The quantitative determination of the nanostructure of electronic devices, subunits of a biological cell, catalyst particles or nanocomposite materials is of utmost importance to understand their function and to foster further developments in nanotechnology, biology, chemistry and materials science. Hard X-ray microscopy is ideally suited to study these systems (Hanke et al., 2008; Carmona et al., 2008; Quitmann et al., 2009; Heinonen et al., 2010), as it allows one to image quantitatively the inner structure of an object without destructive sample preparation or to follow in situ the evolution of a specimen inside of a special sample environment, such as a chemical reactor or a pressure cell. In combination with tomography the reconstruction of the three-dimensional inner structure of a specimen is possible (Withers, 2007).
Today, both full-field (Chu et al., 2008) and scanning hard X-ray microscopy (Carmona et al., 2008; Heinonen et al., 2010) are limited in spatial resolution to a few 10 nm by the numerical aperture of modern X-ray optics (Schroer et al., 2005; Kang et al., 2006; Mimura et al., 2007; Kang et al., 2008; Mimura et al., 2010). By combining coherent X-ray diffraction imaging with scanning microscopy, these limitations in spatial resolution can be overcome (Rodenburg & Faulkner, 2004; Rodenburg et al., 2007; Thibault et al., 2008; Abbey et al., 2008; Giewekemeyer et al., 2010; Schropp et al., 2010). The spatial resolution beyond the diffraction limit of the optics is obtained by extracting the high-resolution information from coherent diffraction data recorded at each position of the scan. In this way, X-ray micrographs with spatial resolutions down to the nanometre range are possible. Recently, this has been demonstrated by microbeam coherent X-ray diffraction imaging, reaching resolutions well below 10 nm (Schroer et al., 2008; Takahashi et al., 2009).
In this paper, we apply this so-called ptychographic scanning microscopy technique to the imaging of a front-end processed microchip (512 Mb DDR2 RAM by Qimonda in 80 nm technology) as it comes out of the production line after passivation and dicing. No sample preparation was necessary. At this stage of fabrication, the mayor nanoscopic components are buried and inaccessible to other microscopy techniques, including soft X-ray microscopy. We demonstrate that the nanoscopic circuits can be quantitatively imaged with high spatial resolution and strongly enhanced signal-to-noise ratio compared to a conventional transmission image. From the knowledge of the elemental composition of the sample measured in parallel to the diffraction data using X-ray fluorescence analysis, quantitative information can be obtained about specific components of the chip, such as the thickness of plugs and interconnects. This allows for a detailed analysis of the electronic circuits.
Figure 1 shows an X-ray micrograph of the microchip recorded with our scanning microscope. The contrast shown here is the phase shift introduced by refraction inside of the sample. The electronic layout of interconnects and contacts are clearly visible and, as shown in the inset, can be schematically redrawn (reverse engineering). The smallest structures of the microchip have a size of 80 nm. The retrieved phase shift covers a range between 0 and −0.7 rad. The contrast is adjusted to highlight smallest and thinnest structures of the microchip, ranging from 0 to −0.4 rad. Features producing a stronger (negative) phase shift are depicted in black.
In Fig. 2, a schematic cross section of a typical microchip is shown. It is a layered structure consisting of various metal (M0, M1, M2) and contact layers (CS: source/drain contact, CG: gate contact, C1, C2). In the top part, a distribution of aluminium conductive paths, silane nitride insulation and empty volumes (voids) is present. The latter are visible in the X-ray micrograph as bright lines (Fig. 1). Due to the small contrast between aluminium and silane nitride as compared to that of tungsten with the surrounding materials, the aluminium structures are not seen in the ptychographic image. They would require a better signal-to-noise ratio, which could be obtained by longer exposure times. However, due to their strong contrast with the surrounding materials, the contacts and interconnects made of tungsten are clearly localized in the layers C1, M0 and CG/CS.
The phase difference of approximately −0.5 rad between a tungsten plug (marked with the number 1 in the inset of Fig. 1) and neighbouring silicon oxide yields a thickness of 570 nm of tungsten, which is in good agreement with the expected value for a C1 contact situated above a CS contact (≈2 × 300 nm). Contacts marked with number 2 in Fig. 1 can be assigned to either a C1, CG or CS contact. The phase shift introduced by conductive paths is approximately −0.08 rad, which corresponds to a thickness of tungsten of 90 nm at the photon energy of 15.25 keV.
The hard X-ray micrograph shown here was recorded with our nano-diffraction setup installed at beamline ID13 of the European Synchrotron Radiation Facility (ESRF). It is designed for scanning microscopy with transmission, fluorescence and (coherent) diffraction contrast. The highest spatial resolution with nanobeam sizes below 100 nm is obtained with nanofocusing refractive X-ray lenses (NFLs) (Schroer et al., 2005; Schropp et al., 2010) in the X-ray energy range between 12 and 30 keV.
In this particular experiment, hard X-rays (E= 15.25 keV) were focused to an approximately 80 × 80 nm2 spot on the sample by two crossed NFLs. The microchip was scanned in the focal plane on a rectangular grid with 80 × 80 steps and a step size of 50 nm both in vertical and horizontal direction. At each position of the scan a far-field diffraction pattern and the fluorescence signal were simultaneously recorded. Full diffraction patterns (without beam stop) were captured with a MAXIPIX detector having 256 × 256 pixels with an area of 55 × 55 μm2, each. The detector was positioned at a distance of 1900 mm from the sample and a tube filled with helium was introduced between the sample and the detector in order to reduce background scattering from air. The diffraction geometry fixes the effective pixel size in the reconstructed image to about 11 nm. The exposure time was 0.1 s at each scan point, and the integrated number of photons was approximately 106 per scan point.
From the 6561 diffraction patterns, the complex transmission function of the object was reconstructed with a phase retrieval algorithm introduced by Maiden & Rodenburg (2009). The ptychographic phase retrieval method allows one to obtain simultaneously the complex transmission function of the object as well as the complex wave field illuminating it. In this way, the wave field in the focal plane was verified to be approximately Gaussian with a full width at half maximum lateral size of 81 × 84 nm2 in vertical and horizontal direction, respectively (Schropp et al., 2010). The inaccuracies in the positioning of the scanning stage visible in Fig. 1 have no influence on the stable convergence of the ptychographic algorithm and lead to slight artefacts, only.
In order to characterize the performance of ptychographic imaging as compared to conventional scanning microscopy, transmission, dark-field, and fluorescence contrast are shown in Figs 3(a–c), respectively, apart from ptychographic contrast [Fig. 3(d)]. The absorption image [Fig. 3(a)] is obtained by integrating for each diffraction pattern the scattering signal over the total size of the two-dimensional detector. If the same procedure is repeated but excluding the very inner part of the diffraction pattern, that is the region of the direct beam, the dark-field image is obtained [Fig. 3(b)]. Compared to the ptychographic contrast in Fig. 3(d) the signal-to-noise ratio is significantly reduced for both transmission and dark-field contrast, illustrating the superiority of phase over absorption contrast. Both transmission and dark-field contrast are not sensitive enough to resolve the conduction layer M0, and even single contacts of type C1 or CG/CS (marked with 2 in Fig. 1) are hardly visible in the short exposure time of 0.1 s per scan point.
X-ray fluorescence yields a strong contrast for tungsten [Fig. 3(a)]. By comparing the spatial resolution of the fluorescence image and the ptychographic reconstruction, the gain in spatial resolution of ptychography over conventional scanning microscopy can be demonstrated. By evaluating the line spread of the edges of the vertical tungsten interconnect lines in Fig. 3(d), a full width at half maximum resolution of 100 nm and slightly below 40 nm can be extracted for the fluorescence and the ptychographic contrast, respectively. Thus, a resolution enhancement of over a factor 2 was obtained with short exposure time.
As the spatial resolution in the diffraction patterns is limited by the coherent dose, higher spatial resolutions can be obtained by increasing the latter (Schroer et al., 2008; Schropp & Schroer, 2010). This is possible both by more efficient focusing and by increasing the exposure time. In this way, more than four orders of magnitude in coherent dose can be gained, potentially improving the spatial resolution to a few nanometres for radiation hard samples like the one under study. To obtain ptychographic reconstructions free of artefacts on the nanometre scale, a more stable and more accurate sample positioning is required together with improved reconstruction algorithms that compensate for small positioning errors computationally (Guizar-Sicairos & Fienup, 2008). An extension to tomographic or laminographic imaging is straight forward and will give access to the three-dimensional structure of the specimen. It is planned to implement these improvements at beamline ID13 of the ESRF and in a future instrument at PETRA III (Schroer et al., 2010).