This article describes the composite current source (CCS) noise modelfor cell-level noise analysis. CCSisa modeling capability in the open-source Liberty cell modeling standardthat encompasses timing, noise and power. It models the transistorbehavior of a cell for accurate calculation of noise-including couplednoise and noise propagation through cells.
Signal-integrityanalysis is critical at geometries of 130nm and below. The impactof crosstalk on delay must beconsidered, as well as potential functional failures due to noiseglitches on nets that should not be switching.
This article describes a cell-level model to enable accurate andefficient functional noise analysis. CCS noise is a new current-basedmodel that enables accurate noise analysis with results very close to Spice simulation.
It precisely models injected crosstalk noise bumps and allows moreadvanced analysis, such as propagated noise bumps and driver weakening,without significant characterization effort.
With CCS noise, the noise immunity of the cells can be obtainedduring analysis by using actual noise bump waveforms, eliminating theneed for expensive cell-noise immunity characterization. This dynamiccomputation of noise propagation and immunity enables librarycharacterization to be 100 times faster than table-based models.
The need for current-based drivermodels
Current-based driver models are necessary for accurate crosstalk noiseanalysis. The desired cell current model for crosstalk noise analysismust be able to interface with arbitrary input noise waveforms andcoupled load interconnect networks. CCS noise is an advanced currentdriver model that captures both static and transient characteristics ofthe cell.
The static component of the CCS noise model consists of a currenttable as a function of I/O voltage levels, which can be providedthrough an efficient DC analysis known as a basic V i V o (V in /V out ) model.
The key advantage of CCS noise is that it uses several dynamicparameters to model the dynamic response of the cell that a staticcurrent table cannot capture. Dynamic parameters are extracted fromtransient analysis measurements that record the response of the cell tocertain input transitions and noise bumps.
CCS noise can accurately model all crosstalk noise analysis effects,including noise calculation, noise propagation, driver weakening andcombined noise propagation and noise injection. In this context, noisecalculation refers to the calculation of an aggressor net's injectednoise bump characteristics, assuming the victim driver is quiet.
Noise propagation refers to the problem of propagating a noise bumpthrough a circuit cell, assuming there is no coupling in the celloutput net. Combined noise propagation and noise injection analysisrefer to the general case where there is noise propagation through thevictim driver and noise injections by aggressors of the victim net.
Driver weakening is a special case of noise combination, where thepropagated noise is very small, but significantly reduces the drivingstrength of the victim driver due to increased injected noise bumpsize.
|Figure1. The green curves are input noise waveforms; blue curves|
Extensive studies have shown that CCS noise is considerably moreaccurate than other models for these crosstalk noise analysis tasks,especially because of its noise dynamic parameters.
For example, Figure 1 aboveshows that propagated noise waveforms computed using CCS noise matchSpice waveforms much better than those computed using the basic V i V o driver model without the dynamic parameters.
Accurate noise analysis at 90nm and below also requires a receivermodel that captures the dependency of effective receiver pincapacitance on input transition time and output load capacitance.Single-value pin capacitance models common in timing models are notsufficient.
The CCS noise analysis flow can take advantage of its powerfultiming receiver model, which is dependent on input-transition andoutput- load. The noise-analysis engine includes varying inputcapacitance effects of victim receivers; no extra receivercharacterization is required. V i V o -based current driver models bestcapture the behavior of a single channel-connected block(CCB).
For complex circuit cells having more than one CCB, thetransistor-level netlist of the circuit cell needs to be broken intomultiple CCBs and a CCS noise model for each of those CCBs.
The transistor-level net list breaking and CCS noise model parameterextraction are performed during cell characterization. Oncecharacterized, the CCS noise model data are stored either on a timingarc or on a pin in the cell library, depending on the topology of thecircuit netlist.
For circuit cells having a single CCB for an I/O pin pair, one noisemodel is extracted and stored on a timing arc. Such single-stage cellsinclude most inverters, and NAND, NOR, AOI and OAI gates etc. Forcircuit cells having two subsequent CCBs, two CCS noise models arestored on a timing arc.
Such two-stage cells include most of the buffers, and AND, OR,AND-OR and OR-AND gates etc. For circuit cells having three or moreCCBs (including most of the flip-flops), full adders, digital macroblocks, CCS noise model data are stored on pins.
|Figure2. When the noise bump is large enough and the propagated noise bump islatched by the sequential cell, it will cause a functional error bychanging the logic value of the sequential cell.|
Victim and aggressor net cross-couplingCross-coupling between a victim net and its aggressor nets induces anoise bump on the victim net when the aggressors switch. When the noisebump is large enough and the propagated noise bump is latched by thesequential cell, it will cause a functional error by changing the logicvalue of the sequential cell (Figure2, above ).
It will also increase the dynamic power consumption induced by noisebumps. The analysis tool should provide flexible reporting of potentialnoise violations.
The designer may want to fix any large noise bump in the design ormay only fix the noise bumps that can propagate all the way to anendpoint such as the D pin of a flip-flop. Crosstalk analysis toolssuch as PrimeTime SI canprovide detailed information about the source of the propagated noiseviolation.
|Figure3. U4 is a highly noise-immune cell and can tolerate noise withoutcausing a failure.|
Consider the scenario shown inFigure 3, above . U4 is a highly noise-immune cell – it cantolerate a certain amount of noise without causing a failure. Itspropagated noise is significantly attenuated such that no functionalerror occurs in FF.
A CMOS circuit cell can tolerate a certain amount of noise withoutcausing a failure at the cell output. This characteristic is callednoise immunity. CCS noise information can be used to calculate thenoise immunity of a cell during analysis.
Using CCS noise and the actual shape of the input noise bump, the noiseimmunity of the cell can be computed as well as the noise slack – i.e.the amount of noise height that needs to be added to the noise bump tocause a failure.
|Figure4.The test circuit has two aggressor nets coupled to a victim net,which for noise propagation analysis.|
When the noise slack is negative, it means that the noise bump hasexceeded the noise immunity parameters for the cell and can beconsidered a failure. When the noise slack is positive, the noise bumpis within the noise immunity parameters.
Figure 4 above shows theresults of a crosstalk noise correlation with Spice. The test circuithas two aggressor nets coupled to a victim net. It has a fan-out netfor noise propagation analysis.
Figure 4b compares CCSnoise and Spice waveforms for a typical case. The difference inwaveforms is barely discernible at the resolution of the plot. Thecell-driving strengths, the input transition times of the aggressordrivers and the lengths of interconnects have been varied to cover manyscenarios for accuracy correlation.
The noise bump height correlation at point A (for noise calculation)is shown in Figure 4c . Thecorrelation at point B (for both noise calculation and propagation) isshown in Figure 4d .
CCS noise delivers accurate noise calculation that includes theeffects of noise propagation. Results with various designs show thatCCS noise matches Spice simulation results with great correlation.
Nahmsuk Oh is R&D Engineer andAlireza Kasnavi is R&D Manager at SynopsysInc .