Circuit IR Drop Detection Using Neural Network Power Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for measuring IR drop in circuit design are inefficient and inaccurate, often missing significant cases of IR drop, which affects the timing and functionality of the circuit.
Innovation Solution
A neural network-based approach using 3D convolutional neural networks (CNN) and regression layers to analyze power maps and coefficient maps, enabling accurate and efficient determination of IR drop at a cell-level granularity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current methods are used to measure IR drop, then the measurement process is simple, but the accuracy is low and significant cases of IR drop are missed
Solution Approach 1:
The patent introduces an intermediary neural network system that bridges the gap between simple measurement methods and accurate IR drop detection. The neural network acts as a mediator that processes power maps and coefficient maps to predict IR drop values, combining the simplicity of input data collection with the accuracy of sophisticated pattern recognition without requiring complex measurement hardware modifications
Solution Approach 2:
The patent replaces traditional mechanical/electrical measurement systems with a computational approach using neural networks. Instead of using complex physical measurement equipment to directly measure IR drop, the system substitutes these with software-based neural network models that process power map data and coefficient map data to predict IR drop values, achieving high accuracy through information processing rather than physical measurement
2Productivity
If current methods are used to measure IR drop, then the implementation is straightforward, but the speed is slow
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing coefficient maps that contain pre-calculated resistance and current information. These coefficient maps are generated in advance through circuit analysis and stored for rapid lookup during IR drop measurement. When measuring IR drop, the system quickly retrieves relevant coefficient map data and combines it with power maps through the neural network, avoiding time-consuming real-time calculations while maintaining accuracy
Solution Approach 2:
The patent uses copying by creating simplified representations of the circuit's electrical characteristics through coefficient maps. These coefficient maps are copied versions of the actual circuit behavior that can be rapidly processed. The neural network operates on these copied representations (power maps and coefficient maps) rather than the full complex circuit model, achieving fast measurement without sacrificing measurement precision
3Measurement precision
If cell-level granularity analysis is performed, then the measurement precision is improved, but the computational complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the circuit into discrete cells and representing each cell's power characteristics as separate power map entries. The neural network processes these segmented cell-level data points individually through the coefficient maps to determine IR drop for each cell. This segmentation enables precise cell-level measurement while the efficient neural network architecture processes the divided data quickly, reducing overall computation time compared to analyzing the entire circuit as one unit
Data Source
AI summary
The IR drop for a portion of a circuit may include a voltage drop across resistance, and may include a product of current I passing through resistance with a resistance value R. In order to determine IR drop for a circuit in a more accurate and efficient manner, a neural network produces coefficient maps (that each indicate a time-varying distribution of power within an associated portion of the circuit), and these coefficient maps are then used by the neural network to determine an IR drop for each of a plurality of portions of the circuit.