Multidisciplinary Investigations of Dielectric Materials in a Real Device: Short-range Ordering, Composition and Dielectric Response

Résumé fourni par la source

Dielectric materials are extensively used in semiconductor devices to isolate different components as an insulator or diffusion mask. Their atomic structure and chemical composition are essential to understand the performance of the devices. With the shrinking of the device, the local structure and composition of the dielectric materials could vary from their bulk counterpart due to the reduced dimensions or process. It becomes critical to characterize the dielectric materials locally within their structure and their interfaces with other materials. However, even with an advanced electron microscope, it is challenging to fully characterize the dielectric materials locally in a real device. In this work, the short-range ordering, chemical composition, chemical states and dielectric response of the silicon oxide and silicon nitride layers of one commercial 3D NAND device were studied by 4D STEM and monochromated electron energy loss spectroscopy (EELS). Firstly, the composition of the 3D NAND was characterized by EELS from a Thermo Fisher Themis Z system. As shown in Figure 1, the layers stack from outer to inner are metal gate (W), TiN/Al2O3, blocking oxide (SiO2), SiN storage layer (SiN), tunnel oxide, poly-Si channel and inner SiO2. The line profile in Figure 1b shows the clear presence of N in the blocking oxide. A thin SiN layer is observed between the tunnel oxide and poly-Si. Besides the composition, chemical states of 3D NAND structure were also analyzed by monochromated EELS. The FWHM of ZLP was around 0.2eV. Using the nitrogen K-edge as an example, its spectra from the blocking oxide, storage layer, are different from TiN (Fig.2). Interestingly, the N K-edge inside the blocking oxide layer has a similar shape as that in SiN storage layer, which indicates the N exhibits the same coordination and is likely only coupled with silicon atoms inside the blocking oxide layer. There is a sharp peak at 402eV in the N K-edge collected from the tunnel oxide regions between the two SiN layers. This peak refers to strong N-N bonds, which indicates the presence of N2 bubbles inside the tunnel oxide layer [1]. The N2 bubble accumulated between the two SiN layers and part of the interface between the blocking oxide and SiN storage layer. Although EELS provides information on the chemical composition and chemical states with high spatial resolution (sub-nm here), there is still lack of structure information on these oxide and nitride layers. Examining conventional electron diffraction patterns provides very little information due to the disordered structure, but specialized techniques can be used to study the short-range ordering. In this work, we used 4D STEM data collected on a Thermo Fisher Talos F200E with a speed enhanced Ceta camera to calculate the local electron pair distribution function (ePDF) within individual layers of the 3D NAND device [2]. Figure 3a shows the virtual bright field (BF) image generated from the 4D STEM dataset. Diffraction patterns from the highlighted regions are summed together and used to generate the ePDFs shown in Figure 3b. The peaks in the ePDF correspond to atomic pair spacings in the amorphous material [3]. The first peak is positioned at 1.66 and 1.63Å in the blocking oxide and inner oxide layers, respectively and can be attributed to Si-O pairs [2]. In SiN storage layer, the first peak shifts to 1.73 Å, corresponding to Si-N spacing [3]. The tunnel oxide and the thin SiN layers too thin to be distinguished with the probe condition used for 4D STEM. The ePDF in the tunnel oxide layer shows an intermediate peak at 1.71 Å indicating a mixture of Si-N and Si-O. The position of the second peak shifts in a similar pattern with position ranging from 2.65 Å in blocking oxide and inner oxide layers to 3.02 in SiN storage layer layers with the tunnel oxide (with SiN) peak having intermediate position at 2.96 Å. The second peak has contributions from O-O, N-N, and 2nd nearest neighbor Si-Si pairs [4, 5]. As the dielectric layers are very thin, we will try different beam conditions to improve the spatial resolution for ePDF. Additionally, fluctuation electron microscopy (FEM) will be used to explore medium-range order (MRO) in the dielectric materials. Dielectric functions of the dielectric layers in the 3D NAND cell were further extracted from valence EELS collected at 80 kV with an energy resolution of about 0.1eV. The electron loss function was first obtained by removing the tail of ZLP, surface and retardation effect followed by a Kramers-Kronig analysis for the dielectric functions. As shown in Figure 4, the blocking oxide and inner SiO2 show similar dielectric function, except the defect state around 4eV in the inner SiO2, even though the outer SiO2 layers have a higher N concentration. It is still not clear why dielectric functions are similar from two materials with different compositions. Two areas from the SiN storage layer (area 2 and area 7) show similar dielectric functions, confirming the uniformity of the layer. It is very interesting that the dielectric functions from the outer and inner surfaces of the poly-Si layer are similar in shape except the peak of 10.6eV in the inner surface although the outer surface contains nitrogen. The dielectric function of oxide and nitride layers matched very well with the published value [6, 7]. Our method is also applicable to bulk Si, however the extracted dielectric function of the 5-6 nm poly Si in the 3D NAND cell is different from the report value of bulk Si [8]. It is likely that the VEELS of poly-Si contains signals from nearby oxide or nitride due to delocalization and the derived dielectric function of is also a mixture. In summary, we used advanced EELS and 4D STEM techniques to fully investigate the dielectric materials within a real semiconductor device at nanoscale. The atomic structure, chemical composition, chemical states and their dielectric functions are obtained at each layer. This local information would help to understand the performance of the device and similar analysis can be applied to other devices to further improve design and functionality. (a) 2D elemental mapping by EELS of a 3D NAND cell in planar view and (b) elemental line profile from outer to inner layers along the arrow in (a). (a) N K-edge in TiN, blocking oxide, SiN, and tunnel oxide. And (b) maps of three different types of N. (a) Virtual BF image of a 3D NAND cell generated from a 4D STEM dataset. (b) The calculated ePDF from each region highlighted in a. The peak in the 1.63 - 1.74 Å range corresponds to Si-(O/N) pairs. The peak in the 2.65 - 3.02 Å corresponds to O-O/N-N and 2nd nearest neighbor Si-Si pairs. The shift and changing shape of the peaks between regions is the result of changes in atomic structure between the different materials. (a) STEM image of a 3D NAND cell. (b) Energy loss functions of different areas marked in (a). (c) real and (d) imaginary part of the dielectric functions of different regions: 1-blocking oxide, 2-SiN storage layer, 3-tunel oxide (with SiN), 4-inner oxide, 5-interface of poly-Si and inner oxide, 6-poly-Si, and 7-2nd area of storage layer.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.