Memory Data Folding Order for Concurrent Host Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Memory systems face increased latency due to the inability to concurrently perform host system access operations and data folding operations, leading to temporary suspension of either operation when they overlap on the same memory die.
Innovation Solution
The memory system reorders data during the folding operation to allow concurrent performance of access and folding operations by temporarily delaying or replacing data in source blocks, enabling simultaneous execution without suspension.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the memory system performs data folding operations on the same memory die that is being accessed by the host system, then data folding can be performed efficiently, but the host system access operation must be temporarily suspended
Solution Approach 1:
The memory system is divided into multiple independent memory dies, allowing different operations to be performed on different dies simultaneously. The host system access operation is directed to one die while the data folding operation is performed on another die, eliminating the need to suspend either operation.
Solution Approach 2:
A controller acts as an intermediary between the host system and the memory dies. The controller receives the host system access command, determines that a data folding operation is in progress on the target die, and redirects the access command to a different die that is available, thereby mediating the conflict between concurrent operations.
2Loss of time
If the memory system suspends data folding operations to allow host system access, then host system access latency is reduced, but overall system productivity decreases
Solution Approach 1:
By segmenting the memory system into multiple independent dies, the system can perform data folding on one die while simultaneously handling host system accesses on other dies. This parallel execution maintains high overall system throughput without compromising host system access latency.
Solution Approach 2:
The system transitions from sequential operation on a single die to parallel operations across multiple dies, adding a spatial dimension to operation execution. This allows data folding and host system accesses to occur concurrently in different spatial locations (different dies), improving both latency and throughput.
3Device complexity
If the memory system uses a single memory die for both host system access and data folding, then device complexity is reduced, but operation concurrency is limited
Solution Approach 1:
The memory system is segmented into multiple dies, each capable of independent operation. This segmentation enables concurrent execution of different operations on different dies, significantly improving operation concurrency while maintaining relatively simple individual die structures.
Solution Approach 2:
Each memory die is designed to be universal and multi-functional, capable of handling both host system access operations and data folding operations independently. This multi-functionality at the die level, combined with multiple dies, enables high operation concurrency without requiring complex specialized structures.
Data Source
AI summary
Methods, systems, and devices for techniques for concurrent host system access and data folding are described. A memory system may determine to transfer (e.g., fold) data from a set of source data blocks to a set of destination data blocks. The memory system may receive a command to access a first source data block of the set of source data blocks concurrent with the data transfer. The memory system may generate a first order for transferring respective portions of the data that is based on a second order associated with a sequential read of the data from the set of destination data blocks. Based on the accessing the first source data block being concurrent with the data transfer, the first order may exclude a first portion of the data from the first source data block such that the data transfer and the accessing may be concurrently performed.