Memory Subsystem Data Placement via Temporal Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory sub-systems face inefficiencies in data placement, leading to increased write amplification, reduced performance, and shorter operational life due to the need for frequent garbage collection and re-writing of data.
Innovation Solution
Implementing a system that uses a trained machine learning model to determine the temporal metric class of data cached in SLC blocks and moves it to groups of non-SLC blocks based on this classification, optimizing data placement and reducing write amplification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is stored in SLC blocks with frequent updates, then write amplification increases and performance decreases, but data placement efficiency is improved when using machine learning classification
Solution Approach 1:
The system performs preliminary classification of data into temporal metric classes using a machine learning model before data is written to memory blocks. This advance classification enables optimal placement decisions to be made beforehand, preventing future write amplification issues and improving data placement efficiency from the outset.
Solution Approach 2:
The patent applies different storage strategies to different types of data based on their temporal metric class. Frequently accessed data is placed in SLC blocks while less frequently accessed data is placed in non-SLC blocks, optimizing the local quality characteristics of each storage region to match the access patterns of the data it contains.
2Reliability
If garbage collection is performed frequently to manage data updates, then memory sub-system performance is maintained, but operational life is reduced due to re-writing operations
Solution Approach 1:
The machine learning model performs preliminary classification to identify data that is unlikely to be updated soon. This allows the system to proactively place such data in non-SLC blocks before update operations are needed, thereby reducing the frequency of garbage collection and extending the operational life of the memory sub-system.
Solution Approach 2:
The patent introduces an intermediate classification layer using machine learning models that act as a mediator between data arrival and storage placement. This intermediary system analyzes temporal metrics and makes intelligent placement decisions, reducing the need for subsequent data movement operations and extending memory device lifespan.
3Loss of energy
If data is moved between SLC and non-SLC blocks based on access frequency, then write amplification is reduced, but system complexity increases due to machine learning model implementation
Solution Approach 1:
The system implements self-service through automated machine learning models that independently classify data and make placement decisions without requiring complex external control mechanisms. The model continuously learns from access patterns and autonomously optimizes data placement, reducing write amplification while keeping the control architecture relatively simple.
Solution Approach 2:
The patent changes the parameter of data classification from traditional access frequency thresholds to a more sophisticated temporal metric class determined by machine learning models. This parameter change enables more accurate prediction of data update patterns, improving write amplification reduction while the model handles the complexity of analysis internally.
Data Source
AI summary
Systems and methods are disclosed including a memory and a processing device operatively coupled to the memory. The processing device can perform operations including identifying a set of logical addresses associated with data stored on the memory devices in one or more blocks of a first type; determining a temporal metric class associated with the set of logical addresses, wherein the temporal metric class is associated with a corresponding range of predicted update characteristic of the data; identifying, based on the temporal metric class, a set of blocks of a second type, wherein a first block of the first type comprises a first plurality of memory cells having a first number of bits per cell, and wherein a second block of the second type comprises a second plurality of memory cells having a second number of bits per cell that exceeds the first number of bits per cell; and moving the data to the identified set of blocks.