What we check
Every timing arc of every cell, at SPICE level and in Verilog through your timing analyzer, with RC network loads and instances of the same cell as fanout, so driver and receiver models are both tested.
Wrong timing values
Entries that disagree with SPICE under the same conditions.
Missing timing arcs
Critical arcs, often state dependent, that the library does not describe.
Input capacitances
Receiver loads that are wrong for the state or the arc.
Slope and load ranges
Tables that compromise delay calculation accuracy or are derated out of range.
Liberty vs Verilog
Models that are not compatible with each other.
Structural errors
Problems in driver and receiver models and in the library structure.
How it works
Functional match
ACDL descriptions are generated from the Liberty library with Syn2acdl and checked against the SPICE netlists.
Test benches
A Verilog and a SPICE test bench per timing arc, driving a DSPF RC network with the same cell as fanout.
Compare
Timing analyzer delays are compared with SPICE for every arc and corner.
Report
Deviations are listed and narrowed down by changing fanout, RC loading or characterization set-up.
Who it is for
- Teams moving to current source models
- You need to know CCS and ECSM timing holds under real RC loads.
- Library developers
- Catch wrong entries and missing arcs before your customers do.
- Design teams using third-party libraries
- Check a library before you build a chip on it.
Questions
Does the library have to be built with SolutionWare?
No. The check is designed for libraries made by other means, including commercial ones.
Is it only for CCS?
No. It was designed for current source models like CCS and ECSM, and it also tests NLDM libraries for interpolation errors.
Which cells?
Core cells, IOs and memories.
How long does it take?
The run time of CCSTEST is a small fraction of the characterization time.
Tell us about your library
Process, number of cells, model types and the timing analyzer you use. We will reply with what a check would involve.