SoC Power Prediction Using Performance Counters for Dynamic Power Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems-on-chip (SoCs) face challenges in accurately predicting power consumption of processors without requiring separate hardware logic for measurement, which limits design flexibility and power management efficiency.
Innovation Solution
A SoC that utilizes a performance monitor to collect count values of selected performance parameters, trains a neural network model to predict power consumption, and employs a power management circuit to adjust power supply based on these predictions, optimizing power usage without additional hardware for measurement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate hardware logic is added to measure processor power consumption, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent introduces an intermediary neural network model that acts as a mediator between performance parameters and power consumption. Instead of directly measuring power consumption with complex hardware, the system uses performance parameters as intermediaries that the neural network processes to predict power consumption accurately, thereby avoiding the need for complex measurement hardware
Solution Approach 2:
The patent replaces the mechanical/physical measurement system (hardware logic for direct power measurement) with an information-processing system (neural network model using performance parameters). This substitution eliminates the need for complex physical measurement hardware while achieving accurate power consumption prediction through software-based processing
2Measurement precision
If more performance parameters are monitored, then power consumption prediction accuracy is improved, but device complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network model offline using comprehensive performance parameters and actual power consumption data. During runtime, the system only needs to collect a limited set of performance parameters that the pre-trained model knows how to process effectively, eliminating the need to monitor all possible parameters continuously and reducing real-time system complexity
Solution Approach 2:
The patent changes the parameter representation by transforming raw performance parameters into features suitable for neural network processing. The system selectively monitors specific performance parameters and transforms them into appropriate input formats for the neural network, optimizing the balance between prediction accuracy and monitoring complexity through parameter transformation rather than brute-force monitoring of all parameters
3Productivity
If power supply is increased to meet peak processor demands, then productivity is improved, but energy consumption increases
Solution Approach 1:
The patent implements dynamics by enabling the power supply to dynamically adjust to actual processor needs rather than maintaining a fixed high level. The neural network predicts future power consumption based on current performance parameters, allowing the power management circuit to dynamically optimize power delivery - increasing power only when and where needed for peak performance while reducing power during lower-demand periods
Solution Approach 2:
The patent establishes a feedback loop where performance parameters are continuously monitored, fed into the neural network model, and the predicted power consumption is used to adjust power supply in real-time. This closed-loop feedback system ensures power is optimized based on actual processor demands, improving productivity when needed while minimizing energy consumption during normal operation
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
A system-on-chip (SoC) includes a first processor including a first performance monitor configured to perform monitoring on a plurality of first performance parameters including first performance parameters. The SoC also includes a power prediction circuit configured to predict a power consumption of the first processor based on count values of the first performance parameters collected by the first performance monitor, and a power management unit configured to manage power supplied to the first processor based on a prediction result of the power prediction circuit.