g-Hung Chen

Influence of data volume and EPC on process window in massively parallel e-beam direct write

Multiple e-beam direct write lithography (MEBDW), using >10,000 e-beams writing in parallel, proposed by MAPPER, KLA-Tencor, and IMS is a potential solution for 20-nm half-pitch and beyond. The raster scan in MEBDW makes bitmap its data format. Data handling becomes indispensable since bitmap needs a huge data volume due to the fine pixel size to keep the CD accuracy after e-beam proximity correction (EPC). In fact, in 10,000-beam MEBDW, for a 10 WPH tool of 1-nm pixel size and 1-bit gray level, the aggregated data transmission rate would be up to 1963 Tera bits per second (bps), requiring 19,630 fibers transmitting 10 Gbps in each fiber. The data rate per beam would be <20 Gbps. Hence data reduction using bigger pixel size, fewer grey levels to achieve sub-nm EPC accuracy, and data truncation have been extensively studied. In this paper, process window assessment through Exposure-Defocus (E-D) Forest to quantitatively characterize the data truncation before and after EPC is reported. REBL electron optics, electron scattering in resist, and resist acid diffusion are considered, to construct the E-D Forest and to analyze the imaging performance of the most representative layers and patterns, such as critical line/space and hole layers with minimum pitch, cutting layers, and implant layers, for the 10-nm, and 7-nm nodes.

Contour-based kernel modeling and verification for E-Beam lithography

In E-beam lithography, the double or multiple Gaussian kernels used to describe the electron scattering behavior have been discussed extensively for critical dimensions (CDs) larger than the e-beam blur size. However in e-beam direct write on wafer, CD dimensions are close to the beam blur size because of requirements in both resolution and throughput. This situation gives rise to a severe iso-dense CD bias. Hence the accuracy of the modeling kernel is required to achieve a larger common process window. In this paper we present contour-based kernel modeling and verification for e-beam lithography. The edge contours of CD-SEM images of the contact hole array pattern with duty ratio splits are used in this Gaussian kernel modeling study. A 2-step optimization sequence is proposed to improve the fitting efficiency and robustness. In the first step, roundness is the primary and the most effective index at the corner region which is sensitive to determine the beam blur size. The next step is to minimize the deviation of the through-pitch proximity effect by adjusting the ratio of the electron backscattering to the electron forward scattering. The more accurate cost index, edge placement error, is applied in the subsequent optimization step with constrained beam blur sizes extracted from the previous step. The optimum modeling kernel parameters can be obtained by the lowest cost deviation of the simulation contours and the CD-SEM extracted edge contours after optimization iterations. For early study of the proximity impact on future EBDW systems, the exposure experiment is performed on an EBM-8000 mask writer to build the modeling kernel. The prediction accuracy of the optimum modeling kernel on 60-nm features with different pattern densities is also verified experimentally to be within 1.5 nm.

Impact of proximity model inaccuracy on patterning in electron beam lithography

Electron beam lithography is a promising technology for next generation lithography. Compared to optical lithography, it has better pattern fidelity and larger process window. However, the proximity effect caused by the electron forward scattering and backscattering in the resist and the underlying substrate materials has a severe influence on the pattern fidelity when the required critical dimensions (CD) are comparable to the electron beam blur size. Therefore, an accurate electron scattering model and a proper proximity correction play a vital role in electron beam lithography. In this paper, we describe the model accuracy of electron scattering in terms of multiple Gaussian kernels with an in-house proximity error correction to reduce proximity error with much better accuracy and more self-consistency than the double Gaussian kernel on the 100-keV electron energies. The impact of various Gaussian kernels used in the proximity correction on the lineation of typical patterns is also addressed.