Spiking Neuromorphic FEM Circuits for Energy-Efficient Mesh Solving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing finite element methods (FEM) face challenges in scalability and energy efficiency due to the resource demands of modern supercomputers, which are approaching sustainability limits, and neuromorphic computing has not been effectively integrated with traditional parallel scientific computation.
Innovation Solution
A spiking neuromorphic circuit is developed to instantiate a finite element mesh, where spiking neurons represent mesh nodes, and synaptic weights change iteratively to solve linear systems, with hierarchical FEM meshes operating concurrently and interacting via weight matrices to achieve independent convergence rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If traditional supercomputing is used to solve large-scale finite element problems, then computational accuracy is maintained, but energy consumption approaches sustainability limits
Solution Approach 1:
The patent replaces traditional digital computing mechanics with neuromorphic computing mechanics. Spiking neural networks are configured to perform finite element computations using analog-like continuous dynamics, where neuron membrane potentials and synaptic weights naturally evolve to solve the linear system. This substitution eliminates the need for discrete digital operations and memory access patterns that consume high energy in conventional supercomputers, achieving both energy efficiency and computational accuracy simultaneously
Solution Approach 2:
The patent changes the fundamental computational parameters from digital binary states to continuous analog values representing physical quantities. The spiking neurons use membrane potentials as continuous parameters that directly encode solution variables, and synaptic weights represent interaction coefficients. This parameter transformation allows the system to solve finite element problems with energy consumption proportional to the physical dynamics rather than digital operation counts, resolving the energy-productivity contradiction
2Measurement precision
If FEM mesh resolution is increased to improve solution accuracy, then computational precision improves, but memory bandwidth demands and device complexity increase
Solution Approach 1:
The patent segments the computational domain into a finite element mesh where each element is represented by spiking neurons. The segmentation allows local computations to be performed independently at each mesh node and element, with interactions captured through weighted synaptic connections. This distributed segmentation eliminates the need for centralized memory access patterns, reducing memory bandwidth demands while maintaining high solution accuracy through local parallel processing
Solution Approach 2:
The patent adds a temporal dimension to the computational model by using spiking dynamics. Instead of solving the linear system through traditional iterative methods that require repeated memory accesses, the system evolves continuously in time with neuron potentials naturally converging to the solution. This temporal dimension transformation converts a memory-bandwidth-intensive spatial computation into a time-evolution process, reducing device complexity while improving solution accuracy
Data Source
AI summary
A spiking neuromorphic circuit that instantiates a finite element methods (FEM) mesh is provided. The circuit comprises a number of groups of spiking neurons, wherein each group of spiking neurons represents a mesh node in the FEM mesh, wherein the FEM mesh represents a linear system. A bias current represents conditions in the linear system. Interaction weights between adjacent mesh nodes are proportional to the linear system represented by the FEM mesh. The spiking neurons within each group of spiking neurons spike in a manner that flows to a solution variable for the respective mesh node.