Distributed Memory Bank Interconnects for Low-Latency Data Movement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing processor operations on large data blocks suffer from high latency due to scatter and gather operations involving centralized memory access, which can be inefficient and power-consuming.
Innovation Solution
Implementing a distributed memory architecture where processors are directly coupled to local portions of memory through interconnects, allowing for localized data access and reduced latency via direct communication between memory banks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If centralized memory access is used for processing large data blocks, then data can be stored and retrieved through a single memory controller, but latency increases and processing efficiency decreases
Solution Approach 1:
The memory system is divided into multiple memory banks (first memory bank, second memory bank, etc.), each capable of independent operation. Processors are similarly divided into multiple processing units, with each processor directly coupled to specific memory banks through interconnects. This segmentation allows parallel access to different memory regions, eliminating the bottleneck of centralized memory access and reducing latency for large data block operations.
2Use of energy by stationary object
If centralized memory access is used, then memory structure is simplified, but power consumption increases due to longer data transmission paths
Solution Approach 1:
Each processor is directly coupled to specific memory banks through dedicated interconnects, creating localized data access paths. This local quality principle reduces the average transmission distance for data access, as processors can directly access their locally-coupled memory banks without routing through a centralized controller. The result is reduced power consumption for memory operations while the overall system complexity is managed through modular design.
3Loss of time
If distributed memory banks are used with direct processor coupling, then localized data access reduces latency, but interconnect complexity increases
Solution Approach 1:
The interconnect structure is segmented into multiple independent pathways, with each processor having dedicated interconnects to its locally-coupled memory banks. This segmentation of the interconnect structure allows for parallel data transmission paths, reducing contention and latency. While the overall interconnect complexity increases due to the distributed nature of the system, the modular segmented design makes the complexity manageable and scalable.
Data Source
AI summary
In one embodiment, an apparatus comprises: a plurality of banks to store data; and a plurality of interconnects, each of the plurality of interconnects to couple a pair of the plurality of banks. In response to a data movement command, a first bank of the plurality of banks is to send data directly to a second bank of the plurality of banks via a first interconnect of the plurality of interconnects. Other embodiments are described and claimed.