+ 0.005) nm

1184.95

C

e

> m

t

1184.85

¢

#

4)

4

¢

e

e

4

4

e

e

4~

e

#

¢

¢

,X

R 1184.75

(b)

6300

f

3

i

i

5

~

~

i

i

i

]

7 9 11 Run Number

;

i

13

i

i

i

1

15

PLM100 Intensity Reproducibility Average = 6172 (+ 11) Calibrated Units

6200 C

6100 6000 1

3

5

7 9 11 Run Number

13

15

Figure 1. Reproducibility of measurement of a PLMIO0: (a) wavelength, (b) intensity.

0961-1290/99/$ - see front matter © 1999 Elsevier Science Ltd. All rights reserved.

PLmapping

P I!

,z.B

Figure 2. The spatial variation of the PL intensity from a device structure.

talline quality, t e c h n i q u e s such as high s p e e d X-ray d i f f r a c t o m e t r y (XRD) should be used.

Intensity mapping Unlike substrates, the PL intensity can b e used to quantitatively measure the quality of m o s t epitaxial layers. Although f e w device correlations have b e e n published, PL intensity is currently used to assess d e t e c t o r and laser structures and has b e e n successfully used to optimize c h a n n e l quality in high electron mobility transistor (HEMT) devices. In the case of a buried epitaxial layer, interface r e c o m b i n a tion effects also t e n d to limit device p e r f o r m a n c e . Due to the nonlinear r e s p o n s e of the PL intensity w i t h p r o b e and s a m p l e conditions, the attainment of g o o d reproducibility of PL intensity places strict r e q u i r e m e n t s o n the optical design and choice of lasers u s e d to c o n s t r u c t the tool. Figure 1 s h o w s the results of a reproducibility test o f a Philips 'PLMIO0' PL m a p p i n g system for b o t h the e m i t t e d w a v e l e n g t h and the PL intensity.The sample u s e d in this test w a s a PIN d e t e c t o r structure. This test w a s p e r f o r m e d as a series of w a f e r m a p s , w i t h the only o p e r a t o r i n t e r v e n t i o n b e i n g the load and unload of the wafer. Although p a s t a t t e m p t s to quantita-

tively assess epitaxial layers b y PL intensity w e r e h a m p e r e d b y the reproducibility of the m e a s u r e m e n t system, this test d e m o n s t r a t e s that this is no longer the limiting factor. T h e use of intensity as a quantitative screen is normally d o n e b y setting a m i n i m u m a c c e p t a b l e value for the average of the intensity ( w h i c h decreases as the defect density increases) and a m a x i m u m limit for the standard deviation or uniformity. In o r d e r to achieve this, the p r o b e p o w e r and density m u s t b e constant, a situation that is better achieved using an active, p o w er-controlled laser source. Also, b e c a u s e the PL intensity is sensitive to the sample t e m p e r a t u r e , precise m e a s u r e m e n t s m a y n e e d temperature controlled sample holders. Figure 2 s h o w s the spatial variation of PL intensity at 925 n m for an emitting s t r u c t u r e . T h e large variation in intensity was the resuR

of a nucleation g r o w t h p r o b l e m caused b y t e m p e r a t u r e variations across the substrate surface during growth. The numerical effect of this i n h o m o g e n e i t y can b e seen in the large standard deviation of o v e r 20%. In practice, it is also n e c e s s a r y to use an 'exclusion radius' o n the sample data to maintain statistical integrity. T h e exclusion radius allows the user to e x a m i n e the spatial uniformity within the area used for device production only. D e p e n d i n g o n the p r o c e s s this m a y require an edge exclusion of 1 to 5 ram. In all m a p p i n g assessm e n t s of this type it is i m p o r t a n t to k e e p the m e a s u r e m e n t and analysis p a r a m e t e r s the s a m e f r o m run to run. A simple change of varying the exclusion radius can have a significant i m p a c t on the m e a s u r e m e n t of uniformity (Table 1). Here w e have analysed the data f r o m Figure 2 at various exclusion radii. Note the i m p r o v e m e n t in the uniformity, f r o m 23% to 20%, as w e exclude m o r e and m o r e of the edge area. For m a n y processes, however, the uniformity n e a r the edge of the w a f e r is the determining factor in device yield. High-resolution spatial m a p s of PL integrated intensity can also b e used as a sensitive tool for imaging the electrical effects of defects in epitaxial layers. Figure 3 s h o w s a high-resolution spatial m a p of a pair of oval defects in an AIGaAs epitaxial layer. T h e s e measurements were m a d e using an 'SPM210' e q u i p p e d w i t h high-resolution optics. U n d e r these condi-

Table 1. The effect of exclusion radius on statistical results

Edge e x c l u s i o n (mm) 0 1 2 3 4 5

Average intensity

Sigma

Uniformity

816 816 810 804 801 802

193 191 182 175 170 163

23.6 23.4 22.4 21.7 21.2 20.3

(%)

IIl-Vs Review • Vo1.12 No. 2 1999 41

PL mapping

(idm) 100

50

0

50

100

150

200

Figure 3. A high resolution PL map of a pair of oval defects.

tions, the spatial resolution of the data is 0.5 pro. The changes in PL intensity are caused by local changes in defect density. It is interesting to note that PL is one of the few techniques w h o s e contrast and signal intensity improve with material quality: the better the epitaxial layer, the easier it is to see the effects of small defect structures. Similar imaging techniques can be applied to cross-sectional PL intensity measurements. Figure 4 s h o w s the variation in PL intensity for the cross-section of a visible wavelength device structure. Significant effects o n the luminesc e n c e efficiency from a localized defect can be clearly seen in the intensity map. Measurements of this type can give efficient indication of localized g r o w t h or strain problems. This type of measurem e n t should be regarded only as an inspection (qualitative) type screen.

Spectral mapping

Figure 4. A high resolution cross-section map of an AIGaAs device.

Single PL Spectrum 1998 Feb 13 10.05 893 J nr'r 133 4 C I +

(cu) 700 t

600t +

~004 ~ 0 4 PPe~ I n t , ~ y

~00t ..--

F'wHM

100-~ *n=

,

8OO

,

.

•

~,0 Pe~ ~v'4weter~

,

.

.

.

.

900

,

-

.

.

9¢~0

,

lg0

1000

Wa~,~elenglh (nm) Spectral Step L~sar Lena Detector Type

10 nm Green BIG Silicon

Gratm0 Sl~t W=dlh Gain NDF Setling

Vieible O pm 8 N/A O(~

Peak Inlan~dy Peak W~elen~h Full Width at Half Ma~ Upper HalfMa~. W~alen~lth Lower Half Ma~. Wavelen~lth

05rim ~tep, 1mrn elil, 532nm laser [FocueedJ[Peak: Mid 9 5 %

433 5 CU 069 8 n m ~ . 0 nm ~10 1 nm ~ Inm

I

Figure 5. A typical PL spectrum emitted from a multilayer structure showing the parameters normally derived from the measurement.

The most familiar quantitative use of PL is the m e a s u r e m e n t of emission spectra, that is, the measurem e n t of intensity of PL emitted from a sample as a function of wavelength (Figure 5). There are four basic parameters that can easily be extracted from this data: the peak wavelength, the peak intensity, the full w i d t h at halt" m a x i m u m (FWHM) and the u p p e r wavelength at w h i c h the PL has d r o p p e d to half the peak intensity. Care should be taken in the c h o i c e of algorithm for measuring the peak wavelength. Algorithms typically used in the industry vary d e p e n d i n g on the c o m p a n y and the nature of the sample. The simplest algorithm c h o o s e s the wavelength at w h i c h the intensity is a maximum. Although widely used, this is m o r e sensitive to measurem e n t noise than algorithms that use the midpoint of the wavelength b e t w e e n t w o arbitrary intensity levels. Often midpoint 90%

III-Vs Review ° Vo1.12 No. 2 1999 43

PL mapping

or midpoint 85% algorithms are used for LED structures. Even more advanced algorithms such as peak modelling or peak fitting are used with variable results d e p e n d i n g on w h e t h e r the material system is ternary or binary. Whatever algorithm is used, it must be used consistently, as applying different algorithms to the same sample data set gives a noticeable shift in peak wavelength (Table 2). In general, the m o r e sophisticated the algorithm the more sensitive it is to m e a s u r e m e n t conditions and system noise. This is particularly true of algorithms using numerical derivatives to mark the peak or inflection point. The physical processes controlling these four basic parameters are not the same. The spatial changes in Figure 6 are the result of m a p p i n g a quaternary layer and extracting the four basic PL parameters. With the advent of diode array detectors, these maps, w h i c h used to take a few hours to produce, can n o w be d o n e in a few minutes. Statistically, as long as more than 150 spatial points are sampled the uniformity and standard deviation are valid numbers. The patterns seen in the peak intensity (Figure 6a), peak wavelength (6b) and FWHM (6c) are quite different indicating that the spatial variation in each of these parameters is caused by different physical p h e n o m e n a . The peak intensity pattern is controlled by the defect density of the layer and is related to the substrate and the particular cleaning p r o c e d u r e used for

Peak Intensity (mm) 8308.5 + 1087.2 CU

20

,

0,

'

0

'

2'0

Maximum intensity (no smoothing) Maximum intensity (smoothing of 5) Midpoint of 95% Midpoint of 90% Midpoint of 85% Midpoint of 80%

44

IIl-Vs Review • Vo1.12 No. 2 1999

,i

i

40

Peak Width at Half Max (FWHM) 71.7_+ 2.6 nm

Wavelength at Upper Half Max 1337.01 + 9.23 nm

hJ Low I t

I

High

Figure 6. The spatial variation for a detector structure of (a) peak intensity, (b) peak wavelength, (c) FWHM and (d) upper wavelength.

this growth. The peak wavelength variation, in contrast, is caused by changes in the gas c o m p o s i t i o n in the MOCVD reactor and small variations in the surface temperature of the wafer during growth. The FWHM pattern is measuring the h o m o g e n e i t y within the sampled area and is related to mixing and alloying effects. Thus, the one spatial-spectral m e a s u r e m e n t provides three quite different pieces of control intormation for the process. The actual control w i n d o w that is used as the acceptable range for these values varies according to the type of device being produced. For example, the acceptable wavelength variation for wafers to be used li)r 980 n m p u m p lasers is an order of magnitude smaller than that for wafers used for optical detectors. In gen-

Table 2. The variation of peak wavelength with analysis algorithm

Mgorithm

Peak Waveiength t309.84 _+9.74 nm

Wavelength (nm) 990.0 989.0 987.9 986.6 984.9 983.1

eral one uses a measurement criter i o n of (i) average wavelength

within a set window, and (ii) an upper b o u n d on the standard deviation for the unifi)rmitv measurement. It is essential to realize that t~ provide reproducible wavelength measurements, the p r o b e power. p r o b e wavelength, p o w e r density and sample temperature need t~ be maintained to better than 1%. (;nder these closely controlled circumstances, reproducibility ~1 wavelength similar to that s h o w n in Figure lb is achievable. It is not u n c o m m o n to see sample PL wavelengths shift by as m u c h as 0.~ nm for each degree of r o o m temperature change.

Composition mapping For a ternary material, the PI. can be used to provide a measurement of the bandgap energy of the material and thus calculate the sample composition. Like most techniques of this type, this results in a high precision (one can see small changes), low accuracy (depending on the correlation used) measurement technique. The calibrations relating t q wavelength to c o m p o s i t i o n range

mapping

PL

(mm)

80

6 ~

"

(

i

Z0

o,

~6

,b ' 4 0

~

~

i

I

I Figure 7. Spatial uniformity of the Ga content of an InGaP epitaxial layer grown on GaAs.

i

](a)

Imm) Centroid (22,23 mm] VALUE: 7 0 5 "

4O

I 10

O 0

10

20

30 (c)

40

50

(%)

(mm)

26 10 45 4O 2728

35 30 25

28.45

20 15 2962

10 5

6' &, ~ ; 5 Composition

~o2'sgo

~5 4'o 4'8

30 80

- Average:29.53% - Std. Dev: 059% Median: 2960% - 100/o: 28 80% - 90%: 3030%

Figure 8. The spatial uniformity of quaternary composition from PL and X-ray data.

from the simple 'Vegard's Law' approximations (the wavelength is linear with c o m p o s i t i o n change) to more sophisticated models that take into account the changes of

46

IIl-Vs Review • Vo1.12 No. 2 1999

bandgap energy occurring with strain and temperature. Figure 7 s h o w s the spatial variation of gallium c o m p o s i t i o n for an InGaP cpitaxial layer deposited on GaAs.This

particular map was generated using o n e of the c o m m o n calibrations published in the literature. 111 this case, the spatial variation of the gallium concentration is due to small changes in the surface temperature, w h i c h are k n o w n to affect the rate equations governing the MOCVD growth. For this sample each colour change represents a 0.4% change in composition. For a quaternary material, twx) independent pieces of information are needed to identify the c o m p o sition. Typically these are PL emission wavelength (giving the bandgap) and XRD rocking curves (giving the lattice constant). Figure 8 s h o w s the result of taking the Pl_ wavelength map (8a) and the XRD peak separation map (Sb) and combining the t w o pieces of int'ormation to give the c o m p o s i t i o n of o n e of the quaternary elements (8c). The combination of these techniques can be a powertid control m e t h o d to provide the user with information that can be used to control and optimize the cpitaxial growth process. This article has discussed some of the basic m e t h o d s that use PL t~ measure the unitbrmity of epitaxial layers and structures.All of the data discussed in this article w e r e generated using constant probe p o w e r measurements on Philips PL mapping s y s t e m s . W h e n one varies the incident probe p o w e r within PI. measurements, a w h o l e n e w area of device characterization is u n l o c k e d . These types of measurements and the use of optical reflectometry for film thickness will be discussed in a later article in this series. Contact: C h r i s M o o r e

Philips Analytical 10 1 Randall Drive Waterloo Ontario Canada N2V 1C5. Tel: + 1 - 5 1 9 - 7 4 6 - 6 2 6 0 ; Fax: + 1-519-746-8270: E-mail: [email protected]: URL: www.analytical.philips, com.