Volatile Memory Partitioning for Power and Performance Trade-offs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computing devices face challenges in efficiently running applications due to high power consumption, which is exacerbated by bandwidth-intensive operations, and decreasing power consumption limits memory capacity while being costly.
Innovation Solution
Partitioning volatile memory into high and low performance partitions allows for dynamic power management, where high performance data is moved to low performance partitions during power reduction, maintaining power to low performance partitions, and restoring data as needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If memory bandwidth is increased to efficiently run applications, then application performance is improved, but power consumption increases
Solution Approach 1:
The memory system is divided into two distinct partitions: a first memory partition optimized for high bandwidth operations and a second memory partition optimized for low power consumption. The processor selectively accesses data from appropriate partitions based on performance requirements, thereby achieving application performance improvements without uniformly increasing power consumption across the entire memory system.
Solution Approach 2:
Different regions of the memory system are assigned different performance characteristics and power consumption profiles. The first memory partition provides high bandwidth for performance-critical operations, while the second partition provides adequate bandwidth for less demanding operations at lower power consumption. This local differentiation allows the system to optimize power usage based on specific access patterns rather than maintaining high performance uniformly.
2Use of energy by moving object
If power consumption is decreased to extend battery life, then power depletion is reduced, but memory bandwidth usage is limited
Solution Approach 1:
The memory system dynamically selects between the first and second memory partitions based on real-time performance requirements and power management conditions. The processor can switch between high-performance mode (accessing the first partition) and power-saving mode (accessing the second partition), allowing the system to adapt bandwidth usage to actual needs rather than being constrained by a fixed power level.
Solution Approach 2:
The second memory partition provides partial bandwidth capability sufficient for many application workloads without delivering the full bandwidth of the first partition. This partial capability is adequate for extending battery life during normal operations, while the first partition remains available to provide excessive bandwidth when performance-critical tasks require it.
3Use of energy by moving object
If memory capacity is increased to compensate for lower power capabilities, then power depletion is reduced, but cost increases
Solution Approach 1:
The memory system is segmented into two partitions with different performance and power characteristics, allowing the use of cost-effective memory technologies for the second partition while reserving high-performance (and higher-cost) memory for the first partition. This segmentation enables the system to achieve adequate overall capacity without uniformly deploying expensive high-performance memory across the entire system.
Solution Approach 2:
The second memory partition uses cost-effective memory technology that consumes less power but provides lower bandwidth and potentially shorter retention characteristics. This partition handles workloads where extreme performance is not critical, allowing the system to reduce overall cost by using cheaper memory for the majority of storage needs while relying on the first partition only when high performance is required.
Data Source
AI summary
Examples disclose partitioning a volatile memory into a high performance partition and a low performance partition. Further the example discloses retrieving an application with a high performance data and a low performance data from a non-volatile memory to place the high and the low performance data in the high and low performance partitions, respectively. Additionally, the example also discloses receiving a request to decrease power and in response, reduce an amount of power to the high performance partition and maintaining an amount of power provided to the low performance partition.