Nanoscale Three-Dimensional Imaging: An Innovative Tool for Failure Analysis

Neil A. Ives, Martin S. Leung, Gary W. Stupian, Steven C. Moss, Nathan Presser, and Terence S. Yeoh

A new method of cross sectioning and visualization provides detailed images of submicron features. Images can be rendered in movie format to show feature layers virtually melting away.

Microelectronic devices for both terrestrial and spacecraft hardware have been growing ever smaller, with features now measured on the nanoscale—that is, less than 100 nanometers (nm) in size (see sidebar, "Nanoelectronics"). In the past, for larger devices with features on the order of 10 microns, a 100-nm defect would not pose a significant problem. Today, a defect on this scale—such as a void, a misalignment, a nodule, a particle, or a dendrite—could prove catastrophic.

This SEM image shows the trench excavated by the ion beam. The internal structure of the electronic device can be seen on the back wall of the trench.

Visualization of device structures at the submicron and nanometer level has therefore been crucial for improving microelectronic and optoelectronic device performance and for investigating the fundamental causes of device failure. In particular, the emergence of advanced microanalytical techniques such as focused ion beam (FIB) milling has added new dimensions to the applicability of electron microscopy in semiconductor device research and development. FIB milling enables cross-sectional cuts at any location on a semiconductor component with precision and accuracy at the nanoscale. Newer FIB systems are dual-beam instruments that incorporate both an ion beam for cutting and a scanning electron microscope (SEM) beam for imaging the cross section exposed.

This tool has become a standard for failure analysis; however, its full potential has remained largely unexplored. The typical failure analysis involves only one FIB cut through an area of interest and one image from the electron microscope. A single slice contains some, but not all, of the structural and spatial information needed for a comprehensive analysis. Multiple sequential cuts will of course provide more details of internal device structure. But making numerous slices of an object measuring only a few nanometers requires more precision than standard FIB systems can achieve. Moreover, the amount of data generated would require significant processing power to be useful.

Deconstructing an electronic device for 3‑D reconstruction is performed with a focused ion beam. The technique involves both cutting and imaging. The focused ion beam (FIB) cuts off a slice of the electronic device to expose a new face, which is then imaged with the scanning electron microscope (SEM). Cutting and imaging is repeated at regular intervals until the entire structure has been sliced away.

In response to this challenge, Aerospace has developed a new method of cross sectioning, imaging, and visualization. The technique can be used to generate 3‑D models of nanoscale features that can be examined from all angles. This technique has been dubbed nanoscale 3‑D imaging, or nano-3DI. In recent device failure investigations, it's proven to be a crucial tool for determining root cause.

Nano-3DI

In the nano-3DI technique, the ion beam strips away a thin layer of material from the region of interest, and the SEM images the surface of the newly exposed edge. What sets this technique apart is the extreme precision and number of the cuts and images. In fact, Aerospace has developed a special FIB milling technique that can remove material in slices less than 2 nm thick using a standard ion beam roughly 30 nm in diameter. This innovation involves using the change of SEM image contrast and brightness caused by removal of surface carbonaceous deposit as an end point. Thus, the process of cutting and imaging can be repeated at nanoscale increments until the entire structure containing the features of interest is physically deconstructed. It can then be digitally reconstructed from the images taken after each cut.

Illustration showing multiple segmented images from a field-programmable gate array (FPGA) antifuse. The six component objects are obtained from one cross-sectioned and imaged slice. Objects from all slices are extracted in a similar manner and then combined to reconstruct the 3‑D object.

This specialized technique allows researchers to capture a set of highly spatially correlated images of the feature of interest over the entire volume of the failure site, rather than just a few poorly correlated 2-D pictures. The grayscale information contained in the image dataset, used in combination with customized software for extracting the features of interest, enables sophisticated 3‑D volumetric imaging. Given that the SEM has spatial resolution better than 1 nm and the slice thickness is less than 2 nm, the 3‑D volumetric resolution is estimated to be better than 2 nm³. This resolution is at least an order of magnitude better than the current state of the art.

Moving Pictures

While improvement in resolution is highly desirable, the voluminous data generated in the form of SEM images represent a challenge because the features of interest have to be extracted, analyzed, and presented for interpretation. For example, if a structure in a microelectronic device measures 500 nm across, a complete deconstruction would produce between 250 and 500 slices at 1–2 nm per slice. The resulting set of images represents approximately 1–5 gigabytes of data, depending upon the total number of slices and the image resolution. In principle, this set of images contains all of the details of the structure that have been captured. Viewing the images individually or in combination with others, however, does not usually allow the analyst to grasp the essential details (e.g., shape and orientation) of the features in the entire structure.

The electronic device is repeatedly sliced via focused ion beam and imaged with the scanning electron microscope (SEM) through the entire area of interest. Pixel elements from the 2-D SEM image plane are computationally reconstructed with grayscale information from the adjacent planes to derive the voxel that represent the volume element of the 3‑D object.

To overcome this difficulty, Aerospace made use of advanced image-processing techniques to produce movies from all the SEM images collected, making it easy to visualize shape and orientation of the features of interest. The movies render the vast amounts of visual information into a format that would be easy to analyze and interpret. The process can be understood by imagining a deck of playing cards. The face of each card displays an image showing one slice in the complete volume data set. After making the necessary adjustments to ensure precise alignment of the image in each card, the system can display them in flip-book fashion. As one watches the succession of images, various features come and go as they are first exposed then cut away by the ion beam. This mode of data presentation allows the viewer to see the shapes of the features in detail and their spatial relationship with one another. However, even this viewing mode provides only a subjective and unquantifiable mental impression of the features being viewed.

To obtain more quantitative information, Aerospace used the advanced visualization tools of the Amira 3‑D modeling program for model extraction based on voxel reconstruction and segmentation. A voxel (from "volume" and "pixel") is the basic volumetric image element in a 3‑D dataset. Voxel reconstruction is more commonly associated with noninvasive medical imaging methods such as MRI and CT scanning, which generate detailed models of internal organs from a series of individual "slices." Similar datasets on the nanoscale are generated through multislicing FIB deconstruction (a big difference, of course, is that medical imaging is nondestructive, whereas nano-3DI consumes the part during slicing). However, the thickness of each FIB slice is not as precisely controlled as in CT or MRI, so the process of voxel reconstruction is not entirely straightforward.

Voxel Reconstruction

Computer-generated voxel reconstruction of an FPGA antifuse. Data volume (left) with selected planes (middle) for voxel slice viewing (right).

If the FIB milling machine produced a perfect and uniform 1-nm slice every time, then voxel reconstruction of the 3‑D features would simply involve stacking up the images, using the grayscale data for alignment. However, in practice, the FIB cuts are not always the same thickness.

To compensate, Aerospace used an interpolation scheme in which a priori information about the larger features of the device is used to calibrate the thickness between the slices in each region of cuts through the structure. Artificial layers are inserted to keep the apparent spacing uniform and continuous. The computer processing algorithm calculates the grayscale in the artificial layers based on interpolation between regions of similar grayscale in adjacent real layers. In the worst cases, this has required insertion of one or two artificial layers between real layers at a few locations within a structure.

The complete voxel reconstruction with both real and interpolated layers now forms a uniformly spaced 3‑D grid with grayscale data at each grid point, or voxel. This grid can be "virtually" sliced and viewed along any direction at any plane, creating flip-book movies along the three independent spatial axes as well as any other compound axes. These virtual cuts allow visualization of the failure site from any angle—even angles not possible with the real FIB because of limits imposed by the system geometry.

Solid Model Reconstruction

This reconstruction technique also provides a new and more structured 3‑D dataset that can be used to generate solid 3‑D models. In devising the process, Aerospace researchers first made the reasonable inference that the grayscale information in the 3‑D grid must correspond to real features—including the extent of structural changes, phase formation, and separation and voiding—that can be visualized with respect to specific material locations to better understand the chemical and physical mechanisms involved. So, to create a solid model, the features of a 3‑D object are first identified in the 2-D voxel slices according to their grayscale image values. They are then segmented—that is, a boundary is drawn around each one. This step is repeated until all the individual features in the dataset have been segmented. They can then be stacked using a separate image-processing algorithm.

An exploded view at different angles for surface-contour 3‑D reconstruction of the component objects in an FPGA antifuse.

For example, dark voxels indicate a void, while bright voxels are typically associated with a metal. Semiconductors appear as voxels with an intermediate grayscale. Of course, complex material phases may confound these simple distinctions, and imaging artifacts (caused, for example, by the charging of insulators) may also complicate interpretation of the grayscale data. In practice, the investigator usually has some a priori knowledge of device structure and materials composition to guide the segmentation efforts. Further information can be obtained from other microscopic and spectroscopic techniques, which allow identification of not only the elemental composition of features but also chemical-bonding information with nanometer resolution.

Multiple 3‑D solid-model reconstructions of an FPGA antifuse. The model can be sliced along X, Y, and Z planes to visualize more detailed information with respect to the chemical composition of the interior geometries.

Once the features are segmented, individual 3‑D models can be constructed such that each feature uniquely occupies its own space within the 3‑D dataset. Each feature can be assigned a false color to represent its chemical composition and allow viewers to easily distinguish it from other features of different composition. The computer software can display each feature as a solid, a semitransparent object, or a transparent object with a contoured surface. The presentation of the feature as a contoured surface is effective in showing the spatial relationship of one feature to another.

The solid 3‑D models can also be virtually sliced, much like the virtual slicing of the voxel image. This allows investigators to obtain detailed information on the chemical composition of internal features that were present but hidden or obfuscated by adjacent features. For failure investigation, the collection of features that make up the failure site can be presented in an exploded view to show how the individual components of the device fit together. Special effects can be employed to view the individual components stereoscopically to provide perspective and detailed spatial relationships of one feature to another. Once the entire structure exists as a 3‑D interconnected object, it can be imported into various simulation packages that provide an even more realistic model of failure sites. Data without interpretation is of minimal value. When data are displayed in a new and more intuitive fashion, new insights often emerge, and the physics of the root cause failure mechanisms can be more easily conceptualized. Important parameters such as resistivity, diffusivity, and reactivity of materials may also be derived quantitatively from the solid models.

Working with solid models requires customized software, specialized hardware, and raw computer power that is available on multiprocessor workstation-class computers but not on standard desktop platforms. Exporting this new information to a common desktop platform can be accomplished using animation software. The model, rendered as a digital movie file, can then be given special views, rotated, and exploded in a scripted manner to illustrate key points about the morphology using standard media players.

Conclusion

Imaging of nanoscale features in microelectronic and optoelectronic devices is essential for understanding the complex internal workings and failure modes of advanced technologies. By using state-of-the art electron imaging and ion-beam cutting equipment, Aerospace researchers can generate 3‑D models of device features with nanoscale resolution. The nano-3DI volumetric imaging method developed at Aerospace provides valuable insights, otherwise unobtainable, of the internal structure of complex nanoscale devices and could become a standard tool for future reliability investigations for both terrestrial and space hardware.

To Fall 2005 Table of Contents