Dr. Ady Levy, KLA-Tencor Fellow, showed that variations in non-lithographic steps play a substantial role in CD and overlay errors. For instance, the mask writer induces registration, CDU, transmission, and optical proximity compensation (OPC) errors, whereas the etcher impacts line and contact profile, and spacer metrics like height, tilt, and asymmetry. Each technology node has increasingly aggressive overlay and CD budget requirements, and although nanometer level control used to be sufficient, Levy cautioned that “every picometer (pm) counts.” (Figure 1A ) This makes feedback and feed-forward control capabilities crucial, where feed-forward control measures variation at the source of the error and feedback control samples process changes and sends corrections back to the source.
Levy reported that mask registration feed-forward capabilities can improve process control substantially and KLA-Tencor has developed new tools for doing that. One challenge, Levy reported, is that mask registration is feature dependent, and random logic patterns have coverage feature dependencies as well as density dependencies, meaning that measurement of standard alignment marks does not necessarily capture all the registration errors. To address these challenges, the next-generation KLA-Tencor LMS IPRO mask registration metrology tool compares measurements to high accuracy imaging models to construct a mask error fingerprint.
Levy shared case study data for a 2x nm mask set  demonstrating that SRAM and logic cells’ registration signatures are not captured effectively using alignment marks on the mask set (Figure 1B). He reported that feeding forward the device mask registration could improve yield, and showed a 25% reduction in mask overlay contribution is possible using SRAM registration data as compared to overlay mark data (Figure 2A).
Another important overlay contributor comes from errors due to the wafer shape, especially since changes are induced moving through the manufacturing line, Levy explained. To compensate for this, first the KLA-Tencor WaferSight system characterizes process wafer geometry, then a finite element model is able to simulate the geometry impact on defocus and overlay (Figure 2B). Levy revealed that this information could be used for shape-optimized corrections on the scanner to fine-tune exposure parameters for particular wafer geometries.
|Figure 2A. Levy reported that feeding forward the device mask registration could improve yield substantially (left image). Figure 2B. A finite element model is able to simulate the geometry impact on defocus and overlay.|
In his summary, Levy stressed that optimum patterning control is attained with a combination of feedback and feed-forward control loops, and KLA-Tencor has developed new technologies that enable two sources of input for data feed-forward capabilities. He announced that when employed together, these control loops can enable wafer fabs to meet tightened process windows using current-generation process tools.
Tomoyuki Matsuyama, Nikon Strategic Imaging Solutions Section Manager, delivered a complementary presentation describing Design-for-Yield solutions that enable peak tool performance in manufacturing. In his introduction, Matsuyama described the Design-for-Yield process. He explained that scanner-aware process development takes into account scanner-specific information like lens aberrations and Jones Pupils as well as illuminator pupilgrams when constructing EDA solutions like OPC and source mask optimization (SMO) as part of the design-for-manufacturing (DFM) process. He reported that scanner-aware optimization provides the best real-world patterning solution. Using an analogy of the way an orchestra is directed by the maestro to ensure all of the musicians are in harmony, Matsuyama explained that Nikon provides a number of “Masters” – automated software solutions that ensure the scanner is performing at its best. These include LNS (lens) Master, OPE Master, CDU Master, and OVL (overlay) Master (Figure 3A).
LNS Master enables reticle-specific thermal compensation on the scanner. It sends GDS or OASIS mask design information through a fast Fourier transform (FFT) to calculate its effect within the lens, and then feeds that into the LNS Master ThAO system along with pupilgram data. Lens control parameters are calculated and subsequently output to the scanner so that thermal compensation is ready when the reticle is loaded.
OPE Master uses customer test-pattern data and scanner adjustments to provide illumination condition matching for aligning performance across a fleet of scanners and ensuring that one OPC solution works on all of them.
CDU Master provides optimization capabilities that enable the scanner to correct for other process window detractors. It provides high-order dose and focus adjustments to reduce residual CD errors both across the shot and across the wafer-from sources including mask, etch, and the litho process. Matsuyama revealed that CDU Master is used at several customer sites around the world. In one case, CDU 3σ was reduced from 2.3 nm to 1.5 nm after compensation (Figure 3B). In a separate application intra-shot correction was applied to a customer production layer and the 3σ CD error was improved by 20%. Overlay matching plays a central role in multiple patterning applications, and OVL Master enables automated grid and distortion matching, as well as automated reticle expansion correction to maximize yield.
Finally these components are integrated together by the Plug and Play Manager, which can use the various Masters’ software capabilities in conjunction with recipe, mask, and wafer data to deliver optimized scanner exposure parameters that enhance yield on product wafers (Figure 4A).
In closing, Matsuyama informed the audience that Nikon provides an Open Innovation platform for future lithography. The Nikon Design-for-Yield program will include open-source collaboration with EDA vendors as well as metrology systems, tracks, and more to provide optimal production results and flexibility for customers (Figure 4B).