Report to IPC D24D: Intel PCB Transmission Line Loss Characterization Metrology Xiaoning Ye, Key Contributors: Jimmy Hsu, Kai Xiao, et al. 1
Background Current IPC test methods under TM-650 are not adequate to address the quality of measured data for electrical printed boards to meet the demand of high speed applications. Current IPC standard does not encompass common industry methods (such as TRL, etc.) and new/recent developments (2X-thru method such as AFR, SFD approach, etc.) Intel DCG (DataCenter Group) SI Team has been working with Industry Partners to address this issue for several years, including the introduction of Delta-L and Delta-L+ approach. IPC D24D task-force was established 2016 to close the gap in the existing standard. 12/16/2016 2
PCB Tline Loss Measurement at Different Stages Material Selection/Qualification Board Sampling HVM Monitoring A few test samples are made Focus: High accuracy. Typical Participants: Intel, Testing house, OEM/ODM, Material Vendor, PCB manufacturer Quality control Focus: Low cost, fast throughput Typical Participants: PCB manufacturer 12/16/2016 3
Typical PCB Loss Testing Setup 2-line (2L) Method to remove fixture effect Microwave Probing Coaxial Connector DUT DUT 12/16/2016 4
Intel s Strategy - Delta-L & Delta-L+ 3L Method 2L Method Original Delta-L Delta-L 1.0 Delta-L 2.0 Delta-L 3.0 Delta-L+ 1L Method ARO (advanced Root Omega) Fitting Measurement Uncertainty Analysis Material Property Extraction Probe Quality Requirement 12/16/2016 5
Delta-L+: 3L, 2L, 1L Method http://www.intel.com/content/www/us/en/processors/xeon/delta-l-plus-methodology-for-electrical-characterization-guide.html Choose coupons with different length combinations at different stages of PCB Characterization 3L 2L A 10 inch B 5 inch C 2 inch Note: Actual length may vary 1L Best accuracy Cost effective Small Coupon Most suitable for: Material Characterization DK/DF Extraction, Insertion Loss & Surface Roughness characterization Most suitable for: Board quality validation Insertion loss & Impedance validation Most suitable for: HVM monitoring Insertion loss and impedance variation, by one-length approach Typically <5 Boards Typically 5-30 Boards Sample size varies Material Selection Board Sampling HVM Monitoring 12/16/2016 6
Delta-L+ (3L) Three-Line Method A B C 2 inch 5 inch 10 inch (Ade-embedC)/Length (Bde-embedC)/Length f Dk/DF Extraction Surface Roughness characterization 3L Method 1. A de-embedded by C 2. B de-embedded by C 3. db/in comparison of step 1 and step 2 4. Dk/Df extraction, surface roughness characterization, etc. Note: Other de-embedding method (such as TRL, 2X-Thru) can be implemented as well 12/16/2016 7
Delta-L+ (2L) Two-Line Method 2L Method A B 5 inch 10 inch Delta-L 1.0: Direct subtraction of S21 Delta-L 2.0: Direct subtraction of S21 (after moving average) Delta-L 3.0: De-embedding with Eigenvalue Method db/inch 2 Evolution of Delta-L Method 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 0.00E+00 5.00E+09 1.00E+10 1.50E+10 DeltaL 1.0 DeltaL 2.0 DeltaL 3.0 Freq (Hz) 12/16/2016 8
Delta-L+ (2L) Two-Line Method - Multiple Algorithms Eigenvalue Method (default for Delta-L 3.0) (See Next Slide) Delta-L 2.0: http://www.intel.com/content/dam/www/public/us/en/documents/guides/delta-l-methodologyfor-electrical-characterization-guide.pdf K. Xiao, X. Ye, J. Hsu, T. Su, Y. Li Cost-Effective Characterization of Dissipative Loss of Printed Circuit Board Traces, IEEE Intl. Symposium on EMC, July 2016 Closed-form method: In preparation for an IEEE journal publication. Provisional Patent Granted. Thru-Line method B. Chen, X. Ye, B. Samaras, J. Fan, A Novel De-Embedding Method Suitable for Transmission-Line Measurement, IEEE Asia-Pacific International Symposium on EMC, May 2015. (Best student paper award finalist) 2X-Thru method X. Ye, J. Fan, J. Drewniak, New De-embedding Techniques for PCB Transmission-Line Characterization, DesignCon, Jan 2015. (Best paper award finalist) 12/16/2016 9
Delta-L+ (2L) Two-Line Method - Eigen-Value Algorithm Ref Plane Ref Plane Ref Plane Ref Plane DUT T A T B T A T DUT T B L1 (2x-Thru) L2 (2x-Thru+DUT) A & B have same Eigen-Value Acknowledgement: Prof. Jun Fan (M S&T) 12/16/2016 10
Correlation Among 5 different Methods db/inch@ 4GHz 1043 VNA Total Delta-L measurements (across different materials/vendors) 5 and 10 lines Each measurement is post-processed by 5 different deembedding algorithms 12/16/2016 11
Correlation Among 5 different Methods db/inch @ 8GHz 12/16/2016 12
De-embedding Uncertainty Analysis for 2L Method Blue: Raw de-embedded data Red: Fitted curve with ARO method (ARO: advanced root-omega method) Goal: Remove measurement noise in the measurement data Future D24D topic to go over details 12/16/2016 13
Delta-L+ (1L) One-Line Method Material Selection/Board Sampling HVM Monitoring A A B Use two lines to get accurate loss characterization B A Keep track of performance of A Use only one line to monitor the HVM variation A A Criteria: <TBD variation @ Nyquist frequency for at least x samples 12/16/2016 14
Summary Delta-L+ Strategy to address different focus of PCB Loss characterization at various stages of manufacturing We investigated 5 different algorithms to de-embed transmission line losses, and generally good correlations are achieved among them Eigenvalue method is selected as default method for latest Delta-L release (Delta-L 3.0). Other topics to report to D24D in future agenda: Advance Root-Omega (ARO) Method to smooth the measured data Uncertainty analysis Probe quality qualification Close-form Delta-L algorithm 12/16/2016 15