Statistical Leakage-Current Distribution Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting leakage-current variations in semiconductor devices are inaccurate due to the influence of process variations, leading to unrealistic worst-case and best-case estimates, which are not helpful for designers aiming for low power consumption in applications like mobile phones.
Innovation Solution
A method that generates statistical data for representative chip-unit models using Monte-Carlo simulations and statistical parameter estimation to derive distribution parameters from skew-normal or generalized extreme value distributions, providing more accurate leakage-current variation estimates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If worst/best-case corner estimates are used to predict leakage-current variations, then the prediction range covers all possible variations, but the estimates become unrealistic and pessimistic with spans exceeding one order of magnitude
Solution Approach 1:
The patent transforms the statistical modeling approach by changing from traditional log-normal distribution parameters to parameters based on generalized extreme value distributions and skew-normal distributions. This parameter transformation enables the model to capture the true variation behavior of leakage currents while providing realistic estimates that match extensive simulations and detailed leakage measurements.
2Device complexity
If Log-normal distributions are used to predict leakage-current variations, then the modeling approach is simple, but the data does not cover the full range of applied voltages and temperatures and yields inaccurate variation ranges
Solution Approach 1:
The patent replaces the log-normal distribution with generalized extreme value distributions and skew-normal distributions. This distribution transformation allows the model to accurately represent leakage-current variations across the full range of voltages and temperatures while maintaining computational tractability.
Solution Approach 2:
The patent extends the modeling approach to account for multiple dimensions of operating conditions (voltages and temperatures) simultaneously. By using distributions that can capture the full range of variations across these dimensions, the model provides accurate predictions without requiring separate models for each condition.
3Measurement precision
If extensive simulations or detailed leakage measurements are performed to determine realistic distribution width, then accurate variation ranges are obtained, but the process becomes time-consuming and resource-intensive
Solution Approach 1:
The patent performs preliminary statistical analysis to identify the appropriate distribution types (generalized extreme value and skew-normal) that best fit leakage-current behavior. Once the distribution form is established, the model can predict variation ranges directly from distribution parameters without requiring extensive additional simulations or measurements for each design iteration.
Solution Approach 2:
The patent creates a statistical model that copies the essential variation behavior observed in extensive simulations and measurements. This modeled distribution can then be used to predict leakage-current variations for different designs without repeating the extensive simulation campaigns, significantly reducing the time and resources required.
Data Source
AI summary
Disclosed is a method and system for modeling statistical leakage current distribution using logarithmic skew-normal distribution by generating statistical data with a statistical analysis method based on Monte-Carlo simulations or based on a pre-characterization response modeling step for a plurality of representative chip-unit models, deriving a plurality of parameters from said statistical data based on a specific class of statistical distributions, scaling said parameters to values used on realistic chip level, and generating leakage-current variation estimates based on said parameters.