Memory System Energy Throttling via Data Migration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional memory systems face challenges in managing and optimizing the performance and endurance of diverse memory types within a converged memory device, particularly in handling high-energy consumption and temperature-related issues, which can lead to data errors and system damage due to high power usage and temperature vulnerabilities.
Innovation Solution
A memory system is designed with a controller that manages multiple memory groups with different characteristics, utilizing a cache region to throttle energy consumption and temperature by migrating data from high-energy memories to lower-energy ones and controlling access based on refresh cycles and priority requests, thereby reducing operational frequency and energy usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data is stored in high-capacity memory devices, then storage capacity is improved, but energy consumption and temperature increase leading to data errors and system damage
Solution Approach 1:
The memory system is divided into multiple memory groups (first memory group, second memory group, third memory group) with different characteristics. High-capacity memory devices are segmented and organized into groups, allowing the system to distribute data storage across multiple devices with varying energy consumption profiles, thereby managing thermal and power characteristics while maintaining total storage capacity.
Solution Approach 2:
Different memory groups are assigned different local qualities based on their characteristics. The first memory group has high capacity, the second has low energy consumption, and the third has low temperature. Data is selectively stored in specific groups based on access patterns and thermal/energy constraints, giving different parts of the storage system different functional qualities.
2Productivity
If access frequency to memory regions is increased, then productivity is improved, but temperature and energy consumption increase causing reliability issues
Solution Approach 1:
The controller performs preliminary actions by predicting future access patterns and pre-fetching or pre-positioning data in memory regions that will be accessed soon. This reduces the need for frequent high-speed access to the same regions, thereby maintaining productivity while reducing peak temperature and energy consumption that would otherwise cause reliability issues.
Solution Approach 2:
The memory system implements periodic refresh operations and periodic data migration between memory groups. By periodically moving data between high-capacity and low-energy memory groups based on access patterns, the system maintains high productivity when needed while allowing temperature and energy consumption to reset during low-activity periods, thus preserving reliability.
3Quantity of substance
If high-capacity memory devices are used, then storage capacity is improved, but endurance decreases due to limited write cycles
Solution Approach 1:
The system extracts frequently written data from high-capacity memory devices and relocates it to buffer memory or low-endurance memory groups. This separation removes the wear-inducing write operations from the high-capacity storage medium, allowing it to maintain its full storage capacity while extending its operational lifespan by reducing the number of write cycles it undergoes.
Solution Approach 2:
Data is copied between different memory groups based on access patterns and wear leveling requirements. When a high-capacity memory device approaches its write cycle limit, its data is copied to another memory group, allowing the original device to be reset or used for read-only operations, thereby extending the overall system endurance while preserving total storage capacity.
Data Source
AI summary
A memory system may improve the endurance and performance of a plurality of memories included in the memory system mounted on a server system or a data processing system. For example, the memory system may throttle energy of a first memory using a second memory having a different characteristic from the first memory, control accesses to a memory region according to a refresh cycle, and control accesses to memories having different temperatures according to a priority of a request for each of the memories.