- Email: [email protected]
Application of spectrum synthesis method to depth profile analysis
Journal of Electron Spectroscopy and Related Phenomena, 50 (1990) 53-60 Elsevier Science Publishers B.V., Amsterdam - Printed in The Netherlands
APPLICATION OF SPECTRUM DEPTH PROFILE ANALYSIS
SYNTHESIS
METHOD
53
TO
ISA0 KOJIMA, NATSUO FUKUMOTO, MASAYASU KURAHASHI and TETSUYA KAMEYAMA National Chemical Laboratory for Industry, Tsukuba, Zbaraki 305 (Japan) (Received 15 June 1989)
ABSTRACT A spectrum synthesis method based on non-linear least-squares fitting has been applied to the depth analysis of the SiOa thin layer on Si. The present treatment was required because the Auger spectra overlapped between the oxide and the metal species both in Si LVVand in Si KLL regions, and the Auger peak energies shifted to the higher energy side with a decrease in the thickness of the oxide layer. The calculation with appropriate constraining conditions yielded the detailed depth profiles which could distinguish the two chemically different species. A non-uniform distribution in the depth of metal species was observed by the profiles derived from LVV and KLL peaks because of the difference in their electron escape depths.
INTRODUCTION
A spectrum synthesis method in Auger electron spectroscopy (AES ) is useful for advanced analysis such as deconvolution of overlapped peaks and improvement of the accuracy of quantification [ 1,2 1. In a previous paper [ 21, we applied the non-linear least-squares procedure to the spectrum synthesis and obtained satisfactory results for the quantitative analysis of an alloy surface. One of the advantages of our treatment is that the difference in the kinetic energy scale between reference and observed spectra is automatically corrected, whereas such differences should be given as a fixed value in the ordinary linear least-squares calculation. There are several cases in which energy deviations may occur even if an operator chooses the same experimental conditions and identical apparatus. In the first case, the energy of an Auger peak shifts with the change of surface potential owing to the charging effect, particularly for samples with poor electric conductivity. In the second case, the deviation occurs when the adjustment of the analyzing position is wrong since an electron energy analyzer such as a cylindrical mirror analyzer (CMA), which is usually used for AES apparatus, has a restricted focal point, and thus the energy analysis is sensitive to small positional changes. The former case is ex-
0368-2048/90/$03.50
0 1990 Elsevier Science Publishers B.V.
54
petted to occur in the depth analysis of a thin insulator film deposited on a conductive substrate since the electric conductivity of the surface varies considerably relating to the thickness of the insulator. In the present paper, we describe the results of the application of the nonlinear least-squares spectrum synthesis to the depth profile analysis of SiOZ oxide film on Si. It has been known that the reduction of SiOZ occurs by the electron irradiation and that the Auger spectra overlap between the oxide and the metal species [ 31. Furthermore, during the depth analysis, peak energies were found to shift to the higher energy side by about 3 eV with the number of sputtering cycles. Thus, it was necessary to take the energy shifts into account in the spectrum synthesis treatment. The calculation using the non-linear leastsquares spectrum synthesis with appropriate constraining conditions allowed us to obtain the depth profiles in which chemically different species could be distinguished. CALCULATION METHODS
The concept of the spectrum synthesis method is described in a previous paper [2]. Here, we mention additional points which we have recently extended. The first extension is that the calculation of multiregions is included at the same time. In this case, the basic equation which is minimized in the least-squares procedure can be written as follows: M N(m) x2= c 1 wn,i(Ym,i-fm,~)2 m=l
i=l
(1)
where ym,iand fm,iare the observed and calculated intensities at energy Ei for the energy region m, respectively, and W,,i is a weight. A4 is the number of energy regions treated at a time and N(m) is the number of data involved in the energy region m. The synthesized spectrum f+ can be expressed by the sum of the digitally recorded reference spectrum Fm,j
where dE,,j is the energy shift for the jth reference spectrum, and c,,j is its fractional contribution. Here, the parameters to be optimized are LIE,,, and Cm,,. Ym,i and Fm , i are usually pretreated with a digital convolution filter consisting of array coefficients such as first-derivatives, second-derivatives and top-hat functions [ 11. By the second extension, constraints such as fixing parameters at constant values and combining them with a linear relation such as a * Pi - b *P, = c, became available. Here, Pi and Pj are parameters, and a, b and c are constants. Such constraints were especially effective for optimizing the energy shift parameters.
55 MEASUREMENTS
Measurements of the depth profile of SiO,/Si were performed by using a JAMP30 Auger electron microprobe (JEOL Ltd.) equipped with a microbeam ion gun. The CMA is set at the side of the analysis chamber with its axis perpendicular to the electron beam optics. Auger energy spectra were measured under the condition that the primary electron beam was normal to the sample surface. The ion gun was adjusted so that the analyzing position was the center of the sputtered area, and directed at 30” to surface plane. For depth profile measurements, the ion beam was accelerated at 3 kV and rastered in an area of 0.9 x 0.9 mm’. The argon pressure was kept constant by using an autovalve controller ($216, Granville Phillips Co.). A SiOZ thin film deposited on Si with a thickness of 1000 A was supplied from JEOL Ltd., as a standard sample for the correction of ion sputtering rates. The reproducibility of the depth analysis using SiO,/Si samples was less than 5%. RESULTS AND DISCUSSION
Depth profiles of SiOJSi Figure 1 shows the Auger spectra of SiO, and Si surfaces. The oxide spectrum was measured carefully so that reduction by electron irradiation was not
20000.
RltSi LVV)
R2(0
H
R3tS.i
KLL)
H
KLL)
H
16000
c F 12000 g z BOO0 -
I
P Si
0
I 100
I 200
I 300
I 400
I 500
600
KINETIC ENERGY /cV
Fig. 1. Auger energy spectra of SiOp and Si surfaces.
I 15M
1600
1700
56
observed (acceleration voltage= 10 kV, beam current =2 x 10e7 A and beam diameter = 300 pm). For the depth analysis, derivative (dN/dE) Auger spectra for the three energy ranges abbreviated as Rl (62-112 eV), R2 (490-526 eV) and R3 (15901648 eV) in Fig. 1 were measured and stored to use for the further calculation, where Rl, R2 and R3 contain the Si LVV, 0 KLL and Si KLL Auger peaks, respectively. Figure 2 shows the depth profile which was obtained by plotting the difference between an intensity maximum and a minimum for each energy interval. The measurement conditions were the beam current at 5 X 10m7 A and the beam diameter at 10 ,um. Ion sputtering was carried out intermittently for 30 s in each cycle. Figure 3 shows an example of a raw Auger spectrum stored after sputtering eight times in the depth analysis. Although the spectrum was recorded in the oxide layer, peaks due to metallic Si were clearly observed in the Rl region, whereas they were not observed in the R3 region. As shown in Fig. 2, the Rl intensity in the metal layer is remarkably higher than that in the oxide layer, and is also higher than the value expected from the chemical composition of SiOz. This is mainly attributed to the difference of the sensitivity factor between the oxide and the metal species, and partly to the coexistence of different species in the oxide layer. The R3 profile differs appreciably from the Rl profile: since the oxide and metal peaks overlap mark10000 -
so00
-
\.
0 20
NUMBER OF SPUTTERING
y------. 40
_._I
CYCLES
Fig. 2. Depth profiles obtained from the difference between an intensity maximum and a minimum in each energy interval for the three energy regions Rl, R2 and R3 designated in Fig. 1.
57 R2
Rl
R3 (no,observed
oxide
here)
metal
oxide metal
oxide
62 -
112
490
- 526
ENERGY
RANGE
1590
-
1648
(ev)
Fig. 3. Auger spectra for the three energy regions stored after eight cycles of sputtering in the depth analysis shown in Fig. 2.
1632
c
1628
,’ ’
Si KLL(meta0
t
1
::I
OKLL
,
.g e z Si
72 68
1
Si LVWoxide)
1
10
t
I
I
20 NUMBER
LvWmetaO
_...
I
30
I
40
OF S!=UlTERlNG
I
50
I
60
CYCLES
Fig. 4. Energy shifts of the five Auger peaks with the number of sputtering cycles.
edly in the R3 region, more than in the Rl region, the apparent decrease in intensity occurs in the interface layer due to the derivative property of the spectrum. Figure 4 shows the changes in energy of the five Auger peaks designated in
58
Fig. 3 with the number of sputtering cycles. It is seen that the peak energies are increased with the decrease in thickness of the oxide layer. The shifts are almost equal for all peaks and most significant at the interface region. This indicates the occurrence of charging, and thus the spectrum synthesis method is necessary to obtain the precise depth profiles. Depth profiles after spectrum synthesis The five signals, Si LVV(oxide), Si LVV(metal), Si KLL(oxide), Si KLL (metal) and 0 KLL, are considered in the calculation. The oxide spectrum given in Fig. 1 was used as a reference of SiOz and the last Auger energy profiles stored in the depth analysis were used as a reference of Si. The energy deviations can be assumed to be equal for the three energy regions in each spectrum. Thus, the following conditions are reasonable: G{dESi
LVV(S)
>-s{AEO
KJX(S)}=~
(3)
G{dESi
LVV(S)
>-G{dESi
KLZ.,(S)}=o
(4)
where S denotes a chemical state such as SiO, (oxide) or Si (metal). As shown in Fig. 4, the energy shifts during the depth measurement are approximately equal for both oxide and metal in the Rl region, and thus the following assumption is possible: dESi
LVV(oxide)
-&3i
~vvb~tal)
=conStant
(5)
Prior to the least-squares calculation, the values of yi in eqn. (1) and Fi in eqn. (2) were pretreated with a convolution filter consisting of the five point firstderivative coefficients which were given by Savitzky and Golay [ 41. A set of raw energy spectra of the depth analysis in Fig. 2 was calculated with common constraining conditions as stated above. The calculated profiles are given in Fig. 5 (a) for all the species considered. The contributions arising from the oxide species show almost identical curves. On the other hand, it may be seen that for the metallic species, the rising point of Si KLL curve (5) is shifted slightly to the lower number side than that of Si LVVcurve (2). It is of interest that this shift corresponds approximately to a thickness of 30 A which is comparable to the difference of electron escape depths at the two Auger energies [5]. Furthermore, Si LVV(meta1) intensity is larger than that of Si KLL (metal) in the oxide layer, indicating that the reduction by primary electron irradiation occurs at the topmost layer. Figure 5(b) shows the changes in fractional concentration of components against depth. The concentration was estimated by the simple assumption that the contribution to the reference given in Fig. 5 (a) could be divided into fractional amounts of an element constituting the referred compound. For example, using the Si LVV-derived peaks, the Si concentration can be calculated by
59
CUBER OF SPUTTERING CYCLES
0
--“-,--4--I
0
I
1000
500
t
1500
DEPTHI i
Fig. 5. Depth profiles obtained by the spectrum synthesis method: (a) for fractional contributions: 1, Si LVV(oxide); 2, Si LVV(meta1); 3, 0 KLL, 4, Si KLL(oxide); 5, Si KLL(meta1); (b) for
fractionalconcentrations:1, (1/3)C~i~~~(oxl~+C~I~~~(metal);2, (Z/~)~OKU; 3, ~1/3h*~~,~~~id~~ + CsiI(Lt(metaf)v Here, c is the fractional contribution which appears in eqn. (2).
the sum of one-third of the Si LVV(oxide) contribution and Si LVV(meta1) contribution. Thus obtained profiles represent well the concentrations of all the constituents. The non-uniform distribution of constituents in depth can be observed by the discrepancy of Si concentrations estimated from LVV- and ICI&-derived peaks, since the electron escape depths were different between the two peaks.
REFERENCES 1 T. Sekine, Y. Ando and H. Tokumasu,
J. Vat. Sci. Technoi. A, 4 (1986) 1557.
60 2 3 4 5
I. Kojima and M. Kurahashi, J. Electron Spectrosc. Relat. Phenom., 46 (1988) 185. ASTM, E983-84, Surf. Interface Anal., 10 (1987) 173. A. Savitzky and M.J.E. Golay, Anal. Chem., 36 (1964) 1627. S. Tanuma, C.J. Powell and D.R. Penn, Surf. Interface Anal., 11 (1988) 577.
APPLICATION OF SPECTRUM DEPTH PROFILE ANALYSIS
SYNTHESIS
METHOD
53
TO
ISA0 KOJIMA, NATSUO FUKUMOTO, MASAYASU KURAHASHI and TETSUYA KAMEYAMA National Chemical Laboratory for Industry, Tsukuba, Zbaraki 305 (Japan) (Received 15 June 1989)
ABSTRACT A spectrum synthesis method based on non-linear least-squares fitting has been applied to the depth analysis of the SiOa thin layer on Si. The present treatment was required because the Auger spectra overlapped between the oxide and the metal species both in Si LVVand in Si KLL regions, and the Auger peak energies shifted to the higher energy side with a decrease in the thickness of the oxide layer. The calculation with appropriate constraining conditions yielded the detailed depth profiles which could distinguish the two chemically different species. A non-uniform distribution in the depth of metal species was observed by the profiles derived from LVV and KLL peaks because of the difference in their electron escape depths.
INTRODUCTION
A spectrum synthesis method in Auger electron spectroscopy (AES ) is useful for advanced analysis such as deconvolution of overlapped peaks and improvement of the accuracy of quantification [ 1,2 1. In a previous paper [ 21, we applied the non-linear least-squares procedure to the spectrum synthesis and obtained satisfactory results for the quantitative analysis of an alloy surface. One of the advantages of our treatment is that the difference in the kinetic energy scale between reference and observed spectra is automatically corrected, whereas such differences should be given as a fixed value in the ordinary linear least-squares calculation. There are several cases in which energy deviations may occur even if an operator chooses the same experimental conditions and identical apparatus. In the first case, the energy of an Auger peak shifts with the change of surface potential owing to the charging effect, particularly for samples with poor electric conductivity. In the second case, the deviation occurs when the adjustment of the analyzing position is wrong since an electron energy analyzer such as a cylindrical mirror analyzer (CMA), which is usually used for AES apparatus, has a restricted focal point, and thus the energy analysis is sensitive to small positional changes. The former case is ex-
0368-2048/90/$03.50
0 1990 Elsevier Science Publishers B.V.
54
petted to occur in the depth analysis of a thin insulator film deposited on a conductive substrate since the electric conductivity of the surface varies considerably relating to the thickness of the insulator. In the present paper, we describe the results of the application of the nonlinear least-squares spectrum synthesis to the depth profile analysis of SiOZ oxide film on Si. It has been known that the reduction of SiOZ occurs by the electron irradiation and that the Auger spectra overlap between the oxide and the metal species [ 31. Furthermore, during the depth analysis, peak energies were found to shift to the higher energy side by about 3 eV with the number of sputtering cycles. Thus, it was necessary to take the energy shifts into account in the spectrum synthesis treatment. The calculation using the non-linear leastsquares spectrum synthesis with appropriate constraining conditions allowed us to obtain the depth profiles in which chemically different species could be distinguished. CALCULATION METHODS
The concept of the spectrum synthesis method is described in a previous paper [2]. Here, we mention additional points which we have recently extended. The first extension is that the calculation of multiregions is included at the same time. In this case, the basic equation which is minimized in the least-squares procedure can be written as follows: M N(m) x2= c 1 wn,i(Ym,i-fm,~)2 m=l
i=l
(1)
where ym,iand fm,iare the observed and calculated intensities at energy Ei for the energy region m, respectively, and W,,i is a weight. A4 is the number of energy regions treated at a time and N(m) is the number of data involved in the energy region m. The synthesized spectrum f+ can be expressed by the sum of the digitally recorded reference spectrum Fm,j
where dE,,j is the energy shift for the jth reference spectrum, and c,,j is its fractional contribution. Here, the parameters to be optimized are LIE,,, and Cm,,. Ym,i and Fm , i are usually pretreated with a digital convolution filter consisting of array coefficients such as first-derivatives, second-derivatives and top-hat functions [ 11. By the second extension, constraints such as fixing parameters at constant values and combining them with a linear relation such as a * Pi - b *P, = c, became available. Here, Pi and Pj are parameters, and a, b and c are constants. Such constraints were especially effective for optimizing the energy shift parameters.
55 MEASUREMENTS
Measurements of the depth profile of SiO,/Si were performed by using a JAMP30 Auger electron microprobe (JEOL Ltd.) equipped with a microbeam ion gun. The CMA is set at the side of the analysis chamber with its axis perpendicular to the electron beam optics. Auger energy spectra were measured under the condition that the primary electron beam was normal to the sample surface. The ion gun was adjusted so that the analyzing position was the center of the sputtered area, and directed at 30” to surface plane. For depth profile measurements, the ion beam was accelerated at 3 kV and rastered in an area of 0.9 x 0.9 mm’. The argon pressure was kept constant by using an autovalve controller ($216, Granville Phillips Co.). A SiOZ thin film deposited on Si with a thickness of 1000 A was supplied from JEOL Ltd., as a standard sample for the correction of ion sputtering rates. The reproducibility of the depth analysis using SiO,/Si samples was less than 5%. RESULTS AND DISCUSSION
Depth profiles of SiOJSi Figure 1 shows the Auger spectra of SiO, and Si surfaces. The oxide spectrum was measured carefully so that reduction by electron irradiation was not
20000.
RltSi LVV)
R2(0
H
R3tS.i
KLL)
H
KLL)
H
16000
c F 12000 g z BOO0 -
I
P Si
0
I 100
I 200
I 300
I 400
I 500
600
KINETIC ENERGY /cV
Fig. 1. Auger energy spectra of SiOp and Si surfaces.
I 15M
1600
1700
56
observed (acceleration voltage= 10 kV, beam current =2 x 10e7 A and beam diameter = 300 pm). For the depth analysis, derivative (dN/dE) Auger spectra for the three energy ranges abbreviated as Rl (62-112 eV), R2 (490-526 eV) and R3 (15901648 eV) in Fig. 1 were measured and stored to use for the further calculation, where Rl, R2 and R3 contain the Si LVV, 0 KLL and Si KLL Auger peaks, respectively. Figure 2 shows the depth profile which was obtained by plotting the difference between an intensity maximum and a minimum for each energy interval. The measurement conditions were the beam current at 5 X 10m7 A and the beam diameter at 10 ,um. Ion sputtering was carried out intermittently for 30 s in each cycle. Figure 3 shows an example of a raw Auger spectrum stored after sputtering eight times in the depth analysis. Although the spectrum was recorded in the oxide layer, peaks due to metallic Si were clearly observed in the Rl region, whereas they were not observed in the R3 region. As shown in Fig. 2, the Rl intensity in the metal layer is remarkably higher than that in the oxide layer, and is also higher than the value expected from the chemical composition of SiOz. This is mainly attributed to the difference of the sensitivity factor between the oxide and the metal species, and partly to the coexistence of different species in the oxide layer. The R3 profile differs appreciably from the Rl profile: since the oxide and metal peaks overlap mark10000 -
so00
-
\.
0 20
NUMBER OF SPUTTERING
y------. 40
_._I
CYCLES
Fig. 2. Depth profiles obtained from the difference between an intensity maximum and a minimum in each energy interval for the three energy regions Rl, R2 and R3 designated in Fig. 1.
57 R2
Rl
R3 (no,observed
oxide
here)
metal
oxide metal
oxide
62 -
112
490
- 526
ENERGY
RANGE
1590
-
1648
(ev)
Fig. 3. Auger spectra for the three energy regions stored after eight cycles of sputtering in the depth analysis shown in Fig. 2.
1632
c
1628
,’ ’
Si KLL(meta0
t
1
::I
OKLL
,
.g e z Si
72 68
1
Si LVWoxide)
1
10
t
I
I
20 NUMBER
LvWmetaO
_...
I
30
I
40
OF S!=UlTERlNG
I
50
I
60
CYCLES
Fig. 4. Energy shifts of the five Auger peaks with the number of sputtering cycles.
edly in the R3 region, more than in the Rl region, the apparent decrease in intensity occurs in the interface layer due to the derivative property of the spectrum. Figure 4 shows the changes in energy of the five Auger peaks designated in
58
Fig. 3 with the number of sputtering cycles. It is seen that the peak energies are increased with the decrease in thickness of the oxide layer. The shifts are almost equal for all peaks and most significant at the interface region. This indicates the occurrence of charging, and thus the spectrum synthesis method is necessary to obtain the precise depth profiles. Depth profiles after spectrum synthesis The five signals, Si LVV(oxide), Si LVV(metal), Si KLL(oxide), Si KLL (metal) and 0 KLL, are considered in the calculation. The oxide spectrum given in Fig. 1 was used as a reference of SiOz and the last Auger energy profiles stored in the depth analysis were used as a reference of Si. The energy deviations can be assumed to be equal for the three energy regions in each spectrum. Thus, the following conditions are reasonable: G{dESi
LVV(S)
>-s{AEO
KJX(S)}=~
(3)
G{dESi
LVV(S)
>-G{dESi
KLZ.,(S)}=o
(4)
where S denotes a chemical state such as SiO, (oxide) or Si (metal). As shown in Fig. 4, the energy shifts during the depth measurement are approximately equal for both oxide and metal in the Rl region, and thus the following assumption is possible: dESi
LVV(oxide)
-&3i
~vvb~tal)
=conStant
(5)
Prior to the least-squares calculation, the values of yi in eqn. (1) and Fi in eqn. (2) were pretreated with a convolution filter consisting of the five point firstderivative coefficients which were given by Savitzky and Golay [ 41. A set of raw energy spectra of the depth analysis in Fig. 2 was calculated with common constraining conditions as stated above. The calculated profiles are given in Fig. 5 (a) for all the species considered. The contributions arising from the oxide species show almost identical curves. On the other hand, it may be seen that for the metallic species, the rising point of Si KLL curve (5) is shifted slightly to the lower number side than that of Si LVVcurve (2). It is of interest that this shift corresponds approximately to a thickness of 30 A which is comparable to the difference of electron escape depths at the two Auger energies [5]. Furthermore, Si LVV(meta1) intensity is larger than that of Si KLL (metal) in the oxide layer, indicating that the reduction by primary electron irradiation occurs at the topmost layer. Figure 5(b) shows the changes in fractional concentration of components against depth. The concentration was estimated by the simple assumption that the contribution to the reference given in Fig. 5 (a) could be divided into fractional amounts of an element constituting the referred compound. For example, using the Si LVV-derived peaks, the Si concentration can be calculated by
59
CUBER OF SPUTTERING CYCLES
0
--“-,--4--I
0
I
1000
500
t
1500
DEPTHI i
Fig. 5. Depth profiles obtained by the spectrum synthesis method: (a) for fractional contributions: 1, Si LVV(oxide); 2, Si LVV(meta1); 3, 0 KLL, 4, Si KLL(oxide); 5, Si KLL(meta1); (b) for
fractionalconcentrations:1, (1/3)C~i~~~(oxl~+C~I~~~(metal);2, (Z/~)~OKU; 3, ~1/3h*~~,~~~id~~ + CsiI(Lt(metaf)v Here, c is the fractional contribution which appears in eqn. (2).
the sum of one-third of the Si LVV(oxide) contribution and Si LVV(meta1) contribution. Thus obtained profiles represent well the concentrations of all the constituents. The non-uniform distribution of constituents in depth can be observed by the discrepancy of Si concentrations estimated from LVV- and ICI&-derived peaks, since the electron escape depths were different between the two peaks.
REFERENCES 1 T. Sekine, Y. Ando and H. Tokumasu,
J. Vat. Sci. Technoi. A, 4 (1986) 1557.
60 2 3 4 5
I. Kojima and M. Kurahashi, J. Electron Spectrosc. Relat. Phenom., 46 (1988) 185. ASTM, E983-84, Surf. Interface Anal., 10 (1987) 173. A. Savitzky and M.J.E. Golay, Anal. Chem., 36 (1964) 1627. S. Tanuma, C.J. Powell and D.R. Penn, Surf. Interface Anal., 11 (1988) 577.