Introduction and Comparison of an Alternate Methodology for Measuring Loss Tangent of PCB Laminates

Size: px

Start display at page:

Download "Introduction and Comparison of an Alternate Methodology for Measuring Loss Tangent of PCB Laminates"

Valerie Phelps
5 years ago
Views:

Introduction and Comparison of an Alternate Methodology for Measuring Loss Tangent of PCB Laminates Kevin Hinckley, Sun Microsystems, Burlington, MA Doug Winterberg, Sun Microsystems, Burlington, MA

1 Introduction and Comparison of an Alternate Methodology for Measuring Loss Tangent of PCB Laminates Kevin Hinckley, Sun Microsystems, Burlington, MA Doug Winterberg, Sun Microsystems, Burlington, MA Mike Ballou, Agilent Technologies Gustavo Blando, Sun Microsystems, Burlington, MA Jason R. Miller, Sun Microsystems, Burlington, MA Roger Dame, Sun Microsystems, Burlington, MA Alexander Nosovitski, Sun Microsystems, Burlington, MA Gregory Truhlar, Sun Microsystems, Burlington, MA Shelley Begley, Agilent Technologies Istvan Novak, Sun Microsystems, Burlington, MA 1

2 AGENDA Introduction The Capacitance Gradient Method (CGM) Laminate Study Subject, purpose, scope Copper-clad laminates: Low Frequency Bare laminates: High Frequency Unreinforced laminates Composite test results Potential sources of errors Conclusions, acknowledgement DesignCon 2010, Santa Clara, CA. February 2,

3 AGENDA Introduction The Capacitance Gradient Method (CGM) Laminate Study Subject, purpose, scope Copper-clad laminates: Low Frequency Bare laminates: High Frequency Unreinforced laminates Composite test results Potential sources of errors Conclusions, acknowledgement DesignCon 2010, Santa Clara, CA. February 2,

4 INTRODUCTION Laminate loss is becoming more important Df measurement options direct impedance measurements resonance-based methods wide-band model-based signature tests Multitude of IPC standards for Df testing No agreed-upon method followed by the majority Data (if available) may be conflicting and confusing DesignCon 2010, Santa Clara, CA. February 2,

5 AGENDA Introduction The Capacitance Gradient Method (CGM) Laminate Study Subject, purpose, scope Copper-clad laminates: Low Frequency Bare laminates: High Frequency Unreinforced laminates Composite test results Potential sources of errors Conclusions, acknowledgement DesignCon 2010, Santa Clara, CA. February 2,

THE CAPACITANCE GRADIENT METHOD (1) 1.E+03 1.E+02 1.E+01 1.E+00 1.E-01 Impedance magnitude and phase [ohm, deg] Magnitude Phase 100-50 -100 1.E-02-150 1.E+06 1.E+07 1.E+08 1.E+09 1.

6 THE CAPACITANCE GRADIENT METHOD (1) 1.E+03 1.E+02 1.E+01 1.E+00 1.E-01 Impedance magnitude and phase [ohm, deg] Magnitude Phase E E+06 1.E+07 1.E+08 1.E+09 1.E What is it and what are the steps Df derived from change in Capacitance with frequency Measure the impedance of a CCL (Copper Clad Laminate) DUT sample Extract capacitance vs frequency Establish the trendline of C(f) Calculate Df(f) from Wideband Debye model Measured and estimated capacitance and Df [F, -] 2.00E Estimated Df 1.75E E E E-10 Measured capacitance Estimated capacitance Log frequency [Hz] DesignCon 2010, Santa Clara, CA. February 2,

7 THE CAPACITANCE GRADIENT METHOD (2) The theory behind it Capacitance can be extracted from Im{Z} The real and imaginary parts of impedance are linked through causality constraints Integral wide-band Debye model needs only one Df(fo) point to define the entire curve Df is proportional to the slope of Dk Im{Z} Re{Z} ε ( ω) r 5.0E+0 4.0E+0 3.0E+0 2.0E+0 ωc Impedance Vector Integral Wide-band Debye model: ' = ε Dk 2 ' ε Δ + m m Dk(f), Df(f) [-] 1 ω2 + ln ω + Df 1 jω 1 jω ln(10) 2.0E-2 1.6E-2 1.2E-2 8.0E-3 1.0E+0 4.0E-3 0.0E+0 0.0E+0 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10 DesignCon 2010, Santa Clara, CA. February 2,

8 THE CAPACITANCE GRADIENT METHOD (3) Assumptions, benefits Error vector Relies on Wideband Debye model: >> input data is OK from any limited frequency band Admittance Vector ωc Df = G / ωc C(f) is closer related to the magnitude of Y, relative measurement error is usually lower C(f) is measured in a convenient low frequency band >> there are fewer error factors 5.0E+0 4.0E+0 3.0E+0 Dk G Dk(f), Df(f) [-] Df 2.0E-2 1.6E-2 1.2E-2 2.0E+0 8.0E-3 1.0E+0 4.0E-3 0.0E+0 0.0E+0 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10 DesignCon 2010, Santa Clara, CA. February 2,

9 THE CAPACITANCE GRADIENT METHOD (4) Limitations Does not work if/where Wideband Debye model is not valid Lowest frequency is set by impedance limit High-frequency limit is set by lowest resonance frequency Impedance magnitude and phase [ohm, deg] 1.0E+03 Phase E Magnitude 1.0E E E E E E E E E E E E E+02 Frequency limits of measurements [Hz] Resonance Limit Impedance Limit 1.0E+01 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 Size of square [mil] 1.0E-02 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10 DUT size limitation Effective working area -150 DesignCon 2010, Santa Clara, CA. February 2,

10 AGENDA Introduction The Capacitance Gradient Method (CGM) Laminate Study Subject, purpose, scope Copper-clad laminates: Low Frequency Bare laminates: High Frequency Unreinforced laminates Composite test results Potential sources of errors Conclusions, acknowledgement DesignCon 2010, Santa Clara, CA. February 2,

11 LAMINATE STUDY: Subject, purpose, scope (1) Subject Df (no Dk) Purpose Correlate CGM against other methods Find a Df test method that fits our needs Scope Glass-reinforced laminates 45% - 55% glass-resin ratio 4-5 mil thickness One glass style DesignCon 2010, Santa Clara, CA. February 2,

12 LAMINATE STUDY: Subject, purpose, scope (2) Material Laminate A FR408HR Laminate B R1566V Laminate C LGC-451HR Laminate D 370HR Thickness (mil) Glass Style Resin Content (%) 5.0 # # # # Vendor Df Freq (GHz) Method (IPC) DesignCon 2010, Santa Clara, CA. February 2,

13 AGENDA Introduction The Capacitance Gradient Method (CGM) Laminate Study Subject, purpose, scope Copper-clad laminates: Low Frequency Bare laminates: High Frequency Unreinforced laminates Composite test results Potential sources of errors Conclusions, acknowledgement DesignCon 2010, Santa Clara, CA. February 2,

14 COPPER CLAD LAMINATES: Low Frequency (1) Laminate A measured with E4294A Impedance Analyzer and 16192A SMD (Surface Mount Device) fixture Measurements taken on unconditioned samples 1.68E-09 Capacitance at mid-side points [F] 1.66E E E E-09 1.E+03 1.E+04 1.E+05 1.E+06 1.E E-02 Loss tangent at mid-side points [-] 7.50E E E E+00 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 DesignCon 2010, Santa Clara, CA. February 2,

15 COPPER CLAD LAMINATES: Low Frequency (2) Comparing the four laminates 2.50E-02 Df(f) of Copper-Clad Laminates 2.00E-02 Laminate D Laminate C 1.50E-02 Laminate B 1.00E-02 Laminate A 5.00E E E E E E E E+07 DesignCon 2010, Santa Clara, CA. February 2,

16 COPPER CLAD LAMINATES: Low Frequency (3) Laminate A, 3 x3 size Correlation to Integral Wideband Debye model is poor Differential form of Wideband Debye model correlates well C( f ) C( f + Δf ) a Df ( f ) = ; a = C( f ) f + Δf ln f π 2 ln(10) 1.00E E-03 Loss tangent at mid-side points [-] Df(f) from 1MHz 1.00E E-03 Loss tangent at mid-side points [-] 5.00E E E-03 Df(f) from 10kHz Df(f) from 100kHz 0.00E+00 1.E+03 1.E+04 1.E+05 1.E+06 1.E E-03 Df(f) from Differential Debye Model 0.00E+00 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 DesignCon 2010, Santa Clara, CA. February 2,

17 AGENDA Introduction The Capacitance Gradient Method (CGM) Laminate Study Subject, purpose, scope Copper-clad laminates: Low Frequency Bare laminates: High Frequency Unreinforced laminates Composite test results Potential sources of errors Conclusions, acknowledgement DesignCon 2010, Santa Clara, CA. February 2,

18 BARE LAMINATES: High Frequency Instrumentation for 1MHz 1GHz: E4991A and 16453A fixture (Parallel Plate) Instrumentation for 10GHz: E8363A and 85072A Split Cylinder Resonator (SCR) Sample Bare laminate sample Fixture creates plate capacitor C(f) and D(f) directly measured Cylinder Q is measured with/without DUT Dk and Df are calculated from shift of resonance frequency and Q DesignCon 2010, Santa Clara, CA. February 2,

19 B-STAGE BARE LAMINATES: High Frequency (1) All 4 laminates measured with PP and SCR Measurements taken on unconditioned samples Df of B-Stage Laminates 2.00E E E-02 Laminate D Laminate C Laminate B Laminate A B-stage: not fully cured resin Df < 1% under 1MHz All laminates show similar trends 5.00E E+00 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10 Significant changes in slope Dashed line indicates frequency range with no data DesignCon 2010, Santa Clara, CA. February 2,

20 B-STAGE BARE LAMINATES: High Frequency (2) Integral Wideband Debye model does not match Differential Debye model correlates well Df B-Stage Laminate B [-] 2.00E-02 10GHz 1GHz 1.60E MHz 1.20E E-03 10MHz 4.00E-03 Df(f) from differential Debye model 0.00E+00 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10 DesignCon 2010, Santa Clara, CA. February 2,

21 C-STAGE BARE LAMINATES: High Frequency All 4 laminates measured with PP and SCR Measurements taken on unconditioned samples Df of C-Stage Laminates 2.00E E E E-03 Laminate D Laminate C Laminate B Laminate A C-stage: fully cured resin Trend significantly different Multiple inflection points Neither integral Wideband Debye nor Differential Debye model correlates well Df(f) from differential Debye model 0.00E+00 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10 DesignCon 2010, Santa Clara, CA. February 2,

22 C-STAGE BARE LAMINATES: Etching Etching Experiment Why did trend change from B-stage to C-stage? 2.0E-02 Df, Laminate D, inner and outer layer etching [-] C-stage were etched cores Impact of inner vs. outer layer etching process is minimal 1.6E E-02 Outer Process Parallel Plate Inner Process Parallel Plate Outer Process SCR Inner Process SCR 8.0E-03 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10 DesignCon 2010, Santa Clara, CA. February 2,

C-STAGE BARE LAMINATES: Soaking Is moisture responsible for

Qualitative measurements by Parallel Plate and SCR methods 1.

0E-02 Parallel Plate Fixture soaked Split Cylinder soaked 8.

0E-03 Parallel Plate Fixture as is Split Cylinder as is 2 Hour

23 C-STAGE BARE LAMINATES: Soaking Is moisture responsible for Df(f) signature? Qualitative measurements by Parallel Plate and SCR methods 1.4E-02 Loss tangent of Laminate A [-] 1.2E E-02 Parallel Plate Fixture soaked Split Cylinder soaked 8.0E E E E-03 Parallel Plate Fixture as is Split Cylinder as is 2 Hour Soak in pressure cooker Moisture absorption increases Df but doesn t change trend 0.0E+00 1.E+07 1.E+08 1.E+09 1.E+10 DesignCon 2010, Santa Clara, CA. February 2,

C-STAGE BARE LAMINATES: Baking Step 1 Step 2 Step 3 Step 4 30min @ 110C 2 hours @ 110C 14 hours @ 110C 2 hours @ 185C Df of Laminate B [-] 2.0E-02 Step 2 1.6E-02 Step 1 1.2E-02 8.0E-03 4.0E-03 0.

24 C-STAGE BARE LAMINATES: Baking Step 1 Step 2 Step 3 Step 4 110C 2 110C C 2 185C Df of Laminate B [-] 2.0E-02 Step 2 1.6E-02 Step 1 1.2E E E E+00 Step 3 Step 4 1.E+06 1.E+07 1.E+08 1.E+09 Baking DOES change Df signature Step 4 shows previous C-stage trend Additional baking had no effect DesignCon 2010, Santa Clara, CA. February 2,

25 AGENDA Introduction The Capacitance Gradient Method (CGM) Laminate Study Subject, purpose, scope Copper-clad laminates: Low Frequency Bare laminates: High Frequency Unreinforced laminates Composite test results Potential sources of errors Conclusions, acknowledgement DesignCon 2010, Santa Clara, CA. February 2,

26 UNREINFORCED LAMINATES (1) Purpose: checking to see if difference in field orientation shows up in isotropic laminates as well 6.00E-02 Df Low Frequency [-] 2.00E-02 Df High Frequency [-] 5.00E E E E-02 Negative Df slope 1.20E-02 Inflection point in Df(f) is still present 2.00E E E E E+00 1.E+02 1.E+03 1.E+04 1.E+05 1.E E+00 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10 Low and high-frequency Df signature of DuPont FR0121A acrylic laminate DesignCon 2010, Santa Clara, CA. February 2,

27 UNREINFORCED LAMINATES (3) Df of Oak Mitsui Technologies BC24M 1/1 (left graph) and experimental laminate (right graph). Blue data points: SUN Microsystems; red data points: courtesy of Oak Mitsui Technologies. 2.5E E-02 Df BC24M Laminate Maybe no inflection point 1.0E E-03 Df Experimental Laminate Maybe no inflection point 1.5E E E E E E E+00 1.E+06 1.E+07 1.E+08 1.E+09 1.E E+00 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10 DesignCon 2010, Santa Clara, CA. February 2,

28 AGENDA Introduction The Capacitance Gradient Method (CGM) Laminate Study Subject, purpose, scope Copper-clad laminates: Low Frequency Bare laminates: High Frequency Unreinforced laminates Composite test results Potential sources of errors Conclusions, acknowledgement DesignCon 2010, Santa Clara, CA. February 2,

29 COMPOSITE TEST RESULTS (1) Df of Laminate B for two different resin contents, measured with Short Pulse Propagation (SPP) method. Data courtesy of Compeq. Df variation based on resin content [-] 2.50E E E E-02 Fitted 68.6% resin Fitted 46.7% resin 5.00E E E E E E E E+01 Frequency [GHz] SPP: Laminated interconnect is measured with a narrow pulse Different length traces are measured Complex propagation constant is calculated from far-end received pulse From cross section data, DC resistance and field-solver data, a first order model is created R(f), L(f), C(f) and G(f) are fitted to match measured response DesignCon 2010, Santa Clara, CA. February 2,

30 COMPOSITE TEST RESULTS (2) Df of Laminate A (left graph) and Laminate B (right graph) with different measurement methods. SPP data courtesy of Compeq and GCE 2.00E-02 Df of Laminate A [-] 2.00E-02 Df of Laminate B [-] 1.50E E E E E E-03 GCE 50.5% SUN PP SUN SCR 0.00E+00 1.E-04 1.E-03 1.E-02 1.E-01 1.E+00 1.E+01 Frequency [GHz] GCE 50.5% 1.20E-02 Compeq 48.66% SUN PP SUN SCR 1.00E-02 1.E-04 1.E-03 1.E-02 1.E-01 1.E+00 1.E+01 Frequency [GHz] DesignCon 2010, Santa Clara, CA. February 2,

31 AGENDA Introduction The Capacitance Gradient Method (CGM) Laminate Study Subject, purpose, scope Copper-clad laminates: Low Frequency Bare laminates: High Frequency Unreinforced laminates Composite test results Potential sources of errors Conclusions, acknowledgement DesignCon 2010, Santa Clara, CA. February 2,

32 POTENTIAL SOURCES OF ERRORS (1) Impact of electrode pressure of 16453A fixture on the measured capacitance and Df Capacitance and Df [F, -] 1.2E E-11 Capacitance 2.5E E E E E-12 Df 1.0E E E-12 Low pressure Medium pressure High pressure 5.0E E E+00 1.E+07 1.E+08 1.E+09 DesignCon 2010, Santa Clara, CA. February 2,

33 POTENTIAL SOURCES OF ERRORS (2) Impact of point averaging in E4991A Impedance Analyzer 4.E-02 Df(f) 3.E-02 2.E-02 1.E-02 0.E E E E E E E+06 '1 pt' '2 pts' '5 pts' '10 pts' '20 pts' '50 pts' '100 pts' DesignCon 2010, Santa Clara, CA. February 2,

34 POTENTIAL SOURCES OF ERRORS (3) Impact of stacking on Df. Laminate D samples were measured in 16453A Parallel-Plate fixture. 2.5E-02 Df of Laminate D 2.3E E-02 Sample 1 Sample 1&2 1.9E-02 Sample 1&2&3 Sample 1&2&3&4 1.7E-02 Sample 1&2&3&4&5 1.5E-02 1.E+06 1.E+07 1.E+08 1.E+09 DesignCon 2010, Santa Clara, CA. February 2,

35 SUMMARY AND CONCLUSIONS Wideband Debye model does not match measured data Multiple inflection points on Df(f) curves CGM can not be used to extrapolate to higher frequencies with no data Differential Wideband Debye model matches measured Df data wherever capacitance can be extracted reliably CGM can be used within the measured frequency range to crosscorrelate data Short Pulse Propagation, Parallel-plate and Split-cylinder methods provide different results DesignCon 2010, Santa Clara, CA. February 2,

36 ACKNOWLEDGEMENT The authors wish to express their thanks to the following companies and individuals for their valuable comments and suggestions, for providing samples and measurement results: LG Innotek (Jay Juon, Mickey An), GCE (Dan Slocum Jr, Dino Chen, Joe Beers), Compeq (Michael Spencer, Richard Tu, Jesse Tsai), Viasystems (Greg Lucas), DuPont (Gerry Sinks, Glenn E. Oliver, David McGregor), Oak Mitsui Technologies (John Andresakis, Jin-Hyun Hwang), Amphenol (Bob McGrath, Tony King), Panasonic (Antonio Senese), Northeastern University (Nian Sun, Xing Xing), Agilent Technologies (Yasuhiro Mori), IBM (Roger Krabbenhoft and Alina Deutsch), CCNi (Don DeGroot), Sun Microsystems (Karl Sauter, Mike Freda). DesignCon 2010, Santa Clara, CA. February 2,

37 THANK YOU 37

Making Sense out of Dielectric Loss Numbers, Specifications and Test Methods

DesignCon 2010 Technical panel: Making Sense out of Dielectric Loss Numbers, Specifications and Test Methods Panelists: Don DeGroot Roger Krabbenhoft Tony Senese Allen F. Horn Istvan Novak* CCNi IBM Panasonic