Practical Electron Microscopy and Database

Figure 3441a shows the simulated Si L_3,2​ core-loss spectra for pure Si, SiO, and SiO₂​ . Each material has its own characteristic Si L_3,2​ edge.

In the simulation, the spectra for the Si L_3,2​ core-loss edges are modeled using Gaussian functions. Gaussian functions are commonly used to simulate spectral peaks because they approximate the shape of many experimental core-loss edges. For each spectrum (pure Si, SiO, and SiO₂), we use two Gaussian functions: one for the L₃ peak and one for the L₂ peak.

The Gaussian function is defined by the equation:

--------------------------------------------------------------- [3441a]

where,

I(E) is the intensity at energy E.
A is the peak amplitude (maximum intensity).
E₀​ is the peak center (the energy at which the peak occurs)
σ is the standard deviation, which controls the width (broadening) of the peak.

For each material (pure Si, SiO, and SiO₂), the total spectrum is the sum of two Gaussian functions representing the L₃ and L₂ peaks.

For pure Si, we have:

--------------------------------------------------------------- [3441b]

• L2 peak:

--------------------------------------------------------------- [3441c]

• The total spectrum for pure Si:

--------------------------------------------------------------- [3441d]

For SiO, we have:

--------------------------------------------------------------- [3441e]

• L2 peak:

--------------------------------------------------------------- [3441f]

• The total spectrum for SiO:

--------------------------------------------------------------- [3441g]

For SiO₂, we have:

--------------------------------------------------------------- [3441h]

• L2 peak:

--------------------------------------------------------------- [3441i]

• The total spectrum for SiO₂:

--------------------------------------------------------------- [3441j]

Figure 3441a indicates that The shifting of the Si L_3,2 ​ core-loss peaks to higher energies as silicon becomes more oxidized — from pure Si to SiO, and then to SiO₂ ​ — is a fundamental phenomenon observed in Electron Energy Loss Spectroscopy (EELS). This shift is primarily due to changes in the chemical environment and the oxidation state of silicon. As silicon oxidizes, oxygen atoms (which are highly electronegative) bond with silicon. This increases the oxidation state of silicon:

As silicon's oxidation state increases, the effective nuclear charge, Z_eff, experienced by the core electrons (like the 2p electrons involved in the Si L_3,2​ edges) also increases. This is because oxygen atoms withdraw electron density from silicon, reducing electron shielding and making the silicon nucleus more "felt" by its electrons. With a higher Z_eff, the core electrons require more energy to be excited or ejected. This results in the core-loss edges shifting to higher energies. On the other hand, oxygen atoms, being more electronegative, pull electron density away from silicon. This charge transfer leads to a partial positive charge on the silicon atoms. The increased positive charge on silicon reduces the screening effect (electron cloud shielding the nucleus). Consequently, the core electrons are held more tightly, necessitating higher energy for excitation. Furthermore, in pure silicon, electrons in the outer shells can better screen the nuclear charge from the core electrons. As oxidation increases, the reduction in electron density around silicon diminishes this screening, leading to higher core-electron binding energies.

Numerous experimental studies shows that the shift of core-loss edges to higher energies with increasing oxidation states. For instance, in EELS studies:

Experimental data of Si EELS are shown below. The Si L_2,3 signal yields a much higher signal-to-noise ratio (SNR) than the Si K edge. In EELS analysis, the main edges which are used in Si characterization are Si L_2,3 edge ( L₃ 99 eV and L₂ 100 eV) and Si K edge (1839 eV) as shown in Figure 3441b.

(a)

(b)

Figure 3441b. Main edges (after removed background) used in Si characterization: (a) Si L2,3 edge ( L3 99 eV and L2 100 eV) obtained with energy dispersion of 0.5 eV/channel , and (b) Si K edge (1839 eV).

It is necessary to note that the Al L_2,3 edge and Si L_2,3 edge in EEL spectra are very close as shown in Figure 3441c.

Figure 3441c. Al L2,3 edge ( L3 73 eV and L2 74 eV) and Si L2,3 edge ( L3 99 eV and L2 100 eV) in EELS.

The EEL spectra in Figure 3441d show the Al L_2,3 edge as well as the Si L_2,3 edge after background subtraction, taken from Al_xSi_yO_z matrix. The threshold value of the Al ionization edge here is about 75 eV indicating that aluminum is bonded to oxygen because the onset energy of the edge of pure aluminum is about 73 eV. This energy difference is called chemical shift. Depending on the environment of Si element, the chemical shift of Si L_2,3 energy loss edge can be up to 6.9 eV.

Many Si-based materials, e.g. some minerals, consist of SiO₄ tetrahedron. These tetrahedra may form crystalline silica polymorphs, as well as amorphous and glassy phases. In Figure 3441e, the Si L_2,3 edges from quartz and the two clay minerals shows very similar edge shapes, characterized by two sharp peaks, i.e. peaks A (106.3 eV) and B (108.4 eV), followed by a sharp peak C (115.0 eV) and a broad peak D (133 eV). Peak A, from quartz, is further split into two peaks separated by 0.45 eV.

The resulting chemical shift can be mapped using the energy filtered imaging taken with a narrow energy slit of 1.8 eV or similar spectrum mapping in STEM mode.

The silicon ‘metallic’ bulk plasmon is found at 17 eV, and it moves up to 23 eV for silicon nitride (Si₃N₄). Figure 3441f shows the Si L_2,3 EEL spectra of Si, SiO₂ and Si₃N₄. The positions of the edge onsets clearly indicate chemical shifts and differences in ELNES present. Table 3441a lists the Si L_2,3 energy loss onsets (E_L2,3) of some Si compounds and chemical shifts (ΔE_L2,3) from that of pure Si. In Figure 3441f, for pure Si, the ELNES shows a first peak at 101.5 eV followed by a small plateau between 105 and 114 eV, and a very broad peak with the maximum at 135 eV. The SiO₂ ELNES has a strong peak at 108.5 eV, which is also followed by a small plateau between 112 and 116 eV. However, the second peak located at 128 eV is far less intensive than the one for Si. An additional small peak presents at 161 eV. The ELNES of Si₃N₄ is very similar to the SiO₂ ELNES. The two main peaks are located at 105 and 124.5 eV.

Figure 3441f. Si L_2,3 EEL spectra of Si, SiO₂ and Si₃N₄. [3]

Table 3441a. Si-L_2,3, Si-L₁, and Si-K energy loss onsets (E_L2,3, E_L1, and E_K) of
some Si compounds and chemical shifts (ΔE_L2,3, ΔE_L1, ΔE_K) from that of pure Si.

In order to distinguish different compounds containing a specific element, e.g. silicon in pure Si, SiO₂ and Si₃N₄, the positions of the energy windows (in three-window method) and respective slit widths have to be chosen according to the ELNES as mentioned above. Table 3441b shows the optimized imaging filter settings in three-window EFTEM mode for Si, SiO₂ and Si₃N₄. For Si mapping in SiO₂ the post-edge window was placed above the second peak at 143 eV (See Table 3441b). The two pre-edge windows were placed in the small plateau region between the first (110 eV) and the second (125 eV) peaks. The optimal slit width was found to be 5 eV. The filter settings for Si₃N₄ were determined in a similar way by considering the shift of the Si L_2,3 spectrum towards lower energy loss energies. Therefore the pre-edge windows were placed at 107 and 110 eV. The post-edge window was placed at 114 eV in order to avoid interference with Si and SiO₂ signals. A slit width of 3 eV was used. Those settings were used to take Si images in a MOSFET structure as shown in Figure 3441g. Note that in those images, SiO₂ and Si₃N₄ can be distinguished without mapping O and N elements. It is clearly shown that the polycrystalline Si gate (a) is separated by a thin dielectric layer (b) from the single crystalline Si (c). The spacers (d) consisting of two different types of material can be seen on both sides of the gate. Between the polycrystalline gate and the single crystalline Si, the gate oxide layer is also seen clearly. The spacers consist of two layers, namely a thin SiO₂ layer and a thicker Si₃N₄ layer. Note that the bright contrast at the top of image (c) can be an artifact because of the signal of some elements overlapping the Si energy windows of SiO₂.

Table 3441b. The imaging filter settings of EFTEM for Si, SiO₂ and Si₃N₄. [3]

The valence state of Co can be indirectly determined by mapping the Si ELNES characteristics for the cases of CoSi_x (x = 1 or 2). Figure 3441h shows the as-measured and background-subtracted Si L_2,3 EELS spectra of CoSi and CoSi₂. The spectra for both silicides have Si L_2,3 edge onsets located at 100.6 eV, while the Si ELNESs are slightly different. The CoSi₂ ELNES has one broad peak with a characteristic pre-edge peak, while the CoSi ELNES does not show this pre-edge peak. Based on those characteristics, the parameters for three-window EFTEM method in Table 3441e are used. Here, in order to distinguish between CoSi and CoSi₂, the pre-edge peak of CoSi₂ was used, while the third peak of CoSi at 145 eV was used for the post-edge window. Using those settings, the Si ELNES maps of CoSi and CoSi₂ were obtained together with TEM image as shown in Figure 3441i. Note that the Ti, Co, and Si elements have overlapped layers as indicated in Figure 3441i (a). Although the Ti film also gave some artefacts in the maps, it is very important that the grains consisting of CoSi and CoSi₂ were still distinguished and the images revealed that a CoSi₂ layer was formed between CoSi and Si substrate.

Figure 3441h. The as-measured (a), and background-subtracted (b) Si L_2,3 EELS spectra of CoSi and CoSi₂. Adapted from [3]

Table 3441c. Three-windows imaging filter settings for Si L_2,3 ELNES for CoSi and CoSi₂. [3]

Figure 3441i. The TEM image (a), CoSi map (b), and CoSi₂ map (c) of Ti/CoSi/CoSi₂ layer stack. Adapted from [3].

Figure 3441j shows that the chemical shifts are revealed by spatially resolved EELS of Si-L_2,3 edges. Figure 3441j (a) shows the 2-D (two-dimensional) spectrum image obtained by an energy filter. Note that this image shows the image contrast before background subtraction. The vertical axis corresponds to the location (y) in the layered materials of SiON. Si₃N₄, SiO₂ and pure Si. The horizontal axis corresponds to the energy loss E. Energy dispersion used here was 0.051 eV per pixel. The energy range ΔE recorded in the 2-D spectrum image was 52 eV (= 88 - 141 eV in energy loss). Both the shift of the bright lines (shown by red arrows) in Figure 3441j (a) and the vertical lines in Figure 3441j (b) indicate the chemical shift of Si-L_2,3 edges. Figure 3441j (b) shows energy loss spectra of each layer extracted from Figure 3441j (a). Here, the backgrounds of the spectra were subtracted. It is necessary to point out that the spectra in Figure 3441f were not background-subtracted, while they were background-subtracted in Figure 3441j (b). Another difference of ELNES between the two “pure” Si spectra in Figure 3441f and 3441j (b) probably originated from the quality of the silicon materials or their impurities.

Figure 3441k shows the spatially resolved EELS of Si-K edges:  (a) 2D spectrum image and (b) spectra of Si₃N₄, SiO₂ and Si. Those spectra are about two orders of magnitude weaker than that of the L_2,3 edges so that the experimental conditions are different. For instance, the energy dispersion for Si-K edges is decreased to 0.30 eV (0.051 eV for Si-L_2,3 edge described above) per pixel.

Figure 3441k. The spatially resolved EELS of Si-K edges:  (a) 2D spectrum image and (b) spectra of Si₃N₄, SiO₂ and Si. [4]

Figure 3441l shows the plasmon regions and Ni–M_2,3 edges of the three nickel silicide phases in EELS. The three phases can be distinguished by their bulk plasmon energy position (E_p). The energy positions are shifted from 19.4 eV for NiSi₂, to 20 eV for NiSi and 22.2 eV for Ni₂Si. The Ni–M_2,3 core edge shifts can also be obtained from 65.8 eV for Ni₂Si and 66.8 eV for NiSi to 67.5 eV for NiSi₂.

Brown and Ho [6] used an EELS spectrometer with an incident electron energy of 10 eV, consisting of a double-pass cylindrical monochromator and single-pass cylindrical analyzer, to analyze the interaction between methyl chloride and Si( 100) 2 x 1. Figure 3441m shows the EEL spectra obtained after 1 L exposures of methyl chloride (CH₃CI) onto the Si(100) surface held at 85 and 300 K. The origin of the peaks, related to the existence of molecularly adsorbed species, is listed in Table 3441f.

Figure 3441m. The EEL spectra obtained after 1 L exposures of methyl chloride onto the Si(100) surface held at 85 and 300 K. [6]

Figure 3441n shows experimental ELNES spectra at the Si-L_2,3 edge taken from amorphous, C49, and C54 TiSi₂. All the three spectra consist of a small peak located at around 102 eV superposed on a very broad peak. The clear difference between three spectra is the width of the first peak. This peak is sharper in the C54 sample than in the C49 sample, while the amorphous sample is similar to the C49 sample.

In many cases, the deep level states in semiconductors can be analyzed by EELS measurements. For instance, the pronounced shoulder below 100 eV indicated by the arrow in Figure 3441o presents a deep level state, above Fermi level in band gap, produced by symmetric segment a 5-fold coordinated atom in CSL junction in polycrystalline silicon. The shoulder in the Si-L_2,3 ELNES was detected only from the symmetric segment of the {112} Σ3 CSL boundary near the CSL junction, but not from the bulk, {112} and {111} Σ3 CSL boundaries. The CSL junction would be electrically active and affects the electrical conductivity in polycrystalline silicon.

Single scattering distribution (SSD) spectra can be extracted from experimental spectra by deconvolution of elastic/quasi-elastic phonon and multiple scattering effects. For instance, this deconvolution can be done by removing multiple scattering with Fourier-log deconvolution method. For some materials (especially wide band gap materials), the band gap (E_g) from low-loss spectra obtained by EELS technique can be determined by eye observation of the SSD spectrum. In this case, E_g is defined as the energy corresponding to the first onset in the spectrum, for instance, 0.67 and 1.12 eV for Ge and Si respectively, in the schematic spectra in Figure 3441p.

Figure 3441p. Schematic illustration of Si and Ge SSD spectra in the low-energy range.

As an example of the binning effects on the energy resolution and the background of EEL spectra, the EELS of Si is shown in Figure 3441q. The two spectra were taken at energy dispersions of 0.44 eV/channel and 3.82 eV/channel, respectively. The full-width at half-maximum (FWHM) of the small peak marked by the red arrow is only 2 eV; however, the energy of each channel (pixel) with energy dispersion of 3.82 eV/channel is 3.82 eV. Therefore, such small peak is invisible even though the same EELS system and electron microscope are used for both spectra. Furthermore, the height of the peak marked by the blue arrow is smaller for the greater energy dispersion (3.82 eV/channel).

(a)

(b)

Figure 3441q. EEL spectra of Si taken at energy dispersions of 0.44 eV/channel and 3.82 eV/channel, respectively. The two figures (a) and (b) present the same spectra, but in different format. Binning in a spectrum is performed by using greater energy dispersions. Here, 3.82 eV/channel is binning of 0.44 eV/channel by about 9x.

[1] I. Sieber, R. Schneider, I. Doerfel, P. Schubert-Bischoff, S. Gall, W. Fuhs, Preparation of thin polycrystalline silicon films on glass by aluminium induced crystallisation—an electron microscopy study, Thin Solid Films 427 (2003) 298–302.
[2] Laurence A. J. Garvie, Alan J. Craven, Use of electron-energy loss near-edge fine structure in the study of minerals, American Mineralogist, 79, (1994) 411-425.
[3] M. Worch, H.J. Engelmann, W. Blum, E. Zschech, Cross-sectional thin film characterization of Si compounds in semiconductor device structures using both elemental and ELNES mapping by EFTEM, Thin Solid Films 405 (2002) 198–204.
[4] Koji Kimoto, Tomoko Sekiguchi and Takashi Aoyama, Chemical shift mapping of Si L and K edges using spatially resolved EELS and energy-filtering TEM, Journal of Electron Microscopy 46(5): 369-374 (1997).
[5] S. Pokrant, R. Pantel, M. Cheynet, Physical characterization by valence electron energy loss spectroscopy, 83 (11–12) 2364–2367 (2006).
[6] Kyle A. Brown and W. Ho, The interaction of methyl chloride and Si( 100) 2 x 1, Surface Science 338 (1995) 111-116.
[7] Mizoguchi T., Tanaka I., Mizuno M., Adachi H., Hashimoto T., Inui H., Yamaguchi M., Defect and electronic structure of TiSi2 thin films produced by co-sputterings. - Part II: Chemical bonding and electron energy-loss near-edge structures, Acta mater. 49 (2001) 2321–2328.
[8] Norihito Sakaguchi, Makito Miyake, Seiichi Watanabe, and Heishichiro Takahashi, EELS and Ab-Initio Study of Faceted CSL Boundary in Silicon, Materials Transactions, 52(3) (2011) 276.