Outline Introduction TSV Array Cross Talk Analysis Return Path - - PDF document

outline
SMART_READER_LITE
LIVE PREVIEW

Outline Introduction TSV Array Cross Talk Analysis Return Path - - PDF document

4/25/2014 3D IC-Package-Board Co-analysis using 3D EM Simulation for Mobile Applications Darryl Kostka, CST of America Taigon Song and Sung Kyu Lim, Georgia Institute of Technology CST COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013


slide-1
SLIDE 1

4/25/2014 1

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

3D IC-Package-Board Co-analysis using 3D EM Simulation for Mobile Applications

Darryl Kostka, CST of America Taigon Song and Sung Kyu Lim, Georgia Institute of Technology

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

  • Introduction
  • TSV Array Cross Talk Analysis
  • Return Path Discontinuity Modeling
  • 2.5D Link Analysis
  • 3D IC Link Analysis
  • Conclusion

Outline

slide-2
SLIDE 2

4/25/2014 2

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

Introduction to 3D Integration

Source: Cadence Design Systems, Inc. “3D ICs with TSVs—Design Challenges and Requirements”

3D IC System-on- Chip (SoC) System-in- Package (SiP) 2.5D Stacking

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

D=100μm, R=15μm, L=100μm, dox=0.1μm, εSiO2=3.9 (tand=0.001), εSi=11.9 (cond=10S/m)

TSV Signal-Ground Pair

Sharp slope due to transition from slow wave to quasi-TEM mode Losses due to displacement currents in Si

Insertion Loss (dB)

slide-3
SLIDE 3

4/25/2014 3

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

Cross Talk: TSV Array

Baseline: D=100μm, R=10μm, L=200μm, dox=1μm, εSiO2=3.9 (0.001), εSi=11.9 (10S/m)

1 2 3

Aggressor via Neighboring Victim via Shielded Victim via Return vias Victim vias

  • 5x5 TSV array with 1 driven aggressor and two

victim vias;

  • Aggressor

(TSV1) is driven with a pulse (risetime=100ps, amplitude=2V) using a 50Ohm source resistor.

  • Far end of aggressor TSV and both sides of all other

signal TSVs are terminated in 50Ohms

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

Cross Talk: TSV Array

NEXT – Frequency domain

Neighboring victim Shielded victim

NEXT – Time domain

  • Thicker oxide liners help reduce cross talk for low resistivity substrates
  • High resistivity substrates act as low loss dielectrics and therefore help reduce cross talk
  • Low resistivity substrate has a larger peak voltage and longer coupled noise duration (ISI)
  • Chip-package co-design since TSV response in the chip stack can propagate into package
slide-4
SLIDE 4

4/25/2014 4

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

Interposer: Return Path Discontinuities

Surface current distribution (Glass) @30GHz showing cavity resonance

  • Microstrip-to-microstrip transition causes

a change in the reference plane (RPD)

  • Results in large SSN voltage induced

between planes at resonance frequencies

  • Increased insertion loss

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

Eye Diagrams for Silicon and Glass Interposers

Glass Silicon

282.6p s

0.29V

17.89p s 295.2p s

0.31V

6.7ps 3.2 Gbps 210-1 PRBS stream

  • Jitter and eye opening are considerably improved in the Silicon interposer, in

comparison with the Glass interposer

  • Performance of glass interposer can be improved by using decoupling capacitors
slide-5
SLIDE 5

4/25/2014 5

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

2.5D Link Test Case

Die size: 1 mm x 1 mm, 250 um thickness (Die 1 is identical to Die 2) Double Sided Silicon Interposer size: 40 mm x 40 mm, 300 um thickness u-bump dimensions: 60 um diameter , 52 um height, 200 um pitch TPV dimensions: 40 um diameter , 300 um height, 200 um pitch

Signal Signal

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

IC Design

I/O Pad Map

GDSII Import from Cadence Virtuoso

Silicon IC BEOL

  • NCSU FreePDK 45nm technology library
  • 10 metal layers, 12um thick Cu backend
  • 11 signals total

M1 M10

u-bump

slide-6
SLIDE 6

4/25/2014 6

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

Double Sided Silicon Interposer Design

IC 1 IC 2 Interposer

(Eps = 2.51, tanD = 0.004)

20mm

Microstrip-to-microstrip signal routing used to maximize RPD effects (worst case scenario)

40mm CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

Method 1: Complete chip-interposer-chip link co-simulation

2.5D Link: Analysis Methodology I

  • Pos: Highest level of accuracy since all 3D coupling effects between the ICs and

interposer are captured

  • Neg: Not computationally feasible (3D full-wave analysis) due to the complexity of the

IC design and the aspect ratios involved

slide-7
SLIDE 7

4/25/2014 7

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

Method 2: Decoupling / Cascading approach

2.5D Link: Analysis Methodology II

Reference Plane Electric Wall used as Discrete Port Reference Electric Wall used as Discrete Port Reference

+

port

  • Center of the u-bump array is used as the reference plane (electric wall)
  • All signal, power and ground nets need to be terminated to maintain

return current path continuity

S-parameters (Die 1) S-parameters (Interposer) S-parameters (Die 2) CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

2.5D Link: Analysis Methodology II

IC Interposer

PEC sheet reference PEC sheet reference

slide-8
SLIDE 8

4/25/2014 8

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

2.5D Link: Analysis Methodology III

Method 2: Decoupling / Cascading approach

Reference Plane Reference Plane

The reference plane is selected along a uniform section of the signal traces to ensure TEM propagation mode.

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

Excitation Port Definition

Die 1

Port 1 Port 2 Port 3 Port 4 Port 5 Port 6 Port 7 Port 8

Die 2

Die 1 and Die 2 are identical

slide-9
SLIDE 9

4/25/2014 9

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

IC: Surface Current Distribution

Coupling to surrounding (non neighboring) nets can be observed at higher frequencies

100 MHz 2 GHz

Signal Excitation Port Signal Excitation Port

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

S-Parameter Results

Port 5 Port 1 Port 6

FEXT NEXT IL RL

… … … …

Port 2 Port 7 Port 3 Port 8 Port 4

  • For the “fewer pins” case,
  • nly

the PWR/GND pins neighboring the signal pins were included

  • Results

demonstrate that good correlation can be achieved provided the return current path continuity is preserved

slide-10
SLIDE 10

4/25/2014 10

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

3D IC Link Test Case

Signal Signal

Die size: 1 mm x 1 mm, 250 um thickness Tier 3 = Signal I/Os, PDN distributed between Tier 2 and Tier 1 Double Sided Silicon Interposer size: 40 mm x 40 mm, 300 um thickness u-bump: 36 um diameter , 200 um pitch flip-chip bump: 60um diameter TSV: 12 um diameter , 50 um height TPV: 40 um diameter , 300 um height

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

3D IC Link: Analysis Methodology

Method A Method B Method C Full 3D link

  • Center of the u-bump and flip-chip bump arrays are used for the reference plane locations
  • All signal, power and ground nets are terminated to maintain return current path continuity
slide-11
SLIDE 11

4/25/2014 11

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

S-Parameter Results

  • Stronger coupling between

TSV’s is observed in 3D IC’s and therefore using a decoupling simulation strategy provides inaccurate results especially for cross talk.

FEXT NEXT IL RL

CST – COMPUTER SIMULATION TECHNOLOGY | www.cst.com ECTC 2013

  • Demonstrated 3D full-wave electromagnetic analysis of a chip to

chip channel using a simple IC (few I/O’s) and Si interposer prototype

  • TSV’s demonstrate high levels of cross talk due to the conductive

Silicon substrate

  • For 2.5D applications, it is possible to decouple the IC from the

interposer and obtain accurate results up to around 20 GHz

  • For 3D IC applications, the strong coupling between the TSV’s in the

IC stack makes it impossible to perform a decoupled analysis

Conclusion