High-resolution mask inspection at the wafer plane

Although particles and defects must be avoided at all stages of semiconductor manufacture, defects on the masks (reticles) used to photographically pattern the structures are especially destructive because they affect many chips. High-resolution reticle-plane inspection (RPI) has historically been an effective and efficient means of finding important reticle defects. These defects include small reticle defects in regions that are very sensitive to errors (high mask-error-enhancement-factor (MEEF) regions). In addition, small defects associated with mask-manufacturing-process issues provide an early warning for future yield-limiting defects.

Wafer-plane inspection (WPI) technology meets the needs of advanced-mask manufacturers to identify lithographically significant defects while ignoring other, lithographically unimportant defects.^1,2 WPI detects defects based on a model of how mask features would actually print on the photoresist. The ability to ignore non-printing defects and to effectively increase sensitivity in high-MEEF areas enables development for advanced technology nodes that feature smaller critical dimensions. In addition, the modeling allows the inclusion of important polarization effects that occur in the resist for high-NA (numerical aperture) operation. Finally, the simulation easily allows users to utilize unique or custom scanner-illumination profiles. This allows the more precise modeling of profiles without inspection-system-hardware modification or risk to company intellectual property. WPI has been shown to meet chip makers' critical 32nm-generation defect-sensitivity and inspectability requirements and has been adopted by a leading US-based chip maker.

WPI combines the inherently high sensitivity of high-resolution imaging with sophisticated modeling to find all of the relevant defects and to display those that are most interesting. The process has three main components, as shown in Figure 1.

The first and most critical step is mask-pattern recovery (MPR). A new computational-lithography algorithm converts the transmitted- and reflected-light images from the inspection system into a modeled representation of the actual mask pattern, including pattern defects on the mask.

Aerial-image modeling is then performed from the recovered mask pattern, using a model of the imaging process of the 193nm scanner to generate an ‘aerial’ image of the mask as it appears just above the wafer. This unique modeling method provides a high degree of control and flexibility in the formation of the aerial image, including the ability to use both arbitrary source illumination profiles and actual, measured, scanner-illumination profiles, instead of idealized profiles. Wafer-plane modeling then translates the aerial image into a resist- or wafer-plane image by calculating where the resist is exposed. Only after the system creates full-mask images in the wafer plane is defect detection performed, because only on the resist plane is there a linear relationship between the defect signal and the wafer critical-dimension error for all geometries. With this system, a full-process-window inspection across a broad range of focus and exposure points can be accomplished with a single inspection scan.

Figure 1. Wafer-plane imaging uses longer-wavelength inspection images for mask-pattern recovery (MPR). The recovered mask pattern or a rendered database (DB) is then used to simulate the aerial image at the exposure wavelength and its ultimate printing at the wafer.

Accuracy is critical for a reticle inspector to properly match a wafer print line. To achieve accuracy for hyper-numerical-aperture immersion lithography, which employs index-matching fluid, it is essential to properly account for vector-polarization effects, since the interaction between local mask patterns and light polarization direction strongly affects image contrast. Zeiss, with the breakthrough development of the new AIMS™ 45-193i tool,^{3, 4} has developed and patented a novel means of accounting for these effects. Figure 2 shows an overlay of a profile from this tool, using the polarization-correct scanner mode, with the WPI-generated aerial-image profile for a dense contact layer with hyper-NA immersion lithography. There is excellent agreement between the two profiles. WPI effectively accounts for vector-polarization effects since this is readily accommodated within the lithography simulation.

Figure 2. The normalized aerial image intensity versus position for a cross-section through a dense contact layer for the WPI vector model. This agrees well with the Zeiss AIMS^TM tool using scanner mode.

The WPI Die : Database version, which compares the inspection results to a stored database, will be available this fall. The Die:Die version of WPI, which compares the inspected region to a reference die, was released in May 2008, and has already provided good sensitivity and inspectability on advanced-node 32N and 45N reticles in production use at customer sites. Furthermore, this performance comes at larger pixel sizes than are typically used in high-resolution inspection, which improves throughput. Using high-resolution imaging to drive the simulation at the wafer plane delivers the best signal-to-noise ratio and hence the best sensitivity, while the use of simulations of aerial images provides the best accuracy and flexibility. Overall, detecting reticle defects on the wafer plane provides the best connection to their impact on printed wafers.