Design specific AOCV tables unleash the full potential of stage based on chip variation for both timing and optimization.
Littleton, MA March 31, 2011 — CLK Design Automation Inc. today announced a breakthrough in timing analysis and optimization: Design Specific Stage Based AOCV tables. Design Specific AOCV tables provide unprecedented ability to accurately account for manufacturing variance in timing and optimization; all using existing STA and physical optimization tools and flows. Amber FX AOCV generates a SPICE accurate database for all possible AOCV tables for a library in a fraction of the time it requires other approaches to create one table. Creating a Design Specific AOCV table from the database takes less than a minute.
“Design Specific stage based AOCV tables created with Amber FX AOCV provide the best of both worlds” said Isadore Katz, President and CEO of CLK Design Automation. “First and foremost, they are safe, accurate, and practical. They are SPICE accurate, they are a safe representation of design variance without the excessive pessimism or random guesswork of other OCV approaches, and they are practical, working with any STA or optimization tools that support TSMC’s Reference Flow 11. Second, they are a breakthrough innovation that truly breaks the bounds of traditional corner based methodologies to improve performance, power and yield.”
Design Specific stage based AOCV table generation is a new capability offered in Amber FX AOCV. Amber FX AOCV table generation is the only qualified table generation solution in TSMC’s Reference Flow 11. The previous release of Amber FX AOCV introduced a 100x speedup, making it by far the fastest and only practical, safe solution for table generation.
The Initial Stage Based AOCV Approach has Severe Limitations
Stage based on chip variation is currently being used as a methodology model manufacturing variance during static timing analysis. AOCV derating tables can be used by a variety of STA tools including Synopsys’ PrimeTime™ and Cadence Encounter Timing System™. Amber FX AOCV generates derating tables using cell depth, complexity, and weighting factors as well as design specific statistical information included in the cell library SPICE models.
However, stage-based AOCV tables have two severe limitations. The first is the time it takes to generate tables using SPICE or Fast SPICE. Accurate approaches using Monte Carlo or smart sampling take weeks or months.
The second limitation is unrealistic pessimism. In fact, the tables generated by first generation SPICE solutions can be significantly more pessimistic than traditional OCV derates. This is because the first generation tables found the most pessimistic derate based on worst/worst slew and slope – even if none of the cells in the design exhibit these values.
Other vendors have proposed solutions to the performance and pessimism problem by suggesting that reducing the table size, or guessing at the worst/worst value was sufficient. At best these approaches are SWAGS, and create a table that is both pessimistic and unsafe.
Introducing Design Specific Stage Based On Chip Variation
Design Specific stage based AOCV tables generated with Amber FX AOCV address both of these limitations. First, Amber FX AOCV generates a complete database of all possible table values in a fraction of the time other approaches take to guess at the values. A complete AOCV database for a library can be generated in 3 days using 3 typical multi-core machines. A database can be generated in much less time by adding more machines or cores. Generating a specific table for a design takes less than one minute.
Second, the unrealistic pessimism is eliminated. Rather than using an unrealistic worst/worst load slew value for all cell instances in a design, the table uses actual load/slew values taken from the design. The worst load/slew value for a cell as selected from the design sets the table value. The selection of the load/slew value can be changed depending on the application. For optimization or STA, the critical path cell instances can be used to set the values. For power optimization, non-critical paths may be tapped to reduce power, the entire design load/slew range can be applied to set the range. After an optimization pass, regenerating a table to reflect new loads and slews can be done in seconds.
The design specific tables are accurate, safe and practical. Accurate: using the Amber FX analysis capabilities, the AOCV database is typically within 2% or better of Monte Carlo SPICE. Safe: the design specific tables are based on the actual load/slew cell instances in the design, not an arbitrary guess. Practical: these tables work with all STA and optimization tools that support the TSMC Reference Flow 11 format.
About Amber FX AOCV
Amber FX leverages patented technology to deliver accurate, fast, and practical solutions for high accuracy timing. Amber FX AOCV uses transistor level statistical analysis to build SPICE accurate advanced OCV tables. Amber FX AOCV is one million times faster than Monte Carlo SPICE, making this technology practical for real world designs, and uses industry standard input and output formats to complement existing design flows.
About CLK Design Automation
CLK Design Automation is the technology leader in high accuracy timing solutions for nanometer semiconductor designs. CLK DA was founded in 2004, and is backed by Morgenthaler Venture Partners and Atlas Ventures. Its Board of Directors includes Barry Fidelman from Atlas, Bob Pavey and Paul Levine from Morgenthaler, and noted EDA entrepreneur and visionary Jim Hogan.
Amber and Path FX are trademarks of CLK Design Automation. All other trademarks or registered trademarks are the property of their respective owners.