Decentralized Power Management for Data Processing Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing power management schemes in system-on-chip integrated circuits induce a power-delay trade-off, increasing latency and struggling to scale effectively in larger systems with multiple processing units and peripherals, as they rely on centralized power management and complex predictive models.
Innovation Solution
A decentralized power management approach where slave units and master units use usage signals to proactively switch between power states based on predicted transaction intervals, eliminating the need for complex central controllers and enabling precise power management at the micro-architectural level, with power management signals routed alongside communication signals for scalability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If centralized power management schemes are used, then power consumption is reduced, but system scalability deteriorates
Solution Approach 1:
The patent segments the centralized power management function into distributed power management units at each slave device. Each slave independently monitors its own transaction patterns and controls its own power state, eliminating the need for a complex central controller while maintaining effective power management across the entire system.
Solution Approach 2:
Each slave device performs self-power management by autonomously detecting its own transaction intervals and making power state decisions. The slave uses usage signals from master units to determine when to enter low-power states, eliminating dependency on centralized control and enabling natural scalability.
2Loss of energy
If heuristic power management policies are used, then power consumption is reduced, but latency increases
Solution Approach 1:
The patent implements a feedback mechanism where master units send usage signals to slave units indicating when transactions are expected. The slave uses this real-time feedback to dynamically adjust its power state, transitioning to low-power mode only when certain about upcoming transactions, thereby reducing both power consumption and latency compared to fixed heuristic policies.
Solution Approach 2:
The slave unit performs preliminary power state transitions based on predicted transaction intervals. By receiving advance usage signals from master units, the slave can proactively enter or exit power states before transactions actually occur, optimizing the balance between power savings and response time.
Data Source
AI summary
A device, such as an integrated circuit is described including master units, and slave units connected by an interconnect. In addition to the normal data signals and address signals passed with a transaction, there are also passed usage signals which specify the time interval until a next transaction will be sent to a slave unit. A local slave power controller is responsive to such usage signals to switch into a low power mode and pre-emptively switch back to an operational mode in time to respond to the next transaction to be received.