Storage Device Read Latency Reduction via Cross-Page Data Continuity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Storage devices face latency issues in read operations due to the need for multiple page reads when calculating result values for functions, particularly in neural network computations, which can be computationally intensive and power-constrained environments.
Innovation Solution
A storage device architecture that sequentially stores result values of functions across multiple pages, with the last area of one page being stored in the start area of the next page, allowing for a one-time read operation to calculate activation function results, thereby reducing latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If result values are stored sequentially across multiple pages in conventional manner, then data storage is simple, but read operation latency increases due to multiple page reads
Solution Approach 1:
The patent applies preliminary action by pre-storing the last area of one page in the start area of the next page during the write operation. This advance preparation ensures that when a read operation needs to access function result values that span page boundaries, the data is already positioned contiguously in memory, eliminating the need for complex multi-page read coordination and reducing read latency.
Solution Approach 2:
The patent changes the storage dimensionality by extending data storage from a single-page linear sequence to a multi-page continuous sequence. By storing the last area of page N in the start area of page N+1, the patent creates a cross-page continuity that allows read operations to access spanning data as if it were within a single page, effectively adding a dimensional aspect to data organization that resolves the latency issue.
2Productivity
If multiple page reads are performed for function result calculation, then complete data is obtained, but operational efficiency decreases
Solution Approach 1:
The patent merges the last area of one page with the start area of the next page by storing duplicate copies of the boundary data in both locations. This merging creates a logical single-page view of data that physically spans multiple pages, allowing the system to retrieve complete function result values in a single read operation rather than multiple separate reads, thereby improving operational efficiency.
Solution Approach 2:
The patent uses copying by duplicating the last area data of page N into the start area of page N+1. This copy operation creates redundant data paths that enable faster read access, as the system can retrieve the needed data from either page without requiring coordinated multi-page reads, thus reducing read operation time and improving productivity.
3Speed
If data is stored continuously across page boundaries, then read latency is reduced, but memory management complexity increases
Solution Approach 1:
The patent applies preliminary action in memory management by pre-positioning the last area data of each page into the start area of the following page during the write phase. This advance arrangement eliminates the need for complex runtime memory management operations when reading, as the data continuity is already established, allowing fast read operations without intricate page boundary handling logic.
Data Source
AI summary
A storage device can be designed to reduce latency in a read operation. Such a storage device can include: a memory device including a plurality of pages that include a first page and a second page different from the first page, each page including a plurality of memory cells that are configured to store data; and a memory controller in communication with the memory device and for sequentially storing result values of a function with respect to a plurality of input values in the plurality of memory cells, and controlling the memory device to store a result value in a last area of the first page and a start area of the second page.