sealPurdue News

April 4, 2002

Lasers are key to future chip-processing detective work

WEST LAFAYETTE, Ind. – Engineers at Purdue University have demonstrated concepts that could eventually save computer chip manufacturers millions of dollars in downtime every year by using lasers to quickly identify and pinpoint the sources of contaminating dust and other defects.

Semiconductor manufacturers already use lasers to detect dust particles on expensive silicon wafers, which contain hundreds of chips. But then the manufacturing operation must be shut down while workers try to determine what the dust particles are made of and where they came from, especially when large quantities are found. The source of the dust must be eliminated before production can resume. If an electron microscope is needed to trace the dust, the contaminated wafer is essentially destroyed in the process, adding to the loss.

Each 300 millimeter-wafer, roughly 12 inches in diameter, can yield 200 chips, which might eventually be worth nearly $1 million. If a new wafer rolls off the line every minute, one hour of downtime can cost tens of millions of dollars.

"It would be a huge advantage if you could identify the contaminant and its source in addition to just detecting it during the in-line laser inspection process," said E. Dan Hirleman, a professor and head of the Purdue School of Mechanical Engineering. "As the features in circuits are getting smaller every 18 months or so, the size of a killer defect is getting smaller and smaller. One way to rapidly detect and identify smaller defects is to use laser beams of shorter wavelengths, such as in the deep ultraviolet range."

The same lasers used to detect dust might also be used to identify the particles and begin tracing the contamination source within seconds.

Because circuits in new computer chips are only slightly wider than the particles, the contaminants are large enough to ruin or short-circuit the tiny "wires" in the chips. New chips scheduled to debut in 2006 will contain circuit features only 70 nanometers wide – so tiny that they will be ruined by dust particles as small as 30 nanometers, which is only a few hundred atoms across, or roughly 1/3,000 the width of a human hair. Although the wafers are produced in clean environments that are isolated from the outside world, all dust cannot be eliminated.

"This nanodust could be any number of things, like particles of tungsten, silicon or photoresist material used in the manufacturing process," Hirleman said. "It could be something that flaked off from the side of the walls of the reaction chamber. A big challenge will be to develop online laser-based material identification, which allows you to in seconds start the detective process."

Purdue researchers have developed mathematical models that reveal a dust particle's fingerprint hidden in the precise way in which it "scatters" laser light. The laser reflects off of the silicon wafer's near-perfect mirror surface, but the light does not reflect well from dust particles. Some of the light is said to be scattered as it bounces off in different directions after hitting a dust particle. Different types of particles have specific scattering signatures, which can be modeled mathematically, creating fingerprints that can help identify certain types of particles.

Hirleman and doctoral student Haiping Zhang presented a research paper detailing the findings March 6 during the conference Process Characterization and Diagnostics in IC Manufacturing in Santa Clara, Calif. The conference, the first of its kind, was sponsored by the Society for Photo-optical Instrumentation Engineers.

"A new conference was needed because of the growth in the nanotechnology field and the special needs for measurement and sensor technologies," Hirleman said.

Data from the models might be used by industry to design new types of instruments that reduce downtime. Future technologies might even use lasers to remove contamination, a sort of "laser dry-cleaning" system that would shake off the particle, Hirleman said.

Such a dry-cleaning technique would be needed because wet-cleaning methods often deposit more particles than they remove, actually increasing the contamination. The models developed at Purdue also will be important for designing such instruments.

The research engineers precisely measure how laser light bounces off dust using a new instrument called a scatterometer. Then they use those data to validate complex mathematical modeling software that identifies a particle's composition based on its light-scattering fingerprint. Purdue researchers also are developing a new, more precision deep ultraviolet scatterometer that can detect smaller particles, which is essential to keep up with the constantly shrinking dimensions of computer chip circuitry.

The work is funded by a group of government agencies and companies known as the Consortium for Metrology of Semiconductor Nanodefects and by the Semiconductor Research Corporation.

Writer: Emil Venere, (765) 494-4709,

Source: E. Dan Hirleman, (765) 494-5688,

Purdue News Service: (765) 494-2096;

NOTE TO JOURNALISTS: An electronic copy of the research paper described in this release is available from Emil Venere, (765) 494-4709,


Computation of light scattering from particles on a filmed surface using discrete-dipole approximation

Haiping Zhang and E. Dan Hirleman, School of Mechanical Engineering, Purdue University

Numerical calculation of angle-resolved light scattering characteristics of features with arbitrary shapes, such as particle contaminations and surface defects, on a filmed surface is very useful to the development and calibration of wafer inspection tools. A model and associated code based on the discrete-dipole approximation (DDA) used to compute the light scattering from a particle on a filmed surface is developed. The reflection interaction matrix is modified with the Sommerfeld integrals for filmed surfaces. Three-dimensional fast Fourier transform method is used for accelerating the computation. Model predicted scattering signatures for polystyrene latex (PSL) spheres on a smooth thin layer of silicon dioxide film on silicon substrate are compared with experimental results. The incident beam has a wavelength of 632.8 nm and the incident angle is 70 degrees. The comparison shows very good agreement between the modeling results and experimental results. The model is also checked with another numerical model, which further shows the validity of the model.

* To the Purdue News and Photos Page