Computer Chips Measured Using Optical Scatterfield Technique

Using a microscope that combines standard through-the-lens viewing with scatterfield imaging techniques, researchers have accurately measured silicon wafer features 30 times smaller than the wavelength of light.

The achievement highlights how scattered light, once disregarded, contains a wealth of accessible optical information.

The study was a part of a National Institute of Standards and Technology effort to supply measurement tools that enable the semiconductor industry to continue doubling the number of devices on a chip about every two years, and to help other industries make products with nanoscale features.

An illustration of how tiny changes in wavy images scattered from lines in a gridlike array can be reconstructed when paired with advanced optical and computational techniques. Lines are 15 nm wide, 30 times smaller than the wavelength of light used to measure them. Images courtesy of Bryan Barnes/NIST.

NIST researchers reported that measurements of the etched lines — as thin as 16 nm — on a wafer fabricated by R&D institute Sematech in Austin, Texas, were accurate to 1 nm. With the technique, they identified variations in feature dimensions amounting to differences of a few atoms.

Measurements were confirmed with an atomic force microscope, which achieves subnanometer resolution but is considered too slow for on-line quality control measurements.

Combined with earlier results, the researchers wrote, the optical approach solves a problem facing chip makers and others seeking to commercialize nanotechnology by offering a way to nondestructively measure nanoscale structures with subnanometer sensitivity without sacrificing high throughput.

Optical microscopes can't identify features smaller than the wavelength of light (around 450 nm), at least not in the crisp detail necessary for accurate measurements. However, light does scatter when it strikes subwavelength features and patterned arrangements of such features.

The pattern depicts estimated uncertainties in the experimental data. Coloring corresponds to the magnitude of the variance for specific data points.

"Historically, we would ignore this scattered light because it did not yield sufficient resolution," said Richard Silver, a NIST physicist who led the scatterfield imaging effort. "Now we know it contains helpful information that provides signatures telling us something about where the light came from."

Silver and colleagues methodically illuminated a sample with polarized light from different angles. From this collection of scattered light — which appears to the untrained eye as a sea of wiggly lines &dmash; team extracted characteristics of the bounced lightwaves to, in aggregate, reveal the geometry of features on the specimen.

Light-scattering data are gathered in slices, which together image the volume of scattered light above and into the sample. The slices are analyzed and reconstructed to create a 3D representation in a process akin to a CT scan, except the slices are collections of interfering waves, not cross-sectional pictures.

"It's the ensemble of data that tells us what we're after," said project leader Bryan Barnes. "We may not be able see the lines on the wafer, but we can tell you what you need to know about them &dmash; their size, their shape, their spacing."

Scatterfield imaging has critical prerequisites that must be met before it can yield useful data for high-accuracy measurements of small features. Key steps entail detailed evaluation of the path light takes through lenses, apertures and other system elements before reaching the sample. The path traversed by light scattering from the specimen undergoes the same level of scrutiny.

Fortunately, scatterfield imaging lends itself to thorough characterization of both sequences of optical devices, according to the researchers. These preliminary steps are akin to error-mapping, factoring out sources of inaccuracy.

The method also benefits from the as-designed arrangement of circuit lines on a chip, down to the size of individual features. Knowing what is expected to be the result of the complex chip-making process enables comparison of theoretical and experimental data.

The researchers used standard equations to simulate light scattering from an ideal, defect-free pattern. Using wave-analysis software they developed an indexed library of light-scattering reference models. Once a specimen is scanned, the team relies on computers to compare their real-world data to models and to find close matches. From there, succeeding rounds of analysis hone in on the remaining differences, reducing them until the only ones that remain are due to variations in geometry such as irregularities in the height, width, or shape of a line.

Next steps include extending the technique to even shorter wavelengths of light in the UV range, with an ultimate goal of accurately measuring features as small as 5 nm.

The research will be published in Light: Science & Applications (doi: 10.1038/lsa.2016.38 [PDF download]).

In a related development, Checkpoint Technologies LLC of San Jose, Calif., said it had achieved optical resolution of 125 nm using solid immersion lens technology and laser scanning microscopy, with potential applications in semiconductor failure analysis and optical debugging.

"We believe this is the smallest optical resolution achieved with midrange visible light through the backside of silicon," said G. Xiao, the company's chief technology officer.