At LithoVision, Dr. John Sturtevant, Director of RET Product Development at Mentor Graphics, highlighted the importance of predicting and controlling total edge placement error (EPE). Although the industry has always paid attention to EPE, he warned that there are some new realities and vanishingly small budgets, as well as additional considerations that aren’t really being discussed. Sturtevant explained that although scaling was mainly resolution-limited at previous nodes, edge placement is the limiting factor now. He emphasized the need to reduce the effective k1 factor using multiple patterning in conjunction with minimizing contributions to total EPE (Figure 1A).
Figure 1A. Sturtevant reported the need for multiple patterning combined with EPE reduction to lower the k1eff (left image). Figure 1B. The inter-layer relative EPE (rEPE) is critical to yield.
Inter-layer design rules are set by the minimum physical-electrical requirements and assumptions of in-fab distributions of critical dimension (CD) and misalignment between two layers. In Simulation, EPE equals the simulated contour minus the design target. The Fab EPE isn’t directly measurable, instead only the inter-layer relative EPE (rEPE) can be measured, and rEPE is critical to yield (Figure 1B).
There are a multitude of contributors to process proximity effects including the mask, exposure, post-expose bake and develop, etc. Sturtevant highlighted the suite of specialized Calibre Physical Models to address those effects. He reported that by the year 2020, model accuracy RMS error requirements will be approaching 250 picometers (1σ), and compared that to the silicon to silicon bond distance is a mere 230 pm (Figure 2A).
Figure 2A. Sturtevant emphasized the need for on-going improvements in model accuracy (left image). Figure 2B. Metrology modeling effects are significant and cannot be ignored for 7 nm manufacturing and beyond.
Sturtevant emphasized that metrology modeling is increasingly important. CD SEM metrology effects are becoming an increasingly large percentage of the CDs, and he warned they can’t be ignored for the 7 nm node and beyond (Figure 2B). The resist formulation’s base polymer type affects the amount of shrink, while SEM parameters such as higher voltage/frames increase SEM shrink as well. In addition, the thinner EUV resists also contribute to SEM shrink. Sturtevant showed that in the case of EUV lithography using ~20 nm minimum dense width and space features, CD-SEM shrinkage of up to 5 nm is possible, a huge percentage of the linewidth.
Figure 3A. Sturtevant discussed the increasing challenges and importance of accurate mask representation (left image). Figure 3B.Newly developed modelforms are yielding substantial accuracy improvements.
Accurate photomask representation is vital to the overall imaging. In the past few years we have gone from simply worrying about the dimension of the features on the mask, to concern about their shape in two dimensions, and now with ultra-low k1 lithography, we have to take into account the full three dimensional form of these features–as well as the optical properties of the materials used as absorbers (Figure 3A). Sturtevant also detailed recent work with 4th generation negative tone develop (NTD) modelforms for best focus imaging of holes/spaces. He described newly developed modelforms for mechanisms including resist de-protection shrinkage, different develop Rmin/Rmax, and different optical absorptions. He highlighted significant variation amongst the different materials/processes, and reported 50% accuracy improvement with the 4th generation models compared to previous ones (Figure 3B). Sturtevant explained that it is no longer possible to simply take ‘snapshots’ when modelling processes. He demonstrated how all of the factors discussed previously including mask effects, resist effects, and metrology must be added to very accurate image modelling in order to successfully investigate and characterize EPE in multi-patterning usage and to help establish realistic process controls.
A comparison of EUV versus DUV modeling showed substantial differences. Flare, mask 3D effects, aberrations, through slit effects, and field edge effects all were areas of heightened concern with EUV (Figure 4A). Sturtevant described a possible paradigm shift for EUV hidden in the noise. He detailed a complex impact of aberrations on imaging including CD and image shift behavior, and said that in principle that could be compensated with OPC. Industry practice has always been for a single model to be used at all field locations for all tools in manufacturing, whereas EUV’s behavior will necessitate different models across the reticle field. The differing aberrations on each EUVL tool being significant enough that–at the minimum, EUV tool-slit position specific models and verification are likely to be required. He also said the need for EUV tool-specific sources, OPC models, and masks should be considered and warned of potential cost implications (Figure 4B).
Figure 4A. Stark differences were noted in a comparison between EUV and DUV modeling (left image). Figure 4B. EUV may drive a paradigm shift with added verifications and different models, etc.
A global litho model contains information about how models vary within the full field. In the case of EUV, across slit and across tool contours differ significantly, which may erode some of the EUV advantages over 193 immersion multiple patterning (Figure 5A).
Figure 5A. EUV across slit/across tool behavior may erode some of its advantage over 193i MP (left image). Figure 5B. Sturtevant summarized the critical factors for rEPE control in 2020.
Sturtevant summarized his informative presentation by reminding the audience of the critical success factors in relative EPE control working towards the year 2020 (Figure 5B). He stressed the need to drive model accuracy down to 250 pm RMS, as well as the importance of SMO and OPC to optimize common depth of focus CD control and image placement. In closing, he also reminded the audience of the marked improvements needed in EUV across slit tool-tool aberration matching. Otherwise, the industry will require tool-specific full-chip verification, and perhaps even tool-specific models and masks.