DesignCon East 2004 Advances in Design, Modeling, Simulation and Measurement Validation of High Performance Boardto-Board 5 to10 Gbps Interconnects Brian Vicich, Samtec, Inc. Scott McMorrow, Teraspeed Consulting Group LLC Tom Dagostino, Teraspeed Consulting Group LLC Bob Ross, Teraspeed Consulting Group LLC Rob Hinz, Cider Designs Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 1 TERASPEED CONSULTING GROUP DesignCon East 2004 Introduction • Final InchTM, a method for the design, modeling, simulation and evaluation of high performance board-to-board interconnects. – We will present a collection of methods which, when combined, provide a powerful framework for evaluation and correlating interconnect performance, where: • Everything matters • Everything is modeled • The results speak for themselves Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 2 TERASPEED CONSULTING GROUP DesignCon East 2004 Final Inch™ Modeling and Evaluation Process Materials Measurement Modeling 3D Fullwave Time Domain Simulation 2D Frequency Dependent Extraction 2D Lumped Element S-parameter Generation Hspice W-element Table Model Multi-section Connector Model Samtec Connector Model Library Passivity Correction Spice Conversion Automatic Model Generator Final Inch™ Simulation Deck System Measurement Correlation Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 3 TERASPEED CONSULTING GROUP DesignCon East 2004 Frequency Dependent Modeling • Frequency dependent modeling of significant interconnect elements is necessary for accurate simulation of systems. – Size matters • The longer an element is, the more important that accurate frequency dependent modeling is performed. – Traces, long connectors, flex, cables …. » For short, well controlled elements, such as short board-to-board connectors, losses may be ignored with low error. – Irregularity matters • Irregular and 3-dimensional objects generally have non-TEM propagation modes and require modeling in the frequency domain. – Non-uniform traces, vias, SMA launches, connector transitions, cable transitions, connector breakout regions, antipads …. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 4 TERASPEED CONSULTING GROUP DesignCon East 2004 TEM Modeling Of Uniform Structures • Uniform long structures may generally be modeled using TEM or Quasi-TEM assumptions with 2-D field solvers. – Traces, coax, some connector cross sections …. • But error increases if the field solver does not model frequency dependent conductor and dielectric losses correctly. – Most do not! – Finite field penetration into conductors (skin effect) is often only partially modeled. Usually the resistive portion of skin effect is calculated, while the inductive portion is ignored » Most solvers provide one value for inductance, which is incorrect! Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 5 TERASPEED CONSULTING GROUP DesignCon East 2004 Frequency Dependence of Resistance Modeled with Ansoft Maxwell 2D Actual resistance values will be strongly dependent upon the conductor cross-section. AC resistance Proportional to Sqrt(f) DC resistance Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 6 TERASPEED CONSULTING GROUP DesignCon East 2004 Frequency Dependent Inductance Low frequency Inductance limit Modeled with Ansoft Maxwell 2D Internal Inductance Transition Region External Inductance Asymptote Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 7 TERASPEED CONSULTING GROUP DesignCon East 2004 Frequency Dependence of Inductance Inductance at 350 ps risetime Inductance at 150 ps risetime Variation of inductance in normal operating region is 2% to 3% of extracted value at infinity. Inductance at 35 ps risetime Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC High frequency surface inductance limit. Page 8 TERASPEED CONSULTING GROUP DesignCon East 2004 Final InchTM Trace Modeling • Our approach to trace modeling. – Utilize Ansoft Maxwell 2D. • Finite element quasi-static field solver. • Capable of extracting frequency dependent R and L. – Measure (when possible) substrate material properties across frequency (Er and Loss tangent) and use during parameterization. – Extract trace parameters using a parametric sweep. • Sweep from 10 Hz to 50 GHz for accuracy across all frequency bands. • Utilize Z and Y matrices. – RLCG matrices do not include losses in Ansoft 2D. – Create HSPICE W-element table model. • Automated process to extract Z and Y matrices to create compatible table model. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 9 TERASPEED CONSULTING GROUP DesignCon East 2004 2D Trace Modeling Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 10 TERASPEED CONSULTING GROUP DesignCon East 2004 Parametric Sweep Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 11 TERASPEED CONSULTING GROUP DesignCon East 2004 Snippet of Final W-element Table Model .MODEL final_inch_se W MODELTYPE=table N=1 + RMODEL = final_inch_se_R LMODEL = final_inch_se_L + GMODEL = final_inch_se_G CMODEL = final_inch_se_C * ###R-model### * data type = * R-model .MODEL final_inch_se_R SP N=1 SPACING=nonuniform VALTYPE=real + INTERPOLATION=spline + DATA=32 * ============= ============= ============= * FREQUENCY: + 0.0000000000000000e+000 * TABLE ELEMENTS: * === row 1 === + 5.1907890527286469e+000 * ============= ============= ============= * FREQUENCY: + 1.0000000000000000e+002 * TABLE ELEMENTS: * === row 1 === + 5.1907890900627756e+000 Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 12 TERASPEED CONSULTING GROUP DesignCon East 2004 Synopsys HSPICE W-element Fix ******************************************* * code to force HSPICE W-element time step and bandwidth algorithm * work correctly for slow edge rate signals, and play well with * other Laplacian and lumped element models * vfrog frog 0 pulse (1 0 0 25p 25p 75p 200p) rfrog frog 0 50 The above HSPICE code provides a “fix” for algorithmic problems with the welement. In a nutshell, the HSPICE w-element automatically sets the bandwidth and time step for it’s internal inverse Laplace transformations using the rise time of signals in the system. For almost all normal excitations, this causes the bandwidth to be set too low, resulting in incorrect waveform results in the time domain, and oftentimes instability when interfaced with other elements. This code forces the w-element to adjust its bandwidth to accommodate 25 ps rise times and results in extraordinary waveform accuracy, as will be seen later in the presentation. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 13 TERASPEED CONSULTING GROUP DesignCon East 2004 TEM vs. Non-TEM Modeling Of Non-Uniform Structures • Non-uniform structures require modeling with a 2.5-D or 3-D full wave approach. – Fields generally do not meet TEM or Quasi-TEM assumptions. • Electric and Magnetic fields are not reasonably orthogonal. • Lumped and/or distributed model approximations are no longer accurate. – Network parameters (S-parameters) are generally the best way to model the broadband performance of these structures. • Full wave field solvers and simulators like CST Microwave Studio can be used for the extraction of these structures. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 14 TERASPEED CONSULTING GROUP DesignCon East 2004 Connector Breakout Region (BOR) Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 15 TERASPEED CONSULTING GROUP DesignCon East 2004 SamArrayTM 3-D Modeling Top View Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 16 TERASPEED CONSULTING GROUP DesignCon East 2004 SamArrayTM 3-D Modeling Via Stack Side View Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 17 TERASPEED CONSULTING GROUP DesignCon East 2004 Single-ended S-parameters 1 2 System 3 4 S11, S22, S33, S44 = energy reflected back from ports (Return Loss) S31 = energy transferred from port 1 to port 3 (Near End Crosstalk) S41 = energy transferred from port 1 to port 4 (Far End Crosstalk) S21, S12, S34, S43 = energy transferred along through paths (Insertion Loss) The paths that are defined to be insertion loss are the through paths from the inputs to the outputs of a system. The actual port combinations will change for each design, depending upon the numbering convention. When there is only one path, two ports, it is customary to label them 1 and 2. But for multipath systems, this is not always the case. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 18 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Mixed Mode S-parameters Port 1 node 1 node 2 System node 3 node 4 Port 2 Single-ended ports may be grouped together logically to represent differentially excited ports. Four single-ended ports may be combined into two differential ports, as a simple linear mapping operation. Since each port contains two nodes, two modes of excitation may be described: Differential Mode and Common Mode. These are normally annotated using “D” for differential and “C” for common. Sdd11 – port 1 differential mode return loss Scc11 – port 1 common mode return loss Sdd21 – port 1 to port 2 differential mode insertion loss Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 19 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Mixed Mode S-parameters Excitation of a differential port can be completely described by linear combinations of even and odd mode excitation of the pair elements. Formally this means that the excitation is a linear combination of the excitation vectors: ?1 ? 1? ?v1 p v 2 p ? ?1 1 ? ? ?v1m v 2m? ? ? ? ? ? ?v1d ? v1c ? v 2d ? v2c ?? where subscripts p and m are plus and minus terminal voltages for the differential pair, and subscripts d and c are differential and common mode voltages. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 20 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Mixed Mode S-parameters The formulation of mixed mode S-parameters involves a linear transformation of the natural S-parameters: S mm ? MS nat M ? 1 Smm and Snat are the mixed mode and natural S-parameter matrices respectively. The work of the linear transformation is done with the matrix M. Let’s consider for a moment a 4-port single-ended network: Port 1 Port 2 v1 v3 4-port Network v2 v4 Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Port 3 Port 4 Page 21 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Mixed Mode S-parameters To convert the natural S-parameters of this network to mixed mode, matrix M is constructed as follows: ?1 ? 1 ? 1 ?0 0 M ? ? 2 ?1 1 ? ?0 0 0 0? 1 ? 1? ? 0 0? ? 1 1? Smm then becomes: Smm dd ? S ? MS nat M ? 1 ? ? cd ?S S dc ? ? S cc ? Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 22 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Mixed Mode S-parameters Sdd and Scc are the pure differential and common mode S-parameters, and Sdc and Scd are the mixed mode differential to common and common to differential Sparameters respectively. ?S11dd S12dd ? Sdd ? ? dd dd ? ?S 21 S22 ? ?S11dc S12dc ? Sdc ? ? dc dc ? ?S21 S22 ? ?S11cd S12cd ? Scd ? ? cd cd ? ?S21 S22 ? ?S11cc S12cc ? Scc ? ? cc cc ? ?S21 S22 ? Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 23 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Mixed Mode S-parameters Extending the M matrix to N ports is simply a matter of adding the necessary even and odd mode excitations. For example the network below: Port 1 Port 2 Port 3 Port 4 v1 v2 v3 v5 8-port Network v6 v7 v8 v4 Port 5 Port 6 Port 7 Port 8 The M matrix to combine ports 1 and 2, 3 and 4, 5 and 6, and 7 and 8 into differential pairs will be: ?1 ? 1 ?0 0 ? ?0 0 ? 1 ?0 0 M ? ? 2 ?1 1 ? ?0 0 ?0 0 ? ?0 0 0 0 1 ?1 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 1 ?1 0 0 0 0 0 0 1 1 0 0 0 0? 0 0 ?? 0 0? ? 1 ? 1? 0 0? ? 0 0? 0 0? ? 1 1? Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 24 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Mixed Mode S-parameters Like the 4-port example, Smm is: S mm ? MS nat M ?1 ?S dd ? ? cd ?S S dc ? ? S cc ? but now: ?S11dd ? dd S Sdd ? ? 21 dd ?S31 ? dd ?S41 S12dd S13dd S14dd ? dd dd ? S23 S 24 ? dd dd ? S33 S34 ? dd dd S43 S 44 ? ?S11dc ? dc S Sdc ? ? 21 ?S31dc ? dc ?S 41 S12dc S13dc S14dc ? dc dc dc ? S22 S23 S24 ? dc dc dc ? S32 S33 S34 ? dc dc dc S42 S43 S44 ? ?S11cd ? cd S Scd ? ? 21 cd ?S31 ? cd ?S41 S12cd S13cd S14cd ? cd cd cd ? S22 S23 S24 ? cd cd cd ? S32 S33 S34 ? cd cd cd S42 S43 S44 ? ?S11cc ? cc S Scc ? ? 21 cc ?S31 ? cc ?S41 S12cc S13cc S14cc ? cc cc cc ? S22 S23 S 24 ? cc cc cc ? S32 S33 S34 ? cc cc cc S42 S43 S 44 ? dd S22 dd S32 dd S42 Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 25 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Mixed Mode S-parameters ?S11 ? S21 ? S12 ? S22 ? 1 ?S31 ? S41 ? S32 ? S 42 Sdd ? 2 ?S51 ? S61 ? S52 ? S62 ? ?S71 ? S81 ? S72 ? S82 S13 ? S23 ? S14 ? S 24 S15 ? S25 ? S16 ? S26 S33 ? S43 ? S34 ? S44 S35 ? S45 ? S36 ? S 46 S53 ? S63 ? S54 ? S64 S55 ? S65 ? S56 ? S66 S73 ? S83 ? S74 ? S84 S75 ? S85 ? S76 ? S86 ?S11 ? S21 ? S12 ? S22 S13 ? S23 ? S14 ? S24 ? 1 ?S31 ? S41 ? S32 ? S42 S33 ? S43 ? S34 ? S44 Scc ? 2 ?S51 ? S61 ? S52 ? S62 S53 ? S63 ? S54 ? S64 ? ?S71 ? S81 ? S72 ? S82 S 73 ? S83 ? S74 ? S84 S15 ? S 25 ? S16 ? S 26 ?S11 ? S 21 ? S12 ? S22 ? 1 ?S31 ? S 41 ? S32 ? S42 Sdc ? 2 ?S51 ? S61 ? S52 ? S62 ? ?S71 ? S81 ? S72 ? S82 S15 ? S 25 ? S16 ? S 26 S13 ? S23 ? S14 ? S24 S35 ? S 45 ? S36 ? S 46 S55 ? S 65 ? S56 ? S66 S75 ? S85 ? S76 ? S86 S17 ? S27 ? S18 ? S28 ? S37 ? S47 ? S38 ? S48 ?? S57 ? S67 ? S58 ? S68 ? ? S77 ? S87 ? S78 ? S88 ? S17 ? S 27 ? S18 ? S28 ? S37 ? S 47 ? S38 ? S48 ?? S57 ? S67 ? S58 ? S68 ? ? S77 ? S87 ? S78 ? S88 ? S73 ? S83 ? S74 ? S84 S75 ? S85 ? S76 ? S86 S17 ? S27 ? S18 ? S28 ? S37 ? S47 ? S38 ? S48 ?? S57 ? S67 ? S58 ? S68 ? ? S77 ? S87 ? S 78 ? S88 ? ?S11 ? S21 ? S12 ? S22 S13 ? S23 ? S14 ? S24 S15 ? S25 ? S16 ? S26 ? 1 ?S31 ? S41 ? S32 ? S42 S33 ? S43 ? S34 ? S44 S35 ? S45 ? S36 ? S 46 Scd ? 2 ?S51 ? S61 ? S52 ? S62 S53 ? S63 ? S54 ? S64 S55 ? S65 ? S56 ? S66 ? ?S71 ? S81 ? S72 ? S82 S 73 ? S83 ? S74 ? S84 S75 ? S85 ? S76 ? S86 S17 ? S27 ? S18 ? S 28 ? S37 ? S47 ? S38 ? S48 ?? S57 ? S67 ? S58 ? S68 ? ? S77 ? S87 ? S78 ? S88 ? S33 ? S43 ? S34 ? S44 S35 ? S 45 ? S36 ? S46 S53 ? S63 ? S54 ? S64 S55 ? S 65 ? S56 ? S66 Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 26 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Mixed Mode S-parameters The mixed-mode formulation can be extended to any number of ports, limited only by memory and processing power. Port 1 Port 2 v1 v2 4 N-port Network V(n-1) Port N-1 V(n) Port N For Final InchTM S-parameter processing, 48-port single-ended S-parameter files are commonly processed to produce the necessary mixed-mode S-parameters. The following slides show the resulting output plots for the previous SamArrayTM Final InchTM breakout region (BOR). Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 27 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Insertion Loss Insertion loss for thin board (0.093” in this case) shows less variation across multiple pairs with differing via stub lengths, except at extremely high frequencies. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 28 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Return Loss Return loss is especially sensitive to via stub length. Longer stubs causing increased return loss. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 29 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential Crosstalk < 1% Crosstalk up to 10 GHz Mixed mode calculations show extremely low differential crosstalk throughout breakout region. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 30 TERASPEED CONSULTING GROUP DesignCon East 2004 Differential to Common Mode Conversion < 4% mode conversion below 5 GHz Differential to common mode conversion loss is primarily dependent upon trace routing and breakout skew. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 31 TERASPEED CONSULTING GROUP Non-TEM Modeling Of Vias DesignCon East 2004 – Via structures, inherent in breakout out regions (BOR) of connector pin fields, can have a significant impact on signal quality. • These structures can be extracted in a complete BOR full-wave extraction. • The via stub effect can be easily seen through measurements, and its affect on insertion loss can be easily ascertained. – For thin boards (.063” for example) via stub signal degradation is generally limited to extremely high frequencies and should not be confused with other possible sources of resonance. • The via stub effect is the most likely cause for documents erroneously stating that trace loss follows a 2rd order polynomial curve, rather than a 1st order polynomial. – We have found no material loss mechanisms responsible for any loss that is proportional to f2. » Most probable cause for erroneous loss equations are quarter and half-wave resonance phenomena in structures, whether in connectors, vias, or cables. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 32 TERASPEED CONSULTING GROUP DesignCon East 2004 Via Stubs Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 33 TERASPEED CONSULTING GROUP DesignCon East 2004 Insertion Loss of Small Connector Stubs and Vias QSH/QTH Via Stub Length Insertion Loss Comparison f 0 f Internal connector resonance pattern is independent of via stub length. -10 0.063” PCB with .010” and .050” stubs Insertion Loss (dB) -20 Short Stub f 3-D full wave modeling will capture these effects -50 short stub long stub -30 -40 short stub 2 long stub Long Stub -60 0 2 4 6 8 10 12 14 16 18 20 Frequency (GHz) Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 34 TERASPEED CONSULTING GROUP DesignCon East 2004 SMA Launches • Test and evaluation boards require methods to accurately measure material and interconnect. – Off the shelf SMA connectors are attractive. • Easy to interface to. • Deceptively simple to place on boards. • But do they work well? – Generally the answer is no, without detailed transition development. – SMA connector launches are extremely complex 3-D non-TEM structures, which lend themselves to full-wave analysis and optimization. • If not optimized, SMAs will obscure measurement of real system performance. – The goal is to make the connection to instrumentation as transparent as possible. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 35 TERASPEED CONSULTING GROUP DesignCon East 2004 SMA Launch Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 36 TERASPEED CONSULTING GROUP DesignCon East 2004 There’s No Such Thing As A Free Launch 80 Ohms A very bad launch. 70 Ohms A very typical launch. Bad launches tend to ring and obscure the interconnect under test for a very long time. 60 Ohms 50 Ohms 40 Ohms Teraspeed Launch 50 Ohm +/- 2 Ohm (Better than SMA M/F mate) Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 37 TERASPEED CONSULTING GROUP DesignCon East 2004 Optimized SMA Launch SMA M/F Connection Anti-pad Transition 51 Ohms PCB Trace Impedance 48 Ohms PCB Via Transition With good launch transparency, details of the interconnect system are not lost. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 38 TERASPEED CONSULTING GROUP DesignCon East 2004 Optimized SMA Launch S-parameters Launch is transparent to greater than 20 GHz lending extreme confidence in measurements!!! Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 39 TERASPEED CONSULTING GROUP DesignCon East 2004 Material Characterization – Accurate modeling and simulation of an interconnect requires reliable modeling of materials. • FR4, Polyimide, BT … • Most materials are specified at the blazing fast frequency of 1 MHz – Useless for high performance modeling. • Methods for accurate modeling of materials vary. – – – – Measurement of capacitance. Measurement of perturbation of microwave cavity. Measurement of resonance. Measurement of delay and attenuation. – What is the easiest and most accurate method for measurement of normal PCB and flex laminates? Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 40 TERASPEED CONSULTING GROUP DesignCon East 2004 IPC-TM-650 Test Methods Manual – Number 2.5.5.3, “Permittivity (Dielectric Constant and Loss Tangent (Dissipation Factor) of Materials (Two Fluid Cell Method)” – Number 2.5.5.5.1, “Stripline Test for Complex Relative Permittivity of Circuit Board Materials to 14 GHz” – Number 2.5.5.7, “Characteristic Impedance and Time Delay of Line on Printed Boards by TDR” describes a time domain measurement technique on a long microstrip. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 41 TERASPEED CONSULTING GROUP DesignCon East 2004 Flex Material Cross Section Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 42 TERASPEED CONSULTING GROUP DesignCon East 2004 FR4 Material Cross Section Fiberglass Bundles Er = 6.6 Epoxy Er 3.6 Non-homogeneous material has dielectric constant variation due to ratio of glass to epoxy Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 43 TERASPEED CONSULTING GROUP DesignCon East 2004 Dupont™ Pyralux® FR Material Specifications vs. Measurement • Specified – Er = 3.5 @ 1 MHz – Loss Tangent 0.020 • Measured – Effective Er ranges from 3.05 to 3.5 – Loss Tangent .0127 to .0163 • The difference is left on the table during design modeling, simulation and design trade-off! Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 44 TERASPEED CONSULTING GROUP DesignCon East 2004 Trace Width Etch Factor Determination by DC Measurement In order to accurately create models of substrate traces, it is necessary to know the actual trace width of traces produced. When cross sectional data is not readily available, a method using the ratio of DC resistances may be used within reasonable error. Or (w1 – x) R1 = (w2 – x) R2 x = (w1 R1 – w2 R2) / (R1 – R2) Where R1 and R2 = measured resistances for different widths w1 and w2 = corresponding specified widths. By utilizing several different length and width of traces on the same substrate, it is possible to determine the average trace reduction (etch factor) used during the etching process. Accurate determination of the etch factor allows for much more accurate determination of target trace widths, for impedance control and modeling purposes. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 45 TERASPEED CONSULTING GROUP DesignCon East 2004 Trace Width Etch Factor Determination by DC Resistance Measurement Drawn Design Width (mils) 5 mil 10 mil 15 mil 1 inch 0.15 ohms 0.06 ohms 0.04 ohms (adjusted to 0.0375 ohms) 2 inches 0.30 ohms 0.12 ohms 0.08 ohms (adjusted to 0.075 ohms) 4 inches 0.60 ohms 0.24 ohms 0.15 ohms Calculated Etching Adjusted Widths 3.33 mils 8.33 mils 13.33 mil Length (inches) Etch factor for this process was 1.67 mils. Longer trace lengths will allow for more accurate determination of R Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 46 TERASPEED CONSULTING GROUP DesignCon East 2004 Characteristic Impedance and Delay Design Width (mils) 5 mil drawn 10 mil drawn 15 mil drawn 3.33 mil etched 8.33 mil etched 13.33 mil etched Length (inches) 1 inch 53.5 ohms 36.6 ohms 28.9 ohms 2 inches 53.4 ohms 36.7 ohms 29.2 ohms 4 inches 53.1 ohms 36.9 ohms 29.3 ohms Field Solver Impedance 54.46 ohms 34.65 ohms 25.46 ohms Error in Z with wide traces most likely due to variation in dielectric thickness. Adhesive “squish” is proportional to the area void of copper. In the area of wider or more tightly packed traces, dielectric is thicker, causing increased Z. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 47 TERASPEED CONSULTING GROUP DesignCon East 2004 Effective Relative Dielectric Constant and Loss Measurement – Two basic measurement methods were used and evaluated for dielectric constant and loss measurements. • Resonator method with multiple resonator types. – Found to be extremely sensitive but noisy. – Not reliable for broadband measurements and lossy substrates like Polyimide and FR4. • Trace delay method. – Found to be easy to use and accurate. – Will be presented here. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 48 TERASPEED CONSULTING GROUP FR4 Effective Er Measurement Delay Method DesignCon East 2004 FR4 Er from 14.5 in Line vs Freq. (Hz) 4.5 Curve follows a shape that generally corresponds to delay variation due to internal inductance and does not necessarily indicate a change in Er of the laminate. 4.3 4.1 3.9 3.7 3.5 -1.0E+09 1.0E+09 3.0E+09 5.0E+09 7.0E+09 9.0E+09 1.1E+10 1.3E+10 1.5E+10 Trace delay vs. frequency is measured by de-embedding the cables and SMAs with a VNA but does not measure Er directly. Trace delay is a function of Er and a function of the internal inductance of the conductors. Thus we call this the effective relative dielectric constant. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 49 TERASPEED CONSULTING GROUP DesignCon East 2004 FR4 Quadratic Fit Loss Extraction FR4 Quadratic Fit for 14.5 in Line Attenuation and Conductance Loss vs SQRT(scaled Freq.) y = -1.033x 2 - 0.1425x - 0.013 R2 = 0.9946 0 -0.5 -1 -1.5 -2 -2.5 -3 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 Freq GHz Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 50 TERASPEED CONSULTING GROUP DesignCon East 2004 FR4 Total Attenuation & Conductance Loss (dB/in.) vs. Freq. (Hz) FR4 Attenuation and Conductance Loss (dB/in) vs. Freq. (Hz) 0 -0.5 -1 -1.5 -2 -2.5 -3 0.0E+00 5.0E+09 1.0E+10 1.5E+10 Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC 2.0E+10 Page 51 TERASPEED CONSULTING GROUP DesignCon East 2004 FR4 Loss Tangent vs. Freq. (Hz) FR4 Loss Tangent vs. Freq. (Hz) 0.030 0.025 0.020 0.015 0.010 0.005 0.000 0.0E+00 5.0E+09 1.0E+10 1.5E+10 Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC 2.0E+10 Page 52 TERASPEED CONSULTING GROUP Time and Frequency Domain Transformation DesignCon East 2004 Fourier Transform F ? j? ? ? ? ? ?? The value of f(t) at each instance in time has implications for all frequencies. f ?t ?e ? j? t dt The value of F (j? ) at each instance in time has implications for all frequencies. Inverse Fourier Transform 1 f ?t ? ? 2? ? ? F ? j? ?e j? t ?? Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC d? Page 53 TERASPEED CONSULTING GROUP DesignCon East 2004 Passivity ? ?1 ? Z ? ? 2?1 ?? 2 ? 2?1 ? 2 2 2 2 2? ? 2 ? 2? 2 ?? ? 2 ? ? 2 ? 1 ? ? ?? 2 .414 ? 3.414? ? 1 ? ??? 3.414 ? 2.414?? 2 ? 1 ?? Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC 1 1 -3.41 ? ? S?? ? ?? Page 54 TERASPEED CONSULTING GROUP DesignCon East 2004 Passivity 1 1 ?1 ? j ? j ? Z?? ? ? j j ? ? 1 ? 3 ? 2 j ? 6 ? 4 j? S? ? ? ? 6 ? 4 j ? 1 ? 8 j 13 ? ? Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 55 TERASPEED CONSULTING GROUP DesignCon East 2004 Passivity S error Original S-parameters 1 ? 2 .9 ? 1 .8 j ? 5 .9 ? 3 .6 j ? ? ? ? 13 ?? 5.9 ? 3.6 j ? 1.2 ? 7.3 j ? ? 1.1 ? j ? 0.1 ? j ? Z?? ? ?? 0.1 ? j 0.1 ? j ? Measurement or Extraction Error 1 -100m 1 ? 3 ? 2 j ? 6 ? 4 j? S? ? 13 ?? 6 ? 4 j ? 1 ? 8 j ?? Active Circuit 1 Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 56 TERASPEED CONSULTING GROUP DesignCon East 2004 Formal Condition for Passivity I ? S * S' Has Eigenvalues with non-negative real parts Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 57 TERASPEED CONSULTING GROUP DesignCon East 2004 Simple Linear Network Model of Simulation Stability Feedback Network Oscillation in simulation occurs whenever loop gain is greater than one. (negative eigenmodes) Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 58 TERASPEED CONSULTING GROUP DesignCon East 2004 Eigenvalue Display Non-passive near DC due to error in low frequency measurement, poor resolution, or limited extraction run time. Non-passive at high frequencies, usually due to under damped resonance. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 59 TERASPEED CONSULTING GROUP DesignCon East 2004 S-parameter Scaling Computation of S-parameter Scale Factors Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 60 TERASPEED CONSULTING GROUP DesignCon East 2004 Simulation of Original vs. Passivity Corrected and Nudged Model Instability is a very bad thing! Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 61 TERASPEED CONSULTING GROUP DesignCon East 2004 Simulation of Original vs. Passivity Corrected and Nudged Model Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 62 TERASPEED CONSULTING GROUP DesignCon East 2004 Simulation of Original vs. Passivity Corrected and Nudged Model Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 63 TERASPEED CONSULTING GROUP DesignCon East 2004 Putting the Final InchTM Together SMA Connectors Transition type Traces Breakout Vias Top PCB 1 QSE Edge Bottom PCB 2 QTE Breakout Vias Traces SMA Connectors Final InchTM test and simulation environment Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 64 TERASPEED CONSULTING GROUP DesignCon East 2004 QTE/QSE Final InchTM Connector Only vs. Connector + BOR Connector only Connector + BOR Pulse Response Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 65 TERASPEED CONSULTING GROUP DesignCon East 2004 QTE/QSE Final InchTM Connector Only vs. Connector + BOR Connector only Connector + BOR 10 Gbps Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 66 TERASPEED CONSULTING GROUP InchTM DesignCon East 2004 QTE/QSE Final Connector Only vs. Connector + BOR + 8” Total Trace Length Connector only Connector + BOR + 8” Trace Pulse Response Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 67 TERASPEED CONSULTING GROUP InchTM DesignCon East 2004 QTE/QSE Final Connector Only vs. Connector + BOR + 8” Total Trace Length Connector only Connector + BOR + 8” Trace 10 Gbps Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 68 TERASPEED CONSULTING GROUP InchTM DesignCon East 2004 QTE/QSE Final Connector Only vs. Connector + BOR + 8” Total Trace Length + SMAs Connector only Connector + BOR + 8” Trace + SMAs Ripple due to older less-optimized version of SMA launch Pulse Response Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 69 TERASPEED CONSULTING GROUP InchTM DesignCon East 2004 QTE/QSE Final Connector Only vs. Connector + BOR + 8” Total Trace Length + SMAs Connector only Connector + BOR + 8” Trace + SMAs 10 Gbps Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 70 TERASPEED CONSULTING GROUP InchTM DesignCon East 2004 QTE/QSE Final Field Solver Modeled vs. VNA Measured Model Simulation of a Trace Red – Field solver modeled Blue – VNA measurement based model 10 Gbps Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 71 TERASPEED CONSULTING GROUP InchTM DesignCon East 2004 QTE/QSE Final Field Solver Modeled vs. VNA Measured Model Simulation of a Trace Red – Field solver modeled Blue – VNA measurement based model Total delay error less than 0.6% Note slightly different trace impedance. Complete system sensitivity studies can be performed with multiple trace trace models at various impedances. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 72 TERASPEED CONSULTING GROUP InchTM and DesignCon East 2004 QTE/QSE Final 1-Meter EQCD Coax, with Accelerant Networks AN6425 PAM-4 Serdes PAM-4 eye pattern for Accelerant Networks AN6425 at 6.22 Gbps with Samtec QSE/QTE Final Inch™ test board and a 1-meter long 38 AWG micro coax assembly showing excellent eye opening. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 73 TERASPEED CONSULTING GROUP DesignCon East 2004 10 Gbps Technology Demonstration – Demo > 10GBS reliable data transfer over the QTE/QSE connector. – Use existing low cost parts. – Ability to instrument all data lines. – Show total performance including crosstalk. • 12 - 10 Gbps drivers and receivers. • 2 - differential pairs with SMAs for crosstalk measurement. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 74 TERASPEED CONSULTING GROUP DesignCon East 2004 Simplified 10 Gbps Concept Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 75 TERASPEED CONSULTING GROUP DesignCon East 2004 Description – Each differential pair will be driven by a 9.95 Gbps serial PRBS 7 data stream. – System allows for external data streams. • Ability to attach BERT for additional testing capability. – Standard QSE/QTE connector, HFEM flex and EQCD cable systems. • No expensive or exotic parts and materials. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 76 TERASPEED CONSULTING GROUP DesignCon East 2004 QTE/QSE 10 Gbps Serdes Demonstration Board SMA Connectors for Instrumentation Primary Receive board Primary Transmit board Bi-directional transmit and receive boards shown with Twin-ax cable attached. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 77 TERASPEED CONSULTING GROUP DesignCon East 2004 Instrumentation Noise Floor Less than 5 mV instrument noise floor. Measurement noise floor. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 78 TERASPEED CONSULTING GROUP DesignCon East 2004 Crosstalk Measurement Less than 20 mV total xtk. Negligible differential crosstalk with 12 simultaneous 10 Gbps pseudorandom data transmissions. Operational crosstalk is only slightly above measurement noise. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 79 TERASPEED CONSULTING GROUP DesignCon East 2004 Crosstalk Averaging Crosstalk averaging shows no uncorrelated data dependent crosstalk and verifying random aggressor patterns. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 80 TERASPEED CONSULTING GROUP DesignCon East 2004 Measurement vs. Simulation PRBS 7 Simulation of Modeled Interconnect w/o Driver PRBS 7 Measurement PRBS pattern as transmitted through connectors and PCB only. (9.95 Gbps actual data rate) Measurement differences due to additional loss and jitter in MAX3952 driver (9ps random jitter). Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 81 TERASPEED CONSULTING GROUP DesignCon East 2004 65 GHz vs. 20 GHz Sampling PRBS pattern as transmitted through connectors and PCB only, with 65 GHz and 20 GHz sampling heads. (9.95 Gbps actual data rate) Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 82 TERASPEED CONSULTING GROUP Board to Board With and Without Equalization DesignCon East 2004 Binary eye pattern for MAX3952 PRBS, PCB trace, QTE/QSE connectors, and MAX3805 adaptive equalizer shows excellent eye opening. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 83 TERASPEED CONSULTING GROUP DesignCon East 2004 Boards With 5” Flex With and Without Equalization PRBS pattern as transmitted through connectors, 5” flex and PCB only. Equalized Binary eye pattern for MAX3952 PRBS, PCB trace, QTE/QSE connectors, 5” HFEM flex assembly and MAX3805 adaptive equalizer. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 84 TERASPEED CONSULTING GROUP DesignCon East 2004 Boards With 10” Flex With and Without Equalization Binary eye pattern for MAX3952 PRBS, PCB trace, QTE/QSE connectors, 10” HFEM flex assembly and MAX3805 adaptive equalizer. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 85 TERASPEED CONSULTING GROUP DesignCon East 2004 Boards With 6” Coax With and Without Equalization PRBS pattern as transmitted through connectors, 6” cable and PCB only. Equalized Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 86 TERASPEED CONSULTING GROUP DesignCon East 2004 Boards With 0.5 Meter Coax With and Without Equalization Binary eye pattern for MAX3952 PRBS, PCB trace, QTE/QSE connectors, 0.5-meter long 38 AWG EQCD micro-coax and MAX3805 adaptive equalizer shows excellent eye opening. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 87 TERASPEED CONSULTING GROUP 1 Meter Coax With and Without Equalization DesignCon East 2004 Binary eye pattern for MAX3952 PRBS, PCB trace, QTE/QSE connectors, 1-meter long 38 AWG EQCD micro-coax and MAX3805 adaptive equalizer. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 88 TERASPEED CONSULTING GROUP DesignCon East 2004 2 Meter Twin-ax With and Without Equalization PRBS pattern as transmitted through connectors, 2 m Twin-ax and PCB with equalization. Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 89 TERASPEED CONSULTING GROUP DesignCon East 2004 For More Information – Final InchTM – A method for the design, modeling, simulation and evaluation of high performance boardto-board interconnects. • Where: – Everything matters – Everything is modeled – The results speak for themselves • www.samtec.com – [email protected] • www.teraspeed.com – [email protected] Copyright © 2004 Samtec, Inc Copyright © 2004 Teraspeed Consulting Group LLC Page 90 TERASPEED CONSULTING GROUP