Memory Buffer Flushing and Folding for Unavailable Dies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Memory systems face increased latency due to unavailable destination memory dies, leading to buffer overflow and inability to accommodate incoming data, especially when linked data is received out-of-sequence or at a rate faster than flushing can occur.
Innovation Solution
Implementing a fold-focused technique to flush linked data to memory dies out-of-order when a destination die is unavailable, followed by in-order folding to high-density portions, and a flush-focused technique to delay flushing if data is received non-sequentially below a threshold rate.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data is flushed to memory dies in strict in-order sequence, then data integrity and parallel reading capability are maintained, but buffer overflow occurs when destination dies are unavailable and incoming data rate exceeds flushing capability
Solution Approach 1:
The patent segments the buffer management process into two distinct phases: a flush phase that writes data to low-density portions of memory dies, and a fold phase that redistributes data to high-density portions. This segmentation allows the system to handle data in smaller, manageable units and applies different strategies to different phases, resolving the contradiction between maintaining data integrity and accommodating high incoming data rates.
Solution Approach 2:
The patent performs preliminary actions by first flushing data to low-density portions of memory dies before performing the fold operation. This preliminary flush phase ensures that data is temporarily stored in an available location even when the final destination is unavailable, preventing buffer overflow while maintaining the ability to perform subsequent fold operations for optimization.
2Manufacturing precision
If the memory system waits for destination dies to be available before flushing data, then data is written to correct locations, but latency increases and buffer space becomes unavailable for new data
Solution Approach 1:
The patent introduces low-density portions of memory dies as intermediary storage locations. These intermediary locations serve as temporary destinations during the flush phase, allowing data to be written even when the final high-density portions are unavailable. The intermediary enables the system to proceed with data flushing without waiting for the final destination to be ready, thereby reducing latency while maintaining the ability to perform accurate placement through the subsequent fold operation.
3Productivity
If linked data is received out-of-sequence or faster than flushing can occur, then throughput increases, but buffer overflow prevents accommodation of new data
Solution Approach 1:
The patent implements dynamic buffer management by switching between two operational modes based on real-time conditions: flush mode when destination dies are unavailable and fold mode when data needs to be redistributed. This dynamic adaptation allows the system to handle variable incoming data rates and maintain buffer availability, resolving the contradiction between high throughput and reliable buffer fullness control.
Data Source
AI summary
Methods, systems, and devices for buffer management techniques for a memory system are described. A memory system may store, in a buffer, a set of data for a parallel read operation that involves concurrently reading from a first die and from a second die, the set of data comprising first data and second data. The memory system may write the first data to a low-density portion of the first die based on writing the second data to the low-density portion of the first die and based on the second die being unavailable for writing the first data. The memory system may copy, based on the first data being included in the set of data for the parallel read operation, the first data from the low-density portion of the first die to a high-density portion of the second die.