Compare NLDM and CCS for multivoltage scaling - which is the right choice and why?
From PDVerse PrimeTime STA Interview Guide · pdVerse Mentor Guide
Ten-second Interview Answer
CCS is the recommended model for scaling library groups: it represents the cell as a current source and captures nonlinear voltage dependence, whereas NLDM's voltage-independent delay and slope tables drift from SPICE across a wide voltage sweep.
Complete Technical Explanation
NLDM describes a cell with delay and output-slope tables that are indexed by input slew and output load, but that carry no real voltage dependence of their own. Within a narrow voltage band that is fine. Stretch it across a wide sweep and the single-slope assumptions behind those tables progressively diverge from what SPICE says the cell actually does, because the shape of the driving waveform, not just its scale, changes with supply voltage. CCS takes a different approach: it models the cell as a current source, so the driver behaviour is described in a form that reproduces the nonlinear dependence on supply voltage much more faithfully. The recommendation to use CCS for scaling library groups follows directly from that. Multivoltage analysis is, by definition, the business of running the same cells at several different voltages and interpolating between characterized points - which is precisely the regime where NLDM's error is largest and where CCS earns its extra characterization and runtime cost. Put another way, the accuracy gap between the two models is smallest exactly where you do not need scaling and largest exactly where you do. If you are building voltage scaling library groups for a DVFS design, treating NLDM as good enough imports an error that grows with the very voltage range you added scaling to cover.
In Pd / Signoff
At library setup for any multivoltage or DVFS signoff, you check what the scaling library group is actually made of. If the vendor delivered only NLDM data, you know your interpolated corners carry model error that gets worse the further you are from a characterized voltage, and you either request CCS libraries or tighten the spacing of characterized voltage points. It also gates DVD analysis, which needs CCS noise data.
Common Trap
Assuming interpolation accuracy depends only on how many voltage points were characterized, and so building a scaling library group from NLDM data on the argument that the points are close together. The model form is the dominant error source across a wide sweep, not the point spacing.
Expect Next
Why does modelling the driver as a current source rather than a delay table capture voltage dependence more accurately?
Topic And Primetime Commands
Multivoltage / General. No specific PrimeTime command is required for this conceptual answer.
Keep building interview depth
Continue practising STA
Continue learning free
Get a practical low-power chapter
Receive the existing “Low Power and Multivoltage Fundamentals” PDF chapter and its download link by email.
Continue practising