Improved SRAM detection through scattered light collection

10/01/2011

Reuven Barel, Keren Shachar, Yakir Bechler, Nir Horesh, Applied Materials Israel, PDC, Rehovot, Israel;

Hsien-Tsung Chiang, To-Yu Chen, Taiwan Semiconductor Manufacturing Company (TSMC), Hsinchu, Taiwan ROC

A new optical scheme for inspection of advanced SRAM patterns offers enhanced defect signal-to-noise ratio (SNR), enhanced defect capture rate, and higher throughput.

Accelerated design rule shrinkage over the past years has continuously introduced new challenges to the world of yield monitoring and optical wafer inspection. One of the main challenges is the detection of ever smaller defects of interest (DOI) below the optical resolution limit of the WI systems. Evident to any large scale manufacturer today, these DOI, typically in the size range of the technology node, are correlated to yield, and their detection is therefore crucial to the production yield.

Traditional bright field (BF) systems detection sensitivity depends on the optics resolution and pattern contrast, which in turn is correlated with ??/NA (?? is the illumination wavelength and NA is the numerical aperture of the objective lens). In advanced DR, when the pitch size is much smaller than the optical spot size on the wafer, BF resolution reaches its limit and certain DOI cannot be detected by traditional BF inspection.

Scattered light detection, which involves collection of scattered light outside of the illumination cone, and detection on a dark background, overcomes the resolution barriers, since the detection is determined by the signal-to-noise ratio (SNR) and not by pattern contrast, which is impacted by the optics resolution.

Unresolved dark scenarios are typical of repetitive dense arrays, where distinct diffraction lobes at specific locations are formed and blocked by spatial filtering. Typically, when such lobes can be blocked by a spatial mask, dark background detection scenarios are enabled. This is usually the case of advanced DR, line-space structures, such as DRAM and Flash, where the lobes are either outside of the collection aperture or are filtered out. In such cases, this detection method offers higher sensitivity than traditional BF.

However, on logic patterns, this method is typically not applied because of the pitch size and the pattern complexity formed by multidirectional random features. In general, logic devices consist of two main pattern types that require different inspection strategies. The first pattern type is the logic circuitry, and is typically characterized by random patterns. The second type of pattern is the SRAM, which is typically characterized as a repetitive pattern. Scattered light detection on SRAM patterns, based on scattered light and pattern suppression was typically not performed due to the plurality of the diffraction lobes and of the complexity of the required filters. On the other hand, BF imaging has limited resolution for detection of small defects in the size range of the DR. This situation often leads to reduced sensitivity on critical structures.

In this work, we present a new scheme involving tailored illumination and collection to enable effective pattern suppression. To demonstrate the new scheme, we used a laser-based DUV system having both brightfield (BF) and scattered light collection channels (UVision, Applied Materials). BF light is collected through the illumination aperture and scattered light (grayfield - GF) is collected outside of the illumination aperture.

Experiment

Two 28nm DR wafers at two process steps, via layer after resist development (via after develop inspection), and SiGe deposition at the gate stage (SiGe deposition) were analyzed in this work. In the via ADI layer, the SRAM areas were inspected in scattered light mode, with pixel size (PS) of 100nm and 140nm using tailored masks. In the SiGe DP layer, the SRAM areas were inspected using the scattered light mode with PS of 100nm, BF mode with PS of 100nm and 70nm, and in the new SRAM mode using PS of 140nm and 190nm. Table 1 summarizes the configurations used for the two layers. Random defects from all inspection maps were reviewed on an SEM tool.

Results and discussion

Via ADI. In order to compare the SNR between the BF and the scattered light configurations, two particle defects the size of ~50nm, which were detected by both configurations, were chosen. The SNR was calculated based on the inspection images, as shown in Fig. 1.

Figure 1. Top: SNR comparison for two particle defects, for the scattered light @100nm PS (left columns, Blue) and new SRAM mode @140nm PS (right columns, Red). Bottom: Scan and difference images of particle #2. Left image of each pair correspond to scattered light @100nm PS. Right image of each pair corresponds to the new SRAM mode @140nm PS.

From the comparison in Fig. 1, it can be clearly seen that the SNR is significantly higher in the new SRAM mode compared to the scattered light mode. The SNR difference is understood by comparing the scan images in Fig. 1. While both images have a similar defect signal, the pattern, seen in the scattered light mode images, is effectively filtered in the new SRAM mode, hence improving the SNR.

SiGe DP. SRAM pattern areas were inspected using the configurations shown in Table 1. Capture rates of three defect types were compared between the inspection configurations (Fig. 2): 1) missing pattern, 2) bridge, and 3) particles.

Figure 2. Capture rate comparison of the three defect types for all the inspected configurations, with missing pattern and bridge defect illustrations.

From the comparison Pareto shown in Fig. 2, it can be clearly seen that the new SRAM modes are the only modes to capture the missing pattern defect type, and that these modes also have a better CR of the bridge defect type compared to the other modes.

The difference in the CR between the resolved configurations to the new SRAM modes can be explained by analyzing the missing pattern defect scan images. Since the defect signal is low, the reduction in background noise achieved by the new SRAM mode significantly enhances the defect SNR.

The new SRAM modes have additional benefits in addition to the increased CR and SNR. The ability to use larger PS (140nm and 190nm) vs. 70nm and 140nm, as demonstrated in the two examples above, enables higher WI productivity, and therefore lower cost-of-ownership.

Conclusion

A new inspection scheme was proposed and demonstrated for enhanced sensitivity on SRAM patterns for 28nm DR, via ADI and SiGe DP layers. The scheme is based on tailored illumination and collection. The new inspection scheme has demonstrated higher sensitivity over scattered light and BF inspection. The ability to filter SRAM pattern increases the defects' SNR. The SNR increase enhances the CR of these defect types and enables capturing defects with lower signal. Using the new SRAM modes enables not only higher sensitivity, but also allows using large PS, thereby enabling higher throughput and productivity.

Acknowledgment

UVision is a trademark of Applied Materials.

Reuven Barel is a brightfield Product Specialist at Applied Materials Israel, PDC, 9 Oppenheimer Street, Rehovot, Israel; email [email protected]

More Solid State Technology Current Issue Articles

More Solid State Technology Archives Issue Articles