BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20101117T202400Z
DTEND:20101117T204200Z
LOCATION:399
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: We develop a parallel branch-and-bound method for a nonconvex=0Aoptimization problem arising in semiconductor lithography. After=0Adescribing the lithography printing problem and its physical=0Abackground, we motivate the posing of the optimization problem as a=0Asearch over the surface of a sphere in high dimension. Previous work=0Aon this problem and similar challenges is surveyed. We explore several=0Anaive methods for parallel optimization as well as a mixed-integer=0Aquadratic program. As an original contribution, we develop a serial=0Abranch-and- bound implementation that achieves five orders of=0Amagnitude performance improvement over commercial solvers. We then=0Adescribe an implementation of a parallel branch-and-bound method using=0Anovel bounding and branching strategies and provide computational=0Aresults on the IBM BlueGene/P supercomputer architecture, showing=0Astrong scaling to 80% efficiency on up to 4096 processors.
SUMMARY:Parallel Strategies for Nonlinear Mask Optimization in Semiconductor Lithography
PRIORITY:3
END:VEVENT
END:VCALENDAR