Determining S-parameters With Agilent’s ADS Via EM Modeling · Determining S-parameters With Agilent’s ADS Via EM Modeling With today’s high levels of both component and platform

Determining S-parameters With Agilent’s ADS Via EM Modeling

Withtoday’shighlevelsofbothcomponentandplatformintegration,measuringsignalsthatcan’tbephysicallyaccessedisbecomingthenorm ratherthantheexception.Previously, this series of papers has dealt with generating S‐parameters to determine device characteristics. The common thread has been that the device has been directly accessible to the probe. It has been shown how to use S‐parameters in both the time and frequency domains. However, the pressure on the electronics industry towards miniaturization and integration shows no signs of abating. In fact, new technologies such as 3D integrated circuit (ICs), non‐planar, multi‐level wafers, <15‐nm gate technologies and ultra‐low power designs will all but eliminate direct point measurements. Therefore, in many cases today and certainly true going forward, the only practical way to acquire transmission line parameters will be by simulation. As with the pervious papers in this series, Agilent offers a flexible and prolific platform of solutions that will make this transition as painless as possible. The current offering matrix is show in Figure 1. For this discussion, the software utilized will be Agilent’s InfiniiSim de‐embedding software, coupled with the Infinium real‐time oscilloscope, and EMPro.

The PreMaking aand, in mfrequencoppositeand cond

The Infinmeasurea fully intarray (FPand softwequal precomponethe techn

Figure 1. Ag

edicament a measuremmost cases, scy, what is co is true. Theditions that s

iium real tim signals fromtegrated priPGA), to a sinware can accecision. The ents betweeniques that w

gilent’s EEsofEDA

ent by attactraightforwaollected is acre is any numskew the col

me scope witm any point onted circuit ngle transmicomplish thimeasureme

en the devicework for phy

A line of offering

ching the proard. With thctually whatmber of extrllected data.

th InfiniiSimon the devicboard (PCB)ssion line ors from both ent consists oe and the proysical device

gs.

obe directly e exception t is there. Wiraneous effe.

de‐embeddce under test) or a highly r via in multivirtual andof two segmobe. When tes also work

to the compof some issith inaccessiects from an

ding softwart (DUT). Theintegrated fi‐layer layer physical me

ments – the atrying to detfor virtual d

ponent or cirues with loaible measureny number of

e allows thee DUT can befield‐programsubstrates .easurement actual devicetermine devdevices.

rcuit is reliabading or ement pointf componen

e engineer toe anything frmmable gate The hardwapoints with e and the ice paramet

ble

ts the nts

o rom e are

ters,

This paper will address the process of collect data from physically inaccessible “virtual” test points. This is defined as collecting data from a physical launch or other accessible test location, then factoring out the effects of the “stuff” whatever it may be, between the probe point and the measurement point of interest. The first part of this paper shows how to use 3D geometry, material properties, and electromagnetic simulation to create the de‐embedding information even for points of interest that are inaccessible. The second part will discuss exploring the design space without the expensive and time consuming “cut and try” method of building and measuring many different prototypes. The discussion will use measured data from already existing hardware that has been built. It will then use electronic data automation (EDA) simulation tools too quickly and cost‐effectively determine the characteristics of candidates on the drawing board. As can be seen from figure 1, there are a number of challenges and solution – too many to address, in‐depth, in one paper. Therefore this paper will focus on electromagnetic (EM) properties and high speed digital signal integrity using Advanced Design System (ADS) and Electromagnetic Professional (EMPro). Electromagnetism is an extremely important issue. It is involved in nearly every aspect of our lives. Ignoring I,t when trying to design high‐speed digital circuits, is a recipe for disaster. Analog engineers are quite comfortable with EM. Digital engineers, on the other hand, didn’t have a real need to understand it until speeds ramped up and integration became the norm.

EM 101 Before diving headlong into the effects and analysis of EM, a short refresher course is warranted. Electromagnetism can be characterized by Maxwell’s equations. Maxwell developed these quite a while ago, primarily, from work done by Oliver Heaviside, in conjunction with Willard Gibbs. These equations define the physical lines of force (see Figure 2). Fundamentally, EM is about forces and the motion of charged particles, i.e.; electrons. These electrons have an E‐ and M‐field, that interacts with other EM fields under various conditions. But once the electrons start to move, things get a bit more complex. Basically, when a potential energy is used to excite these fields, the first charge to receive the energy sends out what is referred to as a B‐field, which is eventually sensed by another charge further down the transmission path. This energy is transferred from atom to atom, which creates current flow as the E‐and B‐fields couple to each other. As this current flows though the circuit, the related E‐ and M‐fields become coupled by Faraday’s law (see Figure 3.)

Until nowthe E‐ fiefields. Thchangingappears tquantitat An electrdistance)changinggenerate

Figure 2. Mfrom his bThird Edit

w, things havelds of all othhis is where tg B‐field alsoto have realtively differe

rostatic field) around a clg B, Kirchoff’e E, but it tur

Maxwell’s equatbook “A Treatisetion.

ve been fairlhers chargesthings get tro generates Ephysical effent.

d is conservalosed path iss Voltage Larns out that

tions for wave pe of Electricity an

ly simple. Sts as a force. icky. Static cE. While somfect ‐ and the

tive, in the ss zero. This iaw is no longchanging E r

ropagation. Taknd Magnetism. V

ationary chaIn addition mcharge isn’t tmewhat conve effect is qu

sense that ns defined byger valid (seereturns the f

en Vol II,

Fig

arges generamoving charthe only forcvoluted, thisualitatively d

et energy chy Kirchoff’s Ve Figure 5). Nfavor.

ure 3.Faraday’s

ate E‐fields, rges generatce that genes mathematidifferent not

hange (that Voltage LawNot only doe

Law

and they sete and senseerates E‐fieldical concept t just

is force time. However, wes changing

nse e B‐ds: a

e with B

Consideris magnemiddle? experimemissing, called thwhere is might argpositivelyBut whatexists evedisplacem The conspermits ethe magnexample makes thof space.equation

Figure 4. M

r this case wetic field genDoes the maentally that na changing Ee displacemthe currentgue that they charged cat if the spaceen then. It sment and no

sequences apelectromagnneto‐static cof an electrhe equations. In earlier tins, few situat

Maxwell’s modifie

here a consterated by thagnetic fieldno, the magE‐field generent current.? If the space electron cloapacitor plate between theems as if tho current.

ppear quite netic waves, coupling conomagnetic ws hard to solvmes, using ations actuall

ed version of Am

tant current he current o disappear? netic field isrates B‐field It’s a strangce between toud of each te until the rhe plates is ahe E‐field its

profound. Ttraveling at stants, whicwave. And thve. The locaarchaic toolsy existed of

mpere’s law.

is ramping un either sideMaxwell gres continuous. Maxwell adge term becathe plates is atom in the restoring fora vacuum? Eelf is capabl

The bi‐directone over th

ch turns out his wave, spl dynamics cs such as penclosed‐from

up the voltae of the dieleeat insight ws. Where thedded a termause what isfilled with a dielectric drce of their nEmpirically te of generat

ional couplinhe square roto be the spreading throcauses wavencil and papm solutions.

ge on this caectric, but wwas to propoe conventionm to Ampere’s it that’s disa polarizableisplaces slignuclei balanche displacemting a B‐field

ng of the E‐ ot of the elepeed of lightoughout spaes to spread er and parti

apacitor. Thwhat about inose and confnal current is’s Law whichsplacing? Ane dielectric, ohtly toward ces out the fment currend with no

and B‐fieldsectrostatic ti. Light is an ce, is what throughoutal differentia

ere n the firm s h he d one the force. nt

s mes

all al

Today, hokeep tracspace. One popstaggeretechniqudifferencsomewha

The SolThere aredomain ((FEM) anhardware

Figure 5. Ye

owever, mock of some b

ular methodd grids of poe, the evoluce time domat for efficie

ution – Moe several sol(FDTD) analynd method oe and power

ee Cell Diagram

dern high‐spbounded par

d is based onoints in spaction of the fain solvers, ency (see Fig

odern Comlutions availaysis has beenof moments (rful software

peed compurt of space a

n a grid of soe where loofields over tialthough moure 5).

mputationaable for comn one of the (MoM) are ve.

uters and sopnd how the

o‐called Yee ps of E surrome can be todern codes

al Mathemmputational emore populvery effectiv

phisticated pfields influe

cells. These ound a B poitracked. Thiss like EEsof’s

matics electromagnlar but of latve with today

programs haence each ot

are comprisint and vice s is the basiss EMPro refi

netic. Finite te, finite eley’s fast com

ave the abilither within t

sed of two versa. With s of many finne the grid

difference tment methoputational

ty to his

this nite

time‐ods

Depending on the problem any or all of these approaches will work. However, each has advantages and disadvantages. For example, if the 3D structure is a multilayer assembly such as a PCB or chip package, the first choice would be MoM, because of its speed. However, and although MoM is being extended outside of the pure multilayer constraint – certain bond wire model for example – it doesn’t handle arbitrary 3D structures as well as FEM and FDTD. If the target is a connector, dielectric brick or ball grid array (BGA) break out, the choices are between FEM and FDTD. To determine which is the best method simply depends upon the characteristics of the target. For example, a high‐Q structure favors FEM, while for an electrically large FDTD is the better choice. Ultimately the method chosen is determined by the number and type of characteristics of the structure.

ADS and Other Options. Advanced Design System (ADS) Momentum is the industry’s leading MOM simulator. However, it tuned for schematics and multilayer layout structures so it doesn’t work as efficiently on arbitrary 3D geometries like connectors, ball grid array (BGA) breakouts, dielectric bricks, bond wires, shields, and so on. For these types of structures, EMPro was developed. It consists of a state‐of‐the‐art 3D drawing environment and both FEM and FDTD simulators. It is fully integrated with the ADS design flow. ADS layouts can be exported to EMPro and import parameterized 3D structures into ADS. The FEM engine can even be run directly from ADS, without opening the EMPro user interface. It is a very smooth design flow (see Figure 6).

EM AnaNow thatthe desig(CAD) fordesired. ports. If abidirectioare definautomatFigure 7)

Figure 6. EM

alysis t the groundgn flow. If a srmats, it is aIf the structuat all possiblonal in flow.ned in the seically divide .

MPro functionali

dwork has bestructure is a simple taskure doesn’t le use waveg Waveguidense of incidespace into a

ty blocks

een laid, it isalready avaik to import itexist, createguide ports se ports give tent, reflectea mesh or gr

s time to do lable in one t EMPro. It ce it. The firstsince EM wathe best accued, and transrid similar to

the actual E of the commcan be editedt task in a neave are distruracy and msmitted wavo the Yee cel

EM analysis. mon compud in the userew structureributed in namimic the waves. The toolls mentione

First, a lookter‐aided der interface (Ue is to set up ature and ay S‐paramet can d above see

k at esign UI) if the

ters

e

Next, andvisualize be gainedstructurestructure There is aobtainingpaper, it environm

3

B

H

A

Im

A

A

Figure 7. EM

d at this pointhe resultind from thesee to resize ore can be imp

a demo avaig the S‐parais available

ment:

D drawing e

oolean oper

Hierarchical e

Add material

mporting a c

Add and setu

Align simulat

M flow block diag

nt there is ag fields and e graphics, wr move. If thported into a

lable that wmeter file. Ifhere: www.

environment

rations: subt

editing: Und

properties a

complete str

p waveguide

ion boundar

gram

n option to scurrents supwhich will hee result is saany tool (like

will step throf the reader agilentxxxx

t: Basic extru

tract “tool” f

oing any ste

and change

ucture

e ports

ry with meas

set up sensoperimposed elp explore tatisfactory, ae InfiniiSim) t

ugh the creawould like t. Basically th

uded shapes

from “blank”

ep, not just t

appearance

surement pl

ors if desired on your struthe design spa multiport Sthat accepts

ation of a neto view this he steps are

s (cube, sphe

”

he last step

(This can be

ane

d, run the simucture. A lotpace by suggS‐parameters CITI or Tou

ew CAD projebefore proceas follows w

ere)

by using the

e omitted if

mulation andt of insight cgesting whatr model of thchstone form

ect and eeding with within the EM

e hierarchy

time is shor

d can t he mat.

this MPro

rt)

Set up simulation

Visualize results: Bode plot, Smith chart, fields

Export s‐parameters

Advanced visualization 3D field animation etc.

The Exercise The quickest way to start a project is to import a CAD file. However, before doing that it would be prudent to do some basic manipulation from scratch, of objects so the reader has some background to work with. The first object is a cube. In this case, the cube represents and extrusion. The cube can be modified (its parameters) by, say changing the offset. Next, two more cubes are added at the origin. These cubes reference the Yee cell discussion earlier. Assume that a loop of B‐field around this cube edge and an E‐field being generated at the face center. There is an interpenetrating loop of the E‐field at the center and a B‐fields on the edge of this cube. This illustrates the object in terms of geometry. The objects can be increased in complexity by adding and subtracting the cube from the blank. The cube becomes the current tool. This operation will demonstration the subtract results. The object can be made as complicated as your design demands to zoom to the extents. The next step will be to subtract a sphere. The sphere tool is chosen by selecting: Boolean→ existing parts→ select my tool, and then simply subtract. Note, if it is discovered that a mistake has been made several steps earlier, there is a unique undo option available that offers the ability to undo incongruent steps, say undo step two, even though the present step is six. This is done by going into the desired Boolean operation and stepping through this hierarchy to open up the particular extrusion. For example, if too much of the extrusion was removed, it can be reworked easily. EMPro remembers all of the steps that were done to create that object and can alter any of the steps used to create the part. So far, this object is simply a lump of nondescript material. To give it some properties, it can be further defined, either from the custom menu, or from the default materials library. In this case the defining element will be copper. To give an object “copper” properties simply drag “Copper” onto the part. As well, the copper can be redefined to add a custom property such as a different shad of the stock color. This is an example of how to use the drawing environment.

Importing and Modifying a CAD File Objects This exercise will import a CAD file of higher complexity. This file consists of a PCB, a microstrip , a chip on a package, vias underneath the package, the package itself, bond wires, etc. There are a number of variables that can affect the attributes of components within the structure. For example, invisibility can be applied to any object to make viewing hidden objects easier. For the purpose of this exercise, assume there is as a real physical object that was created from the drawing within the CAD file. Physically, a probe can be placed at the end of port number one, but port number two is physically inaccessible. However, the objective is to view the eye

diagram attached to the real physical object as if the probe were placed physically, at port number two. To do so requires factoring out (de‐embedding) some extraneous elements that are picked up at port number one, that are between the two ports. The process involves placing to physical probe at port number one, then calculating the S‐parameters from port number one to port number two, using EM simulation. Once calculated, InfiniiSim is used to subtract off this structure. The result is what actually appears at port number two, even though port number one was the probe location. The process is to first define the ports and, add a new waveguide port. Step on is to select the desired face. For this discussion, accept all defaults except on the “Properties” tab, to name the port. Name it wp1. For the Waveguide Port Definition select 1‐W Modal Power Feed. The 1‐W Model Power Feed is a pre‐defined component. If desired, to define the impedance, use the Power/Voltage method. Since voltage is path‐dependent, specify an impedance line that will be used to measure the voltage along the two end points. With a microstrip the parameter is the maximum voltage, center‐to‐center. This defines the impedance line for the impedance calculation. Do the same for port 2, which is identical. There is a caveat with these ports. The software gives a warning that the waveguide is not on the simulation bounding box. The reason for this is that the ports have to be on the outer surfaces of the simulation because fields cannot be simulated beyond the port. That is corrected by using FEM Padding – simply eliminate the free space. Once implemented, the caution symbols disappear and the boundary of the simulation is correct on the port planes.

Ready, Set, Simulate Once the designer is satisfied with the port definitions, the next step is to simulate. For this exercise, FEM is the method of choice, which differed slightly from finite difference time Yee cells. This method uses tetrahedral meshing, which is an automatic adaptive mesh. Instead of using cubes, it uses tetrahedrons. To being, use the iterative solver. The default is the direct solver but in this case a high‐end processor isn’t necessary. The iterative solver uses a less memory than the direct solver. And this particular structure does not have any convergence problems so there is no disadvantage to using the iterative one. If the structure is very complicated, then the more powerful direct solver is the tool of choice, but will also require a more powerful computer. And to keep this computation manageable, the delta error has been altered, which is basically the change threshold in the S‐parameters that kicks off a further iteration. This is like a 2% magnitude change in the S‐parameters and will be the criterion for the iterations to end. Once the S‐parameters change gets less than 2%, it will be halted because the convergence is now within that error on the Smith chart. Now, create and queue the simulation. Once the results have been generated, the project is named. And pick the results; in this case, frequency and S‐parameter. These now can be plotted on a linear graph and the forward and result loss of the s‐parameters seen. As well a Smith chart can be created if desired. If the

results arimported

MovingThe S‐pato “see” designer and checaccidenta Other fungridding,lowest frmuch onhoweverattenuatavailableleft for th

Figure 8. Ge

re acceptabld into any to

g Beyond trameters givinside the sthas the abil

cked for connally have be

nctions inclu or any giverequency, po sync acrossr, it shows thion behavioe to the desighe reader fo

etting the real m

le, they can ool, including

he Black Bve you a gootructure withity to orbit anectivity. Then added to

ude such thinn plane, andort one, mods the entire shat the field r of its field gner at this pr further inv

measurement

exported tog InfiniiSim.

Box od black boxh “advancedaround the shis tool is vero the structu

ngs as the abd animate thde one and estructure. If tpattern breapatterns. Thpoint. Howevestigation a

a CITI file fo

model of thd visualizatiostructure in ry useful for re.

bility to highe fields. Forexamine thethe designeaks up by thhere are numever, to keepat the approp

ormat. From

he structure on.” With ad3D. Any partlocating mi

hlight the bor example, the E‐field. Thisr examines te reflectionsmerous othep this discusspriate time.

m that format

but there isdvanced visuticular planenute gaps th

ond wires anhe designer s port’s fieldthe highest fs corresponder options ansion manage

t, they can b

s also the abalization, the can be selehat might

nd look at thecan select thd appears prefrequency, ding to the nd functions eable, they a

be

ility e ected

e he etty

are

Once theof the DUessence osubtract is the datprobe an

Playing Another developeparametevarying aintegrate

A new itestructurethat deliv

Figure 9. Ex

e S‐parameteUT, without of this discuor factor outa at the desnd measurem

What‐If approach is ed. To do thier file. Onceany parameted with the o

eration of thes can take uvers a new e

trapolating the “

er file has behaving to plassion (see Fiut (de‐embedsired measument point

to extrapolas, again calie the file haser i.e.: the goriginal eye

he structure up to 30 mineye diagram

“what‐if”

een generateace the probigure 8). Thisd) the effectrement poin

ate “what‐if”brate the sys been genegeometry or diagrams in

can be run anutes, or longfor the scen

ed, it can bebe at the exas process wit of this channt minus the

” scenarios ostem and rurated from tthe materiaADS. (See Fi

as often as dger to recomnario. This vi

e used to detact physical ill be detailennel elemene effects of t

once the firsun the simulathe prototypal propertiesigure 9).

desired (howmpile) to “burtualization

termine the point of theed using Infint. The net efhe elements

st prototypeation to genpe, the EM ss. The new E

wever, note tuild” a new voffers the a

characteriste DUT – the niiSim to ffect is that s between th

e has been erate the S‐imulator alloM results ca

that complevirtual structbility to “see

tics

this he

ows an be

x ture e”

where fiethicknessmatter ocorrectlyvariationprototyp

An ActuThis is camodificaresult is timpedanplane of conventi Currentlyto createetch littlesimulatio

elds are, or wses, for examf performingy. This “virtuns and their re.

ual Case Stse will addretion to add ato alter the ice altering mthe microstronal design.

y, there are e a bottoms‐e spirals out on bounding

where to plample, withoug a single caualization” aresults almo

tudy ess ground padditional efimpedance omodificationrip structure.

no commerc‐up EM simuof the grou

g box.

ace stitchingut the need tlibration of ability saves tst immediat

plane designffective induof a microstrn. Such modie, can often s

cially availablation (see Fnd plane, be

vias, via strto build adda real structtime and motely, rather t

n, specificallyuctance or carip, create bifications, wsignificantly

ble circuit deFigure 10). Oelow the 50Ω

uctures, anditional physture to insuroney by rendthan waiting

y a defectedapacitance tband‐rejectiohich appeary reduce the

esign tools toOne the left iΩ transmissi

d where to cical prototype the simuladering any ng a weeks to

d ground struto a microstron or add so as defects ostructures f

o do this. This the DG S. on line. On t

change the wpes. It is simation is set uumber of analyze ano

ucture (DGS)rip. The desime other on the grounfootprint ove

he designer hThe processthe right is t

wave ply a up

other

) ired

nd er

has s is to the

Figure 11FTDT, anpredictiocorrelatio

Figure 12return an

Figure 10. T

Figure 11.

1 shows the d the red is on of the banon to measu

2 presents thnd so on. In t

The structure fo

Measurement v

measurememeasured dnd stop resoured data.

he field plotsthe pass reg

or creating the E

s. simulated plo

ent versus thata. The goanances at ~2

s. The field pgion the field

M simulation

ts.

he simulational is to make2 and ~5 GH

plots give insd is going thr

n. The blue ie band stop fz. This show

sight by illusrough; in the

is the FEM, tfilters and thws a very hig

strating the te stop region

the green is here is a corh degree of

trapping, gron the field is

the rect

ound s

trapped the ~2 G

Figure 13patented

Figure 12. F

because of tHz anti‐reso

3 presents and convolutio

Field plots of the

these spiral cnance.

n ADS graphn algorithm.

DGS at 1.985 G

cut outs. The

hic that inclu. The graphic

Hz

e graphic at

de S‐paramec confirms th

the bottom

eters in the he correlatio

left is the e

time domaion between

lectric field

n via our measured d

at

data

and simusimulatoThe meapart of thThe bottois from thcorrelatio

Figure 14included

Figure 13. F

ulated data, r. This simulsurement whe graphic shom left showhe statisticaon with mea

4 presents adin the simul

ield plots of the

with time doation and m

was performehows simulaws the overlal mode. Bothasured data.

dditional melation. It con

DGS at 1.985 G

omain eye dmeasured resed using Agilation result uaid plot of sih bit‐by‐bit a

easured datanfirms good

Hz

diagrams in Asults are of tlent DCA‐J fousing channeimulated anand statistic

a. In this casagreement

ADS, using ththe Tyco chaor a 210 bit pel simulator d measuredal simulatio

se, time withbetween the

he Agilent channel for a 1pattern lengtfor a 1010 b

d data. The bn shows exc

h a feed‐forwe measured

hannel 6” trace length. The top rit pattern lebottom rightcellent

ward equalizdata and th

gth. right ngth. t plot

zer e

simulatiotrace lenmeasureshows thstatisticadata. Agadata.

ConclusThis papevia simulphysical This metstructureinaccessiInfiniiSimgeometry

Figure 14. S

on. This grapgth at 5 Gbpment was phe measuremal simulator. ain, both bit

sion er has preseation. The dmeasureme

hod presente and use theble location

m. It can alsoy.”

upplemental da

phic shows thps. In this caerformed usment results The bottomby bit and s

nted the bediscussion invnt point is a

ted will allowe acquired d. The data is

o be shown h

ta with a feed‐fo

he simulatiose four‐tap,sing Agilent using DCA‐J

m right plot sstatistical sim

st method fovolved how t some dista

w the designdata to deters then fed tohow to explo

orward equalize

on and measFFEs were uDCA‐J for a 2J. The bottomhows the ovmulation sho

or determinto de‐embeant location

ner to probe rmine the aco and analyzeore the desig

er in the simulati

ured plots oused to equa210 bit pattem left plot shverlaid plot oows excellen

ing accurateed, or factor than the de

in a convenctual parameed via de‐emgn space by v

ion

of the Tyco calize the receern length. Thows resultsof simulatednt correlation

e measuremout errant dsired measu

ient locationeters seen ambedding sovarying a sim

channel for aeived signal.he top left ps from ADS d and measun with meas

ent paramedata when thurement poin

n on the t the desiredoftware like mulated “vir

a 16” . The plot

red sured

ters he nt.

d,

rtual

Such methodologies and their related tools are extremely useful to the designer because they save time, money and bring products to market via economies of scale. The final paper in this series will present to the designer the concluding process of taking S‐parameter files, whether simulated or measured, and view them on an oscilloscope.