Semiconductor Yield Prediction Using Covariance Matrix

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

It is challenging to accurately estimate the fail rate of semiconductor devices during mass production due to limited experimental data, making it difficult to predict the yield and characteristics of devices to be produced.

Innovation Solution

A method and apparatus that collect data from both mass-produced semiconductor devices and experimental samples to create a covariance matrix and mean vector, allowing for the prediction of characteristics and yield of future semiconductor devices by using a measurer, calculator, and random number generator to generate prediction data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If experimental samples are used to verify performance, then device characteristics can be tested, but the number of experimental data is limited making it difficult to estimate fail rate accurately

Engineering Contradiction:

Improvefail rate estimation accuracyVSAvoidnumber of experimental data

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent applies preliminary action by collecting data from mass-produced devices before mass production begins. The system gathers first data from a plurality of mass-produced semiconductor devices and combines it with second data from experimental samples to build a covariance matrix and mean vector, enabling accurate fail rate prediction before full-scale production starts

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent merges two different data sources: first data from mass-produced semiconductor devices and second data from experimental samples. By combining these datasets, the system creates a comprehensive statistical model (covariance matrix and mean vector) that leverages both the volume of mass production data and the quality of experimental validation data to accurately predict fail rates

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS11152235B2Apparatus and method for manufacture of semiconductor devices

Publication Date: 2021.10.19 SAMSUNG ELECTRONICS CO LTD

AI summary

A method for predicting characteristics of semiconductor devices includes collecting first data for a plurality of first characteristics from first semiconductor devices already in mass production, and collecting second data for the first characteristics and third data for a plurality of second characteristics from at least one second semiconductor device manufactured as an experimental sample before beginning the mass production. A covariance matrix is then obtained based on the first, second, and third data, and a mean vector for third semiconductor devices to be in the mass production is determined. Prediction data for third semiconductor devices is then generated based on the covariance matrix and the mean vector.