Monitoring Yield-Critical Defects in DRAM Structures

Inspection strategy

At its 300 mm facility in Dresden, Qimonda AG (formerly Infineon Technologies) is currently running 90 nm design rule in production. Process changes, especially in the front-end-of-line (FEOL), were causing several yield issues and required a method of more intense monitoring to prevent yield busts.¹ Two issues occurring in the FEOL DRAM stack will be discussed (Fig. 1 ). The issues occurred in the late deep trench module and isolation trench (IT) module. The critical process steps involved are buried strap etch and shallow trench isolation (STI) CMP.

One issue was caused by residue in HAR structures in the nearly filled trenches after the buried strap etch step. Such residue appeared either attached around part of the trench wall or as a bridge-like structure at the bottom of the trench structure (Fig. 2 ). The residue affected a number of trenches across a large area, and could be easily detected as previous layer defects in the subsequent IT etch step. Therefore, the intention with the new inspection step on buried strap etch was to detect the defect at the exact process step in order to facilitate a much faster feedback loop.

1. Critical process steps in the FEOL DRAM stack include the buried strap etch and the STI CMP.

STI voids were the second yield issue that required monitoring. Tiny oxide voids occurred between the active areas (AA isles) of the STI module (Fig. 3 ). These defects grow increasingly more critical with decreasing design rules. In subsequent process steps, such voids get filled in, causing shorts and degrading device reliability.

2. In a trench cluster (left), the after buried strap etch residue appears around the trench wall (A) or as bridges at the bottom of the trench structure (B), or may occur in isolation (right).

3. 30-70 nm STI voids typically occur between the active area (AA) isles. They usually originate in the oxide fill process and become exposed by CMP.

After buried strap residue

The detection of the HAR structure residue raised two main concerns. The residue appeared as clusters with multiple trenches affected, forming an elongated signature. The first concern was uncertainty of the average size of the clusters detected. In addition, the capture rate of residue defects in isolated trenches was unclear.

4. Cluster size distribution of residue-affected trenches shows a higher capture rate of larger clusters.

5. Residues in isolated trenches: Inspector patches relevant for detection (defect encircled) with the associated signal to noise (left) and related SEM images (right).

6. After buried strap residue in the trenches, with the affected signature perpendicular to those shown previously in Figures 2 and 4 .

7. Typical defect types at the buried strap inspection are small particles and residues on the surface (A, B), residues in isolated trenches (C, D) and residues affecting multiple trenches and forming specific signatures (E, F).

Figure 5 shows three typical defects from that class. The inspection compares the defective defect patch image with its reference patch from the adjacent dies. Both patches can be saved routinely on the tool. These patches are very helpful, as they can also be used later for inline defect classification (iADC) setup or to assist with defect location during SEM review. It could be shown that signal-to-noise values of these defects range from 2 to 7. Comparison of the detection parameter characteristics from each defect and the threshold parameters used in the detection algorithm reveal that these defects can be detected by a fair margin with good capture rate.

The second concern was related to the orientation of the laser beam incidence with respect to the symmetrical axis of the trench, as well as a special polarization scheme of the illumination and collecting light, which were both essential to capture the signature. It was demonstrated that, for the orientation where the laser beam incidence was perpendicular to the symmetrical axis of the trenches, the S polarization in the illumination path and the S polarization or no polarization filter in the detection path could reveal these defects.

For the laser beam incidence parallel to the symmetrical axis, a polarization combination (180° shifted, P polarization) in the illumination path and only the P polarization in the detection path could reveal the defects. All other polarization combinations in the inspection were not able to detect this special residue defect type, which was located in the trenches and, thus, hard to review optically and with the SEM.

Given that all residue-affected trench clusters exhibited an elongated appearance (Fig. 4), there was also a concern that the orientation of the signature itself could potentially impact the detection. However, evidence revealed that the signature of residue-affected trenches could be detected if, under certain process circumstances, the signature appears orientated perpendicular to the symmetrical axis of the trench structure (Fig. 6 ).

Through the detailed investigations above, the inspection was well characterized and well suited to serve as a tool to drive and monitor the process efforts to resolve the problem. Although systematic process variation issues could be dealt with given quick inspection feedback, there will always be the possibility of the residue issue returning as a tool excursion. Thus, the inspection purpose changed from a pure HAR residue monitor to a baseline inspection with integrated residue excursion monitoring.

Figure 7 shows the main defect types caught in this baseline monitoring step. These defects include surface particles and residue in the trenches (Fig. 7, C, D) that cause the signatures (Fig. 7, E, F) and those that are at the surface and are randomly distributed (Fig. 7 , A, B).

STI oxide voids

Selective sensitivity and high-throughput inspection are key to effective STI oxide void capture. The main focus of the investigation was to provide a fast monitoring inspection for possible void excursions on the wafer. Although other defects — such as bridging, small particles, pattern defects and scratches — can be detected in this inspection step, void excursions tend to increase the defect count significantly. Thus, monitoring total defect count is an easy way to flag potential void excursions.

All defect types can be revealed with any inspection magnification, as shown in Figure 9 . All throughput and magnification modes investigated had sufficient sensitivity toward the void defect type to capture all the relevant signatures. Even more important, the void detection could be realized at a very low SEM non-visible (NV; nuisance) rate.

8. There are two different types of voids occurring; the main focus is the voids on the diagonal between the AA isles. The liner voids at the short edge of the AA isle typically appear in large clusters.

9. Puma 9000 was able to detect the voids, maintaining a low nuisance rate, with all the tested pixel sizes.

With sufficient capture of voids in all pixel sizes, a very cost-effective inspection could be created using the largest pixel size.

Comparing these sL10 Puma inspection results (~17 wph) with the optical sensitivity benchmark of a KLA-Tencor 2365 brightfield pattern wafer inspection tool reveals, as expected, a much higher STI void capture with the 0.12 µm pixel size 2365 inspection (~1 wph). Figure 10 shows a one-die comparison, where all defects reviewed were STI voids. The 2365 inspection revealed more small voids, highlighting the trade-off of sensitivity and throughput.

10. Full-die comparison of Puma 9000 sL10 inspection (right) and 2365 0.12 µm pixel size inspection at 25% coverage. Puma inspection provides sufficient and very specific sensitivity toward voids at much higher throughput without additional nuisance filtering.

Although the large-pixel Puma inspection had a reduced void count capture, it was less susceptible to previous layer noise, which is prevalent when applying brightfield inspection technology on this layer. This applies basically throughout all the magnifications on the Puma 9000. Figure 9 shows a similar nuisance rate for all magnifications, with Puma's high-sensitivity pixel (sL60) at ~2× higher capture rate of voids compared with Puma's larger pixel (sL10). With the more sensitive brightfield inspection, the nuisance rate has to be reduced by repeated filtering, on the tool and in the yield management system, to obtain the results shown in Figure 10 .

11. Even smaller STI oxide voids at an earlier process step occur, here shown in a cluster. Afterwards, they get decorated through several doping and clean steps, and thus become easier to detect.

Conclusions

Deep-trench DRAM production beyond the 100 nm design rule node requires the introduction of new processing and design concepts. New processes and designs always imply yield challenges in the development and ramp phase. Additionally, especially in the FEOL, known defect mechanisms like oxide-filling voids become increasingly critical, because printed pattern features in the die range are the same size dimensions as the defects (e.g., voids, particle). In this work, we demonstrated that a novel darkfield inspection technology that offers selective sensitivity at high throughput can very effectively address two major yield-limiting process issues in advanced DRAM production and, therefore, enable a much faster development and ramp.