Time-based power analysis
Systems, machine readable media and methods are described for analyzing one or more physical systems using techniques that recognize patterns in underlying data and use the patterns to efficiently compute outputs using the patterns to reduce computations. The physical systems can be simulated with an estimation (e.g., an estimated power versus time waveform) that can be efficiently computed and then the estimation can be analyzed to detect patterns in the data. The detected patterns can each be analyzed with, in one embodiment, higher accuracy than the estimation to provide data that can be combined across multiple instances of each pattern to provide a higher accuracy evaluation of the system with a lower computational overhead.
1. A non-transitory machine readable medium storing executable instructions which when executed by a data processing system cause the data processing system to perform a method for performing an analysis of a circuit, the method comprising:
generating a representation of a power waveform that is an estimation of power consumption by a design of an integrated circuit (IC) over a period of time, the estimation based on switching activities in the design of the IC over the period of time;
recognizing one or more patterns in the representation of the power waveform, each of the one or more patterns being a unique pattern;
computing power consumed for each unique pattern to derive a set of pattern based power consumptions;
computing power consumed in the design of the IC based on a weighted sum of the set of pattern based power consumptions;
receiving a revised design of the IC to reduce power consumption in response to the computed power consumed exceeding a desired goal;
providing data about the revised design of the IC to a manufacturing facility to fabricate the IC.
2. The non-transitory machine readable medium as in claim 1 , wherein the method further comprises: filtering the power waveform with one or more low pass filters to smooth the power waveform to create a smoothed power waveform.
3. The non-transitory machine readable medium as in claim 2 , wherein the method further comprises: quantizing values in the power waveform to discretize power values in the power waveform into quantized bins.
4. The non-transitory machine readable medium as in claim 1 , wherein the method further comprises:
identifying a set of one or more steady state portions of the power waveform;
identifying a set of one or more transient portions of the power waveform.
5. The non-transitory machine readable medium as in claim 4 , wherein the method further comprises one or both of:
determining, within each of the steady state portions, a number of instances of each unique pattern and wherein the power consumed in the IC is computed based on a weighted sum of the number of instances in the steady state portions; and
determining, within each of the transient portions, a number of instances of each unique pattern and wherein the power consumed in the IC is computed based on a weighted sum of the number of instances in the transient portions.
6. The non-transitory machine readable medium as in claim 1 , wherein the one or more patterns are recognized within assigned time periods based on algorithms that use one or more inputs comprising: one or more statistical values, a number of peaks within a time period, a number of troughs in the time period, locations within the time period of the peaks and troughs, durations of the peaks and troughs, and magnitudes of peaks and troughs.
7. The non-transitory machine readable medium as in claim 1 , wherein the power consumed for each unique pattern is computed with more accuracy than the estimation.
8. The non-transitory machine readable medium as in claim 7 , wherein the power waveform is estimated from a toggle simulation data of populated nets in a netlist and a load capacitance of each of the populated nets.
9. A machine implemented method for performing an analysis of a circuit, the method comprising:
generating a representation of a power waveform that is an estimation of power consumption by a design of an integrated circuit (IC) over a period of time, the estimation based on switching activities in the design of the IC over the period of time;
recognizing one or more patterns in the representation of the power waveform, each of the one or more patterns being a unique pattern;
computing power consumed for each unique pattern to derive a set of pattern based power consumptions;
computing power consumed in the design of the IC based on a weighted sum of the set of pattern based power consumptions;
receiving a revised design of the IC to reduce power consumption in response to the computed power consumed exceeding a desired goal;
providing data about the revised design of the IC to a manufacturing facility to fabricate the IC.
10. The method as in claim 9 , wherein the method further comprises: filtering the power waveform with one or more low pass filters to smooth the power waveform to create a smoothed power waveform.
11. The method as in claim 10 , wherein the method further comprises:
quantizing values in the power waveform to discretize power values in the power waveform into quantized bins.
12. The method as in claim 9 , wherein the method further comprises:
identifying a set of one or more steady state portions of the power waveform;
identifying a set of one or more transient portions of the power waveform.
13. The method as in claim 12 , wherein the method further comprises one or both of:
determining, within each of the steady state portions, a number of instances of each unique pattern and wherein the power consumed in the IC is computed based on a weighted sum of the number of instances in the steady state portions; and
determining, within each of the transient portions, a number of instances of each unique pattern and wherein the power consumed in the IC is computed based on a weighted sum of the number of instances in the transient portions.
14. The method as in claim 9 , wherein the one or more patterns are recognized within assigned time periods based on algorithms that use one or more inputs comprising: one or more statistical values, a number of peaks within a time period, a number of troughs in the time period, locations within the time period of the peaks and troughs, durations of the peaks and troughs and magnitudes of peaks and troughs.
15. The method as in claim 9 , wherein the power consumed for each unique pattern is computed with more accuracy than the estimation.
16. The method as in claim 15 , wherein the power waveform is estimated from a toggle simulation data of populated nets in a netlist and a load capacitance of each of the populated nets.