CPU Turbo Yield Analysis Using Explainable AI and WAT Factors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for enhancing Central Processing Unit (CPU) turbo yield, such as using the Pearson coefficient square, often result in low R-square values and fail to identify WAT items that genuinely influence yield, leading to inefficient improvements or even decreases in performance due to power consumption limits.
Innovation Solution
Employing an Explainable Artificial Intelligence (XAI) framework that integrates a training model, such as XGBoost, and a WAT importance analysis module, such as SHAP, to accurately predict CPU turbo yield and quantify the contribution of each WAT parameter, enabling targeted adjustments to enhance performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional Pearson coefficient square method is used to identify WAT items, then the analysis process is simple, but the association between selected WAT items and CPU turbo yield is weak resulting in low R-square values
Solution Approach 1:
The patent replaces the conventional Pearson coefficient square statistical method with an Explainable AI framework combining gradient boosting models and SHAP analysis. This substitution transforms the analytical approach from simple correlation calculation to a sophisticated machine learning system that can capture non-linear relationships and interactions among WAT items, thereby significantly improving the association strength and R-square values while maintaining interpretability through SHAP explanations.
2Ease of operation
If on-current of WAT items is increased based on conventional analysis, then the analysis is straightforward, but turbo yield may decrease due to power consumption limits
Solution Approach 1:
The patent implements a feedback mechanism where the XAI model continuously analyzes the impact of WAT item modifications on both turbo yield and power consumption. The SHAP explanations provide actionable insights that guide modification directions, and the model evaluates the expected outcomes before implementation. This closed-loop feedback system ensures that modifications improve turbo yield while staying within power consumption constraints, avoiding the pitfalls of conventional straightforward but potentially harmful adjustments.
3Loss of time
If conventional methods are used to find influential WAT items, then the process is quick, but genuine contributors to CPU turbo yield may be missed
Solution Approach 1:
The patent employs preliminary action by using the gradient boosting model to pre-process and rank WAT items based on their predicted importance to turbo yield before detailed analysis. The SHAP analysis then provides preliminary explanations for top candidates, allowing engineers to focus on the most promising items first. This two-stage preliminary approach efficiently identifies genuine contributors without exhaustive analysis of all WAT items, balancing speed and accuracy.
Data Source
AI summary
A central processing unit (CPU) turbo yield boosting method includes acquiring a plurality of Wafer Acceptance Test (WAT) parameters from a foundry, acquiring a CPU turbo yield from a final test stage, generating a predicted CPU turbo yield for the plurality of WAT parameters using a training model based on the plurality of WAT parameters and the CPU turbo yield, and generating a plurality of importance values corresponding to the plurality of WAT parameters using a WAT importance analysis module based on the predicted CPU turbo yield and the plurality of WAT parameters. The training model and the WAT importance analysis module are integrated into an explainable Artificial Intelligence (AI) framework.