Semiconductor manufacturing defect dynamic random sampling method using AI model

Abstract

The invention discloses a semiconductor manufacturing defect dynamic random sampling method using an AI model, and relates to the technical field of semiconductor defect random sampling. The method comprises the following steps: collecting the information of a target product; carrying out AI modeling on the information of the target product to generate simulation AI model training-1; collecting main defect problems and associated data of the target product; carrying out AI model training establishment of defect root causes according to the collected data to generate AI model training-2; deploying the AI model training-1 and the AI model training-2 in an AI model operation platform for operation; establishing butt joint between the AI model operation platform and an existing sampling system; and executing a sampling result in the sampling system. The method has the effect of high efficiency.

Description

technical field [0001] The invention relates to the technical field of random sampling of semiconductor defects, more specifically, it relates to a method for dynamic random sampling of semiconductor manufacturing defects using an AI model. Background technique [0002] In the semiconductor manufacturing process, there are certain defects on the semiconductor. Defects are the most important reason for the reduction of chip manufacturing yield and reliability, so the management and control of defects is particularly important in the semiconductor manufacturing process. At present, the main control methods for defects are: 1) In the manufacturing process, select sampling inspection sites based on experience; 2) Set corresponding static sampling rules for each inspection site through the experience of engineering personnel, that is, the mantissa of the product batch number The number is used as the basis for sampling or not. [0003] However, because the current method is a s...

AI Technical Summary

Problems solved by technology

[0003] However, because the current method is a static sampling method and relies heavily on the experience of engineers, it has a relatively large deficiency in the coverage of defect problems, and has a relatively large waste of limited detection resources.

For the selected inspection site, the selection basis is basically the copy of the previous generation of products, and there is no research on the defect distribution characteristics of the product itself, so it is easy to ignore the uniqueness of the product in the early stage; for static sampling rules , because no other factors are considered, only random sampling is performed, so only systematic and common defects can be found, and the detection rate of accidental time due to production equipment, personnel errors, etc. is extremely low

Images