AOCV Analysis using PrimeTime

• Integrated circuit designers have been using a global derating value for years, adding design margin to account for on-chip variations (OCV).
• An AOCV analysis is a natural extension of OCV analysis
• AOCV uses variable derating factors, considering location and logic depth of each path
• AOCV reduces design margins and timing violations by using context-specific derating values.
• This pessimism-reducing approach minimizes over-design and enables faster timing closure.
• PrimeTime uses derating tables for AOCV information, calculating variable derating factors based on path location and depth.
• The PrimetTime AOCV solution integrates into graph and path-based analysis capabilities
• This enables a successive refinement strategy similar to traditional static timing analysis (STA).

Graph-Based AOCV Analysis

• Graph-based AOCV analysis delivers fast, design-wide results using the update_timing command.
• Pessimism reduces in graph-based AOCV compared to OCV analysis allowing it to be enough for signoff
• Graph-based AOCV serves as a prerequisite for further analysis and margin reduction through path-based AOCV analysis on critical paths.

Path-Based AOCV Analysis

• Path-based AOCV analysis reduces pessimism via path-specific recalculation on paths still violating after graph-based AOCV.
• Path-based AOCV analysis uses path-specific path-depth and location-based bounding box values to determine path-specific derating values
• Path slack is not worsened in path-based AOCV.
• Graph-based AOCV analysis before path-based analysis is essential
• Path-based AOCV analysis is computationally intensive, scaling complexity with analyzed paths, potentially exponentially

Significance of AOCV Metrics

• Metrics like path-depth and location-based bounding box are used to derive a context-specific AOCV derating factor that replaces global derating.
• PrimeTime selects conservative path-depth and location-based bounding box values for graph-based AOCV analysis
• Graph-based analysis is, by design, more conservative than path-based analysis to prevent optimistic calculation; depth counts are limited to prevent optimistic results.

Path-Based AOCV Metric Calculation

• Path-based AOCV calculates metrics on a per-path basis
• Path depth counts objects (leaf cells or nets) in a path, with potential differences between net and cell path depths.
• Systematic variation between cells on a die is a function of the distance between cells (in nm), and PrimetTime assumes net locations do not affect systematic variation, calculating path distance separately for cells and nets.
• Cell path distance involves computing the diagonal of a box encompassing all cells in the path.
• Similarly, net path distance computes the diagonal of a box encompassing all nets in the path
• PrimeTime assumes zero variation in the common path, because launch and capture signals are the same. Thus, it defaults to excluding net or cell objects in the common path for AOCV metric computations.

Graph-Based AOCV Metric Computation

• Graph-based analysis involves many paths for each object (cell and net)
• Depth and distance metrics are computed conservatively to avoid optimism
• Depth values vary for data and clock networks,
• Arcs in clock networks can have separate depth and distance values for launch, capture, and data

Combinational Convergence and Divergence:

• Objects at convergence inherit pessimistic metrics.
• Objects at divergence inherit pessimistic metrics
• SolvNet article 022963 illustrates path distance calculation.
• AOCV metrics are measured from the clock reconvergence pessimism removal (CRPR) common point in Path-based analysis
• PrimeTime selects the conservative common point for analysis to avoid optimism, if there are multiple possible common points for a cell.

AOCV Flow Phases

• AOCV flow has setup and reporting phases
• Setup includes:
• Reading the design
• Loading parasitics with coordinate information
• Applying constraints
• Enabling AOCV analysis
• Specifying AOCV information
• Reporting includes deratings and analysis results

AOCV Analysis Setup Steps

• Read in the design (step 1)
• Import SPEF with physical locations (step 2)
• Enable AOCV analysis (step 3)
• Specify AOCV information (step 4)

Importing SPEF With Physical Locations

• Path distance calculation requires coordinates, while path depth does not
• Coordinates of net nodes, pins, and ports must be read into PrimeTime from SPEF or SBPF files. Set read_parasitics_load_locations to true, then use the read_parasitics command.

Enabling AOCV Analysis

• Graph-based AOCV analysis is enabled by setting timing_aocvm_enable_analysis variable to true

Specifying AOCV Information

• The command read_aocvm aocv_file_name loads AOCV derating information from a disk file. The derating factors must be specified in a table using the Synopsys AOCV file format for the values to be directly applied to the arcs in the paths of interest
• If there is incorrect Syntax PrimeTime issues and error message
• AOCV tables can be annotated onto one or more design objects
  • library cells
  • hierarchical cells
  • designs For cell arcs, the precedence of data tables is the design derating table, the hierarchical cell derating table, and the library cell derating table (lowest to highest)
• Cells or net derating factors will be defined using path depth and path distance.
• When dealing with values missing on the exact path depth/distance; PrimeTime will use linear interpolation

AOCV Reporting Phases

• Reporting AOCV Deratings
• Reporting AOCV Analysis Results

Reporting AOCV Deratings

• The command report_aocvm displays AOCV information including derating table data
• This will also show objects annotated with early, late, rise, fall, cell, or net derating tables
• It displays computations of graph and path-based AOCV metrics. -A summary can be displayed by using the command report_aocvm, that shows the numbers of lead cells annotated with derate tables. If you specify a timing path argument, PrimeTime displays path metrics (distance, launch depth, and capture depths)
• Note: The path depths have been scaled by the AOCV coefficients if they exist

Reporting AOCV Analysis Results

• All graph based reporting commands reflect upon the AOCV results after AOCV analysis is enabled.
• Below contains some commands to report results : -report constraint -report clock timing - derate -report_timing -nworst 2000000 -max_paths 2000000 -derive For path-based AOCV analysis, command report timing with pba_mode path option displays the timing report with AOCV derating factors

Using get_timing_paths:

• You can query and obtain an AOCV-adjusted path by using the get_timing paths command with the pba_mode path option
• Prime time does not perform a path search to determine whether those paths are recalculate
• The path ordering reports matches AOCV timing reports in its based path ordering. By setting the command get_timing_paths -pba_mode exhaustive option time Prime invokes its extensive path that finds and correctly orders
• Note: There's a derate column that displays both graph-based and path-based. AOCV analysis

Exhaustive Search

• Run report_timing command with the -pba__mode exhaustive option find all correctly ordered
• Requires the pba_exhaustive_endpoint_path_limit variable avoids the runtime if we terminate exhaustive path search.
• Strong suggestion, set pba_aocvm_only_mode variable set in True to reduce the pessimism

AOCV File Format

• Specifies AOCV derating tables in the AOCV file format
• Supported Table Types:
• One dimensional tables
• Two dimensional tables for both depth and distance

Syntax Definition

• Described Section requires you to specify field for each line

version: aocv_format_version

• Used for back compatibility
• Valid version numbers 1.0, 2.0, 3.0
• Set the path type to 2.0
• Set the group name to 3.0

group_name: name

• (Optional field) that specifies AOCV table for that table group assigned.

object_type: design | cell | lib_cell

• You can specify object types,
• Cell and Net tables are design specific
• Cell and Net are tables for hierarchical cells using cell object
• Cell tables for library cells use the lib_cell type

rf_type: rise | fall | rise fall

Requires to specify what you need to select

delay_type: cell | net | cell net

• You need to specify either the cell, net, or cell net.
• PrimeTime can annotate them

derate_type: early | late

• you can specify either way

object_spec: [patterns] [-filter expression]

• Definiton that specifies the object name, its expression which is based attribututes.
• (Same as regexp tcl command)

voltage: floating_number

• Suppoorts for the multivoltage AOCV derating command

path_type: data | clock | data clock

• This is an optional field if your specifying on tables depending paths with clock/data
• version: 2.0 field must be specified

Depth: list_of_floating_numbers

• Specify path for depth set of the M point
• M can be set to zero

Distance: lsit_of_floating_numbers

• Specify point if the paths distance set
• N can be zero

Table: array_of_NxM- numbers

• Factors for their derating have to be specific in a table
• There is also some special cases of code

Comment line

• To indicate comments is double slash (\) at start of each line

Additional ADCov Capabilities

• Clock and data network AOCV analysis
• Clock network restrictuions
• Single Pass metrics for all objeccts
• Combining launch and capture
• Seperate value for their pass in clocks
• AOCV Coefficients
• Guard-Banding in AOCV
• Using OCV Deratings
• Incremental Timing Deratings
• Applying Voltage-Dependent
• AOCV Table Groups
• Encrypting AOCV

Interactions With Other PrimeTime Features

• Pessimistic Removal (CRPR)
• OC Behavior
• Distributed Analysis
• Hierarchial Flows
• Delta Delay And Timing
• Combination and Aware Analysis

Enable CRPR

Report Timing Command

• This has a reduced runtiime using its smaller n - worst with setting
• Main benefit helps the reduce/improve results accuracy for the designs

AOCV Table Group

• Using AOCV tables for designs and providing methods for specific tables contains different ratings values to portions. With controlled designs these help lead into their margining needs

CRPR

• Set variable in Timing_remove_clock_reconvergence_pessism too True if you needed graph base values used for past base for AOCV analysis. If removed than common portions is not derived for base analysis