Neuromorphic Synapse Core On-Demand Connectivity Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In neuromorphic computing systems with a large number of neurons, storing connectivity information requires significant memory space due to sparse and random connections, leading to storage inefficiencies.
Innovation Solution
The system generates connectivity information on-the-fly using seed numbers and finite field mathematical functions, such as Galois field functions, instead of storing all addresses, and stores synaptic weights in a compact manner to reduce memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If connectivity information of all neurons is stored in memory, then complete connectivity data is available, but storage space requirements become excessively large
Solution Approach 1:
The patent extracts only the essential seed numbers from the complete connectivity information, storing merely these compact seeds in memory rather than the full connectivity matrix. The actual connectivity data is regenerated on-demand from these seeds, dramatically reducing storage requirements while preserving complete connectivity information when needed.
Solution Approach 2:
The system pre-generates seed numbers that encode connectivity information in a compressed form. These seeds are stored in advance in memory, and when connectivity information is needed, it is regenerated on-demand from these pre-stored seeds through deterministic algorithms, eliminating the need to store complete connectivity data.
2Speed
If all neuron addresses are stored in the synapse core, then fast address lookup is achieved, but the synapse core size increases significantly
Solution Approach 1:
The patent extracts only the critical seed numbers from the full connectivity information and stores them in the synapse core. When address lookup is needed, the system regenerates the complete address information from these compact seeds using deterministic algorithms, achieving fast lookup without storing all address data in the core.
Solution Approach 2:
Instead of storing actual neuron addresses in the synapse core, the system stores compact seed numbers that can be used to generate addresses on-demand. This creates a virtual copy mechanism where addresses are regenerated from seeds when needed, reducing physical storage requirements in the synapse core.
3Ease of operation
If connectivity information is stored in a conventional manner, then data access is straightforward, but memory bandwidth consumption increases
Solution Approach 1:
The patent extracts only the essential seed numbers from complete connectivity information and stores them in memory. When connectivity data is needed, it is regenerated on-demand from these seeds, dramatically reducing the amount of data that needs to be transferred over memory bandwidth while maintaining ease of access through deterministic regeneration.
Data Source
AI summary
A neuromorphic computing system is provided which comprises: a synapse core; and a pre-synaptic neuron, a first post-synaptic neuron, and a second post-synaptic neuron coupled to the synaptic core, wherein the synapse core is to: receive a request from the pre-synaptic neuron, generate, in response to the request, a first address of the first post-synaptic neuron and a second address of the second post-synaptic neuron, wherein the first address and the second address are not stored in the synapse core prior to receiving the request.