SEMATECH benchmarks OPC software, mask-writers
09/01/1997
Sematech benchmarks OPC software, mask-writers
One of the highlights of the KLA-Tencor Yield Management Seminar held during SEMICON/West was a summary of the SEMATECH J111 project, presented by Hua Zhang of Rockwell Semiconductor Systems. The goal of the J111 project was to benchmark current Optical Proximity Correction (OPC) software as well as the capability of different mask-writing tools. Five different developers of OPC software were asked to apply corrections of five different levels of complexity to the same test pattern. The five converted test patterns from each of the five suppliers were then combined into a single job deck and fabricated using six different mask writers on 12 different reticle plates. The quality of the fabricated masks was judged by measuring the "pullback" of line ends and the "deviant area" of a critical feature on each plate and by counting the number of defects using die-to-die inspection. The reticles could be inspected successfully with current technology. The number of defects in the six test patterns found in die-to-die inspections varied from 29 to 546 depending on the OPC supplier.
OPC corrections were made (and masks generated) for two different plan-of-record resist processes, one for i-line and one for DUV. To judge the effect of the OPC on the wafer pattern, critical dimensions and deviant areas were measured on wafers after etch for each of the two processes and each of the fabricated masks. The figure shows typical mask and resist patterns. The nominal uncorrected mask pattern is at the top of the left column, with four different OPC technologies below. The right-hand column shows the resulting etched patterns. Some of the OPC methods improved pattern fidelity, while others did not. Combining the best OPC technology with the most-capable mask-writer produced significant improvements in pattern fidelity and CD control. The "deviant areas" in two-dimensional patterns could be reduced by less than 50%, and the pull-back of line ends could be eliminated. More aggressive OPC methods did not necessarily produce the best results; rather there was an optimum degree of correction for each of the suppliers` methods, which depended on the mask-writing technology.
One surprising result was that the most sophisticated mask writing technology (a MEBES 4500 with GHOST e-beam proximity correction) did not produce dramatically better mask or wafer results. Laser mask writers (CORE 2564 and ALTA) performed as capably as the MEBES 4500 e-beam system in these tests. The dramatic differences in the performances of the different OPC schemes was perhaps less surprising. OPC technology has not been standardized, and the J111 project was the first attempt at benchmarking. One must expect the results obtained to be highly process-dependent, and fitting the correct OPC parameters to an unfamiliar process will continue to be an art. Once a process is matched to an OPC scheme, the process and OPC can evolve together more stably. - M.D.L.
 |
Mask patterns (left) and etched wafer patterns (right) corresponding to different OPC schemes (a-d) The uncorrected pattern is at the top.