Neural network training method in hyperbolic wavelet transformation space for recognition of photomasks

Abstract

During deployment of existing hardware-software complexes for controlling of photomasks quality in the conditions of small production lots of microelectronics leads to unjustified quality cost increase. In case of lowering of requirements to the software of such complexes, noise on the photomasks images inevitably appear, which leads to the decrease of credibillity of recognition of reference signs, which are located on the images, and to the decrease of credibillity of control results. Therefore, with the purpose of increase of credibillity of recognition it is proposed to execute recognition of reference sings on the bases of nose-eliminating of neural network training in the space of hyperbolic wavelet transformation.