Memory Bank Diversity via Request Splitting and Arbitration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current memory architectures, such as SDRAM and DDR-DRAM, face inefficiencies due to bank conflicts and write-to-read turnaround conflicts, which reduce memory performance, especially in multimedia applications requiring real-time and high-quality data transfer.
Innovation Solution
A method of memory management that splits access requests into sub-requests to minimize bank conflicts by ensuring each sub-request targets a different bank, combined with a re-ordering arbitration mechanism to prioritize and sequence requests, thereby increasing bank diversity and reducing conflicts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If consecutive access requests are made to the same bank, then the memory controller can maintain simple access patterns, but bank conflict occurs resulting in increased latency
Solution Approach 1:
The patent segments a single access request to bank B0 into multiple sub-requests targeting different banks (B0, B1, B0, B1...). By dividing the original request into smaller units and distributing them across multiple banks, the system avoids bank conflicts while maintaining data integrity through proper reassembly at the destination
Solution Approach 2:
The patent introduces bank diversity as an additional dimension to the access pattern. Instead of sequentially accessing the same bank (one-dimensional pattern), the system distributes requests across multiple banks (multi-dimensional pattern), transforming the access strategy from time-sequential to space-parallel operations
2Reliability
If the memory controller waits for tRC latency to complete bank conflicts, then correct memory operations are ensured, but memory efficiency is reduced
Solution Approach 1:
The patent performs preliminary actions by pre-charging banks in advance before they are needed for subsequent requests. The memory controller proactively manages bank states (activating, pre-charging, or idle) ahead of time, allowing seamless transitions between bank operations without waiting for full tRC latency periods
Solution Approach 2:
The patent ensures continuous useful action by eliminating idle waiting periods. While one bank is being accessed, the controller simultaneously prepares other banks (pre-charging or activating them), ensuring that the memory system continuously performs useful work without interruption or wasted latency periods
3Productivity
If access requests are split into multiple sub-requests, then bank diversity is increased and conflicts reduced, but request processing complexity increases
Solution Approach 1:
The patent implements self-service mechanisms where the system automatically tracks bank usage states and dynamically determines optimal bank assignments for each sub-request. The memory controller autonomously manages the complexity of request splitting, bank state tracking, and sub-request reassembly without external intervention, making the complexity management transparent to higher-level systems
Data Source
AI summary
A method of memory management is disclosed. The invention increases bank diversity by splitting requests and is also integrated with re-ordering and priority arbitration mechanisms. Therefore, the probabilities of both bank conflicts and write-to-read turnaround conflicts are reduced significantly, so as to increase memory efficiency.