22 RFIC AND MMIC TECHNOLOGYto give lines a thickness in a 2 ½ Dsimulator if <strong>on</strong>e stacks two lines <strong>on</strong> topof <strong>on</strong>e another and meshes them togetherjudiciously with vias. It is important toexperiment with this approach <strong>on</strong> simplestraight line geometries, to make sure <strong>on</strong>ehas the right separati<strong>on</strong> and via count.• Passivati<strong>on</strong> layers: — The circuit isusually passivated by placing a thin layerof passivating material <strong>on</strong> top of thechip. The <str<strong>on</strong>g>inductor</str<strong>on</strong>g>’s self res<strong>on</strong>ance canbe affected by this layer; it is thereforeimportant to include its effects in themodel.• C<strong>on</strong>ductor losses: — These areimportant for correct calculati<strong>on</strong> of theQ. Simulati<strong>on</strong> software will corporatec<strong>on</strong>ductor loss in <strong>on</strong>e of two ways.Either it will use impedance boundaryc<strong>on</strong>diti<strong>on</strong>s <strong>on</strong> the surfaces of thec<strong>on</strong>ductors, or it will mesh the interiorof the c<strong>on</strong>ductors in a 3D fashi<strong>on</strong>.Both ways have reported to have beensuccessful if carefully used.• Frequency effects: — The softwareshould include frequency effects for RFICapplicati<strong>on</strong>s. There is software availablethat gives DC inductance and resistancevalues, but this is of limited use becauseof the frequency dependent nature of theinductance and the capacitance.• Current return: — It is important tounderstand how the software is settingup sources that excite the <str<strong>on</strong>g>inductor</str<strong>on</strong>g> andwhere the return current is. For example,many software packages excite thestructure with wave ports, sometimescalled edge ports. These ports exist at theedge of the simulati<strong>on</strong> boundary, and areexcited with an electromagnetic wave.Let’s assume <strong>on</strong>e is using an impedanceboundary c<strong>on</strong>diti<strong>on</strong> ground which is setto that of the substrate. How does <strong>on</strong>eknow that the return current is comingback <strong>on</strong> this plane? For example, itmay really be going <strong>on</strong> the sidewalls ofthe box – if there is <strong>on</strong>e. Also, even ifthe current is in the substrate, it will inreality not be as simulated because itwill actually flow between two substratec<strong>on</strong>tacts. Therefore, care must be takenin setting up the excitati<strong>on</strong>.• Extracti<strong>on</strong> of model parameters: — Sparameters are the normal output of thesoftware. The modeler is then faced withfitting the model parameters to the data.(Of course, he or she also has the opti<strong>on</strong>of inserting the S parameter block intothe circuit simulati<strong>on</strong> software directly.)This can be an error pr<strong>on</strong>e process. Thetwo most comm<strong>on</strong> problems are a lackof model sensitivity to a given parameter,and a large number of parameters tofit. If the model is not sensitive to aparameter, it is possible to have wideswings in the parameter’s value. As asimple, c<strong>on</strong>trived example, imagine <strong>on</strong>esimulates an <str<strong>on</strong>g>inductor</str<strong>on</strong>g> with <strong>on</strong>ly <strong>on</strong>e port– the other port being open. It will bevirtually impossible to see the inductanceas no current will be going through it(although it might be a good way to lookat capacitance). As another more realisticexample, when <strong>on</strong>e excites the <str<strong>on</strong>g>inductor</str<strong>on</strong>g>with a wave, there is an impedance to thetwo ports. Make sure that the parametersyou are trying to extract give impedancesthat are the same order of magnitude ofthe port impedances. Otherwise, you willnot be able to get an accurate answer.For example, if you use ports that haveimpedances of 100 Ohms, and thecapacitance has an effective impedanceof 10,000 Ohms, it will look like anopen, and be very difficult to get anaccurate answer.• C<strong>on</strong>vergence of simulati<strong>on</strong>: — Howdoes <strong>on</strong>e know the simulati<strong>on</strong> hasc<strong>on</strong>verged? The simplest answer is– experience and intuiti<strong>on</strong>. For simulatorswith iterative meshers, <strong>on</strong>e must alwaysask what the c<strong>on</strong>vergence criteria are.It is very comm<strong>on</strong> to have the simulatorc<strong>on</strong>verge <strong>on</strong> the S parameter changes.The problem is that the S parametercould be totally dominated by theinductance. The mesh may not yet beadequate for an accurate resistance value.Therefore, care must be taken whenspecifying an error criteri<strong>on</strong>.Measurement issuesMany companies build and measure<str<strong>on</strong>g>inductor</str<strong>on</strong>g>s to test models, checksimulati<strong>on</strong>s, and develop model libraries.Comm<strong>on</strong> problems to be avoided are:• Calibrati<strong>on</strong>: — Make sure that thesystem has been properly calibrated.The probes, and test structures haveinherent inductance, capacitance, andresistance. It is important that theseeffects be understood, and removed if thefinal model parameter values are to beaccurate.• Current Return: — Understand wherethe return current is. The inductancevalue measured is the loop inductance,which includes the return current.• Parameter Sensitivity: — Make surethat the measurements are sensitive tothe parameters you are interested in. Forexample, if the inductance is completelydominating the resistance, it is unlikely agood resistive value can be determined.Software availableSeveral excellent EM software toolsare available through the AWR EMSocket interface, and AWR c<strong>on</strong>tinuesto work with other emerging EM andspiral <str<strong>on</strong>g>inductor</str<strong>on</strong>g> <str<strong>on</strong>g>modeling</str<strong>on</strong>g> tool vendorsto integrate their soluti<strong>on</strong>s into theAWR design platform through its EMSocket technology Links to the vendors’web sites can be found at: http://www.appwave.com/sales/alliance/partners/EM.• AWR – EMSight – Method ofmoments simulator.• S<strong>on</strong>net – Method of momentssoftware.• IE3D – Method of momentssoftware. There is a discussi<strong>on</strong> of<str<strong>on</strong>g>inductor</str<strong>on</strong>g> <str<strong>on</strong>g>modeling</str<strong>on</strong>g> <strong>on</strong> their TechnicalNotes page.• EM3DS – MEM Research – Methodof moments software. Uses volumecurrents or surface currents.• OEA <str<strong>on</strong>g>Spiral</str<strong>on</strong>g> –– the product isspecifically tailored for spiral<str<strong>on</strong>g>inductor</str<strong>on</strong>g> analysis. It also hassynthesis capability, wherein thedesigner inputs the inductance value,and the layout is created.• Optimal – SI applicati<strong>on</strong>s, includingO-Wave, a 3D full-wave extracti<strong>on</strong>tool.Author biographyDr. John Dunn is a senior engineeringc<strong>on</strong>sultant at Applied Wave Research,Inc., specializing in signal integrityissues. His areas of expertise includeelectromagnetic <str<strong>on</strong>g>modeling</str<strong>on</strong>g> and simulati<strong>on</strong>for high-speed circuit applicati<strong>on</strong>s.Before joining AWR, he was a professorof electrical engineering at the Universityof Colorado for 15 years.Dr. Dunn is the author of twentypapers in technical journals, as well asnumerous c<strong>on</strong>ference publicati<strong>on</strong>s, andseveral invited talks. He received hisPh.D. in applied physics from HarvardUniversity in 1984 and is a seniormember of IEEE.Company Informati<strong>on</strong>Applied Wave Researchwww.appwave.comMicrowave Engineering Europe ● May 2006 ● www.mwee.com