HAADF and EELS Study of ULK Dielectrics

Impact of CMP on Microstructure and Electronic Properties

  • Fig.1: High Angle Annular Dark Field images of cross-sectioned TEM foils, illustrating the successive metal layers deposited for capped sample (a) and CMP sample (b).Fig.1: High Angle Annular Dark Field images of cross-sectioned TEM foils, illustrating the successive metal layers deposited for capped sample (a) and CMP sample (b).
  • Fig.1: High Angle Annular Dark Field images of cross-sectioned TEM foils, illustrating the successive metal layers deposited for capped sample (a) and CMP sample (b).
  • Fig. 2: Zero-loss peaks and plasmon regions of two low-loss line scans (20 spectra) recorded across ULK layer: a) and c) correspond to capped ULK layer; b) and d) correspond to CMP ULK layer.
  • Fig. 2: Zero-loss peaks and plasmon regions of two low-loss line scans (20 spectra) recorded across ULK layer: a) and c) correspond to capped ULK layer; b) and d) correspond to CMP ULK layer.
  • Fig. 2: Zero-loss peaks and plasmon regions of two low-loss line scans (20 spectra) recorded across ULK layer: a) and c) correspond to capped ULK layer; b) and d) correspond to CMP ULK layer.
  • Fig. 2: Zero-loss peaks and plasmon regions of two low-loss line scans (20 spectra) recorded across ULK layer: a) and c) correspond to capped ULK layer; b) and d) correspond to CMP ULK layer.
  • Fig.3: Typical band gap and defects signature deduced from calculated SSD spectrum: a) capped ULK, b) and c) bottom and top region of CMP ULK.
  • Fig.3: Typical band gap and defects signature deduced from calculated SSD spectrum: a) capped ULK, b) and c) bottom and top region of CMP ULK.
  • Fig.3: Typical band gap and defects signature deduced from calculated SSD spectrum: a) capped ULK, b) and c) bottom and top region of CMP ULK.
  • Fig.4: Mapping of the band gap and defects signature across capped ULK (a) and CMP ULK (b) layers.
  • Advanced Metal Interconnection integrating copper and Low K dielectrics (composition of 3 STEM images at variable detection angels), by courtesy of R. Pantel STMicroelectronics Crolles France

The ITRS requires the integration of dielectric materials with effective dielectric constant (k) lower than 2.8. This is achieved using porous SiOCH. Unfortunately during integration in the devices, damages are introduced in the low-k layer by CMP. The impact of these damages on the microstructure and the electronic properties are studied using HAADF imaging and Valence Electron Energy Loss Spectroscopy in TEM environment. Results are compared to low-k capped with an etch stop layer.

The physical properties of the materials involved in semiconductor devices are pushed to their limits for each new technology node. One critical example is the requirements on the properties of the insulator filling the shrinking gap between metal lines. The dielectric constant needs to be as small as possible, but the insulator should still be compatible with chip fabrication steps like chemical mechanical polishing (CMP), etching or annealing. Ultra low-k materials based on porous SiCOH compounds are reported as promising candidates to replace the dense SiO2 [1]. In bulk form a reduction of the dielectric constant from 4 to 2.8 is obtained for such materials. The main reasons for achieving lower dielectric constant is the lower polarizability of Si-C or Si-CH3 bonds than Si-O bond. Unfortunately, during standard fabrication processes like CMP, ULK material is susceptible to be damaged, leading to the degradation of the dielectric properties. To minimize these damages an alternative consists in the deposition of a denser oxide layer on the top of the ULK before the CMP treatment. In this article, we compare the microstructure and the insulating properties of CMP and capped ULK at 7 Metal Level in a finished device.

Experiments

Microstructure and electronic properties of ULK are determined by the combination of HAADF imaging (High Angle Annular Dark Field) and EELS analysis (Electron Energy Loss Spectroscopy) in a TEM environment i.e. FEI TECNAI G2 [2]. Both specimens are prepared by focused ion beam (FIB) milling. First, cross-sections were cut out of a 300 mm wafer using the in-situ lift-out method. Then, the final thinning is performed using low kV gallium at 30 kV, 5 and 2 kV successively, to reduce surface amorphization.

The insulator layer sequence is either SiCN/ULK/capping material/SiCN or SiCN/ULK/SiCN for capped and CMP samples respectively (fig. 1). The relative thicknesses, calculated from low loss spectra recorded in the middle of the two SiCN layers, located be low and above of the low-k layer, are found to be equal. This indicates that the TEM foil thickness across the low-k layer is constant, as expected from FIB thinning. Since contrast variations in HAADF images are proportional to the product of the density, thickness (t) and atomic number (Z 3/2), HAADF imaging contrasts can be thus only attributed to chemical or density effects. Regarding EELS analysis, all experiments are performed in the low energy domain <100 eV, in line-scan mode across the stack. Spectra are recorded with short exposure time to prevent contamination and irradiation effects and are summed to get a final spectrum with a high signal to noise ratio. In this energy range, EELS signal results from individual and collective valence electrons excitations and gives information on chemical as well as physical properties. The zero-loss peak recorded simultaneously to the low-loss spectrum informs about the experimental resolution and the material density. The calculated single scattering distribution (SSD) spectrum is proportional to the product of the matrix element between VB and CB and the Joint Density of States (JDOS) [3]. Hence, almost each structure in the SSD Spectrum low energy region (<10 eV) is related to an electronic transition. In particular, the first intensity jump can be attributed to the band gap.

Microstructure of Capped and CMP ULK

Figure 1 displays HAADF images of capped and CMP sample. In both cases, the interfaces between the different deposited layers can be observed by contrast variations, showing the ability of HAADF imaging to be an accurate metrology tool. In HAADF images the contrasts are reversed in comparison to bright field TEM imaging, heavy materials appear white and light element or porous material dark. The contrast variations across the metal layer sequences reveal that density increases from low-k, to SiCN, to capping material and Cu. In addition, for CMP sample, we observe that ULK is darker at the bottom than at the top, whereas the same contrast is observed across the capped ULK, indicating that CMP step introduces either density or chemical variations. In figure 2 zero-loss peak and plasmon region line-spectra recorded across the capped and CMP ULK are displayed. The shape and the intensity of the zero-loss peak as well as the plasmon region appear identical at any level in the capped ULK layer, indicating that neither significant density nor chemical composition changes occur. On the contrary, for CMP sample, the zero loss and plasmon loss region intensities increase by a factor of 2, from the bottom to the top, but no change is observed neither in the shape nor in the plasmon peak energy. The contrast variation across the CMP low-k layer can be thus attributed to density variation effects. ­SiOCH compounds are known to be porous materials. Thus a likely interpretation is that the chemical solution used for CMP step, penetrates into ULK and slightly dissolves ULK. Pore volume fraction is thus increases and density decreases [4].

Band Gap Measurement of ULK

Figure 3 displays the calculated SSD spectra representative for ULK layer in capped sample and for ULK at the top and bottomregions in CMP sample. For each one of these spectra two significant intensity jumps are observed within the 0-10 eV range. For the capped ULK, the first jump is located at 2.4 eV and the second at 8.5 eV while for CMP ULK, the first is located at 1.1 ± 0.1 eV and the second at 8 and 8.5 eV depending on the acquisition place. To valid the method, the same band gap determination is applied to spectra recorded from the well known phases present in the stack i.e. SiCN and amorphous-SiO2. For both, band gap and low-loss structures energies are in agreement with values reported in the literature. The presence of any artefacts in the ULK SSD spectra due to Cerenkov effects or surface coupling effects can be excluded [5], since refractive index of SiCOH compound is reported lower than 1.4 and TEM foil thickness is measured to be 55 +/-5 nm. This confirms the reliability of our results. Comparing ULK, SiO2 and capping material SSD spectra, we find that they are not very different in the energy range above 10 eV, where plasmon contributions dominate the spectrum. The typical structures located at 10.5, 14.6 and 18.1 eV observed in am-SiO2 spectrum are also particularly well reproduced, indicating that SiOCH electronic behaviour is close to the SiO2 one. The only difference is that the SiOCH plasmon energy position is shifted about 1 eV to lower energy. This is in agreement with the expected chemical composition effects resulting to C and H atoms addition to SiO2. The interpretation of the spectra in the energy range below 10 eV is less straight forward. The signatures located at 8 and 8.5 eV for CMP and capped ULK are naturally identified as the band gap since it is the case for the intensity jump in am-SiO2 located at 8.9 eV. While as expected, the intensity falls down to zero in am-SiO2 for energy lower than 8.9 eV, a surprisingly significant signal remains in this energy range from 1.1 to 8 eV and 2.4 to 8.5 eV for CMP and capped ULK respectively.
Taking account of these signal intensity level, it would be unrealistic to attribute it to band tail effects. The most probable explanation to electron states in the band gap region, are dangling bonds created at the surfaces of the pores in the ULK. Therefore, to our knowledge no calculations supporting these assumptions have actually been carried out. After processing each individual spectrum of line-scans across capped and CMP ULK, a mapping of the band gap and defects signatures as a function of the depth can be established (fig. 4). While in the capped ULK both signatures are constant across the layer, a significant decrease of the band gap from the bottom to top occurs for the CMP ULK. Moreover, the defect signatures energy position is always about twice lower in CMP than in capped ULK, independently of the location in the layer, like as this signal was not impacted by the density variations. Additional experiments are carried out to shine light on this result.

Conclusion

The impact of CMP on ULK material in a state of the art device is investigated using HAADF imaging and EELS in HR-TEM. STEM-EELS line profiles allow an accurate mapping of three fundamental features of ULK material: density, band gap and electronic defect signatures. We have first proven that the capping layer plays its protection role since no densification variations are detected and a flat band gap profile is observed across the capped ULK. In the contrary, a significant modification of the contrasts and electronic properties are observed in the CMP ULK layer. As a consequence, a negative impact on the dielectric constant of ULK unprotected before CMP is thus expected.

References:
[1] Maex K. et al.: J. Appl. Phys. 93, 8793-8800 (2003)
[2] www.fei.com/products/families/tecnai-family.aspx
[3] Egerton R. F.: Electron Energy Loss Spectroscopy in Electron Microscope, 2nd Edition, Plenum Press, 131-194 (1996)
[4] Miranda, P. A. et al.: Microelectronics and Electron Devices, 2004 IEEE Workshop, 85-88 (2004)
[5] Erni R. et al.: Ultramicroscopy 108, 84-89 (2008)


Authors:
Dr. Marie Cheynet, Grenoble University - CNRS, SIMaP-PHELMA, France (corresponding author)
Dr. Fabien Volpi, Grenoble University - INP, SIMaP, France
Dr. Simone Pokrant, Product manager, A Carl Zeiss SMT AG Company, Oberkochen, Germany
Dr. Roland Pantel
Dr. Mohammed Aimadedinne
Dr. Vincent Arnal, St. Microelectronics, Crolles, France

 

Contact

Grenoble University - CNRS / SIMaP-PHELMA
1130 rue de la Piscine
38402 ST M Grenoble / St. Martin D´Heres Cedex
France
Phone: +33/4/76826608

Register now!

The latest information directly via newsletter.

To prevent automated spam submissions leave this field empty.