Semiconductor Image Processing Using Exposure-Condition GAN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semiconductor image processing methods using Generative Adversarial Networks (GAN) struggle to generate reliable images without considering exposure conditions, leading to variations in semiconductor device quality during manufacturing.
Innovation Solution
A semiconductor image processing apparatus that includes an image input unit, an exposure condition input unit, a generator, and a discriminator, which perform a learning process to generate a second semiconductor image that resembles a third semiconductor image by considering exposure conditions, using a Conditional GAN technique like pix2pix to extract and adjust feature amounts, ensuring the generated image accurately represents the intended semiconductor device.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If GAN is used to generate semiconductor images from CAD drawings, then image generation capability is improved, but reliability of generated images deteriorates when exposure conditions are not considered
Solution Approach 1:
The patent applies parameter changes by incorporating exposure conditions (exposure amount, focus condition) as additional input parameters to the GAN model. This transforms the image generation process from considering only CAD drawing data to considering both CAD drawing data and exposure condition parameters, thereby improving the reliability of generated images to match actual captured images under various exposure conditions.
2Device complexity
If exposure conditions are not considered in image generation, then processing simplicity is improved, but manufacturing precision deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-collecting and storing actual captured images along with their corresponding exposure conditions and CAD drawing data. This pre-prepared dataset is then used to train the GAN model, enabling it to learn the relationship between CAD drawings, exposure conditions, and actual image outcomes. This preliminary preparation allows the model to generate reliable images without requiring complex real-time adjustments during the generation process.
3Measurement precision
If GAN training with actual captured images is performed, then image accuracy is improved, but learning time and computational resources increase
Solution Approach 1:
The patent applies partial action by using a discriminator that evaluates specific features of generated images against actual captured images, rather than requiring the entire image generation process to be perfectly accurate. The discriminator focuses on identifying whether a generated image matches the characteristics of an actual captured image under corresponding exposure conditions, allowing for efficient training with targeted feedback rather than exhaustive optimization of all image parameters.
Data Source
AI summary
A semiconductor image processing apparatus has an image input unit inputs a first semiconductor image, an exposure condition input unit configured to input exposure conditions, a generator performs a process of extracting a feature amount in consideration of the exposure conditions while reducing resolution of the first semiconductor image and thereafter use the extracted feature amount to increase the resolution to generate a second semiconductor image, and a discriminator configured to discriminate whether the input image is the second semiconductor image or a third semiconductor image provided in advance. The generator performs learning so that the discriminator erroneously discriminates the second semiconductor image as the third semiconductor image based on a result discriminated by the discriminator. The discriminator performs learning so as not to erroneously discriminate the second semiconductor image as the third semiconductor image, and not to erroneously discriminate the third semiconductor image as the second semiconductor image.