Ultra Fast AOCV Table Generation
With its breakthrough variance engine, AOCV FX generates advanced on chip variation tables in a fraction of the time required by approaches based on Monte Carlo SPICE simulation.
The FX variance engine is 1 million times faster than Monte Carlo SPICE but still easily within 2% of the results for total path delay.
AOCV FX can generate tables for a full library – every cell, every arc, for clock and data cells – in hours, not days or weeks like other approaches.
|MC SPICE||MC Fast SPICE
|38 Simple Cells||3 Weeks
|882 Cell Library||103.5 Years
Why stage-based on chip variation?
At 40nm and below single value value “global” OCV derate values are overly pessimistic in most cases and dangerously optimistic for short paths. Stage-based advanced on chip variation (AOCV) reduces pessimism and improves margin without impacting quality during static timing analysis and optimization.
AOCV models manufacturing variation. The AOCV tables include separate derate factors for each cell based on the depth of the logic, or stages, for both clock and data paths. Depth helps to take into account local on-die variation independently from global die to die variation.
Stage-based AOCV tables work with all of the major static timing and optimization tools, and can be adopted with no changes to the primary tool flow. AOCV and can be applied by traditional timing and optimization tools without requiring full statistical timing analysis.
Clock and data, every cell, every arc
AOCV FX generates AOCV tables for every cell in your library – sequential or combinational – for both clock and data paths.
AOCV FX doesn’t take risky shortcuts during derate calculation like cherry picking load/slew points or eliminating complicated cells because of the number of timing arcs. AOCV FX is able to calculate derate values at all specified depths instead of extrapolating values based on a single stage.
With AOCV FX, you aren’t limited to applying derate values for a few cells. Instead, you’ll be able to see the value of reducing global OCV and applying AOCV on the whole design early on in your design flow.
|Derate Method||WNS (ns)||TNS (ns)||Per Path Gain|
|Clock Only AOCV||-0.386||-2380||±10ps|
|Data & clock AOCV||-0.328||-1625||79ps|
|Design specific AOCV||-0.274||-1074||134ps|
|Instance specific AOCV||-0.257||-881||154ps|
Design Specific AOCV
AOCV FX builds a unique variation database for continued refinement of your AOCV tables – removing pessimism safely at each step in the process.
Start with standard AOCV tables providing a safe, worst-case, derate value. Next run a simple Tcl script in your timing tool and capture the actual load and slew values for cell instance in your design. Finally generate new AOCV tables with derate values based on how your cells are actually used.
Design specific AOCV table generation is fast – a few seconds for all of the cells in your library. Flip a switch and generate instance specific tables for unique derate values for each cell instance.
What about ECOs? No problem. Generating new tables based on your latest design changes is just as fast – just a few seconds – you won’t be saddled with out of date data or waste time waiting for updates.
Validated by TSMC
AOCV FX was introduced in TSMC Reference Flow 11.
Design specific tables were added in Reference Flow 12.
Benefits of AOCV FX for AOCV Table Generation
AOCV FX is threaded and distributed and scales based on the number of CPUs available. It’ll run great on an 4 or 8 CPU machine but if you’ve got a grid or “cloud” at your disposal you’ll have AOCV tables even faster.
Contact email@example.com for more information and to set up an evaluation today.