Dynamic Memory Power Management via Voltage Droop Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional memory sub-systems face inefficiencies due to pre-defined peak power management settings that do not account for varying host system designs and power delivery networks, leading to reduced throughput and stability during parallel memory operations.
Innovation Solution
A training phase is implemented where memory dies execute multiple operations to determine voltage parameter levels, generating tokens that guide power management actions to optimize power consumption based on actual host system demands, allowing for dynamic power management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pre-defined peak power management settings are used, then power consumption is controlled within limits, but throughput is reduced and stability deteriorates during parallel memory operations
Solution Approach 1:
The patent implements dynamic power management by transitioning from static pre-defined peak power settings to a training-based approach where voltage parameter levels are determined through actual execution of memory operations. The system dynamically adjusts power management parameters based on detected voltage droop during training phases, allowing optimal throughput while maintaining stability under varying operational conditions.
Solution Approach 2:
The patent changes the power management parameters from fixed pre-defined values to dynamically determined voltage parameter levels. During a training phase, the system executes memory operations and detects actual voltage droop to establish appropriate power management thresholds, thereby adapting parameters to real system behavior rather than relying on theoretical or worst-case assumptions.
2Adaptability or versatility
If pre-defined peak power management settings are used, then power consumption is controlled, but adaptability to varying host system designs and power delivery networks is reduced
Solution Approach 1:
The patent implements feedback-based power management where the system executes memory operations during a training phase and detects actual voltage droop in the power delivery network. This feedback information is used to determine appropriate voltage parameter levels and power management settings specific to each host system design and power delivery network configuration, thereby achieving adaptability while maintaining proper power consumption control.
Solution Approach 2:
The patent performs a training phase before normal operation where memory operations are executed to characterize the power delivery network and determine voltage parameter levels. This preliminary action allows the system to adapt to specific host system designs and power delivery networks before actual workloads are processed, ensuring optimal power management for each configuration.
3Measurement precision
If pre-defined peak power management settings are used, then power consumption is controlled, but measurement precision of actual voltage parameter levels is reduced
Solution Approach 1:
The patent uses feedback from actual voltage droop detection during memory operation execution to precisely determine voltage parameter levels. Instead of relying on pre-defined or theoretical values, the system measures actual voltage changes in the power delivery network and uses this precise measurement to establish accurate power management thresholds specific to each system configuration.
Data Source
AI summary
A request is received from a host system to execute a portion of a memory management operation associated with a memory cell of a plurality of memory cells of one or more memory devices. A voltage parameter level associated with execution of the portion of the memory management operation is identified. A determination is made that a comparison of the voltage parameter level with a voltage parameter level threshold satisfies a condition. A power management action is performed in response to the condition being satisfied.