Abstract:
A mixer (1) capable of uniformly mixing gases with different weights, a thin−film manufacturing device with the mixer (1), and a thin−film manufacturing method using film forming gas formed by mixing the gases in the mixer, the mixer (1) comprising a mixing chamber (2) having the gas inlets (5a, 6a) of two gas inlet pipes (5, 6) positioned so as to be opposed to each other and a mixing gas diffusion chamber (3), wherein a partition plate (4) is installed between the mixing chamber and the diffusion chamber so that the volume of the diffusion chamber is larger than that of the mixing chamber, and a gas outlet (7) is provided in the partition plate at a specified position on the vertical lower side of a straight line connecting two gas inlets to each other; the thin−film manufacturing device wherein a cylindrical sleeve member (61) is fitted to the side wall of a stage (53) inside a vacuum reaction chamber (2), and the height of the stage is set at a position where the volume of a second space on the lower side of the stage connected to a vacuum exhaust means is larger than that of a first space so that exhaust gas can be isotropically exhausted through a clearance between a sleeve member and the inside wall surface of the reaction chamber without causing convection in the first space between a gas head and the stage, whereby the gas flow of the mixing gas can be uniformized by suppressing the occurrence of turbulent flow, convection, and thermal convection.
Abstract:
A three-dimensional shape aligning method involves a feature that a tangent plane is used as an image plane and a local distance image described on a log-polar coordinate system is used. A created image is converted into a power spectrum by Fourier series expansion and the indefiniteness of the angle axis around the normal is normalized to convert it into a quantity invariable to rotation. The power spectrum is expanded in an eigen space by using an eigen vector to compress the dimension. The dimension-compressed space is searched for a correspondence point by nearest neighbor to determine a point correspondence relation. An erroneous correspondence, if any, is eliminated by verification to determine the position relation between distance images. Verification by cross-correlation and RANSAC is carried out to obtain reliable correspondence relations by screening, and a tree structure representing a coherence relation of the distance images is created. The result of this position aligning is used as the initial value, and simultaneous alignment technique is applied to the distance images of tree structure, thereby making a shape model.