Modeling of Integrated RF Passive Devices Sharad Kapur and David E. - - PowerPoint PPT Presentation

modeling of integrated rf passive devices
SMART_READER_LITE
LIVE PREVIEW

Modeling of Integrated RF Passive Devices Sharad Kapur and David E. - - PowerPoint PPT Presentation

Modeling of Integrated RF Passive Devices Sharad Kapur and David E. Long www.integrandsoftware.com Custom Integrated Circuits Conference, 2010 Outline Introduction and motivation Three topics EM simulation using Integral methods


slide-1
SLIDE 1

Modeling of Integrated RF Passive Devices

Sharad Kapur and David E. Long

Custom Integrated Circuits Conference, 2010

www.integrandsoftware.com

slide-2
SLIDE 2

 Introduction and motivation  Three topics

  • EM simulation using Integral methods
  • Modeling layout dependent effects
  • Circuit models and component synthesis

 Experiments  Conclusions

Outline

2

slide-3
SLIDE 3

Applications

 Applications

  • Mobile
  • Wireless (WiFi, WiMax)
  • Wired (Ethernet)
  • Storage (Hard disks)

 Passive components found

these devices

 A decade ago inductors,

MIM capactors

 Now a whole range of

components and devices are being used

3

slide-4
SLIDE 4

Tech trends

 Thick metals

  • 3um to 8um copper

 High-resistivity

substrates

  • 10 -cm to 1000-cm

 Fine feature sizes

  • 0.1m width at 65nm

 Many metal layers

  • High density MOM

capacitors instead of MIM (2fF vs 4fF)

4

slide-5
SLIDE 5

 IC processes offer tight tolerances  Low variability  Chip real estate is expensive  Integration offers big cost savings

  • No extra packaging
  • Much tighter tolerances and better yield

 When a device can be built with reasonable

quality compared to an off-chip or an LTCC structure it will be integrated

 Devices that used be considered exotic are

now routinely used

IC processes

5

slide-6
SLIDE 6

Passive components

6

inductor ind + shield MOM cap Transformer

slide-7
SLIDE 7

Full circuits

7

Full circuit1 Diplexer2 VCO3: inductors + interconnect+ capacitor bank

  • 1. SiGe Semiconductor on IBM BiCMOS
  • 2. STATS ChipPAC IPD technology
  • 3. Wipro on TSMC 90nm
slide-8
SLIDE 8

 Increased prevalence of passives has made fast

and accurate modeling critical

 Two aspects to modeling

  • 1. Electromagnetic (EM) simulation to evaluate

candidate designs (and possible refinement)

  • 2. Converting EM simulation results that can be

used in higher level simulators (like Spice).

Modeling

8

slide-9
SLIDE 9

 Two broad categories for solving Maxwell‟s

equations

  • 1. Differential formulations
  • Finite-difference, Finite element (FEM), FDTD
  • 2. Integral Formulations
  • Method-of-Moments, Boundary element

(BEM), Integral equation solvers

EM simulation

9

slide-10
SLIDE 10

Differential vs Integral

Diff formulations

 Flexible  Imposes no constraint

  • n shape of metals,

dielectric regions

 Need to enforce

Maxwell‟s equations everywhere surrounding the object

 Leads to large sparse

matrix solve Integral formulations

 Planar dielectrics,

conductors

 Need to enforce

Maxwell‟s equations

  • nly on conductors

(Green‟s theorem)

 Leads to smaller dense

matrix to solve

 Many techniques

developed recently

 For IC passives this

approach is the best

10

slide-11
SLIDE 11

3D Integral Formulation in EMX

 EMX is a 3D EM

simulator

 3D volume integral

formulation (time harmonic)

 Unknowns (charge

and currents)

  • Surface charges
  • Volume currents
  • 3D: Current

vectors can be in x-y-z directions

11

slide-12
SLIDE 12

Matrix formulation

 Suppose that N elements in the mesh  Conventional approach O(N3) time and O(N2)

memory

  • Cost is prohibitive
  • Double the size of the problem 8X time

12

(continuous form)

B Ax 

(A is a dense matrix)

slide-13
SLIDE 13

 Iterative methods were developed in the

numerical analysis community (GMRES, Yale, 1986)

 Matrix vector products instead of matrix

inversion

 This reduced the time to O(N2)

Innovations in numerical methods (GMRES)

13

b Ax 

} ,..., , , {

2

b A b A Ab b K

n n 

slide-14
SLIDE 14

 The Fast Multipole Method was developed in

1987

 Developed N-body problem  Applied to capacitance byWhite at MIT for

  • FastCap, Fast Henry, (1990s).

 Applied to 2.5D field solution

  • IES3, Bell Labs, Kapur and Long, (1990s)

 These sorts of problems can be solved in linear

time (with a large constant)

 All recent effort is decreasing this cost

Innovations in numerical methods (FMM)

14

b Ax 

slide-15
SLIDE 15

Exploiting regularity

15

 Mesh generation was regarded

as an orthogonal sub problem (typically unstructured Delauny triangulation)

 Layout has a lot of structure  This structure can be imposed on

the mesh

 Identical interactions are

repeated all over

slide-16
SLIDE 16

Adaptive frequency sweep

 An adaptive frequency

sweep

 Reduced order model

using Krylov subspace methods

  • Methods developed in

the „00s (Bell Labs, MIT, CMU, Intel)

 The reduced order

model from a small set

  • f EM solutions

 Only few simulations

need to be done

16

slide-17
SLIDE 17

Time and Memory scaling

 Single frequency

simulation (including iterative solve)

 Compare speed and

memory for 1, 2, 4, 8, …, 64 inductors

17

1 inductor 64 inductors

slide-18
SLIDE 18

Examples: Spiral Inductor

18 3D mesh Current Courtesy: TSMC. 65nm RFCMOS, 9LM thick metal technology. Published at RFIC 2009

“Including Pattern-Dependent Effects in Electromagnetic Simulations of On-Chip Passive Components”, Integrand and TSMC

Standard high Q spiral inductor used in lots of circuits. Thick copper and large size.

slide-19
SLIDE 19

Stacked Inductor

19 3D mesh Current Courtesy: TSMC. 65nm RFCMOS, 9LM thick metal technology. Published at RFIC 2009

“Including Pattern-Dependent Effects in Electromagnetic Simulations of On-Chip Passive Components”, Integrand and TSMC

Small stacked inductor. High inductance low Q (used in Chokes). Upto 20-30nH in small area

slide-20
SLIDE 20

MOM (finger)Capacitor

20 3D mesh of 0.6pF Cap Courtesy: TSMC. 65nm RFCMOS, 9LM thick metal technology. Published at RFIC 2009

“Including Pattern-Dependent Effects in Electromagnetic Simulations of On-Chip Passive Components”, Integrand and TSMC

High-density MOM caps (at 40nm can be 4fF/square micron). Important to parasitic inductance to get SRF.

slide-21
SLIDE 21

Transformer

21 3D mesh Current Courtesy: UMC. 90nm RFCMOS, 8LM thick metal technology. Published at CICC 2007

“Synthesis of Optimal On-Chip Baluns”, Integrand and UMC

Used as a part of matching network. Can get reanonably high coupling “k” values of 0.8-0.9 in a standard thick metal CMOS process

slide-22
SLIDE 22

Balun (with MiM caps)

22 3D mesh Current Courtesy: UMC. 90nm RFCMOS, 8LM thick metal technology. Published at CICC 2007

“Synthesis of Optimal On-Chip Baluns”, Integrand and UMC

A balun is a passive component that transforms power from a BALanced to an UNbalanced port. FOM is usually insertion loss. 1dB insertion loss and is very competitive with off chip baluns.

slide-23
SLIDE 23

BiCMOS Diplexer (with Thru Silicon Vias)

23 3D mesh Current Courtesy: SiGe Semiconductor. IBM BiCMOS 5PAE.

Diplexer has a high-band and low band filter in the same circuit. Used in chips which operate in multiple bands. Fully passive circuit with inductors, resistors, capacitors, TSVs. Need to model all effects and coupling.

slide-24
SLIDE 24

IPD Diplexer

24 3D mesh Current Courtesy: STATSChipPAC, IPD technology (8um Cu on high resistivity Si substrate)

Built on a “lossless” substrate so coupling is very strong between components. Methodology of design is to create inductors with EM simulation. Tune with caps. Resimulate and re design. Simulation time of about 1 hour for full circuit.

slide-25
SLIDE 25

CMOS VCO

25 3D mesh (inductor+ capacitor bank) Courtesy: Wipro, TSMC90nm, 1P5M

High-frequency VCO. Inductor and 66 MiM capacitor bank. Inductor is small so coupling between inductor and interconnect must be considered. Block by block model fails to predict VCO behavior.

slide-26
SLIDE 26

 Multi-threaded EMX is 2-3X faster for small

examples and 5-7X faster for larger examples

  • n an 8 CPU machine.

 The memory for the multi-threaded version

goes up at a slower rate than the speedup.

Benchmark Summary

26

slide-27
SLIDE 27

 In IC processes the width, thickness and

resistance of wires vary depending on the width and spacing of the surrounding wires

 These effects are called pattern dependent

effects

 Width and spacing dependence in the process

description

 EMX modifies the layout to mimic the

fabrication process

 Leads to improved simulation and modeling

accuracy

Layout dependent effects

27

slide-28
SLIDE 28

 Fabricated metal width varies as a function of

width and spacing of wires

 Physical width vs drawn width as a function of

width and spacing

 Width can vary by 50% from drawn width

Pattern dependent effects

28

slide-29
SLIDE 29

 Sheet resistance varies as a function of width

and spacing of wires

 Sheet resistance as a function of width and

spacing

 Sheet resistance can vary by 200%

Pattern dependent effects

29

slide-30
SLIDE 30

 EMX automatically

modifies IC layout to mimic fabrication effects

 When you have uniformly

spaced wires “width” and “spacing” have intuitive meanings

 Need to come up with a

definition of width and spacing for general layout.

 What happens when you

have non-uniform layout?

Mimic fabrication effects

30

slide-31
SLIDE 31

 Fabricated width is

different from drawn width according to rules provided by foundry by a “bias” amount

 Shaded regions

represent original drawn geometry

 Lines represents

modified “grown” geometry based on local width and spacing

Modifying layout

31

slide-32
SLIDE 32

EMX simulation of MOM capacitors

The iRCX width-and-spacing dependence is more critical for structures that are not at minimum dimensions. The accuracy

  • f EMX using iRCX is increased since the fabrication process is

mimicked more closely.

slide-33
SLIDE 33

EMX simulation of Stacked inductors

slide-34
SLIDE 34

 Once you have S-parameters from simulation

  • Can use Harmonic Balance and do higher-

level simulations in the frequency domain

 For transient analysis need time domain

representation

 Best to have a true Spice Model (RLCK) with

positive values, etc.

 The way it can be done is to fit to a prescribed

topology using certain constraints.

 Often “scalable” models are required which are

parameterized models based on layout parameterized by geometry.

Circuit Modeling

34

slide-35
SLIDE 35

I.

Create an automated layout generator

  • II. Run EM simulations over the design space
  • III. Create a scalable model
  • IV. Use an optimizer to determine the optimal

layout and tuning capacitor values based on designer inputs

Optimal balun syntheis

35

slide-36
SLIDE 36

I.

Create an automated layout generator

  • II. Run EM simulations over the design space
  • III. Create a scalable model
  • IV. Use an optimizer to determine the optimal

layout and tuning capacitor values based on designer inputs

Optimal balun syntheis

36

slide-37
SLIDE 37

Layout generation

 Parameterized layout

generator for transformers

 The design space

  • turns ratio
  • number of turns
  • width
  • outer diameter

 Create about 1000

transformer layouts

37

slide-38
SLIDE 38

EM solution

38

Simulate the layout from DC to 20GHz.

3D Mesh of Balun current flow

slide-39
SLIDE 39

 The topology for the

scalable model was derived from intuition

  • Two center-tapped coils
  • The specific form is

based on physical intuition (e.g., main series resistance is proportional to diameter and inversely proportional to width)

Building a scalable transformer model

39

slide-40
SLIDE 40

Scalable Model synthesis

 1000s of S-parameter

files

 Continuum uses  A specialized non-linear,

least squares optimizer

 Guaranteed passive with

circuit constraints

 Error histogram shows <

2% error model vs simulation

40

Model playback vs simulation

slide-41
SLIDE 41

Balun silicon verification (Insertion Loss)

41

CICC 2007

slide-42
SLIDE 42

 Modern EM simulators like EMX are fast and

accurate enough for modeling integrated RF components and circuits

  • EM simulation of passives
  • Circuit Modeling of passives
  • Discussed Layout dependent effects

 Industry trend

  • more wireless devices
  • more integration
  • Technology of high-resistivity, thick metals,

TSV and 3D structures

 All these will make this area of research critical

Conclusions

42

slide-43
SLIDE 43

 Taiwan Semiconductor Manufacturing Corp (TSMC).

  • Joint work on layout dependent effects

 United Microelectronics Corp. (UMC)

  • Joint work on balun modeling and synthesis

 SiGe Semiconductor

  • Example of IC diplexer

 STATSChipPAC

  • Example of IPD Diplexer

 Wipro (Newlogic)

  • Integrated VCO design

Acknowledgements

43

slide-44
SLIDE 44

References

44

slide-45
SLIDE 45

45

Extra Slides

slide-46
SLIDE 46

EMX innovations layout regularity

Layout is regular

  • 1. Wires are paths of constant

width.

  • 2. Distance between adjacent

routing is constant Routing is at 45 or 90 degrees

  • 3. Components, spiral

inductors, capacitors, are symmetric

Layout “space” is actually a very small subset of all possible routing

46

slide-47
SLIDE 47

 EMX exploits parallelism in various ways

  • Multipole setup (direct interaction

computation)

  • Independent solves at different frequencies

(has memory cost)

 EMX determines parallel scheduling on the fly,

preferring “higher level” splitting when possible

Multithreading

47

slide-48
SLIDE 48

48

Designing full RF blocks

 Step 1:

  • Rough design of inductors and baluns
  • Run EM simulation

(seconds to minutes; sometimes in a scripting loop along with a layout generator)

 Step 2:

  • Simulate full structure with

interconnect and large number of internal ports (e.g., 20) for capacitors. Tune design for caps.

 Step 3:

  • Re-simulate final structure with caps

included (few ports and high discretization). Accounts for coupling and effects of interconnect.

48

802.11b/WIMAX balanced diplexer

slide-49
SLIDE 49

 A natural data structure for

representing local distances is a Voronoi diagram

 Example MOM capacitor

layout

 Partition plane into non-

  • verlapping regions
  • Shaded regions are wires
  • Between wires are

segments which represent Voronoi boundaries

Voronoi Diagrams

49

slide-50
SLIDE 50

 The 4 characterized Baluns have excellent

characteristics!

  • Insertion loss of less than 1.5dB
  • Return loss of about 16dB
  • Phase imbalance of less than 0.25 degrees
  • Amplitude imbalance of less than 0.25dB

Balun summary

50