Semiconductor Power Prediction via Action Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting energy and power usage in semiconductor devices are slow, limited in analysis coverage, and do not effectively identify root causes or work with software simulators, failing to explore different solutions through simulation.
Innovation Solution
The method involves recognizing integrated circuit actions within a signal activity stream, determining associated values, and generating models of power use based on these actions, using processors to modify power consumption, with options including direct signal analysis, formal automata, and machine learning, allowing for faster and more accurate power prediction and analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing methods rely on analysis of transistor gate activity, then power prediction can be performed, but the analysis is slow and has limiting coverage
Solution Approach 1:
The patent segments the power analysis process into distinct phases: (1) RTL-level simulation to generate action streams, (2) action recognition and classification, (3) power model generation, and (4) power prediction. This segmentation allows each phase to operate at its optimal level of abstraction, achieving both speed and accuracy.
Solution Approach 2:
The patent introduces a new dimension of analysis by moving from traditional transistor gate-level analysis to action-based analysis at the RTL level. This dimensional shift enables parallel processing and avoids the exponential complexity of gate-level simulation while maintaining predictive accuracy through action recognition.
2Measurement precision
If existing methods use transistor gate activity analysis, then power consumption can be measured, but they do not identify root causes
Solution Approach 1:
The patent introduces action recognition as an intermediary layer between RTL simulation and power analysis. This intermediary extracts meaningful actions (e.g., memory reads, arithmetic operations) from simulation signals, enabling both accurate power measurement and identification of root causes through action classification and attribution.
Solution Approach 2:
The patent performs preliminary action recognition and classification before power calculation. By identifying and categorizing actions in advance, the system establishes traceability to root causes while preparing data for efficient power modeling, avoiding the need for post-hoc analysis.
3Ease of manufacture
If existing methods do not work with software simulators, then hardware analysis can be performed, but different solutions cannot be explored through simulation
Solution Approach 1:
The patent creates a universal action recognition framework that works with multiple simulation types (RTL, gate-level, software co-simulation) and hardware configurations. The action-based model serves as a common interface, enabling the same power analysis methodology to be applied across different simulation fidelities and software/hardware scenarios.
4Measurement precision
If detailed transistor-level analysis is performed, then accurate power prediction can be achieved, but the process becomes extremely slow
Solution Approach 1:
The patent implements a dynamic, multi-level analysis approach that adapts the simulation fidelity to the analysis needs. The action recognition system operates at RTL level for speed, while power models incorporate detailed transistor-level characteristics when needed, dynamically balancing accuracy and speed based on the specific analysis context.
Data Source
AI summary
Methods for modifying power use of a semiconductor device include receiving, at one or more processors, an activity stream of a simulation of a semiconductor device, the activity stream comprising a stream of signals. Using the one or more processors, integrated circuit actions are recognized from the activity stream, each integrated circuit action representing an abstraction of work done by the semiconductor device. The processor(s) determine one or more values associated with the integrated circuit actions. A model of power use is generated for the semiconductor device, the model based at least in part on the recognized integrated circuit actions and the associated values. Based on an output of the model, power use of the semiconductor device is modified. Other methods and systems related to determining, modeling, and predicting power/energy use of semiconductor devices are also disclosed.