Recursive Array Layout for Transposable Synaptic Weight Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional memory access methods in neuromorphic and synaptronic computation systems, such as standard SRAM and eDRAM, are limited to one-dimensional access, leading to inefficiencies in power consumption and performance when handling synaptic weights in neuromorphic networks, particularly in event-driven architectures that require transposable access for learning rules like STDP.
Innovation Solution
The implementation of a recursive array layout, such as the Morton layout, allows for transposable access to synaptic weights in neuromorphic networks, enabling efficient row and column read/write operations by distributing information across blocks, reducing the number of memory accesses required and optimizing power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If traditional one-dimensional memory access methods (standard SRAM and eDRAM) are used, then the memory structure is simple and easy to implement, but power consumption increases and performance decreases when handling synaptic weights in neuromorphic networks
Solution Approach 1:
The patent transforms traditional one-dimensional memory access into two-dimensional transposable access by organizing synaptic weights in a recursive array layout. This allows the same physical memory structure to support both row-wise and column-wise access patterns, enabling efficient implementation of learning rules like STDP without increasing physical complexity
Solution Approach 2:
The recursive array layout enables the memory structure to perform multiple functions: it can access synaptic weights in both row and column directions using the same physical array, eliminating the need for separate memory structures for different access patterns and reducing overall system complexity
2Productivity
If traditional one-dimensional memory access methods are used, then the memory structure is simple, but performance decreases in event-driven architectures that require transposable access for learning rules like STDP
Solution Approach 1:
By adding a dimensional aspect to memory access organization, the patent enables transposable access patterns required for high-performance neuromorphic computing. The recursive layout allows efficient row and column access without requiring physically separate memory arrays for each access pattern
Solution Approach 2:
The patent divides the memory access structure into hierarchical blocks that can be independently accessed. This segmentation allows selective access to specific rows or columns of synaptic weights, improving performance by enabling parallel operations and reducing access conflicts in event-driven architectures
3Productivity
If recursive array layout is implemented for transposable access, then power consumption is reduced and performance increases, but the memory access structure becomes more complex
Solution Approach 1:
The recursive array layout organizes memory addresses in a space-filling curve pattern that maps two-dimensional array indices to one-dimensional physical memory locations. This transformation enables transposable access patterns while maintaining compatibility with standard one-dimensional memory hardware, achieving performance improvements without proportionally increasing physical complexity
Data Source
AI summary
Embodiments of the invention relate to providing transposable access to a synapse array using a recursive array layout. One embodiment comprises maintaining synaptic weights for multiple synapses connecting multiple axons and multiple neurons, wherein the synaptic weights are maintained based on a recursive array layout. The recursive array layout facilitates transposable access to the synaptic weights. A neuronal spike event between an axon and a neuron is communicated via a corresponding connecting synapse by accessing the synaptic weight of the corresponding connecting synapse in the recursive array layout.