PrimeTime® VX

User Guide
Version B-2008.12, December 2008

Comments?
Send comments on the documentation by going
to http://solvnet.synopsys.com, then clicking
“Enter a Call to the Support Center.”
Copyright Notice and Proprietary Information
Copyright © 2008 Synopsys, Inc. All rights reserved. This software and documentation contain confidential and proprietary
information that is the property of Synopsys, Inc. The software and documentation are furnished under a license agreement and
may be used or copied only in accordance with the terms of the license agreement. No part of the software and documentation may
be reproduced, transmitted, or translated, in any form or by any means, electronic, mechanical, manual, optical, or otherwise,
without prior written permission of Synopsys, Inc., or as expressly provided by the license agreement.
Right to Copy Documentation
The license agreement with Synopsys permits licensee to make copies of the documentation for its internal use only.
Each copy shall include all copyrights, trademarks, service marks, and proprietary rights notices, if any. Licensee must
assign sequential numbers to all copies. These copies shall contain the following legend on the cover page:
“This document is duplicated with the permission of Synopsys, Inc., for the exclusive use of
__________________________________________ and its employees. This is copy number __________.”

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.
Registered Trademarks (®)
Synopsys, AMPS, Astro, Behavior Extracting Synthesis Technology, Cadabra, CATS, Certify, Design Compiler,
DesignWare, Formality, HDL Analyst, HSPICE, Identify, iN-Phase, Leda, MAST, ModelTools, NanoSim, OpenVera,
PathMill, Physical Compiler, PrimeTime, SCOPE, Simply Better Results, SiVL, SNUG, SolvNet, Synplicity, Synplify,
Synplify Pro, Synthesis Constraints Optimization Environment, TetraMAX, the Synplicity logo, VCS, Vera, and
YIELDirector are registered trademarks of Synopsys, Inc.
Trademarks (™)
AFGen, Apollo, Astro-Rail, Astro-Xtalk, Aurora, AvanWaves, BEST, Columbia, Columbia-CE, Confirma, Cosmos,
CosmosLE, CosmosScope, CRITIC, DC Expert, DC Professional, DC Ultra, Design Analyzer, Design Vision,
DesignerHDL, DesignPower, Direct Silicon Access, Discovery, Eclypse, Encore, EPIC, Galaxy, Galaxy Custom Designer,
HANEX, HAPS, HapsTrak, HDL Compiler, Hercules, Hierarchical Optimization Technology, High-performance ASIC
plus
Prototyping System, HSIM, HSIM , i-Virtual Stepper, IICE, in-Sync, iN-Tandem, Jupiter, Jupiter-DP, JupiterXT,
JupiterXT-ASIC, Liberty, Libra-Passport, Library Compiler, Magellan, Mars, Mars-Rail, Mars-Xtalk, Milkyway,
ModelSource, Module Compiler, MultiPoint, Physical Analyst, Planet, Planet-PL, Polaris, Power Compiler, Raphael,
Saturn, Scirocco, Scirocco-i, Star-RCXT, Star-SimXT, System Compiler, System Designer, Taurus, TotalRecall,
TSUPREM-4, VCS Express, VCSi, VHDL Compiler, VirSim, and VMC are trademarks of Synopsys, Inc.

Service Marks (SM)

MAP-in, SVP Café, and TAP-in are service marks of Synopsys, Inc.

SystemC is a trademark of the Open SystemC Initiative and is used under license.
ARM and AMBA are registered trademarks of ARM Limited.
Saber is a registered trademark of SabreMark Limited Partnership and is used under license.
All other product or company names may be trademarks of their respective owners.

PrimeTime VX User Guide, version B-2008.12

ii
Contents

What’s New in This Release . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

About This User Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
Customer Support. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

1. Overview
Variation-Aware Timing Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2
Handling Variation by Considering Corner Cases . . . . . . . . . . . . . . . . . . . . . . . 1-2
Handling Variation by Derating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-4
Handling Variation With Distribution Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-6
Variation-Aware and Corner-Case Min/Max Analysis . . . . . . . . . . . . . . . . . . . . 1-10
Choosing Parameters for Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-13
Analysis Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-14
Enabling and Using PrimeTime VX Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-22
Variation Analysis Modes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-22
Libraries, Variations, and Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-23
Quantile and Distribution Reporting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-23
Variation and Crosstalk Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-26

2. Libraries and Interconnect Parameters

Variation-Aware Library Creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2
Using Unified Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3
Full-Chip Variation-Aware Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3
Path-Based-Only Variation-Aware Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-4

iii
 Using Separate (2N+1) Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-6
Full-Chip Variation-Aware Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-6
Path-Based-Only Variation-Aware Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-8
Interconnect Parameter Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-9
Interconnect Parameter Calculations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-10
Reading Interconnect Parameters With Variation . . . . . . . . . . . . . . . . . . . . . . . 2-11
Interconnect Parameters With Temperature Variation . . . . . . . . . . . . . . . . . . . . 2-12
Interconnect Parasitic Corners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-12

3. Variations and Correlations

Parameter Variation and Correlation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2
Creating and Setting Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2
create_variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-3
Normal Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-4
Piecewise-Linear Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-5
Empirical Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-6
Constant Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-6
Uniform Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-6
Lognormal Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-7
Discrete Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-8
set_variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-8
Creating and Setting Correlations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-9
Auto-Correlation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-10
Cross-Correlation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-12
Spatial Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-12
Math Operations on Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-14
Derating Statistical Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-15
Reporting Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-17
Variation Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-18
Variation-Derived Scalar Attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-20
Variations in Timing Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-21
Variations in Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-22
Variations in Nets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-23
Parameter-Specific Net Sensitivity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-24
Variations in Library Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-25

iv
 Variations in Design Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-26
Variations in Parasitic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-27

4. Variation-Aware Timing Reports

Path Timing Report Increments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-2
Clock Timing Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-3
Custom Statistical Reporting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
Variation Distribution Attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-7
Attribute Access Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-9
Session Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-9
Validation Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-12

Glossary

Index

v
vi
Preface FIX ME!

This preface includes the following sections:

• What’s New in This Release

• About This User Guide
• Customer Support

vii
What’s New in This Release
Information about new features, enhancements, and changes, along with known problems
and limitations and resolved Synopsys Technical Action Requests (STARs), is available in
the PrimeTime VX Release Notes in SolvNet.

To see the PrimeTime VX Release Notes,

1. Go to the Download Center on SolvNet located at the following address:

http://solvnet.synopsys.com/DownloadCenter

If prompted, enter your user name and password. If you do not have a Synopsys user
name and password, follow the instructions to register with SolvNet.

2. Select PrimeTime Suite, and then select a release in the list that appears.

About This User Guide

This user guide describes the features and usage flow of PrimeTime VX, a tool that adds
variation-aware analysis capabilities to PrimeTime. A variation-aware analysis takes into
consideration the statistical distribution (rather than simple minimum-typical-maximum
values) of process parameters, thereby increasing the accuracy of the analysis.

Audience
This user guide is for engineers who use PrimeTime for static timing analysis. Prior
knowledge and experience with statistics and probability will be helpful in understanding the
concepts presented.

Related Publications
For additional information about PrimeTime VX, see Documentation on the Web, which is
available through SolvNet at the following address:

http://solvnet.synopsys.com/DocsOnWeb

Preface
viii
You might also want to refer to the documentation for the following related Synopsys
products:

• PrimeTime and PrimeTime SI

• Star-RCXT
• Library Compiler
• Liberty NCX

Preface
ix
 Conventions
The following conventions are used in Synopsys documentation.

Convention Description

Courier Indicates command syntax.

Courier italic Indicates a user-defined value in Synopsys syntax,

such as object_name. (A user-defined value that is
not Synopsys syntax, such as a user-defined value
in a Verilog or VHDL statement, is indicated by
regular text font italic.)

Courier bold Indicates user input—text you type verbatim—in

Synopsys syntax and examples. (User input that is
not Synopsys syntax, such as a user name or
password you enter in a GUI, is indicated by regular
text font bold.)

[] Denotes optional parameters, such as

pin1 [pin2 ... pinN]

| Indicates a choice among alternatives, such as

low | medium | high
(This example indicates that you can enter one of
three possible values for an option:
low, medium, or high.)

_ Connects terms that are read as a single term by

the system, such as
set_annotated_delay

Control-c Indicates a keyboard combination, such as holding

down the Control key and pressing c.

\ Indicates a continuation of a command line.

/ Indicates levels of directory structure.

Edit > Copy Indicates a path to a menu command, such as

opening the Edit menu and choosing Copy.

Preface
x
Customer Support
Customer support is available through SolvNet online customer support and through
contacting the Synopsys Technical Support Center.

Accessing SolvNet
SolvNet includes an electronic knowledge base of technical articles and answers to
frequently asked questions about Synopsys tools. SolvNet also gives you access to a wide
range of Synopsys online services including software downloads, documentation on the
Web, and “Enter a Call to the Support Center.”

To access SolvNet, go to the SolvNet Web page at the following address:

http://solvnet.synopsys.com

If prompted, enter your user name and password. If you do not have a Synopsys user name
and password, follow the instructions to register with SolvNet.

If you need help using SolvNet, click HELP in the top-right menu bar or in the footer.

Contacting the Synopsys Technical Support Center

If you have problems, questions, or suggestions, you can contact the Synopsys Technical
Support Center in the following ways:

• Open a call to your local support center from the Web by going to http://
solvnet.synopsys.com (Synopsys user name and password required), and then clicking
“Enter a Call to the Support Center.”
• Send an e-mail message to your local support center.
- E-mail [email protected] from within North America.
- Find other local support center e-mail addresses at
http://www.synopsys.com/support/support_ctr
• Telephone your local support center.
- Call (800) 245-8005 from within the continental United States.
- Call (650) 584-4200 from Canada.
- Find other local support center telephone numbers at
http://www.synopsys.com/support/support_ctr

Preface
xi
Preface
xii
1
Overview 1
PrimeTime VX adds variation-aware analysis capabilities to PrimeTime. A variation-aware
analysis increases the accuracy of timing analysis by considering the statistical distribution
of process, voltage, and temperature parameters.

This chapter contains the following sections:

• Variation-Aware Timing Analysis

• Choosing Parameters for Analysis
• Analysis Example
• Enabling and Using PrimeTime VX Analysis

1-1
Variation-Aware Timing Analysis
PrimeTime VX increases the accuracy of timing analysis in PrimeTime by considering the
statistical variation and distribution of process, voltage, and temperature parameters.
PrimeTime VX accurately determines the timing behavior of a circuit under varying
parameters, including measured parameters such as channel length and threshold voltage,
process parameters such as implant dose or gate oxidation temperatures, or parameters
derived by principal component analysis from other parameters. Given a set of variation-
aware cell libraries and the distribution of parameter values, PrimeTime VX analyzes path
timing behavior under all combinations of those parameters.

Variation-aware analysis is more accurate than conventional minimum/maximum analysis

because it takes into account the actual distributions of parameters instead of simply
combining the absolute worst-case combinations of parameters. Instead of analyzing a
limited number of corner cases, PrimeTime VX covers all combinations of parameters,
through the full range of parameter values. The calculated slack distributions more
accurately reflect the true results obtained in silicon, resulting in less pessimism in the
analysis.

PrimeTime VX can handle timing analysis for any type of parameter, as long as the
dependency of circuit behavior on that parameter is known and can be characterized.

Handling Variation by Considering Corner Cases

To calculate the delay of a path, PrimeTime must first calculate the delays of the cells and
net parasitics in the path. To calculate the delay of a cell, PrimeTime considers the library
model for the cell and the cell environment: the input slew, output load, process, voltage, and
temperature. Thus, the delay is a function of these variables:

delay = f1(slew_in, load, P, V, T)

PrimeTime also calculates the output slew, which is needed for delay calculation of the next
stage of the timing path:

slew_out = f2(slew_in, load, P, V, T)

The P (process) variable represents all the process-related parameters that can affect the
delay, such as channel length and threshold voltage. To determine the circuit timing under
varying process parameters using conventional min/max analysis, you must characterize the
delays at the most extreme values for those parameters and put that information into a set
of libraries, one library per operating point.

Chapter 1: Overview
1-2
 For example, suppose that the statistical distribution of channel lengths for a process
technology is a curve like the one shown in Figure 1-1. To get the worst-case behavior for
this distribution, you would choose the minimum and maximum ends of the range and
determine the effects at those points.

Figure 1-1 Statistical Distribution of a Process Parameter

Number of typical
Samples actual distribution

min max

Channel length L

If another process parameter has significant variation effects, it must be considered as well.
For example, to consider both channel length and threshold voltage, you might consider a
two-dimensional plot like the one shown in Figure 1-2.

Figure 1-2 Corner Cases Using Two Parameters

“slow” corner: max L, max VT
typical
Threshold
Voltage VT

actual distribution
(darker = more likely)

“fast” corner: min L, min VT

Channel Length L

Variation-Aware Timing Analysis

1-3
 The set of conditions at the most extreme values for two or more different parameters is
called a corner case. For example, in Figure 1-2, the “fast” corner is the place where the
channel length and threshold voltage are both at their minimum values. Similarly, the “slow”
corner is the place where the channel length and threshold voltage are at their maximum
values. This principle can be extended to any number of parameters that have significant
effects.

A typical library set might consist of three libraries characterized at three points: slow, typical,
and fast. However, this analysis is pessimistic because it considers the combination of the
most extreme parameter values, which is very unlikely to occur in practice.

For better accuracy, all the corner cases should be considered. A parameter that affects cell
delays in one way (such as fast) might affect net delays in a different way (such as slow). So
you might consider four corner cases for two parameters, eight corner cases for three
parameters, and so on.

In the past, using a few corner cases has often been adequate due the fairly limited amount
of variation in process parameters. However, in current deep-submicron process
technologies, variations in process parameters are becoming more significant, causing
corner cases to become more pessimistic and making it increasingly difficult to accurately
manage multiple parameter variations.

Handling Variation by Derating

One way to manage small variations in parameters is by derating the corner cases, also
known as “scaling” or “multiplying by K factors.” Figure 1-3 demonstrates how derating can
be used to analyze circuit behavior with known variations in parameters.

Chapter 1: Overview
1-4
Figure 1-3 Derating Delays for Different Channel Lengths

Gate delay D = f(L)

max D
Slope here
typ D determines
derating factor

min D
distribution
of L

Channel
min L typ L max L length L

In this example, the variation in gate delay is a function of transistor channel length. For
simplicity, variations in other parameters that might affect gate delay are ignored.

Characterization of the technology finds that the gate delay varies with channel length as
shown by the curve at the top of Figure 1-3. You create three libraries called min, typ, and
max that have the gate delays min D, typ D and max D. These libraries can be used in
PrimeTime to determine the circuit behavior at the minimum, typical, and maximum channel
lengths.

You want to determine the circuit timing when the channel length is a little less than max L,
as represented by the dotted line in the figure. You can get fairly good results without
generating a new library by specifying a derating factor with the max L library. The slope of
the curve at the maximum operating point gives a good approximation of delay as a function
of L, as long as the value of L is not too far away from max L. Thus, you use the max L library
and apply the derating factor in PrimeTime with the set_timing_derate command. A
separate analysis run is necessary for each derating factor chosen for analysis.

Similarly, you can apply different derating factors to the typ and min libraries to get good
results for values of L near typ L and min L. Expanding derating to two or more parameters
increases the scope of the analysis to areas surrounding the typical case and corner cases,
as illustrated in Figure 1-4.

Variation-Aware Timing Analysis

1-5
Figure 1-4 Corner Cases Using Two Parameters

typical
Threshold “slow” derating coverage
Voltage VT

“typical” derating coverage

“fast” derating coverage

Channel Length L

Because of the nonlinear dependence of D on L, this technique works well only when the
actual value of L is near the minimum, typical, or maximum value of L.

Handling Variation With Distribution Data

In a variation-aware analysis, PrimeTime VX considers the actual distribution of parameter
values, rather than just the minimum and maximum extremes. It considers the functional
dependence of delay, slew, and other circuit behavior on the parameter, rather than just the
few characterization points contained in individual libraries.

For example, in an analysis that considers the effects of channel length on gate delays, you
provide information on the distribution of channel lengths in PrimeTime VX. Then PrimeTime
VX considers the functional dependence of delay on channel lengths and calculates the
distribution of gate delays and slews, and from that information, calculates the distribution of
required times, arrival times, and slacks in the circuit.

To specify the functional dependence of delay or slew on a parameter, the cell libraries need
to characterize the behavior at the typical value and one or two nearby values. For variations
that have highly linear delay and slew response, just the typical data point and one additional
data point, either above or below the typical point, provide sufficient accuracy. An example
is shown in Figure 1-5.

Chapter 1: Overview
1-6
Figure 1-5 Delay as a Function of Parameter P

Gate Delay D = f(P)

Distribution of D

typ D

Distribution
of P

Parameter
typ higher P

Given the distribution of the parameter P, PrimeTime VX calculates the distribution of delays
continuously throughout the range of values, using linear interpolation and extrapolation of
the library-defined functional operating points. In this example, interpolation and
extrapolation between the typical data point and a single data point characterized at the high
end of the distribution provide good accuracy throughout the expected range of the
parameter.

If the parameter variation has a strong effect on delay or slew and is significantly nonlinear,
you should characterize two data points, one on each side of the typical data point to get
higher accuracy throughout the variation range, as shown in Figure 1-6.

Variation-Aware Timing Analysis

1-7
Figure 1-6 Delay as a Function of Parameter Q

Gate Delay D = f(Q)

Distribution of D

typ D

Distribution
of Q

Parameter
lower typ higher Q

The surrounding data points should be close enough to the typical operating point to ensure
good accuracy at the middle of the distribution, where the actual parameter value is most
likely to occur, but also far enough away to get good accuracy at the more extreme parameter
values, where timing violations are most likely to occur.

For within-die variation, characterization of data points located one standard deviation (1σ)
away from the typical value is recommended, to get the best probable average accuracy
along the curve. For die-to-die variations, characterization of data points located three
standard deviations (3σ) away from the typical value is recommended, to get better accuracy
at the tail ends of the distribution where violations are more likely to occur.

To perform variation-aware analysis with two parameters, you must characterize the timing
behavior at three, four, or five operating points: the typical data point, plus one or two more
data points for the first parameter (with the second parameter at its typical value), plus one
or two more data points for the second parameter (with the first parameter at its typical value.
For example, for two parameters L and Vt, you might choose three points like the ones shown
in Figure 1-7 or five points like the ones shown in Figure 1-8.

Chapter 1: Overview
1-8
Figure 1-7 Three Characterization Points for Two Parameters

Threshold
Voltage VT
higher VT

higher
L

typical L and typical VT

Channel Length L

Figure 1-8 Five Characterization Points for Two Parameters

Threshold
Voltage VT
higher VT

lower higher
L L

typical L and typical VT

lower VT

Channel Length L

Variation-Aware Timing Analysis

1-9
 The same principle applies to three or more parameters. If N is the total number of
parameters under consideration, the number of characterization points must be at least N+1,
consisting of the typical data point and plus one additional data point for each parameter; or
it can be as many as 2N+1, consisting of the typical data point plus a lower and a higher
operating point for each parameter.

Characterization of the typical value plus one or two surrounding values for each parameter
is sufficient for accurate timing analysis throughout the expected variation ranges, while
requiring a minimum number of data points to be characterized. Any variations or changes
in parameter values can be handled quickly in PrimeTime VX, avoiding the slow process of
recharacterization and library generation.

Variation-Aware and Corner-Case Min/Max Analysis

Figure 1-9 and Figure 1-10 compare the analysis flows using conventional and variation-
aware timing analysis.

With conventional corner-case analysis, a library must be generated for each corner case.
Typically, at least three libraries are used: minimum, typical, and maximum libraries. At least
three runs are required for a full analysis, one using each of the three libraries. Parameter
variation can be handled by a limited extent by derating, although deciding on an appropriate
derating factor can be difficult. In the end, PrimeTime reports the amount of slack under the
worst-case corner conditions, possibly adjusted by derating.

Figure 1-9 Conventional Corner-Case Timing Analysis

Star-RCXT PrimeRail

Derating Process-related wire Instance-specific

corner conditions voltage annotations

Corner-Case PrimeTime
Libraries (.db)

PVT corner cases

Slack timing
report

Constraints Worst-case slack = 0.8 ns

(SDC) Delay, slew, arrival, required-time min/max values

Chapter 1: Overview
1-10
Figure 1-10 Variation-Aware Timing Analysis

Star-RCXT VX PrimeRail
Parameter
Distributions
Temperature, Voltage
P distributions process, and annotations
wire sensitivity

Variation-Aware PrimeTime VX
Libraries (.db)

PVT dependencies Slack timing

report
Probability

Constraints
(SDC) 1% quantile slack = 1.1 ns
Delay, slew, arrival, required-time,
and slack distributions
1.1
Path slack

With variation-aware analysis, a set of libraries is generated to describe dependence of

delay and slew as a function of the parameters under consideration. PrimeTime VX
considers the full range and distribution of the parameters of interest. In the end, PrimeTime
VX reports the delay, slew, required time, arrival time, and slack based on quantiles of the
value distributions.

When you run a path-based analysis with variation, PrimeTime VX maintains all the
statistical information in memory. Then you can display the statistical results graphically and
query the variation results with the get_attribute command to obtain the mean, standard
deviation, and quantile values for each type of timing result.

The worst-case slack reported by conventional min-max analysis is usually worse (more
pessimistic) because it combines all the most extreme, worst-case delays from each stage
in the path, whereas the variation-aware analysis statistically combines the distribution of
delays from each stage. Figure 1-11 shows how PrimeTime calculates the cumulative delay
along this path using conventional min/max analysis, while Figure 1-12 shows how it
calculates the same cumulative delay using variation-aware analysis.

Variation-Aware Timing Analysis

1-11
Figure 1-11 Conventional Addition of Min/Max Gate Delays

min-max
gate delay min typ min typ max
max min typ max min typ max
.008 .010 .013 .008 .010 .013 .008 .010 .013 .008 .010 .013

cumulative min-max
delay of path segment min typ max
.032 .040 .052
reported reported
min delay max delay

Figure 1-12 Variation-Aware Accumulation of Gate Delays

variation-aware
distribution of
gate delay mean = .010 mean = .010 mean = .010 mean = .010
σ, skewness σ, skewness σ, skewness σ, skewness

combined variation-aware q01 q99

delay of path segment
mean = .042 reported
reported
σ, skewness max delay
min delay

Chapter 1: Overview
1-12
 For this example, we consider only the gate delays and ignore the net delays. Assume also
that the variation in gate delays is random and not correlated between different gates in the
path.

Note:
In this manual, “correlated” means “statistically dependent” and “uncorrelated” means
“statistically independent.”

Using conventional min-max analysis as shown in Figure 1-11, PrimeTime adds the
minimum delays to obtain the total minimum delay of the path segment, and it adds the
maximum delays to obtain the total maximum delay of the path segment. This gives the
worst-case extremes of total delay, without information about the relative likelihood between
these two extremes.

Using variation-aware analysis as shown in Figure 1-12, PrimeTime VX combines the delays
statistically to get the distribution of the total delay for the path segment. This results in a
more realistic, less pessimistic model of total delay for the path segment.

Using variation-aware analysis, for reporting and comparison purposes, PrimeTime VX

considers a low quantile point of the distribution as the worst-case low value, and a high
quantile point of the distribution as the worst-case high value. By default, the 0.135 percent
and 99.865 percent quantiles are used to report the minimum and maximum values,
respectively, corresponding to the –3*sigma and +3*sigma points of a perfectly normal
distribution. You can specify other quantiles (for example, 5 percent and 95 percent) by using
the set_variation_quantile command.

Choosing Parameters for Analysis

In choosing variable parameters for analysis, you should focus on the few dominant
parameters that have the most variation and the largest effects on delay and slack. There
are three main types of parameters that you can characterize and analyze:

• Measured parameters such as channel length and threshold voltage

• First-level process parameters such as implant dose and gate oxidation temperature
• Uncorrelated parameters derived from first-level process parameters by principal
component analysis
PrimeTime VX can accept any type of parameter that can be characterized, including
derived parameters created by principal component analysis. A derived parameter is a
mathematical representation of multiple correlated parameters. Each derived parameter has
no physical meaning by itself. The process of principal component analysis transforms a
number of correlated parameters into a smaller number of uncorrelated parameters. The

Choosing Parameters for Analysis

1-13
 new parameters created by this process are called the principal components of the larger set
of parameters. In this way, a large number of correlated parameters can be reduced to a few
principal components.

A process technology can be characterized by first analyzing the large number of process
parameter variations, followed by principal component analysis to reduce those variations to
a few uncorrelated components. These components can be used in a variation-aware
analysis to account for variations in the analyzed process parameters.

PrimeTime VX can handle multiple components per parameter. For example, a channel-
length variation can be written as a sum of uncorrelated terms such as the following:

Ltotal = Lref + DLfab + DLlot + DLwafer + DLdie

Each term can be further decomposed into a set of uncorrelated components:

DLfab = DLfab1 + DLfab2 + DLfab3

The reference component is the primary component that carries the mean value of the
parameter variation. The other components represent the variation in the parameter that can
be attributed to various causes, such as the fab-to-fab variation, lot-to-lot variation from a fab,
wafer-to-wafer variation in a lot, die-to-die variation in a wafer, within-die variation, and so on.
Each variation component has a different form and magnitude. For example, the lot-to-lot
variation might be larger than the wafer-to-wafer variation. The mean value of each variation
component is zero.

The reference component value is specified in the library, whereas the variation components
are specified in the PrimeTime VX environment. Therefore, you can change the variation
components and analyze the results with a given set of libraries, without going through the
time-consuming process of generating a new set of libraries for each variation change. In
PrimeTime VX, the command for invoking a library and its reference component values is
set_variation_library, using the -reference_value and -values options. The
command for specifying a variation component is set_variation.

Analysis Example
The following example demonstrates typical usage of variation-aware analysis in a
PrimeTime VX session, including the display of arrival time and slack variations in the
PrimeTime graphical user interface (GUI). In this example, the analysis considers the
variation of three process parameters and their effects on delay, slew, and slack.

The flow in this example consists of the following steps in PrimeTime VX:

1. Read in and link the design.

Chapter 1: Overview
1-14
2. Enable variation-aware analysis and set the variation-aware libraries used for analysis.
These libraries characterize cell behavior at different parameter values.
3. Specify the variation and correlation characteristics of the characterized parameters in
the design.
4. Set the timing constraints (clocks, input delays, output delays, and input transition times).
5. Read in the parasitics with variation data. You can either choose a simple normal
distribution for the characterized parameters or explicitly specify the variation and
correlation characteristics.
6. Run update_timing and generate timing reports for the most critical timing paths
based on variation analysis.
7. If you want to further reduce any remaining pessimism in the analysis, run a path-based
variation analysis of the critical paths. You can get detailed statistical results by reading
the timing reports, by viewing graphs and plots in the GUI, and by reading statistical
values directly from the design with the get_attribute command.
Figure 1-13 shows the GUI Timing Analysis Driver window containing a list of the worst-slack
maximum-delay paths in the design. The paths are sorted in order of increasing slack. The
worst path has a reported slack of –0.19 time units. PrimeTime VX calculates the path slack
as a variation, but for listing and comparison purposes, it uses the low quantile of the
variation (0.00135 by default).

Analysis Example
1-15
Figure 1-13 Recalculated Path Table

Right-click

To get a report on the slack variations for the paths in the list, you select those paths, right-
click, and choose New Recalculated Path Comparison Table. PrimeTime VX recalculates the
selected paths and saves all the variation-aware timing parameters (arrival times, required
times, and slacks) for those paths. It displays the results in a table and a graph as shown in
Figure 1-14.

Chapter 1: Overview
1-16
Figure 1-14 Statistical View With Slack PDF Plot

In the slack probability density function (PDF) plot, the curve shows the distribution of slack
values for the path, given the specified variation in the characterized parameters. The total
area under the curve is 1.0. The cross-hatched area under the curve to the left of slack=0.0
represents the probability of timing failure for this constraint. In this case, the probability of
failure is 1.2 percent.

The check boxes below the plot let you display or not display the mean, standard deviation,
bound, and quantile points of the plot. Figure 1-15 shows the plot with all the markers
enabled. The markers are color-coded in the plot.

Analysis Example
1-17
Figure 1-15 Slack PDF Plot With Markers

The mean value is marked with a dark olive color. This particular distribution is nearly
symmetrical, so the mean is very close to the q50 or median value, marked in red. Multiple
standard deviation (s) values around the mean are marked in a light olive color. In this
example, essentially all of the curve lies between plus or minus three standard deviations
from the mean.

The quantile points marked in the distribution represent the arrival times at which the
probability of arrival at less than or equal to that time will be a given value: 1, 50, or 99
percent. For example, the 99 percent (q99) quantile point is at the time value 0.58, which
means that the probability of arrival at or before time = 0.58 is 99 percent.

An alternative type of plot of the same data is a cumulative distribution function (CDF). To
view a CDF plot, select CDF in the Curve Type pull-down option above the plot. See
Figure 1-16. The CDF plot shows the probability that the arrival time will be less than or equal
to each value.

Chapter 1: Overview
1-18
Figure 1-16 Slack CDF Plot

The CDF starts at zero for low values of arrival time and increases to 1.0 for high values. This
plot is the integral (cumulative area under the curve) of the PDF plot.

You can display multiple, overlapping PDF plots for different paths in the same graph. First
select the paths of interest in the Timing Analysis Driver window or by using the Select >
Paths From/Through/To. Then choose Timing > New Recalculated Path Comparison Table.
See Figure 1-17 for an example.

Analysis Example
1-19
Figure 1-17 Multiple PDF Plots in One Graph

You can show or hide each PDF curve by checking or unchecking the boxes in the Show
column of the path list. You can select a path by clicking on its entry in the list or on its PDF
curve in the graph. The PDF curve for the selected path is drawn in magenta and the rest
are drawn in blue. The statistical markers (mean, quantiles, and so on) are shown for the
single selected path only. If multiple paths are selected or no paths are selected, no
statistical markers are shown.

For a particular path, you can display the distributions of arrival times at different points along
the path. From the recalculated path comparison table, right-click the path of interest and
then choose New Path Pin Comparison Table. This displays the arrival PDFs for successive
points along the path, as shown in Figure 1-18. Note that the arrival distributions become
wider and flatter as you go further along the path.

Chapter 1: Overview
1-20
Figure 1-18 Arrival PDFs for Points Along a Path

To view statistical information such as the mean and standard deviation for the arrival time
at one point in the path, click on its distribution curve or select the corresponding driver or
receiver pin in the path pin list. In the foregoing figure, the arrival at pin U3351/I is selected.
The mean arrival time is at 0.50 time units.

To view the plot for just one point in the path, select all of the points, click the Hide Selected
button, and then click the Show box in the path pin list.

To view the slew rather than arrival distribution, select Slew in the Info Type pull-down option
above the plot. For example, Figure 1-19 shows the slew PDF plots for all the pins along the
path.

Analysis Example
1-21
Figure 1-19 Slew PDF Plot

Enabling and Using PrimeTime VX Analysis

Variation-aware analysis is enabled in PrimeTime VX by setting the
variation_enable_analysis variable to true. It is set to false by default. You must set
it to true before you use any variation-related commands. A PrimeTime VX license is
required.

Variation Analysis Modes

When the variation_enable_analysis variable is set to true, a variable called
variation_analysis_mode specifies the scope of variation-aware analysis. The variable
can be set to default or path_based_only.

In the default mode, PrimeTime VX performs variation-aware analysis for both full-chip
analysis and path-based analysis. In path-based analysis, you use the
get_timing_paths and get_recalculated_timing_paths commands to analyze a
selected set of paths with enhanced accuracy.

Chapter 1: Overview
1-22
In the path_based_only mode, PrimeTime VX performs conventional corner-case
analysis for full-chip analysis and variation-aware analysis for path-based analysis. For full-
chip analysis, PrimeTime uses the corner-case libraries specified by the link_path
variable. For path-based analysis, PrimeTime VX uses the variation libraries specified by the
set_variation_library command.

Libraries, Variations, and Correlations

Variation-aware libraries must be read into PrimeTime VX for variation-aware analysis.
These libraries are characterized at selected variation operating points, as illustrated in the
example shown in Figure 1-7 on page 9. Multiple variation-aware can be merged into a
single library to simplify the reading process, or they can be maintained and read in
separately. You specify the information about the variations and operating points in
PrimeTime VX by using the set_variation_library command. For details, see
Chapter 2, “Libraries and Interconnect Parameters.”

You specify the actual parameter values in PrimeTime VX by using the create_variation
command. This command specifies the parameter name, value, and mathematical
distribution of the parameter. To apply a variation to the whole design or a timing object in
the design, you use the set_variation command. The create_correlation and
set_variation_correlation commands specify the correlation relationship between
different instances of the same variation. For details, see Chapter 3, “Variations and
Correlations.”

Quantile and Distribution Reporting

The report_timing command reports the worst-case quantile values for arrival and slack
for a corner-case analysis, or statistical distributions of these values for a statistical analysis.
In the following example, a corner-case analysis is performed first, followed by reslection and
statisitical analysis of the critical path.

pt_shell> set variation_anlaysis_mode path_based_only

pt_shell> set_variation_quantile -quantile_low 0.01 -quantile_high 0.99
pt_shell> report_timing -delay_type max -group wb_clk_i
...
Startpoint: wishbone/rx_fifo/cnt_reg[0]
(rising edge-triggered flip-flop clocked by wb_clk_i)
Endpoint: wishbone/IncrTxPointer_reg
(rising edge-triggered flip-flop clocked by wb_clk_i)
Path Group: wb_clk_i
Path Type: max

Point Incr Path

------------------------------------------------------------------
clock wb_clk_i (rise edge) 0.00 0.00
clock network delay (propagated) 0.37 0.37

Enabling and Using PrimeTime VX Analysis

1-23
 wishbone/rx_fifo/cnt_reg[0]/CP (SEDFCNQD1) 0.00 0.37 r
wishbone/rx_fifo/cnt_reg[0]/Q (SEDFCNQD1) 0.43 & 0.80 r
wishbone/rx_fifo/U71/ZN (NR4D2) 0.12 & 0.92 f
wishbone/rx_fifo/U21/Z (AN2D2) 0.12 & 1.04 f
...
wishbone/U1059/Z (OR2D4) 0.13 & 2.55 f
wishbone/IncrTxPointer_reg/E (SEDFCND0) 0.00 & 2.55 f
data arrival time 2.55

clock wb_clk_i (rise edge) 2.50 2.50

clock network delay (propagated) 0.24 2.74
clock reconvergence pessimism 0.00 2.74
clock uncertainty -0.10 2.64
wishbone/IncrTxPointer_reg/CP (SEDFCND0) 2.64 r
library setup time -0.28 2.36
data required time 2.36
------------------------------------------------------------------
data required time 2.36
data arrival time -2.55
------------------------------------------------------------------
slack (VIOLATED) -0.19

pt_shell> set path1 [get_timing_paths -delay_type max \

-group wb_clk_i -path_type full_clock_expanded]

pt_shell> set path2 [get_recalculated_timing_paths $path1]

pt_shell> report_timing $path2

...
Startpoint: wishbone/rx_fifo/cnt_reg[0]
(rising edge-triggered flip-flop clocked by wb_clk_i)
Endpoint: wishbone/IncrTxPointer_reg
(rising edge-triggered flip-flop clocked by wb_clk_i)
Path Group: wb_clk_i
Path Type: max

Point Incr Path

------------------------------------------------------------------
clock wb_clk_i (rise edge) 0.00 0.00
clock source latency 0.00 0.00
wb_clk_i (in) 0.00 & 0.00 r
wishbone/BUFFD16_G1B1I4_2/Z (CKBD16) 0.19 & 0.19 r
wishbone/BUFFD16_G1B2I135/Z (BUFFD16) 0.17 & 0.36 r
wishbone/rx_fifo/cnt_reg[0]/CP (SEDFCNQD1) 0.00 & 0.36 r
wishbone/rx_fifo/cnt_reg[0]/Q (SEDFCNQD1) 0.42 & 0.78 r
wishbone/rx_fifo/U71/ZN (NR4D2) 0.12 & 0.90 f
wishbone/rx_fifo/U21/Z (AN2D2) 0.12 & 1.02 f
...
wishbone/U1059/Z (OR2D4) 0.13 & 2.48 f
wishbone/IncrTxPointer_reg/E (SEDFCND0) 0.00 & 2.48 f
data arrival time 2.48

Chapter 1: Overview
1-24
 clock wb_clk_i (rise edge) 2.50 2.50
clock source latency 0.00 2.50
wb_clk_i (in) 0.00 & 2.50 r
BUFFD16_G1B1I2/Z (BUFFD16) 0.16 & 2.66 r
wishbone/BUFFD16_G1B2I56/Z (BUFFD16) 0.09 & 2.75 r
wishbone/IncrTxPointer_reg/CP (SEDFCND0) 0.00 & 2.75 r
clock reconvergence pessimism 0.00 2.75
clock uncertainty -0.10 2.65
library setup time -0.28 2.37
data required time 2.37
-------------------------------------------------------------------
data required time 2.37
data arrival time -2.48
-------------------------------------------------------------------
slack -0.11
statistical adjustment (slack) 0.10 -0.01
slack (VIOLATED) (with probability 0.012) -0.01

The first timing report shows the worst-case analysis results using fixed values derived from
the quantiles of the variations. The reported slack is –0.19 time units.

The second timing report shows the variation-aware, path-based analysis of the critical path.
In a path-based analysis with variation, PrimeTime VX retains all of the statistical delay,
transition time, arrival time, and slack distribution information. As in conventional path-based
analysis, pessimism is reduced by a more exact accounting of input conditions.

The slack is initially calculated by subtracting the quantile of the “data arrival time” from the
quantile of the “data required time,” giving a result of –0.11 time units. However, this is not
the same as subtracting the distributions of these two time values and then taking the
quantile of the result. The latter method is the correct way to calculate the slack with
variation. Accordingly, an additional “statistical adjustment” of 0.10 time units is applied to
arrive at the correct statistical value. The final result is a reported slack of –0.01 time units
at the q01 quantile of the slack distribution (refer to Figure 1-14 on page 1-17 for a graphical
display of the slack distribution). The probability of failure (slack less than zero) is reported
as 0.012, or 1.2 percent.

By default, the 0.00135 and 0.99865 quantile points of a distribution are considered the
minimum and maximum values for reporting purposes, corresponding to the –3*sigma and
+3*sigma points of a perfectly normal distribution. For example, the 0.00135 quantile point
of a slack distribution is considered the worst-case slack. You can specify other quantiles
with the set_variation_quantile command. For example, the command to use the
0.02 and 0.98 quantile points is:

pt_shell> set_variation_quantile \
-quantile_low 0.02 -quantile_high 0.98

You can change the quantile settings and generate a new timing report without running a
new timing update.

Enabling and Using PrimeTime VX Analysis

1-25
 Variation and Crosstalk Analysis
If you have a PrimeTime SI license, you can perform variation-aware analysis and crosstalk
analysis together. To determine the arrival windows for aggressor transitions, PrimeTime SI
uses the worst-case variation-aware arrival times.

To calculate the worst delta delay and bump height for a crosstalk situation, PrimeTime SI
considers the parameter variations of cells, parasitic data, pin capacitance, and aggressor
slew. It finds the worst-case delta delay based on the distribution of each parameter and the
impact of the parameter on overall coupled stage delay. Each delta delay value shifts the
entire victim arrival distribution by the calculated amount. This shift can affect uncoupled
variation-aware arrival windows downstream from the victim net.

In the current release, the PrimeTime SI composite aggressor feature is not compatible with
variation-aware analysis.

Chapter 1: Overview
1-26
2
Libraries and Interconnect Parameters 2
Variation-aware static timing analysis requires library cells that have been characterized at
multiple points in the variation space. You need to load and link your design to the variation-
aware library or libraries containing the variation data. To read in parasitic data with statistical
variation, you must use the -keep_variations option of the read_parasitics
command.

This chapter contains the following sections:

• Variation-Aware Library Creation

• Using Unified Libraries
• Using Separate (2N+1) Libraries
• Interconnect Parameter Variation

2-1
Variation-Aware Library Creation
A library used for analysis in PrimeTime VX specifies the timing behavior of cells at one or
more variation data points. To determine the circuit behavior under a distribution of
parameter values, the libraries need to characterize the behavior at two or three values of
each variation parameter. Figure 2-1 shows an example. For more information about
selecting characterization points, see “Handling Variation With Distribution Data” on
page 1-6.

Figure 2-1 Delay as a Function of Parameter P, Three Data Points

Gate Delay D = f(Q)

Distribution of D

typ D

Distribution
of Q

Parameter
lower typ higher Q

For timing analysis in PrimeTime VX, you read in the libraries and specify the actual
parameter values with the create_variation command and apply them to the design or
to design objects with the set_variation command, as described in Chapter 3,
“Variations and Correlations.” Given the distribution of the parameter, PrimeTime VX
calculates the distribution of delays and slews continuously throughout the range of values,
using interpolation and extrapolation of data surrounding the library-defined operating
points.

Chapter 2: Libraries and Interconnect Parameters

2-2
 During library characterization, you can create either a single unified library containing all of
the process data points or a set of separate libraries, one for each variation data point. For
a process technology defined to have N process parameters, variation modeling requires up
to 2N+1 sets of characterization data to be read into PrimeTime VX. All this characterization
data contained can be combined into a single unified variation-aware library.

The Liberty NCX characterization tool from Synopsys can create a single unified library and
can also merge multiple existing libraries into a single unified library. Then you can read the
unified library into PrimeTime VX as a single step. PrimeTime VX analysis results and
runtime are essentially the same whether you use a single unified library or separate
libraries. However, using a unified library is preferred because it is simpler and less prone to
error. Unified libraries always use the compact CCS format, whereas separate libraries can
use either the compact or expanded CCS format.

A unified library contains all the information on the characterized variation values. When you
use a unified library, you only need to link in that library and then use the
create_variation and set_variation commands to specify the distribution of
parameter values for timing analysis. On the other hand, if you are using a set of 2N+1
separate libraries, you first link the nominal library as the main library and then use 2N+1
set_variation_library commands to establish the variation parameters, before you
use the create_variation and set_variation commands.

If you are using 2N+1 separate variation libraries, you need to know the characterized
parameter values for each library so that you can specify these same values when you use
the libraries in PrimeTime VX. It is a good idea to name each library in a manner to suggest
the operating point, for example, “lowL3.2.lib” and “highL3.8.lib.”

Using Unified Libraries

Before you can perform a variation-aware analysis, you must read in the cell library data
contained in the unified library. The commands used to read the library data depend on
whether you are performing full-chip variation-aware analysis or conventional corner-case
analysis followed by variation-aware, path-based analysis.

Full-Chip Variation-Aware Analysis

For full-chip variation-aware analysis using a single unified library, put the library in the link
path as shown in the following example:

set link_path "* vaccs.db ... "

...
link_design

Using Unified Libraries

2-3
 # Define variations
set_variation [create_variation -name len \
-parameter_name len -type normal -values {0 1}]
set_variation [create_variation -name vt \
-parameter_name vt -type normal -values {0 1}]
...

# Analysis
set variation_enable_analysis true
update_timing

You can read and use multiple unified variation-aware libraries containing different cells, as
long as the libraries have the same variation parameters, characterized at the same values.
For example, you might have one unified library named va_lib1.db containing high-Vt cells,
and another library named va_lib2.db containing low-Vt cells. To use both unified libraries at
the same time, put them both in the link path. For example,

set_link_path "* va_lib1.db va_lib2.db other_lib.db "

...
link_design

# Define variations
set_variation [create_variation -name len \
-parameter_name len -type normal -values {0 1}]
set_variation [create_variation -name vt \
-parameter_name vt -type normal -values {0 1}]
...

# Analysis
set variation_enable_analysis true
update_timing

Path-Based-Only Variation-Aware Analysis

Some analysis flows use conventional corner-based analysis for the whole chip, followed by
variation-aware analysis of critical paths. This flow is invoked by setting the
variation_analysis_mode variable to path_based_only. In this flow, you link the
maximum library as the main library and set the minimum library relationship for corner-
based analysis. After whole-chip analysis, you select the critical paths for a more accurate
variation-aware, path-based analysis.

Chapter 2: Libraries and Interconnect Parameters

2-4
If you are using a single unified library for the statistical part of this analysis, put the
maximum corner-case library in the link path. To read in the variation-aware library, use the
read_db command. For example,

set link_path "* maxlib.db ... "

...
link_design
# Create the min library relationship
set_min_library -min_library minlib.db maxlib.db
...
# Corner-based analysis
update_timing
...
# For path-based analysis
# Read in the variation library
read_db vaccs.db

# Define variations
set_variation [create_variation -name len \
-parameter_name len -type normal -values {0 1}]
set_variation [create_variation -name vt \
-parameter_name vt -type normal -values {0 1}]
...
# Path-based variation-aware analysis
set variation_enable_analysis true
set variation_analysis_mode path_based_only
report_timing ... [get_recalculated_timing_paths ... \
[get_timing_paths ...]]

The cells in both the CCS-based variation-aware library and the minimum library must match
the cells in the maximum library with respect to cell names, pin names, and timing arcs. The
variation_enable_analysis variable must be set to true prior to the read_db
command.

If you are using multiple unified variation-aware libraries containing different sets of cells, put
the corner-case libraries in the link path. Read in the variation-aware libraries as needed with
the set_variation_library -link_library and read_db commands. For example,

set_link_path "* corner_1.db corner_2.db other_lib.db"

...
update_timing
...
set variation_enable_analysis true
...
set_variation_library -link_library corner_1.db
read_db va_lib1.db
set_variation_library -link_library corner_2.db
read_db va_lib2.db
...

Using Unified Libraries

2-5
 The variation_enable_analysis variable must be set to true before the
set_variation_library command is used.

The following is an alternative way to read in multiple variation libraries after

variation_enable_analysis has been set to true:

set_variation_library -link_library {corner_1.db corner_2.db}

read_db va_lib1.db va_lib2.db

The library association is in corresponding order, as listed in the commands.

Using Separate (2N+1) Libraries

Before you can perform a variation-aware analysis, you must read in the cell library data
contained in the set of 2N+1 separate libraries. The commands used to read the library data
depend on whether you are performing full-chip variation-aware analysis or conventional
corner-case analysis followed by variation-aware, path-based analysis.

Full-Chip Variation-Aware Analysis

If you are using a single set of 2N+1 separate libraries, you can first link the nominal library
as the main library and then use 2N+1 set_variation_library commands to establish
the variation parameters. For example,

set link_path "* nominal.db ... "

...
link_design

# set variation libraries

set_variation_library -parameter_names {len vt} \
-values {0 0} nominal.db -reference_value
set_variation_library -parameter_names {len} \
-values -1 lib_slen_neg1s_tiny.db
set_variation_library -parameter_names {len} \
-values +1 lib_slen_pos1s_tiny.db
set_variation_library -parameter_names {vt} \
-values -1 lib_svt_neg1s_tiny.db
set_variation_library -parameter_names {vt} \
-values +1 lib_svt_pos1s_tiny.db
...
# Define variations
set_variation [create_variation -name len \
-parameter_name len -type normal -values {0 1}]
set_variation [create_variation -name vt \
-parameter_name vt -type normal -values {0 1}]
...

Chapter 2: Libraries and Interconnect Parameters

2-6
# Analysis
set variation_enable_analysis true
update_timing

The set_variation_library command with the -reference_value option specifies

the reference or typical library on which the related libraries are based. This must be the first
set_variation_library command in the session. The subsequent
set_variation_library commands specify the libraries that have been characterized
at surrounding parameter values. Each command specifies the names of the parameters,
the corresponding parameter values at which the library was characterized, and the name
of the library.

You can read and use multiple sets of variation-aware libraries containing different cells, as
long as the libraries have the same variation parameters, characterized at the same values.
For example, you might have one set of 2N+1 libraries containing high-Vt cells, and a similar
set of libraries containing low-Vt cells. In that case, put the nominal libraries in the link path.
Then, in each group of set_variation_library commands, use the -link_library
option to specify the nominal library in the link path associated with the variation library set.
For example,

set_link_path "* va_lib1_nom.db va_lib2_nom.db other_lib.db"

...
set_variation_library –parameter_names "len vth" –values "0 0" \
va_lib1_nom.db –link_library va_lib1_nom.db -reference
set_variation_library –parameter_names "len vth" –values "1 0" \
va_lib1_len_plus1.db
set_variation_library –parameter_names "len vth" –values "-1 0" \
va_lib1_len_minus1.db
set_variation_library –parameter_names "len vth" –values "0 1" \
va_lib1_vth_plus1.db
set_variation_library –parameter_names "len vth" –values "0 -1" \
va_lib1_vth_minus1.db
...
set_variation_library –parameter_names "len vth" –values "0 0" \
va_lib2_nom.db –link_library va_lib2_nom.db -reference
set_variation_library –parameter_names "len vth" –values "1 0" \
va_lib2_ len_plus1.db
set_variation_library –parameter_names "len vth" –values "-1 0" \
va_lib2_len_minus1.db
set_variation_library –parameter_names "len vth" –values "0 1" \
va_lib2_vth_plus1.db
set_variation_library –parameter_names "len vth" –values "0 -1" \
va_lib2_vth_minus1.db

In the foregoing example, there are two sets of variation libraries, va_lib1_*.db and
va_lib2_*.db, and an unrelated library, other_lib.db. The -link_library option in
the nominal-library set_variation_library command associates the current set of
variation libraries to a library in the link path. The set_variation_library commands
for a group must be together, with the nominal library specified first.

Using Separate (2N+1) Libraries

2-7
 Path-Based-Only Variation-Aware Analysis
Some analysis flows use conventional corner-based analysis for the whole chip, followed by
variation-aware analysis of critical paths. This flow is invoked by setting the
variation_analysis_mode variable to path_based_only. In this flow, you link the
maximum library as the main library and set the minimum library relationship for corner-
based analysis. After whole-chip analysis, you select the critical paths for a more accurate
variation-aware, path-based analysis.

If you are using a single set of 2N+1 separate libraries for the statistical part of the analysis,
you use 2N+1 set_variation_library commands to establish the variation
parameters. For example,

set link_path "* maxlib.db ... "

...
link_design
# Create the min library relationship
set_min_library -min_library minlib.db maxlib.db
...
# Corner-based analysis
update_timing
...
# For path-based analysis
# Set variation libraries
set_variation_library -parameter_names {len vt} \
-values {0 0} nominal.db -reference_value
set_variation_library -parameter_names {len} \
-values -1 lib_slen_neg1s_tiny.db
set_variation_library -parameter_names {len} \
-values +1 lib_slen_pos1s_tiny.db
set_variation_library -parameter_names {vt} \
-values -1 lib_svt_neg1s_tiny.db
set_variation_library -parameter_names {vt} \
-values +1 lib_svt_pos1s_tiny.db
...
# Define variations
set_variation [create_variation -name len \
-parameter_name len -type normal -values {0 1}]
set_variation [create_variation -name vt \
-parameter_name vt -type normal -values {0 1}]
...
# Path-based variation-aware analysis
set variation_enable_analysis true
set variation_analysis_mode path_based_only
report_timing ... [get_recalculated_timing_paths ... \
[get_timing_paths ...]]

Chapter 2: Libraries and Interconnect Parameters

2-8
 If you are using multiple sets of 2N+1 variation-aware libraries containing different sets of
cells, put the corner-case libraries in the link path. Read in the variation-aware libraries as
needed with the set_variation_library -link_library and read_db commands.
For example,

set_link_path "* corner_1.db corner_2.db other_lib.db"

...
set_variation_library –parameter_names "len vth" –values "0 0" \
va_lib1_nom.db –link_library corner_1.db -reference
set_variation_library –parameter_names "len vth" –values "1 0" \
va_lib1_len_plus1.db
set_variation_library –parameter_names "len vth" –values "-1 0" \
va_lib1_len_minus1.db
set_variation_library –parameter_names "len vth" –values "0 1" \
va_lib1_vth_plus1.db
set_variation_library –parameter_names "len vth" –values "0 -1" \
va_lib1_vth_minus1.db
...
set_variation_library –parameter_names "len vth" –values "0 0" \
va_lib2_nom.db –link_library corner_2.db -reference
set_variation_library –parameter_names "len vth" –values "1 0" \
va_lib2_ len_plus1.db
set_variation_library –parameter_names "len vth" –values "-1 0" \
va_lib2_len_minus1.db
set_variation_library –parameter_names "len vth" –values "0 1" \
va_lib2_vth_plus1.db
set_variation_library –parameter_names "len vth" –values "0 -1" \
va_lib2_vth_minus1.db

Interconnect Parameter Variation

PrimeTime VX accepts statistical interconnect data from Star-RXCT VX and performs timing
analysis in the presence of variations in parameters such as metal width, metal thickness,
and dielectric thickness. A statistical analysis more accurately predicts design violations
resulting from metal layer mismatch and avoids the pessimism of worst-case corner analysis.

To read in parasitic data with variation sensitivity information generated by Star-RCXT VX,
use the -keep_variations option in the read_parasitics command. You can either
choose to create normal distributions for all interconnect variations, or you can explicitly
specify the interconnect variations with the create_variation and set_variation
commands.

Interconnect Parameter Variation

2-9
 Interconnect Parameter Calculations
Star-RCXT VX supports the extraction of interconnect resistivity variation, via resistance
variation, and temperature variation. PrimeTime VX supports the reading and usage of the
variation-aware resistivity and via resistance information from Star-RCXT VX. The
temperature-dependent data used for timing analysis is based on the temperature condition
set before read_parasitics.

In Star-RCXT VX, the interconnect capacitance at a given point is specified as follows:

⎛ ⎞
C = C 0 ⎜ 1 + ∑c i Δv i⎟
⎝ i ⎠
where C0 is the nominal capacitance, ci is the sensitivity coefficient for capacitance with
respect to a given variation parameter vi, and Δvi is the change in that variation parameter.

The interconnect resistance at a given point is specified as follows:

⎛ 1 + ∑n Δv ⎞
⎜ i i⎟
R = R 0 ------------------------⎟
⎜ i
⎜ ⎟
⎜ 1 + ∑d j Δv j ⎟
⎝ j ⎠
where R0 is the nominal resistance, ni is the sensitivity coefficient for resistance with respect
to a given process variation vi that affects the resistance in the numerator, Δvi is the change
in that variation parameter, dj is the sensitivity coefficient for resistance with respect to a
given process variation vj that affects the resistance in the denominator, and Δvj is the
change in that variation parameter.

The interconnect resistance can also depend on temperature as follows:

2
R = R 0 ( 1 + a ΔT + b ( ΔT ) )

where R0 is the nominal resistance, a and b are the temperature sensitivity coefficients for
resistance, and ΔT is the change in temperature.

Chapter 2: Libraries and Interconnect Parameters

2-10
PrimeTime VX supports the reading of parasitic data files to recognize the resistance
variation term niΔvi , thereby fully taking into account the resistivity and via resistance
variation information extracted by Star-RCXT. For static timing analysis, PrimeTime VX
considers the interconnect capacitance and resistance sensitivity to the defined variations.

PrimeTime VX uses the temperature condition set with the set_operating_conditions

command prior to read_parasitics and reads in the variation-dependent resistance data
for that temperature only. If you later write out the parasitic data with the
write_parasitics or write_ilm_parasitics command, PrimeTime VX writes the
data as a temperature corner, using the temperature value in effect at the time that
read_parasitics was used.

Reading Interconnect Parameters With Variation

Star-RCXT VX can generate parasitic data files that contain the variation sensitivity
information each element. PrimeTime VX can read and use this information for variation-
aware analysis. The recommended format for variation-aware parasitic data is the Synopsys
Binary Parasitic Format (SBPF). However, SPEF data with statistical variation is also
supported.

To read in interconnect parameter data with variation sensitivity information, use the
-keep_variations option in the read_parasitics command. For example,

pt_shell> read_parasitics -keep_variations \

-create_default_variations \
-format SBPF my_chip.sbpf

The -create_default_variations option creates normal distributions for all

interconnect variations using the mean and sigma values taken from in the parasitic data file.

To explicitly set the interconnect variations in PrimeTime VX, do not use the
-create_default_variations option. Instead, use create_variation and
set_variation. For example,

pt_shell> set var1 [create_variation -type normal \

-values {0.02 0.0008}
-parameter_name METAL1_WIDTH]

pt_shell> set_variation $var1

To report annotated parasitic data and show the interconnect parameter variation
information, use the -variation option of the report_annotated_parasitics
command. For example,

pt_shell> report_annotated_parasitics -variation

...

Interconnect Parameter Variation

2-11
 ParamID Name Stddev
-----------------------------------------------------
0 PO_T 0.0247
1 PO_W 0.0175
2 M1_T 0.0533
3 M1_W 0.028
4 M12_T 0.036571
...

If you write the annotated parasitics back out using write_parasitics or

write_ilm_parasitics, the statistical information can be written as SPEF or SBPF.

Interconnect Parameters With Temperature Variation

Star-RCXT VX can extract variation parasitics information with temperature as a variation
parameter. In this flow, extraction is performed around a single extraction temperature (the
“global temperature” in the Star-RCXT ITF file) but with additional sensitivity information so
that the RC effects of temperature can be predicted across a temperature range.
PrimeTime VX supports the use of this temperature variation information in its analysis.

When you run read_parasitics -keep_variations and the parasitics file has
temperature variation information, the parasitic information is remapped to the global
analysis temperature specified by set_operating_conditions, even if this temperature
is different from the global temperature used for extraction by Star-RCXT. If the
-keep_variations option is not used, the parasitics information is fixed at the global
extraction temperature. If desired, temperature scaling can also be performed for a non-VX
PrimeTime analysis by using the -keep_variations option. However, note that a
PrimeTime VX license is needed to perform the conversion.

In all cases, the parasitic remapping to the analysis temperature is performed only during
read_parasitics -keep_variations. After this conversion is done, the temperature
sensitivity information is no longer kept in memory. Therefore, the global operating condition
should be defined before the parasitics are read in. Any subsequent changes in operating
condition temperature do not affect in the in-memory parasitics.

Interconnect Parasitic Corners

A parasitic data file generated by Star-RCXT VX can serve for both variation-aware analysis
and conventional corner-case analysis. You can use the set_parasitic_corner
command to establish the default values of the interconnect data used for analysis.

With variation analysis enabled (variation_enable_analysis set to true), the

set_parasitic_corner command establishes the default behavior for any interconnect
parameter that has no variation. You can override the default behavior by using the

Chapter 2: Libraries and Interconnect Parameters

2-12
-create_default_variations option of the read_parasitics command or by
creating interconnect variations with the create_variation and set_variation
commands. For conventional corner-case analysis with variation analysis disabled and for
the conventional corner-case portion of analysis in the path-based-only mode, the
set_parasitic_corner command establishes the fixed interconnect values used for
analysis.

The command works in conjunction with a text file that specifies a set of variation ratios used
to calculate the interconnect parasitic values. For example,

pt_shell> set_parasitic_corner -name CMAX my_corner_file

You need to provide a simple ASCII text file containing the corner definitions as a set of
variation ratios. For example:

CORNER_NAME: CMAX
M1_W 1.2
M2_T 0.4
ILD_c_T -2.0
CORNER_NAME: RCMAX
## my definition of RCMAX
M1_W 0.25
M2_T 0.16
ILD_c_T -0.22
CORNER_NAME: RCMIN
M1_W -0.23
M2_T -0.11
ILD_c_T 0.04
CORNER_NAME: CMIN
M1_W -0.18
M2_T -0.14
ILD_c_T 0.06

This example defines four sets of corner parasitics called CMAX, RCMAX, RCMIN, and
CMIN. For the CMAX corner, the M1_W parameter is changed in the positive direction from
nominal (+1.2 multiplied by the M1_W variation coefficient), the M2_T parameter is changed
in the positive direction from nominal (+0.4 multiplied by the M2_T variation coefficient), and
the ILD_c_T parameter is changed in the negative direction from nominal (–2.0 multiplied by
the ILD_c_T variation coefficient). The variation coefficients are contained within the
sensitivity parasitics file.

Interconnect Parameter Variation

2-13
 The general form of the file is:

CORNER_NAME: name
param_name param_variation_ratio
param_name param_variation_ratio
...
CORNER_NAME: name
param_name param_variation_ratio
param_name param_variation_ratio
...

The file should contain at least one corner definition. A corner definition can specify any
number of parameter names (including zero names) and a corresponding
param_variation_ratio value for each parameter. PrimeTime VX multiplies the
param_variation_ratio value by the variation coefficient specified for the parameter in the
parasitic data file, and then multiplies that result by the nominal parameter value to get the
change in value used for analysis.

To report parasitic corner information that has been set, use the command
report_annotated_parasitics -variation. To report the parasitic sensitivity of
nets, use the report_variation command. For details, see Chapter 3, “Variations and
Correlations.”

To remove parasitic corner information, use the remove_parasitic_corner command.

Chapter 2: Libraries and Interconnect Parameters

2-14
3
Variations and Correlations 3
PrimeTime VX determines the distributions of delays, slews, and slacks based on the
variations of selected parameters. The actual distributions of parameters are specified in
PrimeTime VX with the create_variation and set_variation commands.

This chapter contains the following sections:

• Parameter Variation and Correlation

• Creating and Setting Variations
• Creating and Setting Correlations
• Math Operations on Variations
• Derating Statistical Distributions
• Reporting Variations

3-1
Parameter Variation and Correlation
The cell libraries specify the dependency of cell delays and slews on parameters, but not the
probability distributions of those parameters. You specify the distributions in PrimeTime VX
by using the following commands:

• The create_variation command creates a variation object that is associated with a

given characterized parameter. The variation object specifies the type of statistical
distribution (normal, piecewise-linear, empirical, uniform, lognormal, or discrete) and the
values associated with that distribution (for example, the mean and standard deviation of
a normal distribution).
• The set_variation command assigns a given variation object to the current design or
some objects within the current design, resulting in the specified statistical distribution of
the characterized parameter in the design.
• The create_correlation command creates a correlation relationship object. For
auto-correlation between two instances of a variation, the correlation can be set to 0.0
(uncorrelated), 1.0 (fully correlated), or some number between 0.0 and 1.0 (partially
correlated).
• The set_variation_correlation command assigns a correlation object to a list of
variations or to all variations, resulting in the specified correlation relationship between
different instances of a variation.
This organization of parameter variation data is flexible and efficient. You can easily change
the parameter values, distributions, and correlation characteristics to analyze different
analysis conditions, without changing the cell libraries.

Creating and Setting Variations

To perform variation-aware analysis, you need to specify the statistical variation of the
parameters that affect the timing characteristics of the circuit. To specify the variation of a
parameter, you create a variation object with the create_variation command and then
apply the variation to the design with the set_variation command.

This is the syntax of the two commands:

create_variation
[-name variation_name]
[-parameter_name parameter_name]
-type distribution_type
-values values_list
[-lower_bound lower_bound]
[-upper_bound upper_bound]

Chapter 3: Variations and Correlations

3-2
set_variation
variation_list
[object_list]

For example, the following commands load a variation library having parameters “len” and
“vth,” create a variation named “Lfab” for parameter “len,” and set the variation on the current
design:

pt_shell> set_variation_library \
-parameter_names "len vth" \
-values "100 0.2394" maxvth.db

pt_shell> create_variation -parameter_name len \

-name Lfab -type normal -values {0 1}

pt_shell> set_variation [get_variations Lfab]

create_variation
The create_variation command creates a new variation object and returns a collection
containing that variation. You can specify a variation name with the -name option, allowing
you to reference that variation later with the get_variations command. Any string that
has not already been used can be specified as the variation name.

The -parameter_name option specifies the process parameter associated with the
variation. The specified parameter must be defined with the set_variation_library
command or defined in the parasitic data file read in with the read_parasitics command.
Multiple variations can be associated with one parameter.

If the -parameter_name option is not used, the variation is not associated with any
parameter and has no direct effect on the design. However, it can still be used by the
variation commands such as add_variation, sub_variation, max_variation, and
min_variation.

The -type and -values settings specify the statistical distribution of the variation.
Table 3-1 shows the types of distributions you can specify and the values associated with
each type.

Creating and Setting Variations

3-3
Table 3-1 Distribution Types

Distribution Type Values

normal {mean sigma}

pwl {x1 f1 x2 f2 x3 f3 ...}

empirical {x1 x2 x3 ...}

constant {a b}

uniform {a b}

lognormal {mean sigma}

discrete {x1 p1 x2 p2 x3 p3 ...}

You can optionally specify upper and lower bounds by using the -lower_bound and
-upper_bound options.

To remove a variation that you no longer want, use the remove_variation command. This
is the command syntax:

remove_variation
-all | variation_list

Normal Distribution
The following command creates a normal (Gaussian) distribution with a mean value of 100
and a sigma (standard deviation) of 3.34:

pt_shell> create_variation -name test_normal \

-type normal -values {100 3.34}

Chapter 3: Variations and Correlations

3-4
Figure 3-1 PDF Plot of Normal Distribution

100

By default, the distribution is unbounded at both ends. To truncate the distribution at one end
or both ends, use one or both of the options -lower_bound and -upper_bound. For
example,

pt_shell> create_variation -name test_normal \

-type normal -values {100 3.34} \
-lower_bound 93.0 -upper_bound 107.0

Figure 3-2 PDF Plot of Truncated Normal Distribution

93 100 107

Piecewise-Linear Distribution
The following command creates a piecewise-linear distribution. The pairs of values in the
value list specify the data points in the distribution.

pt_shell> create_variation -name test_pwl \

-type pwl -values {-1.0 0.0 0.0 1.0 1.0 0.0}

Creating and Setting Variations

3-5
Figure 3-3 PDF Plot of Piecewise-Linear Distribution

This particular example is the triangular distribution.

The piecewise-linear distribution has n pairs of numbers xi , fi . The sequence of values xi

must be monotonically increasing and all the values fi must be positive or zero. The first and
last values f1 and fn must be zero. The area under the curve must be 1.0 because it is a
probability distribution function.

Empirical Distribution
The following command creates an empirical distribution. The sequential list of values
(possibly generated by simulation or other empirical means) specifies the distribution. The
likelihood that each sample value will occur is 1.0 divided by the sample size.

pt_shell> create_variation -name test_empirical \

-type empirical \
-values {0.0 91.4 96.3 99.5 102.1 90.0}

Constant Distribution
The following command creates a constant distribution that models the case where there is
no statistical variation. The value is a single number. This distribution is the same as a
discrete distribution having a single value with a probability 1.0. The constant distribution can
be used to mathematically shift another distribution by a fixed amount, using the commands
described in “Math Operations on Variations” on page 3-14.

pt_shell> create_variation -name test_const \

-type constant -values {2.0}

Uniform Distribution
The following command creates a uniform distribution. A single pair of values specifies the
minimum value a and maximum value b of the range. The probability density function has a
constant value equal to 1/(b–a).

Chapter 3: Variations and Correlations

3-6
 pt_shell> create_variation -name test_uniform \
-type uniform -values { 1.0 3.0 }

Figure 3-4 PDF Plot of Uniform Distribution

Lognormal Distribution
The following command creates a lognormal distribution. The first value is mu (μ) and the
second value is sigma (σ).

pt_shell> create_variation -name test_lognorm \

-type lognormal -values { 0.0 0.5 }

Figure 3-5 PDF Plot of Lognormal Distribution

1.0 mu = 0.0
sigma = 0.5

1.0 2.0

A random variable X has a lognormal distribution with parameters μ and σ if log(X) has a
normal distribution with mean μ and standard deviation σ. The lognormal distribution has
support over the range (0.0, infinity).

Creating and Setting Variations

3-7
 The lognormal distribution has a probability density function
2 2
- e –( ln x – μ) ⁄ ( 2σ )
1
f ( x ) = ---------------
xσ 2π

for x greater than zero.

By default, the distribution is unbounded at the upper end. To truncate the upper range, use
the -upper_bound option. For example,

pt_shell> create_variation -name test_lognorm \

-type lognormal -values { 0.0 0.5 } \
-upper_bound 2.5

Discrete Distribution
The following command creates a finite discrete distribution. The list contains a sequence of
paired numbers. In each pair of numbers xi and pi , the first number xi is an outcome of the
variation and pi is the probability of that outcome. The sequence of numbers xi must be
monotonically increasing. The numbers p1 through pn must be strictly positive and must add
up to 1.0.

pt_shell> create_variation -name test_discr \

-values { 0.5 0.25 1.0 0.40 1.5 0.15 2.0 0.20 }

set_variation
The set_variation commands applies one or more variation objects to timing objects in
the design or to the whole design. Each variation object specified in the command must have
been created previously with the create_variation command using the
-parameter_name option. Applying a variation object on a timing object affects the timing
behavior of that object.

This is the syntax of the command:

set_variation
variation_list
[design_object_list]

You specify a single variation, a list of variations, or a collection of variation to apply, and
optionally a list of objects in the design on which to apply the variations, such as a collection
cell arcs. If no object list is specified, the variations apply to the whole design, affecting all
occurrences of the associated parameter.

Chapter 3: Variations and Correlations

3-8
 To remove a variation from an object in the design or the whole design, use the
reset_variation command. This is the command syntax:

reset_variation
[-all]
[variation_list]
[design_object_list]

Creating and Setting Correlations

Different instances of variations can have different levels of correlation. Some variations may
be highly correlated, while other are not correlated at all. For the best possible analysis
accuracy, you should specify the correlation relationships between variations with the
create_correlation and set_variation_correlation commands. This is the
syntax of the commands:

create_correlation
-name string
-constant float | -cross_correlations float_list |
-physical_distance float_list

set_variation_correlation
-name variation_correlation_name
-correlation correlation_name
-all | variation_object_list

The create_correlation command creates a correlation object. It assigns a name to the

correlation and specifies a floating-point value or a set of such values, depending on the
correlation type.

There are three types of correlation:

• constant correlation (also known as auto-correlation) – the correlation between different

instances of the same library variation or parasitic data variation
• cross-correlation – the correlation between different variations within a given instance
• spatial correlation – the correlation between device variations that depend on the physical
distance between devices within a single die.
You must specify the correlation type in the create_correlation command using one of
the following: -constant, -cross_correlations, or -physical_distance.

The set_variation_correlation command applies a named correlation object to one

or more variations previously defined by the create_variation command. The -name
string is an arbitrary name that you assign to the variation correlation assignment. The

Creating and Setting Correlations

3-9
 -correlation string is the name of the correlation object previously given in the
create_correlation command. You must either specify a list of variation objects on
which to apply the correlation or use the -all option to apply it to all variations.

For example, the following commands load a variation library having parameters “len” and
“vth,” create variations named “Lfab1” and “Lfab2” for the two parameters, create a full
correction object, and apply that correlation object to the two variations:

pt_shell> set_variation_library \
-parameter_names "len vth" \
-values "100 0.2394" len_vth.db

pt_shell> create_variation -parameter_name len \

-name Lfab1 -type normal -values {0 1}

pt_shell> set_variation [get_variations Lfab1]

pt_shell> create_variation -parameter_name vth \

-name Lfab2 -type normal -values {0 1}

pt_shell> set_variation [get_variations Lfab2]

pt_shell> create_correlation -name full_corr \

-constant 1.0

pt_shell> set_variation_correlation -name device_len \

-correlation full_corr [get_variations Lfab*]

Auto-Correlation
Auto-correlation is the correlation between different instances of the same variation. You
specify the correlation value by using the create_correlation command with the
-constant option. For example,

pt_shell> create_correlation -name full_corr \

-constant 1.0

The constant is the correlation value between two instances of the variation. A value of 0.0
means that different instances of the variation within a die are statistically independent. This
correlation value is recommended for modeling within-die, on-chip variation because such
random variations are uncorrelated. A value of 1.0 means that different instances of the
variation are all identical within a die. This correlation value is recommended for modeling
die-to-die variations because such variations are highly correlated within a die.

Chapter 3: Variations and Correlations

3-10
 Within-die and die-to-die variations are illustrated in Figure 3-6. In each die is a string of
buffer cells, each having a parameter “A” that varies from cell to cell and from die to die.
Within a given die, the parameter has small random variations from cell to cell. There is also
a larger variation from die to die, but these variations are highly correlated; all of the cells
within a given die tend to have the same higher or lower values.

Figure 3-6 Within-Die and Die-to-Die Variations

A=1.2 A=0.9 A=1.1 A=1.0 A=0.8

Die 1

A=3.1 A=3.0 A=2.8 A=3.2 A=2.9

Die 2

A=2.0 A=1.8 A=2.1 A=1.9 A=2.2

Die 3

This type of variation can be modeled as two components. One component is the small,
uncorrelated within-die variation. The other component is the larger, correlated die-to-die
variation.

A parameter can also be partially correlated, having a correlation value between 0.0 and 1.0.
For a partial correlation, it is assumed that the variation has a normal distribution and that
two instances of the variation have a bivariate normal distribution with a correlation equal to
the specified constant value. If a correlation is not specified on a variation, it is assumed to
be a constant correlation with a value of 1.0.

Creating and Setting Correlations

3-11
 Cross-Correlation
Cross-correlation is the correlation between different variations applied to the same timing
object. To specify a cross-correlation, you first specify the partial correlation properties by
using the create_correlation command with the -cross_correlation option. Then
you apply the correlation object to the parameter variations. For example,

pt_shell> create_correlation -cross_correlations \

{-0.85 0.45 0.09} -name xcor_A_B_C

pt_shell> set_variation_correlation -name xcor \

-correlation xcor_A_B_C {A B C}

The variations applied to each device or parasitic variation are assumed to be jointly normal.
The values specified with the -cross_correlations correspond to the number of
parameters in the lower triangular correlation matrix. If (A, B, C, D) are four variations, the
sequential list of values corresponds to the correlations

{ corr(B,A) corr(C,A) corr(C,B) corr(D,A) corr(D,B) corr(D,C) }

The number of cross-correlation terms entered in the command must correspond to the
number of variations to which the correlation object is applied; N parameters require
N(N-1) / 2 coefficients. In addition, the matrix must be positive definite, which implies that the
set of variations is the minimal set, without a linear relationship between any members of the
set, and which cannot be reduced to a lower number of primitive variations. Joint normality
of a vector of variations implies normality of each variation.

You can define multiple sets of cross-correlated variations. However, a given variation can
belong to no more than one cross-correlated set. All parameters within one cross-correlation
set must have the same auto-correlation value, either all constant 0.0 or all constant 1.0.

Spatial Correlation
Spatial correlation is correlation between device variations that is a function of the physical
distance between the objects on the die. This within-die correlation applies only to device
variations only, not to parasitic data variations.

Spatial correlation is described by a correlation function ρ (d). For a variation X(u) at the
physical location u, ρ (d) = Corr(X(u), X(u+d)). The position u is in a two-dimensional plane
and d is the Euclidean distance separating the two devices. This function has a value of 1.0
at d = 0.0 and typically decreases and tapers off to 0.0 as d becomes larger.

Chapter 3: Variations and Correlations

3-12
It is assumed that the correlation function is isotropic, or in other words, the correlation
between two devices depends only on the distance that separates them, not on their
respective positions. Under this assumption, the correlation function is symmetric and
independent of the X and Y directions.

The correlation function must be estimated by the user. You enter the correlation information
as a list of points: (d1 ρ 1 d2 ρ 2 ... dn ρ n) that approximate the function. The distances d
must be greater than 0.0 and must be in increasing order, and the values ρ 1, ρ 2, ... ρ n must
be between –1.0 and +1.0. The variations are assumed to have a normal distribution, or in
other words, for all integers n and all locations u1, u2, ..., un, the vector (X(u1),X(u2),...,X(un))
has a multivariate distribution.

The physical location information must be specified in the SBPF data files. The variable
read_parasitics_load_locations must be set to true prior to reading in the
parasitics. You enter the correlation information by using the create_correlation
command with the -physical_distance option, with distance units in nanometers. For
example, the following script creates a spatial correlation named “c_spatial” and applies it to
a variation called “len” in the SPEF data file sptl_corr.spef.

# Create device variation

set_variation_library -parameter_names {len} \
-values {100} lib_len_nom.db -reference_value
set_variation_library -parameter_names {len} -values 95 lib_len_small.db
set_variation_library -parameter_names {len} -values 104 lib_len_large.db

# It has the normal distribution

set_variation [create_variation -name len -parameter_name len \
-type normal -values {0 3.0 }]

# Set variable so physical information is read in

set read_parasitics_load_locations "true"

# SBPF file contains location information

read_parasitics -format SBPF sptl_corr.sbpf -keep_variation
create_correlation -name c_spatial \
-physical_distance { 500000.0 0.9 1000000.0 0.7 \
2000000.0 0.4 4000000.0 0.1 }
# len now has a spatial within-die correlation
set_variation_correlation -name s_spatial -correlation c_spatial {len}

Creating and Setting Correlations

3-13
Math Operations on Variations
You can perform various types of mathematical operations with variations, including addition,
subtraction, minimum/maximum value, and random number generation.

The following PrimeTime VX commands perform these operations:

add_variation var_object_list
sub_variation var_object_list
max_variation var_object_list
min_variation var_object_list
get_random_numbers
-sample_size num_points
-seed seed_value
variation_object

The add_variation command statistically adds two or more variations, producing a new
variation. The object list specified in the command must be a collection of two or more
variations. Note that adding two variations, then adding the resulting sum to a third variation,
can produce a different result from adding the three variations together at the same time.

The sub_variation command statistically subtracts one variation from another. The
object list specified in the command must be a collection of two variations.

The max_variation command returns a variation that is the statistical maximum of the
variations in the specified collection. The result can be different from any of the original
variations in the specified collection. The min_variation command works in a similar
manner, returning a new variation that is the minimum of the variations in the collection.

The get_random_numbers command generates a list of random numbers from a specified

variation object previously created with the create_variation command. The
-sample_size parameter specifies the desired number of data points. The -seed setting
is an integer used to initialize the pseudo-random number generator, allowing you to
generate the same sequence or a different sequence of numbers each time.

The following command creates a variation object named v1 having a normal distribution
with a mean value of 15.0 and a standard deviation of 2.0:

pt_shell> create_variation -name v1 -type normal \

-values { 15.0 2.0 }]

The following command creates a variation object named v2 having a normal distribution
with a mean value of 10.0 and a standard deviation of 5.0:

pt_shell> create_variation -name v2 -type normal \

-values { 10.0 5.0}]

Chapter 3: Variations and Correlations

3-14
 The following command creates a collection containing the variations v1 and v2. It sets the
variable c1 to the new collection:

pt_shell> set c1 [add_to_collection $v1 $v2]

The following command sets the random variable a1 to the sum of v1 and v2 (previously
collected in c1):

pt_shell> set a1 [add_variation $c1]

The following command subtracts v2 from v1 and sets the variable s1 to the result:

pt_shell> set s1 [sub_variation $c1]

The following command sets the variable mx to the larger of v1 or v2 (previously collected in
c1):

pt_shell> set mx [max_variation $c1]

The following command sets the variable mn to the smaller of v1 or v2 (previously collected
in c1):

pt_shell> set mn [min_variation $c1]

The following command generates a statistically independent sequence of 1,000 floating-

point numbers having the probability distribution previously defined in the variation v1. The
variable rnd_lst is set to the list of generated numbers. The seed value is set to 72, so the
same sequence can be generated later by specifying the same seed value.

pt_shell> set rnd_lst [get_random_numbers \

-sample_size 1000 -seed 72 \
[get_variations v1]]

Derating Statistical Distributions

The variation-aware features of PrimeTime VX can accurately account for the effects of
known variations. You can use derating to adjust the variation effects upward or downward,
for example, to model the additional effects of unknown variations, or to model variation
effects for macrocells or IP cells for which there is no variation library data available.

The set_timing_derate command adjusts the calculated delays of the paths in the
whole design or in specified library cells or cell instances. You can get a more conservative
analysis by making long path delays longer and short path delays shorter, or get a less
conservative analysis by doing the opposite.

Derating Statistical Distributions

3-15
 This is the syntax of the command:

set_timing_derate
[-early] [-late]
[-rise] [-fall]
[-clock] [-data]
[-net_delay] [-cell_delay] [-cell_check]
[-static] [-dynamic]
[-scalar] [-variation]
value
[object_list]

It may be desirable to use larger derating factors for corner-case delays, and to use smaller
factors for statistical delays that already account for the effects of known variations. For this
reason, there are separate derating settings for corner-case and statistical delays.

For statistical derating, use the -variation option; or for corner-case derating, use the
-scalar option. For example,

pt_shell> set_timing_derate -scalar -early 0.9

pt_shell> set_timing_derate -scalar -late 1.2
pt_shell> set_timing_derate -variation -early 0.96
pt_shell> set_timing_derate -variation -late 1.08

The early and late scalar derating factors 0.9 and 1.2 apply to corner-case timing and to cells
or nets that do not have variation data, such as macrocells, pad cells, or other cells without
variation models. Early (minimum) path delays are decreased by 10 percent and late
(maximum) path delays are increased by 20 percent. This causes a more conservative
analysis than leaving delays at their original calculated values.

The early and late variation derating factors 0.96 and 1.08 apply to statistical delays,
including cell arcs of cell instances with variation data and net arcs of nets with interconnect
variation data. Early (minimum) path delay distributions are decreased by 4 percent and late
(maximum) path delay distributions are increased by 8 percent.

Derated delays are adjusted according to the following formula:

new_delay = old_delay * derating_factor

For statistical delay variations, PrimeTime VX multiplies the derating factor by the whole
distribution. This causes the mean value, standard deviation, and peak value of the
distribution to change, as shown by the example in Figure 3-7.

Chapter 3: Variations and Correlations

3-16
Figure 3-7 Derated Maximum-Delay Distribution

Original Derated
distribution distribution

Delay multiplied
by1.08

Delay

In the unusual case of negative delay values, a different derating formula is used to
consistently give a more conservative analysis:

new_delay = old_delay * (2 – derating_factor)

When part or all of the delay distribution is negative, the negative part of the delay is adjusted
as shown in Figure 3-8.

Figure 3-8 Derated Maximum-Delay Distribution With Negative Portion

Original Derated
distribution distribution

Negative portion: Positive portion:

multiplied by 0.92 multiplied by1.08

Delay
0

To report derating factors that have been set, use the report_timing_derate command.
Use the -variation option to report statistical derating settings, or omit the option to report
scalar settings.

Reporting Variations
You can report variations used in the design by using the report_variation command.
In the command, you specify a collection of objects to be reported. The collection can
contain paths collected with the get_recalculated_timing_paths command; cells

Reporting Variations
3-17
 collected with the get_cells command; nets collected with the get_nets command; or
library cells collected with the get_lib_cells -of_objects command. If no collection
is specified, the command applies to the whole current design.

You can sort and filter collections based on variation-aware slack, arrival times, and so on by
setting a variable that controls the variation-derived attribute mode:

variation_derived_scalar_attribute_mode

Then you can use the sort_collection and filter_collection commands.

This is the syntax of the report_variation command:

report_variation
[-significant_digits digits]
[-verbose]
[-nosplit]
[-delay_type delay_type]
[-slack_lesser_than slack_limit]
[-nworst num_worst]
[-clock_network]
[-all_cells]
[-all_nets]
[-transition]
[-parametric_sensitivity parameter_list]
[-input_pins]
[collection]

Variation Sensitivity
The statistical sensitivity of a timing distribution attribute (such as arrival time or slack) is a
measure of how much that attribute changes in response to changes in related variations.
The sensitivity values in a design and the sensitivity values of timing arcs in a library can be
reported with the report_variation command.

For any timing distribution “A,” the sensitivity of that distribution is defined to be the
normalized variance of its distribution, calculated as follows:

sensitivity(A) = std_dev(A) / | mean(A) |

The sensitivity of distribution “A” is the standard deviation of “A” divided by the absolute value
of the mean of “A.” If the mean of “A” is zero or very close to zero, a small threshold value is
used instead of the zero or near-zero value.

Chapter 3: Variations and Correlations

3-18
The sensitivity of a constrained timing path is defined as the sensitivity of the path’s slack,
whereas the sensitivity of an unconstrained path (such as a clock path) is defined as the
path’s arrival sensitivity:

sensitivity(constrained_path) = sensitivity(variation_slack)

sensitivity(unconstrained_path) = sensitivity(variation_arrival)

Each timing point within a path has an arrival time sensitivity and a transition time sensitivity:

delay_sensitivity(point) = sensitivity(variation_arrival)

slew_sensitivity(point) = sensitivity(variation_transition)

Library arcs have sensitivity values that indicate the sensitivity of delay and slew with
changes in a random variable rv:

delay_sensitivity(lib_arc) = Δ(delay) / Δ(rv)

slew_sensitivity(lib_arc) = Δ(slew) / Δ(rv)

The random variable rv is defined by the set_variation_library command. Multiple

libraries are usually involved with each random variable, and the delay and slew functions
depend on input slew and load capacitance, so multiple lookup tables represent the delay
and slew sensitivity functions.

Note that the two different types of sensitivities described here are fundamentally different in
what they represent. The path and timing point sensitivity values contained within path
objects are indicators of the design sensitivity to variations in the context of the design.
These values represent the amount of variation relative to the mean value of the measured
path or point attribute. This type of normalized sensitivity is known as “design sensitivity.”

The sensitivity values obtained from library arcs are called “library sensitivity” values. Rather
than representing the amount of normalized variation relative to the arc itself, library
sensitivity represents the amount of variation relative to a specific variation parameter of
interest. Library sensitivity is not normalized; its magnitude depends on how the library arc
timing data changes with a given change in parameter value.

Reporting Variations
3-19
 Variation-Derived Scalar Attributes
When variation-aware analysis is enabled, getting a timing_path attribute called slack
returns the low quantile slack value for the path. For example,

pt_shell> set path1 [get_timing_paths -to reg[31]/D]

...
pt_shell> set myrecalc \
[get_recalculated_timing_paths $path1]
...
pt_shell> get_attribute $myrecalc slack
-0.455260

In this example, the quantile slack obtained by variation-aware analysis is –0.455 time units.

PrimeTime VX uses separate attributes to hold the fixed slack and variation-aware slack.
The name of the variation-aware attribute is variation_slack, made by adding the string
variation_ as a prefix. There are also variables named variation_arrival,
variation_transition, and so on.

The variation_derived_scalar_attribute_mode variable controls how the

distributions in these variation attributes are converted to the corresponding numerical
attributes. By default, this variable is set to quantile during variation-aware analysis, which
causes conservative statistical values from the edge of the distribution to be returned. If you
set it to mean, the mean values are reported, or if you set it to none, the corner-case values
are reported. For example,

pt_shell> set variation_derived_scalar_attribute_mode mean

mean
pt_shell> get_attribute $myrecalc slack
-0.092147
pt_shell> set variation_derived_scalar_attribute_mode none
none
pt_shell> get_attribute $myrecalc slack
-1.042500

Depending on the variable setting, PrimeTime reports the less conservative mean value of
the slack distribution or the more conservative corner-case slack value.

The quantile values returned depend on the -quantile_low and -quantile_high

option settings of the set_variation_quantile command. The default quantile low and
high settings are 0.00135 and 0.99865, respectively, corresponding to the –3*sigma and
+3*sigma points of a perfectly normal distribution. For quantities like delay and slew, the use
of the low or high quantile value depends on whether the most conservative value is on the
low or high side of the distribution. Slack values always use -quantile_low because lower
slack values are more conservative.

Chapter 3: Variations and Correlations

3-20
The variation_derived_scalar_attribute_mode setting only affects the reporting
of attributes in the database, not the actual values stored. It affects the behavior of
get_attribute, sort_collection, and filter_collection. For example, with the
variation-derived scalar attribute mode set to mean, the sort_collection command sorts
a collection of recalculated timing paths in order of mean rather than quantile slack.

Variations in Timing Paths

To get a report on the variations used in timing paths that have been recalculated with
variation-aware analysis enabled, create a collection of those paths and apply the
report_variation command. For example, the following script creates a collection of the
two worst timing paths, recalculates the timing of those paths with variations, sorts the paths
by quantile slack, and generates a summary report on the variations used in the paths.

set paths [get_timing_paths -nworst 2]

set variation_derived_scalar_attribute_mode quantile
report_variation $paths
set var_paths [sort_collection $recalc_paths "slack"]
set recalc_paths [get_recalculated_timing_paths $paths]
report_variation $var_paths

A standard report on path variations looks like this:

Path startpoint endpoint quantile mean stddev sensitiv

---------------------------------------------------------------------
0 in ff1 3.051712 3.054828 0.001437 0.047031
1 ff1 ff2 9.868665 9.886533 0.007124 0.072059

The report shows the path startpoint, path endpoint, slack at the low quantile, mean slack,
standard deviation of slack, and the slack sensitivity. Sensitivity values are reported in
percent:

[ std_dev(A) / | mean(A) | ] * 100

A verbose report on path variations (using the -vebose option) looks like this:

Path startpoint endpoint quantile sensitiv mean stdd

--------------------------------------------------------------------
1 ff1 ff2 9.868665 0.072059 ... ...

Path attribute quantile sensitiv mean stdd incr

--------------------------------------------------------------------
arrival 0.219267 3.440833 0.202071 ... ...
slack 9.868665 0.072059 9.886533 ... ...
required 10.084497 0.018024 10.088605 ... ...
startpoint_clock_latency 0.065878 2.346163 0.062349 ... ...
endpoint_clock_latency 0.089497 1.942685 0.093604 ... ...

Reporting Variations
3-21
 Point arrival quantile sensitiv mean std inc
---------------------------------------------------------------------
clk (in) 0.000000 0.000000 0.000000 ... ...
b01/ZN (INVD1) 0.009146 2.494643 0.008588 ... ...
b02/ZN (INVD1) * 0.033515 3.722505 0.030891 ... ...
b11/ZN (INVD1) 0.048736 2.773270 0.046092 ... ...
b12/ZN (INVD1) 0.065878 2.346163 0.062349 ... ...
ff1/CP (DFD1) 0.065878 2.346163 0.062349 ... ...
ff1/Q (DFD1) * 0.200045 3.751771 0.183375 ... ...
b21/ZN (INVD1) 0.219267 3.440833 0.202071 ... ...
...

The verbose report includes the quantile and sensitivity values, mean, standard deviation,
and incremental standard deviation for the arrival time of each timing point along the path;
and the quantile, sensitivity, mean, and standard deviation values for the whole-path arrival
time, slack, required time, startpoint clock latency, and endpoint clock latency, where
applicable. The asterisk characters (*) in the list indicate the stages having the highest
incremental standard deviation of the arrival time in the launch path, data path, and capture
path. Use the -input_pins option if you want the variation report to include the arrival
times at cell inputs as well as at the cell outputs along the path.

Variations in Cells
To get a report on the sensitivity of one or more cells in the design, create a collection of
those cells and apply the report_variation command. For example, the following
commands report the variations in three specified cells in the design:

pt_shell> set mycells [get_cells {u13 u24 u02}]

...
pt_shell> report_variation $mycells -significant_digits 6
...
cell lib_cell slack delay mean stddev
------------------------------------------------------------------------
u13 AN2D1 0.542629 0.114985 0.096514 0.007269
u24 OR2D1 0.493780 0.090585 0.075847 0.005648
u02 BUFFD1 0.493780 0.086466 0.075197 0.004733

pt_shell> report_variation $mycells -verbose -significant_digits 6

...
Cell delay variation: u13 (AN2D1)
from to sense slack ... ... stddev
----------------------------------------------------------------------------
u13/A1 u13/Z r positive_unate 0.553502 ... ... 0.006643
u13/A1 u13/Z f positive_unate 0.556909 ... ... 0.005430
u13/A2 u13/Z r positive_unate 0.542629 ... ... 0.007269 *
u13/A2 u13/Z f positive_unate 0.546749 ... ... 0.005876

Chapter 3: Variations and Correlations

3-22
Cell delay variation: u24 (OR2D1)
from to sense slack ... ... stddev
----------------------------------------------------------------------------
u24/A1 u24/Z r positive_unate 0.542162 ... ... 0.005648 *
u24/A1 u24/Z f positive_unate 0.571965 ... ... 0.004758
u24/A2 u24/Z r positive_unate 0.533186 ... ... 0.004827
u24/A2 u24/Z f positive_unate 0.493780 ... ... 0.005454

Cell delay variation: u02 (BUFFD1)

from to sense slack ... ... stddev
----------------------------------------------------------------------------
u02/I u02/Z r positive_unate 0.533186 ... ... 0.003757
u02/I u02/Z f positive_unate 0.493780 ... ... 0.004733 *

The default report lists the slack, the scalar arc delay, and the mean and standard deviation
of the statistical arc delay for the worst timing arc through the cell. The worst arc is the arc
with the highest standard deviation of arc delay among all timing arcs of the cell. The slack
of an arc is the slack of the worst-case timing path through the arc. In order to report slack
values, the variable timing_save_pin_arrival_and_slack must be set to true prior to
update_timing.

The verbose report shows the same information for all the timing arcs through the cell, with
the worst arc marked with an asterisk. (In the foregoing example, some columns of numbers
were omitted so that the report would fit on the page.)

Variations in Nets
To get a report on the sensitivity of one or more nets in the design, create a collection of
those nets and apply the report_variation command. For example, the following
commands report the variations in nets whose worst slack is negative:

pt_shell> report_variation [get_nets] -slack_lesser_than 0

...

net slack delay mean stddev

--------------------------------------------------
udat/in3 -4.0884 0.0100 0.0071 0.0012
udat/w5 -3.7312 0.0088 0.0071 0.0007
udat/w8 -3.6901 0.0067 0.0053 0.0006
udat/in1 -3.5159 0.0064 0.0052 0.0005

Reporting Variations
3-23
 pt_shell> report_variation [get_nets] -slack_lesser_than 0 -verbose

Net delay variation: udat/in3

from to slack delay mean stddev
------------------------------------------------------------------
ffi3/Q udat/u4/A1 r -2.9094 0.0100 0.0071 0.0012 *
ffi3/Q udat/u4/A1 f -2.1000 0.0092 0.0069 0.0009
ffi3/Q udat/u2/A1 r -3.9189 0.0094 0.0071 0.0010
ffi3/Q udat/u2/A1 f -4.0884 0.0088 0.0070 0.0008

Net delay variation: udat/w5

from to slack delay mean stddev
------------------------------------------------------------------
udat/u1/ZN udat/u6/A2 r -2.7033 0.0083 0.0071 0.0005
udat/u1/ZN udat/u6/A2 f -3.7312 0.0082 0.0071 0.0005
udat/u1/ZN udat/u4/A2 r -3.6901 0.0088 0.0071 0.0007 *
udat/u1/ZN udat/u4/A2 f -3.5759 0.0086 0.0071 0.0007

Net delay variation: udat/w8

from to slack delay mean stddev
------------------------------------------------------------------
udat/u4/Z udat/u5/A2 r -3.6901 0.0067 0.0053 0.0006 *
udat/u4/Z udat/u5/A2 f -3.5759 0.0065 0.0052 0.0006

Net delay variation: udat/in1

from to slack delay mean stddev
------------------------------------------------------------------
ffi1/Q udat/u1/A1 r -3.5142 0.0063 0.0051 0.0005
ffi1/Q udat/u1/A1 f -3.5159 0.0064 0.0052 0.0005 *

The default report lists the slack, the scalar net arc delay, and the mean and standard
deviation of the statistical net arc delay for the worst timing arc of the net. The worst arc is
the arc with the highest standard deviation of net arc delay among all timing arcs of the net.
The slack of an arc is the slack of the worst-case timing path through the net. In order to
report slack values, the variable timing_save_pin_arrival_and_slack must be set to
true prior to update_timing.

The verbose report shows the same information for all the timing arcs through the net, with
the worst arc marked with an asterisk.

Parameter-Specific Net Sensitivity

The report_variation command can report the sensitivity of net delay to specified
variation parameters. This type of reporting helps you find the nets that have the most delay
sensitivity to given parameters, which can provide guidance on where to fix or reroute nets
to reduce the delay sensitivity and produce a more reliable design.

Chapter 3: Variations and Correlations

3-24
For example, the following report shows the delay sensitivity of net n1 to all variation
parameters:

pt_shell> report_variation -parametric_sensitivity *\

[get_nets n1]
...
net parameter sensitivity
-----------------------------------------------
n1 metal2_T 0.035652
metal2_W 0.026661
metal3_T 0.025091
metal3_W 0.017360
metal1_T 0.015018
metal1_W 0.011877

The reported sensitivity values show the standard deviation of the net delay contributed by
each variation parameter, based on a calculation using changes to the net resistance and
capacitance with changes in the specified parameters.

Variations in Library Cells

To get a report on the sensitivity of library cells, create a collection of those cells and apply
the report_variation command. For example, the following script creates a collection of
all library cells and generates a summary report on the variation sensitivities of the cells.

pt_shell> set my_lib_cells \

[get_lib_cells -of_objects [get_cells]]

pt_shell> report_variation $my_lib_cells

A report on the library cell sensitivities looks like this:

Library sensitivity for cell: INVD1

delay sensitivity for cell INVD1 arc I->ZN

timing sense r negative_unate f negative_unate
----------------------------------------------------------------------
len: min/avg/max 0.000127/0.001236/0.003896 0.000066/0.000678...
vth: min/avg/max 0.040482/0.409654/1.425395 0.021974/0.286770...

slew sensitivity for cell INVD1 arc I->ZN

timing sense r negative_unate f negative_unate
----------------------------------------------------------------------
len: min/avg/max -0.000214/0.000842/0.003220 -0.000149/0.000278...
vth: min/avg/max 0.012500/0.268289/0.972281 -0.062368/0.102758...

Reporting Variations
3-25
 In this example, there are two independent random variables, len and vth, specified by the
set_variation_library command. The report shows the minimum, average, and
maximum sensitivity of each arc’s rise/fall delay and slew. Note that the library sensitivity
report collects only the static library information, which does not directly reflect the actual
behavior of the cells in the design.

In order to have slack values appear in the report, the

timing_save_pin_arrival_and_slack variable must be set to true before
update_timing.

Variations in Design Objects

The report_variation command can be used to report information about the variation
sources used in the design. For example, to report on the whole current design,

pt_shell> report_variation [current_design]

Variation libraries in design: TEST

library par1 par2 par3 par4 par5

--------------------------------------------------------------------
ncx_mc_nom.db 0.0000 0.0000 0.0000 0.0000 0.0000
ncx_mc_-P1.db -1.0000 0.0000 0.0000 0.0000 0.0000
ncx_mc_+P1.db 1.0000 0.0000 0.0000 0.0000 0.0000
ncx_mc_-P2.db 0.0000 -0.3333 0.0000 0.0000 0.0000
ncx_mc_+P2.db 0.0000 0.3333 0.0000 0.0000 0.0000
ncx_mc_-P3.db 0.0000 0.0000 -0.3333 0.0000 0.0000
ncx_mc_+P3.db 0.0000 0.0000 0.3333 0.0000 0.0000
ncx_mc_-P4.db 0.0000 0.0000 0.0000 -0.3333 0.0000
ncx_mc_+P4.db 0.0000 0.0000 0.0000 0.3333 0.0000
ncx_mc_-P5.db 0.0000 0.0000 0.0000 0.0000 -0.3333
ncx_mc_+P5.db 0.0000 0.0000 0.0000 0.0000 0.3333

Cell variation for design: TEST

parameter name type mean stddev

------------------------------------------------------------------
par1 par1 normal 0.0000 0.3330
par2 par2 normal 0.0000 0.3330
par3 par3 normal 0.0000 0.3330
par4 par4 normal 0.0000 0.3330
par5 par5 normal 0.0000 0.3330

Parasitic variation for design: TEST

parameter name type mean stddev

-------------------------------------------------------------------
FOX_T FOX_T normal 0.0000 0.0583
poly_T poly_T normal 0.0000 0.0333
poly_W poly_W normal 0.0000 0.0283
ILD_T ILD_T normal 0.0000 0.0510
metal1_T metal1_T normal 0.0000 0.0333
...

Chapter 3: Variations and Correlations

3-26
 The report is divided into sections. The first section reports all variation libraries with their
corresponding parameters for each variation source. These libraries and parameters are
defined by the set_variation_library command. The second section summarizes the
global device variation that is present in all cell arcs in the entire design. For each distribution
variable, the column labeled component corresponds to the -name specified in the
set_variation command; and the columns labeled type, mean, and stddev
correspond to the -type and -values settings in the create_variation command.
The succeeding sections report the variations present in each individual cell or arc, in
addition to the global variation settings. The “parasitic variation” section shows random
variables that represent variation sources in parasitic networks.

Variations in Parasitic Data

The -variation option of report_annotated_parasitics generates a report of the
variation sensitivity of a net. For example,

pt_shell> report_annotated_parasitics -variation \

[get_nets n1]

ParamID Name Stddev

--------------------------------------------------------
0 FOX_T 0.05882
1 poly_T 0.03333
...

The -variation option of report_annotated_parasitics generates a report of the

variation sensitivity of a net. For example,

pt_shell> report_annotated_parasitics -variation \

-list_annotated [get_net n1]
...
C in pF Node Pin/Port Sensitivity
---------------------------------------------------------------------
0.0003 1 ffc/QN (driver) FOX_T(0.006) poly_T(0.000) ...
0.0002 2 i1/A (load) FOX_T(0.001) metal1_T(0.046)...
...
---------------------------------------------------------------------
0.0064 Total

R in Kohm Left Node Right Node Sensitivity

---------------------------------------------------------------------
0.0091 1 ffc/QN (driver) 2 i1/A (load) metal2_T(0.320) ...
---------------------------------------------------------------------
0.0092 Total

Reporting Variations
3-27
 | | | | RC | Not |
Net Type | Total | Lumped | RC pi | network |Annotated|
------------------- +-------+--------+---------+---------+---------+
Internal nets | | | | | |
- Pin to pin nets | 1 | 0 | 0 | 1 | 0 |
- Driverless nets | 0 | 0 | 0 | 0 | 0 |
- Loadless nets | 0 | 0 | 0 | 0 | 0 |

...

The report lists the variations in the SBPF file, not the variation values set with the
set_variation command.

The report_net, report_timing -capacitance, and

report_delay_calculation commands show only the fixed values resulting from
nominal variation settings.

Chapter 3: Variations and Correlations

3-28
4
Variation-Aware Timing Reports 4
PrimeTime VX increases the accuracy of timing analysis and offers detailed reports on
distributions of timing results, such as delay and slack, for a given set of parameter
variations. Several reporting methods are available to get information about the distributions
of the timing results.

This chapter contains the following sections:

• Path Timing Report Increments

• Clock Timing Reports
• Custom Statistical Reporting
• Session Examples
• Validation Flow

4-1
Path Timing Report Increments
In a text-format timing report produced by the report_timing command, the arrival times
shown in the tables are fixed values based on the quantiles of the arrival time distributions.
In the default report format, there are columns labeled Incr (Incremental delay) and Path
(arrival time in the path), as shown in the following example.

pt_shell> report_timing $path2

...
Startpoint: wishbone/rx_fifo/cnt_reg[0]
(rising edge-triggered flip-flop clocked by wb_clk_i)
Endpoint: wishbone/IncrTxPointer_reg
(rising edge-triggered flip-flop clocked by wb_clk_i)
Path Group: wb_clk_i
Path Type: max

Point Incr Path

------------------------------------------------------------------
clock wb_clk_i (rise edge) 0.00 0.00
clock source latency 0.00 0.00
wb_clk_i (in) 0.00 & 0.00 r
wishbone/BUFFD16_G1B1I4_2/Z (CKBD16) 0.19 & 0.19 r
wishbone/BUFFD16_G1B2I135/Z (BUFFD16) 0.17 & 0.36 r
wishbone/rx_fifo/cnt_reg[0]/CP (SEDFCNQD1) 0.00 & 0.36 r
wishbone/rx_fifo/cnt_reg[0]/Q (SEDFCNQD1) 0.42 & 0.78 r
wishbone/rx_fifo/U71/ZN (NR4D2) 0.12 & 0.90 f
wishbone/rx_fifo/U21/Z (AN2D2) 0.12 & 1.02 f
...
wishbone/U1059/Z (OR2D4) 0.13 & 2.48 f
wishbone/IncrTxPointer_reg/E (SEDFCND0) 0.00 & 2.48 f
data arrival time 2.48

clock wb_clk_i (rise edge) 2.50 2.50

clock source latency 0.00 2.50
wb_clk_i (in) 0.00 & 2.50 r
BUFFD16_G1B1I2/Z (BUFFD16) 0.16 & 2.66 r
wishbone/BUFFD16_G1B2I56/Z (BUFFD16) 0.09 & 2.75 r
wishbone/IncrTxPointer_reg/CP (SEDFCND0) 0.00 & 2.75 r
clock reconvergence pessimism 0.00 2.75
clock uncertainty -0.10 2.65
library setup time -0.28 2.37
data required time 2.37
-------------------------------------------------------------------
data required time 2.37
data arrival time -2.48
-------------------------------------------------------------------
slack -0.11
statistical adjustment (slack) 0.10 -0.01
slack (VIOLATED) (with probability 0.012) -0.01

Chapter 4: Variation-Aware Timing Reports

4-2
 The variation_report_timing_increment_format variable affects the display of
paths which have been recalculated within the context of a variation-aware timing analysis.
The transition times, delays and arrival times associated with these paths are statistical in
nature. This variable lets you control the display of the delays appearing in the Incr column.

The variable can be set to effective_delay (the default) or delay_variation. When

it is set to effective_delay, each increment displayed is the scalar difference between
the arrival values appearing in the Path column. When it is set to delay_variation, the
quantile or mean value of the statistical increment is displayed instead, according to the
current setting of the variation_derived_scalar_attribute_mode variable.

The arrival times and transition times shown in the report are also scalar representations
based on the underlying distributions, converted into scalar values based on the current
setting of the variation_derived_scalar_attribute_mode variable.

Clock Timing Reports

The report_clock_timing command generates a report on clock timing in the design.
You can report the same-clock skew, interclock skew, latency, or transition time attributes of
a given clock network or sub-network. You can choose a summary report showing the worst
instances of skew, latency, or transition time; or a verbose report showing a trace through a
source-to-sink path through a clock network, including the accumulation of the reported
attribute (skew, latency, or transition time) along the path.

If you use the -variation option of the command, the report includes variation
information, including the quantile, mean, and standard deviation (for skew) or variation
sensitivity (for latency or transition time). To get a clock timing report with variation
information, the variation_analysis_mode variable must be set to the string
detailed_clock_timing.

For example, an ordinary summary clock timing report starts with the worst latency values,
like this:

pt_shell> report_clock_timing -type summary

...
Clock: wb_clk_i
------------------------------------------------------------------------
Maximum setup launch latency:
wishbone/bd_ram/mem1_reg[252][12]/CP 0.43 rp-+

Minimum setup capture latency:

wishbone/bd_ram/mem0_reg[198][7]/CP 0.23 rp-+

Minimum hold launch latency:

wishbone/bd_ram/mem0_reg[198][7]/CP 0.23 rp-+

Clock Timing Reports

4-3
 Maximum hold capture latency:
wishbone/bd_ram/mem1_reg[252][12]/CP 0.43 rp-+
...

By using the -variation option, the standard deviation, mean, and quantile values for the
worst latency values are shown in the report:

pt_shell> report_clock_timing -type summary -variation

...
Clock: wb_clk_i Stddev Mean Quant
------------------------------------------------------------------------
Maximum setup launch latency:
wishbone/bd_ram/mem1_reg[252][12]/CP 0.03 0.36 0.43 rp-+

Minimum setup capture latency:

wishbone/bd_ram/mem0_reg[198][7]/CP 0.02 0.29 0.23 rp-+

Minimum hold launch latency:

wishbone/bd_ram/mem0_reg[198][7]/CP 0.02 0.29 0.23 rp-+

Maximum hold capture latency:

wishbone/bd_ram/mem1_reg[252][12]/CP 0.03 0.36 0.43 rp-+
...

The quantile values are the same as the plain worst-case latency values shown in the
conventional report: the 99% quantiles for launch latency values and the 1% quantiles for
capture latency values. The quantiles used for reporting are affected by the setting of the
variation_derived_scalar_attribute_mode variable.

In a verbose report showing the point-by-point timing along a clock path, a variation-aware
report shows the mean, quantile, and sensitivity of the delay values along the path. For
example,

pt_shell> report_clock_timing -type latency -verbose -variation

...
Clock: wb_clk_i

Endpoint: wishbone/bd_ram/mem1_reg[248][14]
(rising edge-triggered flip-flop clocked by wb_clk_i)

--- Delay ---

Point Trans Sensit Mean Quant Path
--------------------------------------------------------------------------
clock wb_clk_i (rise edge)
clock source latency 0.00 0.00
wb_clk_i (in) 0.00 0.00 & 0.00 r
CKBD12_G1B1I5/I (BUFFD12) 0.03 0.03 0.00 0.00 & 0.00 r
CKBD12_G1B1I5/Z (BUFFD12) 0.51 0.07 0.16 0.19 & 0.19 r
BUFFD12_G1B2I266/I (BUFFD12)
0.51 0.03 0.00 0.00 & 0.19 r
BUFFD12_G1B2I266/Z (BUFFD12)
0.08 0.13 0.10 0.13 & 0.32 r

Chapter 4: Variation-Aware Timing Reports

4-4
 wishbone/bd_ram/CKBD6_Id_024/I (CKBD6)
0.08 0.00 0.00 0.00 & 0.32 r
wishbone/bd_ram/CKBD6_Id_024/Z (CKBD6)
0.18 0.07 0.10 0.12 & 0.43 r
wishbone/bd_ram/mem1_reg[248][14]/CP (SEDFQD1)
0.18 0.00 0.00 0.00 & 0.43 r
VA latency adjustment -0.00 0.43
total clock latency 0.43
--------------------------------------------------------------------------
latency (quantile) 0.43
latency (mean) 0.36
latency (sensitivity) 0.08

For information about variation sensitivity, see “Variation Sensitivity” on page 3-18.

Custom Statistical Reporting

PrimeTime VX can provide detailed statistical reports on delays, arrival times, and slacks.
These types of information are available as timing point, timing path, variation, and
distribution attributes. They can also be reported by the PrimeTime graphical user interface
in the form of tables and graphs.

The statistical variation attributes of the timing_point, timing_path, and variation

object types are available for paths with statistical timing data. These attributes are listed in
Table 4-1 through Table 4-3. You can extract the information stored in these attributes by
using the get_attribute and get_variation_attribute commands.

Table 4-1 Timing Point Object Attributes

Name and type Description

variation_arrival, collection Arrival time variation of the timing point

variation_slack, collection Slack time variation of the timing point

variation_transition, collection Transition time variation of the timing point

Table 4-2 Timing Path Object Attributes

Name and type Description

sensitivity, float Variation sensitivity of the path

variation_arrival, collection Arrival time variation of the path

Custom Statistical Reporting

4-5
Table 4-2 Timing Path Object Attributes (Continued)

Name and type Description

variation_common_path_pessimism, collection The variation of the clock reconvergence

common path pessimism.

variation_endpoint_clock_latency, collection Capture-clock arrival time variation

variation_endpoint_hold_time_value, collection The variation of the register hold time at the

timing endpoint.

variation_endpoint_recovery_time_value, The variation of the recovery time at the timing

collection endpoint.

variation_endpoint_removal_time_value, The variation of the removal time at the timing

collection endpoint.

variation_endpoint_setup_time_value, collection The variation of the register setup time at the

timing endpoint.

variation_required, collection Required-time variation of the path

variation_slack, collection Slack variation of the path

variation_startpoint_clock_latency, collection Launch-clock arrival time variation

Table 4-3 Variation Object Attributes

Name and type Description

full_name, string Name of the variation

parameter_name, string Name of the associated parameter defined in the

library or parasitic data file

mean, float Mean of the distribution of the variation

std_dev, float Standard deviation of the distribution of the

variation

variance, float Variance of the distribution of the variation

skewness, float Skewness of the distribution of the variation

summary, string Text information about the distribution of the

variation, including mean, standard deviation,
skewness, and five quantile points

Chapter 4: Variation-Aware Timing Reports

4-6
Table 4-3 Variation Object Attributes (Continued)

Name and type Description

cdf A single Cumulative Distribution Function value

obtained with get_variation_attribute.

cdf_values A sequence of successive Cumulative

Distribution Function value obtained with
get_variation_attribute.

pdf A single Probability Density Function value

obtained with get_variation_attribute.

pdf_values A sequence of successive Probability Density

Function values obtained with
get_variation_attribute.

quantile A single quantile value obtained with

get_variation_attribute.

Variation Distribution Attributes

You can use the get_attribute command to get information about a statistical
distribution in a variation object. For example, to get the standard deviation of variation p1:

pt_shell> get_attribute [get_variations p1] std_dev

0.310000

These are the distribution attributes you can get with the get_attribute command from
a variation:

• mean – The mean value.

• std_dev – The standard deviation.
• variance – The variance.
• skewness – The skewness.
• summary – A string containing a statistical summary of the distribution, consisting of the
mean, standard deviation, skewness, and quantiles q01, q05, q50 (median), q95, and
q99.
You can query this same type of information from a path. For example, for the path stored in
the variable “a_path” with the get_timing_paths command:

pt_shell> set path_slack \

Custom Statistical Reporting

4-7
 [get_attribute $a_path variation_slack]

you can get its mean value:

pt_shell> get_attribute $path_slack mean

-0.092147

and its statistical summary:

pt_shell> get_attribute $path_slack summary

mean= -0.092147
sdev= 0.134774
skew= -0.125206
q01= -0.455260
q05= -0.312938
...

Some variation attributes require a value to be specified before you can retrieve them, such
as quantiles. You can access these attributes by using the get_variation_attribute
command. For example, to get the q93 quantile value:

pt_shell> get_variation_attribute \
[get_variations p1] quantile .93
0.350722

These are the distribution attributes you can get with the get_variation_attribute
command:

• quantile q_value – The value of the distribution at the q-value quantile point, where
q_value is a floating-point number greater than 0.0 and less than 1.0.
• cdf cdf_value – The value of the cumulative distribution function at a CDF-value point,
where cdf_value is a floating-point number.
• cdf_values number_of_values – A list of distribution function values at the number
of data points specified by number_of_values (an integer), centered around the mean.
• pdf pdf_value – The value of the probability function at a PDF-value point, where
pdf_value a floating-point number between 0.0 and 1.0.
• pdf_values number_of_values – A list of probability density function values at the
number of data points specified by number_of_values, centered around the mean.

Chapter 4: Variation-Aware Timing Reports

4-8
 Attribute Access Examples
The following script creates two procedures, report_my_attr and
report_point_attr, which demonstrate some examples of accessing attributes in the
design.

proc report_my_attr { path } {

set points [get_attribute $path points]

foreach_in_collection point $points {

# report each timing point
report_point_attr $point
}
}

proc report_point_attr { point } {

# get string of pin
set pin [get_attribute [get_attribute $point object] full_name]
puts "Pin: $pin "

# get rise or fall

set rise_fall [get_attribute $point rise_fall]
puts "type: $rise_fall"

# for two variables arrival and transition

foreach v {arrival transition} {
# get corresponding random variable
set val [get_attribute $point $v]
set var variation_$v
set rv [get_attribute $point ${var}]
set mean [get_attribute $rv mean]
set sdev [get_attribute $rv std_dev]
puts "$v mean: $mean sdev: $sdev"
puts "99 percent quantile "
set q [get_variation_attribute $rv quantile 0.99]
puts -nonewline [format "q%02d = $q " $p]
}
}

Session Examples
The following script is a variation-aware analysis session using two variations, channel
length and threshold voltage.

set variation_enable_analysis true

set_variation_library -parameter_names "len vth" \

-values "90 0.22" -reference_value typical.db

Session Examples
4-9
 set_variation_library -parameter_names "len vth" \
-values "90 0.24" vthplus.db
set_variation_library -parameter_names "len vth" \
-values "90 0.20" vthminus.db
set_variation_library - parameter_names "len vth" \
-values "95 0.22" lenplus.db
set_variation_library - parameter_names "len vth" \
-values "85 0.22" lenminus.db

create_variation -name len -parameter_name len \

-type normal -values {0.0 10.0}]
create_variation -name vth -parameter_name vth \
-type normal -values {0.0 0.05}]
set_variation [get_variations len]
set_variation [get_variations vth]

update_timing -full

report_timing -nworst 2

set p1 [get_timing_paths -nworst 2]

report_timing $p1

set p2 [get_recalculated_timing_paths $p2]

report_timing $p2

The following script performs a variation-aware analysis using four variations named A, B,
C, and D.

# set variable defaults

set report_default_significant_digits 3
set variation_enable_analysis true

# define operating conditions

set_operating_conditions -analysis_type on_chip_variation

# set +/- sigma variation libraries

set_variation_library -parameter_names {A B C D} \
-values {0 0 0 0} \
lib_snom.db -reference_value
set_variation_library -parameter_names {A} \
-values -1 libA_neg.db
set_variation_library -parameter_names {A} \
-values +1 libA_pos.db
set_variation_library -parameter_names {B} \
-values -1 libB_neg.db
set_variation_library -parameter_names {B} \
-values +1 libB_pos.db
set_variation_library -parameter_names {C} \
-values -1 libC_neg.db
set_variation_library -parameter_names {C} \
-values +1 libC_pos.db

Chapter 4: Variation-Aware Timing Reports

4-10
set_variation_library -parameter_names {D} \
-values -1 libD_neg.db
set_variation_library -parameter_names {D} \
-values +1 lib_sD_pos1s.db

# set parameter variations

create_variation -name A -parameter_name A \
-type normal -values {0 1}
create_variation -name B -parameter_name B \
-type normal -values {0 1}
create_variation -name C -parameter_name C \
-type normal -values {0 1}
create_variation -name D -parameter_name D \
-type normal -values {0 1}
set_variation [get_variations A]
set_variation [get_variations B]
set_variation [get_variations C]
set_variation [get_variations D]

create_correlation -name die_to_die -constant 1

set_variation_correlation -name all_die_to_die \
-correlation die_to_die -all

# read parasitics, constraints, update timing

read_parasitics -keep_capacitive_coupling \
test_astro.spef.gz
read_sdc test.sdc

update_timing -full
# perform statistical reporting
set path1 [get_timing_paths -fall_from wb_adr_i[9] \
-to ethreg1/MODER_1/DataOut_reg[1]/D]
set path2 [get_timing_paths \
-rise_through wishbone/ram_addr_reg[1]/Q \
-rise_to wishbone/bd_ram/mem2_reg[211][23]/E]
set recalc1 [get_recalculated_timing_paths $path1]
set recalc2 [get_recalculated_timing_paths $path2]
report_timing -input_pins $path1
report_timing -input_pins $path2
report_timing -input_pins $recalc1
report_timing -input_pins $recalc2

Session Examples
4-11
Validation Flow
PrimeTime VX produces analysis results that show the distributions of delay and slew along
selected paths. To validate the accuracy of the PrimeTime VX analysis, you can compare
the reported results against a sequence of SPICE simulations of the same path, using a
Monte Carlo statistical representation of the variation distributions. This is the general
procedure for SPICE validation:

1. Perform the variation-aware analysis in PrimeTime VX.

For each timing path analyzed by PrimeTime VX, you can get detailed statistical reports
on the delay and slew attributes at each pin by using the get_attribute and
get_variation_attribute commands. The statistical reports can include the mean,
standard deviation, quantile values, and PDF or CDF delay and slew values.

2. Create the simulation data files for Monte Carlo SPICE simulation. The following
command generates multiple SPICE decks from PrimeTime VX, using randomly
distributed parameter and parasitic element values.
pt_shell> write_spice_deck -output path.spo \
-sample_size 24 \
-sub_circuit_file my_sub.sp \
[get_timing_paths -justify]
...

This example generates 24 SPICE decks in the directory named path.spo, using the
statistical distribution of parameter and parasitic element values defined in the variation-
aware design. Cross-coupling capacitors, if any, are not sampled, but are assigned their
corner values.

3. Start the SPICE simulation runs and collect the statistical results.
The SPICE simulation should be performed n times, where n is large enough to capture
the statistical properties of delay and slew on each pin according to the statistical
properties of each variation component. Each SPICE run must use a different set of
random numbers.

After the Monte Carlo sequence of SPICE runs is finished, generate reports on the delay
and slew values on each pin with n samples. You should create a report on the statistical
distribution of delay and slew on each pin, including the mean, standard deviation,
quantile values, and PDF or CDF plots.

4. Compare the PrimeTime VX results against the Monte Carlo SPICE results. The
statistical distributions should be similar, thus validating the accuracy of the variation-
aware analysis in PrimeTime VX.

Chapter 4: Variation-Aware Timing Reports

4-12
For device variations, the write_spice_deck command writes out the sample values in
the subcircuit portion as shown in the following example:

* .pin(sub_node) for inv1 (INVD1): .I(inv1/I) .ZN(inv1/ZN)

xinv1 inv1/I inv1/ZN INVD1 ssta_l=4.27218 ssta_delvto=-0.0114263
* .pin(sub_node) for inv2 (INVD1): .I(inv2/I) .ZN(inv2/ZN)
xinv2 inv2/I inv2/ZN INVD1 ssta_l=4.27218 ssta_delvto=-0.0114263

Each cell variation defined in PrimeTime VX is written in the form of variation_name =

sample_value. This format works with HSPICE only if the variations are also defined in the
cell subcircuit file, as shown in the following example:

.subckt AN2D0 A1 A2 Z ssta_l=1.0 ssta_delvto=0.0

MM3 M3:DRN M3:GATE M3:SRC M3:BULK nch ad=0.03p as=0.078p
l='0.1u*ssta_l' nrd=0.333 nrs=0.924 pd=0.5u ps=1.12u ps=1.12u
sa=2.6e-07 sb=9e-07 w=0.3u delvto=”+ssta_delvto”
MM1 M1:DRN M1:GATE M1:SRC M1:BULK nch ad=0.072p as=0.039p
l='0.1u*ssta_l' nrd=0.863 nrs=0.715 pd=1.08u ps=0.56u sa=9.2e-07
sb=2.4e-07 w=0.3u delvto='+ssta_delvto'

You can use the generated SPICE decks directly to run HSPICE multiple times. Then you
can compare the collected HSPICE simulation results against the statistical results reported
by PrimeTime VX.

Validation Flow
4-13
Chapter 4: Variation-Aware Timing Reports
4-14
Glossary GL

CDF
See cumulative distribution function (CDF).
corner case
A set of two or more parameter values that is a combination of the most extreme
values of those parameters, such as the combination of the highest threshold voltage
and the longest channel length.
correlation
The strength of linear relationship or co-dependence between two random variables.
For example, there is a very high correlation (nearly 1.0) between the channel lengths
of two adjacent transistors on a die.
cumulative distribution function (CDF)
A mathematical function that represents the cumulative probability that a parameter
will be less than or equal to a given value, across the full range of the parameter; the
integral of the probability density function.
derating
The scaling of delay values accomplished by multiplying the calculated delay values by
a fixed scaling factor, as specified by the set_timing_derate command.
distribution
A mathematical description of the range of values of a variation along with the
likelihood of occurrence of those values. A distribution is typically expressed in terms
of a few parameters, such as the mean and standard deviation of a normal (Gaussian)
distribution.
Gaussian
The normal distribution having a bell-shaped curve.
min/max analysis
Static timing analysis that is based on fixed worst-case minimum and maximum values
rather than a probability distribution.

GL-1
 normal (distribution)
The Gaussian distribution having a bell-shaped curve and described by two
parameters, the mean and the standard deviation.
parameter
A characteristic or property that has a variable numeric value and has a effect on delay
or slew that can be measured or simulated.
PDF
See probability density function (PDF).
piecewise-linear
A function that is made of linear segments connected end-to-end, which can be
characterized by the segment endpoints.
percentile
For a random variable X with continuous CDF function F, the percentile p associated
with value x is:

F ( x ) = P [ X ≤x ] = p

For example, if the slack at a pin follows a normal distribution with mean 0.0 and
standard deviation 1.0, the probability that the slack is negative is 0.50, so 0.50 is the
percentile associated with x-value 0.0.
principal component analysis
A mathematical procedure that simplifies a data set by transforming a number of
correlated parameters into a smaller number of uncorrelated parameters, called the
principal components. The first principal component contains as much of the data
variability as possible, the second component contain as much of the remaining
variability as possible, and so on.
probability density function (PDF)
A mathematical function that represents the likelihood that a parameter will have a
specific value across the full range of values of that parameter; the derivative of the
cumulative distribution function. It is similar to a histogram that has been smoothed to
make a continuous function.
quantile
For a random variable X with continuous CDF function F, the quantile q associated
with value p is given by:
–1
q = F (p)

For example, if a path arrival time A has a 99 percent quantile equal to 4.3, it means
that the probability is 0.99 that an actual arrival (an outcome of A) will be less than or
equal to 4.3.

GL-2
 For any timing distribution “A,” the sensitivity of that distribution is defined to be the
normalized variance of its distribution, calculated as follows:
sensitivity(A) = std_dev(A) / | mean(A) |

sensitivity
For any timing distribution “A,” the sensitivity of that distribution is defined to be the
normalized variance of its distribution, calculated as sensitivity(A) = std_dev(A) /
|mean(A)| .
sigma (σ)
The standard deviation, the most common measure of statistical dispersion; the
square root of the variance. A large sigma means that the data samples are widely
dispersed around the mean. A small sigma means that the data samples are tightly
clustered around the mean.
skewness
A measure of the asymmetry of a random variable. The skewness S of a random
variable X with mean μ and standard deviation σ is defined as
3
E [ ( X – μ) ]
S = -----------------------------
3
σ
where E denotes the expected value with respect to the probability distribution of X. If
the skewness is zero, the PDF plot is symmetric around its mean. If the skewness is
positive or negative, the plot is not symmetric around its mean.
slew
The transition time of a signal; the amount of time a signal takes to change from high
to low or from low to high.
standard deviation
The sample estimate of sigma (σ).
uniform distribution
A distribution that has the same value throughout a given range.
variation
A random variable that represents a process parameter or device parameter that can
be characterized, such as channel length or threshold voltage.
variation-aware static timing analysis
A technique to accurately analyze circuit performance in the presence of variation in
process and environment parameters (such as channel length and threshold voltage).
This approach significantly improves the accuracy of traditional static timing analysis
by using the actual distributions of parameter variations instead of worst-case corner
assumptions.

GL-3
GL-4
Index

A auto-correlation 3-10
cross-correlation 3-12
add_variation command 3-14
spatial correlation 3-12
adding variations 3-14
cross-correlation 3-12
attributes, variation-aware 4-5
cumulative distribution function 1-18
auto-correlation 3-10

D
C derating 1-4
CDF 1-18
derating statistical variations 3-15
commands
derived parameters 1-13
add_varation 3-14
get_random_numbers 3-14 discrete distribution 3-8
max_variation 3-14 distribution
min_variation 3-14 attributes 4-7
remove_parasitic_corner 2-14 constant 3-6
report_clock_timing 4-3 discrete 3-8
report_timing 4-2 empirical 3-6
report_variation 3-17 lognormal 3-7
set_library_variation 2-6 normal 3-4
set_timing_derate 3-15 piecewise-linear 3-5
set_variation 3-2 types 3-4
sub_variation 3-14 uniform 3-6
write_spice_deck 4-12
constant distribution 3-6 E
corner case 1-4
empirical distribution 3-6
correlation 3-9

IN-5
G R
get_random_numbers command 3-14 random number generation 3-14
remove_parasitic_corner command 2-14
report_clock_timing command 4-3
I report_timing command 4-2
interconnect parameter variation 2-9 report_variation command 3-17
calculations 2-10 cells 3-22
read_parasitics 2-11 design objects 3-26
library cells 3-25
nets 3-23
K parameter-specific net sensitivity 3-24
K factors (derating) 1-4 parasitic data 3-27
timing paths 3-21
L
library S
creation 2-2 scaling (derating) 1-4
merged 2-3
sensitivity 3-18
lognormal distribution 3-7
set_timing_derate command 3-15
set_variation command 3-2
M set_variation_library command 2-6
math operations on variations 3-14 spatial correlation 3-12
max_variation command 3-14 sub_variation command 3-14
merged library 2-3
T
N timing report increments 4-2
normal distribution 3-4
U
P uniform distribution 3-6
piecewise-linear distribution 3-5
principal component analysis 1-13 V
validation flow 4-12
Q variables
variation_derived_scalar_attribute_mode
quantile points 1-18 3-20
variation_enable_analysis 1-22
variation

IN-6
 correlation 3-9 variation-aware static timing analysis
interconnect parameter 2-9 enabling 1-22
math operations 3-14 flow example 1-14
reporting 3-17
sensitivity 3-18
variation_derived_scalar_attribute_mode W
variable 3-20 write_spice_deck command 4-12
variation_enable_analysis variable 1-22

IN-7