Chip Design Optimization via Profit-Based Yield Curves
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional chip design and fabrication methods face challenges in optimizing yield as they often require trade-offs between different metrics like performance and power consumption, leading to cost-inefficient design changes that may reduce yield in one metric while improving another, and fail to maximize profit potential.
Innovation Solution
A computer-implemented method that calculates separate joint probability distributions for various metrics, generates discrete yield curves, and applies a profit function to combine these into a profit-based yield curve, allowing for design optimizations that balance multiple metrics to maximize profit potential.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If design changes are made to optimize yield as a function of one metric (e.g., performance), then yield for that metric is improved, but yield for other metrics (e.g., power consumption) may be reduced
Solution Approach 1:
The patent combines multiple separate yield curves (each representing a different metric such as performance, power consumption, area) into a single profit-based yield curve. This merging allows simultaneous optimization across all metrics by integrating them into one unified optimization framework, resolving the contradiction between optimizing for one metric versus another.
Solution Approach 2:
The patent transforms multiple metric parameters into a common profit denominator through a profit function. By changing the parameter representation from separate metric values to unified profit values, the system enables simultaneous optimization of multiple metrics without the trade-offs that occur when optimizing each metric separately.
2Reliability
If design changes are made to optimize yield as a function of any metric, then yield is improved, but the design changes may not be cost-efficient
Solution Approach 1:
The patent introduces a profit function that converts multiple metric values into a common profit denominator. This parameter transformation enables the evaluation of design changes in terms of their impact on profit rather than just yield, allowing for cost-efficient optimization by considering the profitability implications of each design change.
Solution Approach 2:
The profit function serves as an intermediary that bridges the gap between yield optimization and cost-efficiency. It translates diverse metric outcomes into a unified profit metric, enabling designers to evaluate whether yield-improving design changes are also cost-effective, thus resolving the contradiction between yield improvement and cost-efficiency.
3Reliability
If separate yield curves are optimized independently for different metrics, then each metric's yield is maximized, but the overall profit potential is not maximized
Solution Approach 1:
The patent merges multiple independently optimized yield curves into a single profit-based yield curve. This combination allows the system to capture the interactions between different metrics and identify design changes that simultaneously improve yield across multiple metrics while maximizing overall profit potential, rather than optimizing each metric in isolation.
Solution Approach 2:
The profit function provides a universal framework that can evaluate and combine the effects of design changes across multiple different metrics. This multi-functional approach allows a single optimization process to consider performance, power consumption, area, and other metrics simultaneously, maximizing overall profit potential rather than just individual metric yields.
Data Source
AI summary
Disclosed is a computer-implemented method for designing a chip to optimize yielding parts in different bins as a function of multiple diverse metrics and further to maximize the profit potential of the resulting chip bins. The method separately calculates joint probability distributions (JPD), each JPD being a function of a different metric (e.g., performance, power consumption, etc.). Based on the JPDs, corresponding yield curves are generated. A profit function then reduces the values of all of these metrics (e.g., performance values, power consumption values, etc.) to a common profit denominator (e.g., to monetary values indicating profit that may be associated with a given metric value). The profit function and, more particularly, the monetary values can be used to combine the various yield curves into a combined profit-based yield curve from which a profit model can be generated. Based on this profit model, changes to the chip design can be made in order to optimize yield as a function of all of the diverse metrics (e.g., performance, power consumption, etc.) and further to maximize the profit potential of the resulting chips.