Dynamic Peak Power Management for Multi-Die Memory Subsystems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In memory subsystems, peak power operations by multiple memory components can lead to voltage drops and error-prone conditions due to high current demands, necessitating conservative power management that limits concurrent peak power operations to avoid cumulative power overload, thereby forgoing opportunities for more components to perform peak operations when others are inactive.
Innovation Solution
Memory components communicate their activity states to dynamically determine the number of active components, allowing only a threshold to perform peak power operations concurrently, thereby optimizing power consumption and enabling more components to execute peak operations when others are inactive.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conservative power management limits concurrent peak power operations, then voltage drops and error-prone conditions are avoided, but system performance deteriorates due to fewer components performing peak operations
Solution Approach 1:
The power management system dynamically adjusts the threshold for concurrent peak power operations based on real-time monitoring of voltage levels, power consumption patterns, and component activity states. Instead of using a fixed conservative limit, the system adapts the threshold upward when voltage stability is maintained and downward when voltage drops are detected, thereby optimizing both reliability and productivity simultaneously
Solution Approach 2:
The system implements continuous feedback loops that monitor voltage levels, power consumption, and component states, then use this information to adjust power allocation decisions. The feedback mechanism allows the system to learn from past operations and optimize the balance between preventing voltage drops and maximizing peak operation throughput, resolving the contradiction between reliability and productivity
2Reliability
If all dice are assumed active with worst-case power consumption, then power consumption limit is maintained, but opportunities are forgone to allow more dice to perform peak operations when some are inactive
Solution Approach 1:
Each memory component autonomously communicates its actual activity state to the power management system, eliminating the need for conservative worst-case assumptions. Components that are truly inactive or in low-power states self-report this information, allowing the system to accurately determine available power headroom and safely allocate peak power operations to additional components, thereby increasing throughput without compromising power control
Solution Approach 2:
The system changes the parameter used for power management from a static worst-case power consumption value to a dynamic actual power consumption value based on real-time component states. By monitoring and responding to changes in component activity states, the system adjusts the effective power budget available for peak operations, enabling more components to operate at peak power when conditions permit
Data Source
AI summary
Exemplary methods, apparatuses, and systems include a first die in a power network receiving, from each of a plurality of dice in the power network, a first activity state value indicating that the respective die is in a high current state, a second activity state value indicating that the respective die is a moderate current state, or a third activity state value indicating that the respective die is a low current state. The received activity state values include at least one second or third activity state value. The first die determines, using the received activity state values, a first sum of the activity state values. The first die further selects an activity state based upon the first sum and sends, to the plurality of dice, an activity state value corresponding to the selected activity state.