3.1.6 Test Structure Design for Electrically-Based Calibration of Optical Overlay Measurement Instruments

Will Guthrie

Statistical Engineering Division, CAML

Mike Cresswell,
Richard Allen

Semiconductor Electronics Division, EEEL

Misalignment of layers of circuitry on IC chips results in degraded chip performanceo or defective chips. Thus when selecting chip manufacturing methods or monitoring output, the ability to determine the misalignment of circuit layers (called overlay) is important.

Overlay is currently measured using high-volume optical instruments. A drawback of optical instrumentation, however, is susceptibility to tool-induced shifts (TIS). These systematic errors can be minimized with `shift management techniques', but even so, cannot be eliminated. However, with a test structure allowing comparison of optical measurements with more accurate overlay determinations, reliable corrections should be possible. Not coincidentally, electrically-based methods are good candidates for this application. They have the advantages of being relatively precise and insensitive to many sources of error affecting optical instruments.

Simulated data from a test structure design (currently under assessment), with a proposed data analysis is shown in the accompanying figure. The data from this structure (upper left), is discrete, indicating electrical contact between vias (wires connecting circiut layers) and one of two underlying bars separated by a space. The vias have built-in offsets which, with the unknown overlay (O_f) and noise, control whether or not contact will be made between each via and either bar.

To extract a continuous estimate of O_f, replicate substructures are used in each test structure. This allows the proportion of vias with a particular built-in offset that contact the first bar to be estimated (upper right, lower plot). The proportion of vias contacting the second bar is computed similarly (upper right, upper plot). The contact/noncontact transition points between the vias and each bar estimate $O_f \pm C$ respectively, where C is a constant related to the (unknown) absolute difference in size between the via and the interbar space.

To determine the transition-point values, a nonlinear fit of a logistic regression model (curved line) is first used to cull the data between the asymptotes. Next, this subset of data is transformed to be fit by a linear model using the logistic transformation, $\mbox{log}(x/(1-x))$. Under the linear model, the transition points are given by the x-axis offset values associated with a y-axis value of 0 (lower left plots). Using the linear model allows easier and more accurate computation of uncertainties for the transition estimates.

Finally, the estimated transition points are averaged to extract an estimate of O_f. The transition uncertainties are also combined to get an overall uncertainty estimate, which theoretically should be an approximate 95% confidence interval for O_f.

Figure 6: Simulated data from a prototype test structure with accompanying data analysis plots and results. The dashed lines on the final plot indicate transition points and the solid lines indicate estimated overlay (center line) and uncertainties. The uncertainties are proposed 95% confidence intervals. The estimated overlay in this example is $-12.15 \pm 2.20$ nm. The true overlay, from the simulation input, is -12 nm.

       

Date created: 7/20/2001
Last updated: 7/20/2001
Please email comments on this WWW page to sedwww@nist.gov.