Possible Causes of Discrepancies
While Comparing the Results of
Averaged and Time-Based Power
Analysis
Version 1.0
Copyright Notice and Proprietary Information
© 2023 Synopsys, Inc. This Synopsys software and all associated documentation are proprietary to Synopsys, Inc.
and may only be used pursuant to the terms and conditions of a written license agreement with Synopsys, Inc. All
other use, reproduction, modification, or distribution of the Synopsys software or the associated documentation is
strictly prohibited.
Destination Control Statement
All technical data contained in this publication is subject to the export control laws of the United States of America.
Disclosure to nationals of other countries contrary to United States law is prohibited. It is the reader's responsibility to
determine the applicable regulations and to comply with them.
Disclaimer
SYNOPSYS, INC., AND ITS LICENSORS MAKE NO WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, WITH
REGARD TO THIS MATERIAL, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
Trademarks
Synopsys company and certain product names are trademarks of Synopsys, as set forth at
http://www.synopsys.com/Company/Pages/Trademarks.aspx.
All other product or company names may be trademarks of their respective owners.
Third-Party Links
Any links to third-party websites included in this document are for your convenience only. Synopsys does not endorse
and is not responsible for such websites and their practices, including privacy practices, availability, and content.
Synopsys, Inc.
690 E. Middlefield Road
Mountain View, CA 94043
www.synopsys.com
2
Contents
Introduction ...................................................................................................................................... 4
• Propagation ............................................................................................................................. 4
• SDPD Estimation .................................................................................................................... 4
• Default Annotation................................................................................................................... 5
• X-State / Z-State Handling ...................................................................................................... 5
• Glitch Power ............................................................................................................................ 5
Summary ......................................................................................................................................... 5
3
Introduction
Designers with VCD activity files can analyze the power consumption of their design in
PrimePower, using either the time-based or averaged analysis modes. If gate-level VCD is
available, users will generally perform time-based analysis to analyze the power over time. If RTL
VCD is provided, users will perform either cycle-accurate or averaged power analysis. For a given
VCD file, the results produced by the different modes, may differ slightly due to one or more of the
factors listed below.
• Propagation
To calculate the total power consumption, both averaged and time-based modes require
activity on all the nets in the design. If the design is not fully annotated – as is the case for
RTL VCD files; PrimePower must propagate to obtain activity on the un-annotated nets.
The propagation mechanism differs between the two modes.
In averaged mode, PrimePower uses the toggle rate and static probability of the annotated
nets, to create random vectors which match the given parameters. Zero-delay simulation
of these random vectors is used to derive the toggle rates and static probabilities of the
unannotated nets. If the design is highly correlated the power results will be less accurate.
For example, propagation of activity through XOR trees may show excessive power in
averaged mode, because the data inputs in the design are correlated, while the randomly
generated vectors used for propagation are not.
In time-based, cycle-accurate mode, PrimePower uses zero-delay simulation of the actual
VCD vectors, to determine the activity on the un- annotated nets, per cycle. As such, the
time-correlation between the signals is maintained, thereby producing more accurate
toggle rates and static probabilities on the un-annotated nets.
• SDPD Estimation
Both averaged and time-based modes calculate the internal power of the cells in the design,
by accessing the SDPD (state-dependent, path dependent) library power arcs.
In averaged mode, however, the SDPD toggle rates are estimated because the time-
correlation in the original VCD is lost. Even if the cell is fully annotated, PrimePower must
estimate the toggle rates of the power arcs because the exact event sequence of the VCD
is not available to
identify exactly how often a particular output toggled because of transitions on one input
signal over another. Since PrimePower understands the functionality of combinational and
sequential cells, the SDPD estimated toggle rates are very close to that obtained from
time- based analysis. For black box cells such as memories, however, the SDPD
estimation is less accurate. Without the functionality, PrimePower SDPD toggles
estimation is based purely on the pin activity. This can cause discrepancies in the power
consumption of black box models, if there are large differences in the consumption per arc.
With cycle-accurate analysis, PrimePower processes every event in the VCD file, thereby
maintaining the sequence of events. In this mode, PrimePower can identify the related pin,
and access the appropriate SDPD power arc when calculating the power per cycle.
4
• Default Annotation
In averaged mode, PrimePower will apply default annotation to propagation starting points
– namely primary inputs, black-box outputs and tri-states if they are not available in the
VCD file. These nets are assigned a default annotation based on the default toggle rate
and their related clocks.
In time-based mode, these un-annotated starting points are assumed to have a logic value
of U. They never toggle and the value is not propagated.
Most activity files contain annotation for propagation starting points.
• X-State / Z-State Handling
X states are handled differently in time-based vs. averaged power analysis modes.
In averaged mode, X and Z states are not propagated. When the VCD is converted to SAIF,
transitions to and from these states are considered as glitches which are counted as ½ of a
normal toggle. When calculating the toggle rate for nets with X and Z states, PrimePower
includes the glitch toggles when calculating the toggle rate. This toggle rate is then used to
generate the random vectors for propagation of activity.
In time-based mode, X states are propagated and classified as X transition power.
Transitions to and from Z states are considered a full transition if the original and final logic
values differ, and ignored if they are the same.
Only in designs with a high percentage of X-states, is this a factor.
• Glitch Power
The metric for measuring glitch power differs between the time-based and averaged modes.
As noted in the earlier section, transitions to and from the X state and Z states are
considered as “glitch” toggles in averaged mode, and ultimately considered as part of the
total power.
In cycle-accurate mode, glitches are defined as extraneous activity; wherein there are
multiple toggles on a net within a given cycle. The glitch power is estimated based on the
VCD and propagated activity.
Since there is no comparable metric in averaged mode, including the estimated glitch power
in time-based mode can cause the total power to differ from that calculated in averaged
mode.
Summary
The primary cause of discrepancies in the power consumption in averaged and cycle-accurate
analysis is due to the differences in the activity propagation. The time-based flow propagates the
actual VCD events, while the averaged flow propagates random vectors generated with
equivalent toggle rates and static probabilities. The remaining factors generally affect only very
small portions of the total design and therefore have a lesser impact on the total power
consumption.
