Low-k thin films analyzed using automated SEM sample preparation
10/01/2000
Fred Shaapur, David Griffiths, International Sematech, Austin, Texas Efrat Raz, SELA, Sunnyvale, California
overview
Automated SEM sample preparation avoids many of the pitfalls associated with analyzing today's advanced materials. For example, no matter how carefully they were prepared or what processing tricks were used — ion milling, for instance — conventional manual cleaving of wafers typically resulted in artifacts on samples of HOSP, a contending low-k dielectric material. Automated sample preparation yielded accurate, artifact-free cross-sections, however. These samples were prepared quickly with minimal operator skills.
Scanning electron microscopy (SEM) has become a primary tool for process characterization. The time required to produce meaningful SEM data is an overall contributor to the length of the learning cycle. Sample preparation is a critical aspect of SEM analysis. Quality, precision, reliability, and speed are all becoming increasingly critical as features shrink. Automated SEM sample preparation addresses these issues.
Metal lines in interconnect structures are becoming so closely packed that electromagnetic fields are causing cross-talk between adjacent metal lines. In an effort to reduce this cross-talk, materials with dielectric constants lower than silicon dioxide are being evaluated at International Sematech for use in next-generation semiconductor manufacturing. Candidate materials include organo-silicate glasses (OSGs), whose material properties differ substantially from those of traditional dielectrics.
Hydrido organo siloxane
One OSG currently being evaluated as an inter-layer dielectric to be used in combination with copper is hydrido organo siloxane polymer (HOSP) [1]. Part of our work with this material has addressed challenges encountered while trying to prepare samples containing HOSP films for SEM analysis.
HOSP is a spin-on methlysilsesquioxane (MSQ) that contains high levels of carbon doping. The cagelike structure of HOSP has more internal stress than traditional dielectric materials. These stresses cause most of the problems encountered when trying to prepare samples to be imaged. For example, when HOSP-containing samples are prepared by the traditional method of manually fracturing a silicon wafer, the internal stress within the HOSP can cause the material to tear, delaminate, chip, bow, or otherwise deform.
In our work we found that the "cleaving mechanism" of the SELA MC500 automated microcleaving tool solves some of the problems created by HOSP's high internal stress. From a wafer, this tool automatically creates a high quality cross-section for SEM analysis with very little operator interaction (see "Automated microcleaving" on p. 168); the operator has only to enter identification and segment parameters, and accept or reject the automatic cleaving map calculated by the tool. Significantly, a specific target, such as a contact or transistor, can be cleaved with a target-positioning accuracy of 0.5mm.
Cross-sectioning HOSP
In our work with the difficult to cross-section HOSP, we used several techniques to try to produce acceptable samples from a manual cleave, including liquid nitrogen cooling and, alternately, heating the sample. No technique was able repeatedly to produce cross-sections in which the HOSP was not deformed, however.
 Figure 1. SEM image of vias in HOSP where sample area was 7 in. from an initial scribe to generate a traditional manual cleave. |
With traditional manual-only cleaving, we found that only by choosing a target far from the initial scribe could we produce a reasonable quality cross-section. This technique involves scribing a small cleave-inducing mark 180° from the wafer's notch and then taking a sample from notch end of the cleave. We believe that at this distance from the scribe, the cleave propagates along a silicon crystal plane, reducing the stress. The image from the method still shows a small amount of tearing in the HOSP layer, but the profiles of the vias are unaffected (Fig. 1).
Even leaving additional space between the target and the scribe does not, however, guarantee that the stress in the HOSP layer will not cause deformation of the feature of interest.
In addition, cleaving 7 in. away from a target area is a unique option not usually present when analyzing samples in a manufacturing environment.
 Figure 2. A manual cleave of vias in HOSP. Target was approximately 1 in. from initial scribe line. |
Figure 2 shows the effects that HOSP film stress can have on a via profile under more typical cleaving conditions. The bowl shape of the via is due to the arc shape in the z direction of the HOSP layer. The via is closer to the center on the bottom than it is on the top. No useful information can be obtained from this image due to the high level of distortion. The same complication can be seen in manual-only cleaving of samples containing lines and trenches.
Traditionally, when a material pulls and tears during cleaving, the edge of the sample can be enhanced with focused ion beam (FIB) milling. For instance, samples containing aluminum and copper can be prepared in this manner. HOSP has been found to be unstable under high-energy beams, though. The material shrinks when exposed to an electron beam with an energy level greater than 1kV. Single beam FIBs typically operate at potentials of 30kV. Under these harsh conditions, the material does not retain its original dimensions.
Figure 3 shows a sample prepared with FIB containing vias etched in HOSP. The FIB mill has smoothed the cross-section and made it impossible to differentiate between the layers of dielectric that exist in the sample. Measurements cannot be taken on this sample since the beam interactions have shrunk the HOSP layer. The FIB mill has also created an artifact ("trailing") that creates the illusion of the via's extending more deeply than it really does when the sample is viewed as a true 90° cross-section. Depositing a metal strap over the area prior to milling can usually minimize this phenomenon. The process of depositing the strap will distort the HOSP layer further, however, by causing it to shrink. It should be noted that further development using a dual beam FIB to minimize exposure of the sample to the high-energy ion beam might enhance results on FIB-milled cross-sections.
 Figure 4. A SELA MC500 cross-section of a sample containing vias in HOSP. |
Figure 4 reveals some deformation in the HOSP layer, but it is not enough to modify the profile of the via as seen in Fig. 2. These images show accurate profiles for the vias as well as stop etch —how deeply into the HOSP layer the etch has gone. From this image it is clear the etch went through the silicon nitride layer into the silicon oxide. It has been demonstrated that samples of this quality can be produced repeatably by microcleaving, minimizing the need to prepare additional samples because of deformation caused by the high internal stress of the HOSP film.
Conclusion
Automated microcleaving is becoming essential for failure analysis and process characterization. It is an important component of the effort to automate SEM sample preparation. Its accurate, artifact-free cross-sections, fast sample turnaround, and reduced operator skill requirements all contribute to an acceleration of the yield learning cycle in semiconductor manufacturing operations. We have found particular benefit in using automated microcleaving to prepare cross-sections of advanced materials, such as Low-k dielectrics. We believe that the automation of sample preparation can provide substantial productivity gains by reducing the time-to-yield and time-to-market for new processes and products.
Acknowledgments
The authors thank Alain Diebold, Roger Reyes, and Brendan Foran of International Sematech for contributions to this work. ACCUSPIN is a registered trademark of Allied Signal. Microcleaving is a trademark of SELA.
Reference
- Hydrido organo siloxane polymer is commercially available as ACCUSPIN T-24 from Allied Signal.
Fred Shaapur received a PhD in materials science from University of Southern California. He is a senior materials analyst at International Sematech Inc., 2706 Montopolis Dr., Austin, TX 78741; ph 512/356-7453, fax 512/356-7008, e-mail [email protected].
David Griffiths received his MS in physical chemistry from the University of Notre Dame. At International Sematech, he worked on process characterization and failure analysis. Griffiths is a failure analysis engineer at Infineon Technologies.
Efrat Raz holds a degree in business and management. She is general manager and VP of marketing at SELA, 1030B E. Duane Ave., Sunnyvale, CA 94086; ph 408/736-3700, fax 408/524-5439, e-mail [email protected].
Automated microcleaving
The operator can choose between different processes used to optimize parameters for particular film or wafer types. The microcleaving process offers a choice of cleaving modes:
- Quick cleave — an abbreviated process used for large targets.
- Single cleave — a 5-min process that cleaves a sample to 25mm accuracy, providing excellent edge quality with respect to the crystal plane.
- Submicron cleave — typically a 20-min automated process that can cleave a specific target with 0.5mm accuracy with excellent edge quality.
Once the desired mode is chosen, the system calculates all available cleaving solutions and displays the possible final cleave directions. The operator selects the final cleave direction from one of the recommendations presented by the system and can also select liquid nitrogen cooling to enhance cleave quality through soft materials.
In the single and quick cleave modes, the sample then undergoes a coarse cleave process and is ready for SEM imaging or any further prep, such as wet etches or metal sputter coating, etc. In the submicron cleaving process, the sample undergoes a series of coarse cleaves to an accuracy of ~30mm, ending with the target at a pre-set distance <100mm from the edge. The sample is then prepared by the system for the fine cleave operation and the operator is prompted to confirm the final target location. After the target is identified, the system aligns the fine diamond knife tip and scribes the segment edge. After the scribe is completed, the system induces a finely controlled shock wave from the segment edge opposite the target, resulting in an accurate, high-quality cleaved surface, with no risk of damage to the target by the cleaving tool. When liquid nitrogen cooling is used, even soft materials such as HOSP can be cleaved at a chosen location without damage.