Skip to main content

Ptychographic x-ray tomography of integrated circuits in three dimensions with high resolution

By Arthur L. Robinson April 21, 2017
Reconstructed image of the internal structure of a sample taken from a processor at the transistor level. The sample diameter is about 10 µm. The yellow material represents copper circuit connections that link the individual transistors. Interconnections whose lines could be followed through the volume examined are colored violet and silver for emphasis. The narrowest lines shown are around 45 nm wide. Credit: Mirko Holler, Paul Scherrer Institut

Spurred in part by the growing availability of spatially coherent x-ray beams at synchrotron-radiation sources, so-called “lensless” x-ray imaging techniques under the rubric coherent diffraction imaging (CDI) are burgeoning. Researchers at Switzerland’s Paul Scherrer Institut (PSI), which operates the Swiss Light Source, have now applied a CDI variant that they call ptychographic x-ray computed tomography (PXCT) to the problem of three-dimensional imaging of integrated circuit chips, achieving a spatial resolution of 14.6 nm along with an elemental analysis capability. They report their work in a recent issue of Nature.  

“I think this paper is a beautiful demonstration of the power of ptychographic x-ray imaging, which is a very challenging technique that is being developed by many researchers internationally,” says David Shapiro, an x-ray imaging researcher at the Lawrence Berkeley National Laboratory.

Today’s frontier for x-ray microscopy is moving toward dispensing with optics and its resolution limitations by mathematically reconstructing an image from the pattern of scattered x-rays (speckle pattern) as a coherent beam traverses a sample. Ptychography accumulates enough information to allow image reconstruction by stepping the sample through the beam to generate many speckle patterns from overlapping positions. As developed by the Swiss group for three-dimensional imaging of thick samples using multi-keV x-rays, PCXT combines this approach with sample rotation, as in tomography.

According to Mirko Holler, first author of the Nature article, the resolution for PCXT is no longer set by the beam size and stepping intervals as in scanning microscopy but by the shortest-length density fluctuations in the speckle pattern that results from the interference of scattered waves of the same frequency but different phases and amplitudes. The keys to high resolution reside in accurately knowing where the sample is and precisely controlled scanning, both obtainable with laser interferometry of the positioning. “PCXT at the Swiss Light Source was demonstrated in 2010 for the first time,” says Holler, “After that we started to develop dedicated instrumentation for hard-x-ray PXCT with accurate sample positioning and the option of operating at cryogenic temperatures.”

Using a 6.2-keV coherent x-ray beam, the Swiss group applied PCXT to cylindrical samples ion-beam milled from two kinds of microchips, an application-specific integrated circuit (ASIC) from a silicon detector fabricated using 110-nm complementary metal-oxide-semiconductor (CMOS) technology and an Intel Pentium G3260 processor fabricated using 22-nm fin field-effect transistor (fin-FET) technology. With the ASIC, they achieved a 35-nm resolution, sufficient to image detailed device geometries and interconnections. In addition, elemental identification for each volume element of the image (voxel) was possible from the absolutely calibrated density values that PCXT provides combined with prior knowledge of the materials used in the ASIC, which was designed at PSI. There was a perfect match between the three-dimensional image and the known circuit design. With the proprietary Intel chip, they achieved 14.6-nm resolution and again obtained detailed circuit images but, of course, they did not have the design to compare it to. Subsequent analysis by sequential sectioning with ion-beam milling and electron microscopy of the thin slices produced (FIB/SEM) for both chips resulted in good agreement, although the FIB/SEM images were of lower quality.

While the Swiss group has demonstrated PCXT in materials research ranging from nanocomposites to cement, integrated circuits (ICs) have served as a good test bed for the technique. Metrology of microchips with billions of circuit elements arrayed in precise patterns is a major challenge, both for manufacturers, who need to verify that what comes off their fabrication lines is what they designed, and their customers, who want to know that they are getting what they paid for and with nothing else surreptitiously added.

Reverse engineering is another option for some customers. While x-ray microtomography instruments are now available that can provide nondestructive image resolution down to about 1 µm, to do much better requires destructive sectioning techniques such as FIB/SEM, which not only is time-consuming and destructive but also introduces artifacts during the sectioning. Noting that expensive metrology instruments are already common in IC fabrication facilities, co-author Gabriel Aeppli, who divides his time between PSI, ETH Zürich, and École Polytechnique Fédéral de Lausanne, says “PCXT could be a killer app for IC metrology. If it pans out, chip makers will not hesitate to invest in dedicated beamlines at synchrotron sources, just as pharmaceutical companies already have.” Highly automated x-ray crystallography stations for structure-based development of drug candidates are a common site at synchrotron facilities.

At present, PCXT has two significant limitations for IC metrology. One is the tomography requirement for rotation around an axis normal to the x-ray beam, the reason for milling cylindrical samples from the rectangular chips. They are now implementing a variation of PCXT called laminography in which a planar sample is rotated around a tilted axis, so whole chips can be imaged. The second limitation is that the technique is still slow (about 1 day to accumulate the data for the Intel chip), owing in part to the low flux of coherent x-rays produced at today’s state-of-the-art (third-generation) synchrotron sources. Holler says, “We could increase the speed by a factor of 10,000 with improved synchrotron sources, such as the new fourth-generation sources beginning to come on line, experimental stations, and detectors.”

Read the abstract in Nature.