Power Grid Simulation Using Fast Transform Preconditioners
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Simulating large-scale power delivery networks in integrated circuits is challenging due to the inefficiencies of direct methods and the unpredictable convergence rate of iterative methods, especially when dealing with large, sparse linear systems and irregular power grid structures, which limits the effectiveness of existing preconditioners and parallelization techniques.
Innovation Solution
The use of Fast Transform-based preconditioners that regularize the power grid into a structured form, allowing for efficient solution by Fast Transform solvers, which can be executed on parallel architectures, thereby overcoming the limitations of existing methods by providing a straightforward and inexpensive implementation with improved convergence rates and parallelism.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If direct methods (matrix factorization) are used for solving large-scale power grid linear systems, then robustness is improved, but execution time and memory requirements become prohibitively expensive
Solution Approach 1:
The patent applies segmentation by dividing the large-scale power grid into multiple smaller sub-grids or clusters that can be processed independently. This allows the use of iterative methods on smaller, manageable subsets rather than attempting to solve the entire system at once with direct methods, thereby reducing execution time and memory requirements while maintaining solution accuracy through coordinated updates across sub-grids.
Solution Approach 2:
The patent substitutes direct matrix factorization methods with iterative solution methods. This replacement transitions from a deterministic mechanical approach (direct factorization) to a probabilistic iterative approach that converges to the solution, enabling scalable computation for large-scale power grids by avoiding the cubic complexity of direct methods.
2Ease of manufacture
If direct methods are used with fixed time-step, then reusability of factorization results is improved, but efficiency during long intervals of low activity deteriorates
Solution Approach 1:
The patent implements dynamic time-step adjustment in the iterative simulation process. The simulation automatically adjusts the time-step size based on the activity level in the power grid, using larger time-steps during low-activity intervals to skip unnecessary computation steps, and smaller time-steps during high-activity periods to capture transient behavior accurately. This dynamic approach maintains efficiency while preserving reusability of computed results across varying operational conditions.
3Productivity
If iterative methods are used for large sparse linear systems, then computational and memory efficiency is improved, but convergence rate becomes unpredictable
Solution Approach 1:
The patent applies preliminary action through the use of preconditioning techniques before executing the iterative solver. A preconditioner is constructed that transforms the original linear system into an equivalent system with better spectral properties, ensuring that the iterative method converges rapidly and predictably. This preliminary transformation step addresses the unpredictability of convergence by guaranteeing favorable convergence characteristics before the main iterative computation begins.
Solution Approach 2:
The patent introduces a preconditioner as an intermediary component between the linear system and the iterative solver. This intermediary transforms the original system matrix into a form that is more amenable to iterative solution, acting as a bridge that improves convergence behavior without changing the fundamental solution. The preconditioner serves as a mediator that enables the iterative method to achieve both efficiency and predictable convergence.
4Ease of operation
If general-purpose preconditioners are used with iterative methods, then ease of implementation is improved, but convergence improvement is limited
Solution Approach 1:
The patent applies local quality by developing power-grid-specific preconditioners that are tailored to the unique characteristics of power delivery networks. Rather than using generic preconditioners that treat all linear systems uniformly, the invention creates preconditioners that exploit the specific sparsity patterns, connectivity structures, and physical properties of power grids, thereby achieving superior convergence rates while remaining implementable through systematic construction procedures.
Data Source
AI summary
Systems and methods related to fast simulation of power delivery networks are described. A method is provided for simulating the time-domain responses of a plurality of points of a multi-layer power delivery network, comprising selecting a model of the power delivery network of a circuit, parsing the characteristic data describing the power delivery network, forming a circuit matrix relating to said circuit characteristic data, creating a preconditioner matrix with a specialized structure that allows solution by a Fast Transform solver, simulating the circuit using said circuit and preconditioner matrices by a computer, including a non-transitory computer readable storage medium and at least one processor, but preferably multiple processors, and reporting the responses at selected nodes and branches of the power delivery network.