With the continuous shrinking of minimum feature sizes beyond 193nm wavelength in optical lithography, more and more computationally expensive algorithms are being developed in the field of Optical Proximity Correction (OPC) to improve pattern fidelity and robustness against process variations. Lithography simulation time and image accuracy are proportional to the number of kernels by which the mask is convoluted to generate the intensity map for each OPC iteration. Typically, there is a trade-off between the accuracy of intensity map and computational time which can be minimized by using only one kernel. Nevertheless, the intensity of each pixel tends to be smaller than its actual value and is not accurate enough resulting in intensity error. However, with considering relaxed Edge Placement Error (EPE) conditions, we observed that the error of pixel intensity is not changed much even if the mask is slightly updated. Therefore, in this paper, we exploit this observation to relax the intensity error by constructing intensity difference map in which the differences between one kernel and multiple kernels intensity maps are stored. For each OPC iteration, one kernel is used to generate intensity map, to which the intensity difference map is added to improve its accuracy. Our experimental results show that the proposed algorithm generates mask solutions within a short computational time with almost the same EPE and process variability band obtained using multiple kernels during optimization. 
With the continuous shrinking of minimum feature sizes beyond 193nm wavelength in optical lithography, more and more computationally expensive algorithms are being developed in the field of Optical Proximity Correction (OPC) to improve pattern fidelity and robustness against process variations. Lithography simulation time and image accuracy are proportional to the number of kernels by which the mask is convoluted to generate the intensity map for each OPC iteration. Typically, there is a trade-off between the accuracy of intensity map and computational time which can be minimized by using only one kernel. Nevertheless, the intensity of each pixel tends to be smaller than its actual value and is not accurate enough resulting in intensity error. However, with considering relaxed Edge Placement Error (EPE) conditions, we observed that the error of pixel intensity is not changed much even if the mask is slightly updated. Therefore, in this paper, we exploit this observation to relax the intensity error by constructing intensity difference map in which the differences between one kernel and multiple kernels intensity maps are stored. For each OPC iteration, one kernel is used to generate intensity map, to which the intensity difference map is added to improve its accuracy. Our experimental results show that the proposed algorithm generates mask solutions within a short computational time with almost the same EPE and process variability band obtained using multiple kernels during optimization.

Conference Title

DA Symposium 2014, IPSJ Symposium Series, Vol. 2014

Conference Country

Japan

Conference Date

Aug. 29, 2014 - Aug. 30, 2014

Conference Sponsor

IPSJ