Managing across-wafer process variation
07/01/2008
 |
Alexander Starikov, senior technical contributor, Intel, Santa Clara, CA USA
Step-and-scan and resolution enhancements in optical lithography have enabled printing much smaller usable features. Further reduction is expected in pitch doubling. But with maximum achievable device yield falling industry-wide, there are concerns about patterning marginalities across the lithographic process window [1].
The lithography process window is shrinking to 5% for exposure dose and 120nm for focus. If one-fourth is allocated to across-wafer components, these variations of dose and focus must be controlled to 1.2% and 30nm, respectively. Like registration, dose and focus in step-and-scan are controlled by stage servo loops, which allows for the high-quality patterning of whole wafers using an imaging slit. But, when servo loops are inaccurate, systematic across-wafer variations can occur on 1mm lateral scale.
It is impractical to sample fast systematic across-wafer variations using conventional metrology tools/techniques, but if left unchecked they lead to losses in clock speed and yield (Fig. 1a). We do not have automated inspection of across-wafer dose and focus variation. Yet, with some modifications, optical lithography’s oldest tricks enable valuable insights.
 a) map of yield loss and b) a grating test wafer |
Figure 1b shows a grating test wafer that can provide significant lithographic variability information. A grating reticle was exposed in positive resist exactly as for product wafers, but at a dose high enough for resist to liftoff when defocused. One can observe the same gross failure areas as in Fig. 1a. If we change defocus across the slit by tilting the wafer, segments of resist bands in the direction of scanning will be all that remain. If the segments are out of line, this indicates the magnitude and direction of defocus [2]. Using this technique, the “naked eye” can detect and measure focus variations to <50nm, with the potential to detect down to 1nm. For example, we confirmed the root cause of the variability seen in Fig. 1b: across-wafer focus variation ~100nm through scan, up to 200nm on the interior ends of partial fields, and 600nm near the wafer edge.
For dose variability, we can use a classic “sub-develop” or “sub-E0” test, where a clear reticle is printed at a dose below that required for resist to completely develop (E0). The thickness of resist remaining is modulated by the effective dose, such that light interference in thin resist film creates “Newton’s fringes,” which can be seen by the naked eye. This simple technique reveals effective dose variation to <0.5%, with the potential to resolve 0.01%.
Using simple tests and manual inspection, we have observed across-wafer dose and focus variation in all scanner makes and models. It only takes minutes to localize an issue and to attribute a root cause, or to confirm an effective fix. For the case illustrated in Fig. 1, a single action markedly reduced across-wafer focus variation in all tools used at critical layers. Much higher yield was achieved immediately, without recurring cost.
There are a plethora of systematic across-wafer process variations, but only the most stable and pervasive are detected as gross failure areas in yield maps. Starting from yield data, the vast majority of process variations are hard to identify and control.
What the industry needs is automated inspection of across-wafer lithographic process variation. With capabilities of at least 0.1% for dose and 1nm for focus, this would effectively be a brand-new application and would provide early detection of across-wafer process variations. Once these variations are measured, they can be corrected in the servo loop of a step-and-scan lithography system.
References
- C.N. Berglund, “Trends in systematic non-particle yield loss mechanisms and the implication for IC design,” Proc. SPIE, Vol. 5040; 2003.
- US Patent 7289198.
Alexander Starikov is sr. technical contributor at Intel; 2200 Mission College Blvd., Santa Clara, CA, USA 95054; ph 408/765-7398; e-mail [email protected].