Silicon Valley Test Conference 2013 Configuration Dependent Test Socket Performance Gert Hohenwarter GateWave Northern, Inc. g.hohenwarter@gatewave.com Silicon Valley Test Conference 2013 South East Asia Test Symposium
Silicon Valley Test Conference 2013 Objective Examine effect of different ground configurations on socket performance Use HFSS 3D field simulator model output for socket configurations Use real world PCB model in Mentor HyperLynx for overall performance assessment Silicon Valley Test Conference 2013 2 South East Asia Test Symposium 2
Silicon Valley Test Conference 2013 Simplified socket Generate model output for ‘as tested’ configuration: 3x4 array Silicon Valley Test Conference 2013 South East Asia Test Symposium
Silicon Valley Test Conference 2013 Grounding Ground-Signal-Ground GG-S-S-GG Silicon Valley Test Conference 2013 4 South East Asia Test Symposium 4
Silicon Valley Test Conference 2013 Rationale Insertion loss S21 (f) Test configuration (grounded unused pins) = red All 50 Ohms (except one Ground) = green All open (except one ground) = blue Silicon Valley Test Conference 2013 5 South East Asia Test Symposium 5
Silicon Valley Test Conference 2013 To be avoided Unused pins become resonators Lowest frequency occurs for case of open on one end, short on the other Silicon Valley Test Conference 2013 6 South East Asia Test Symposium 6
Silicon Valley Test Conference 2013 Grounding Performance changes with location and number of grounds Insertion loss S21(f) sig out # grounds sig in Return loss S11(f) Silicon Valley Test Conference 2013 7 South East Asia Test Symposium 7
S-parameters handy for diagnosis, but…. … eye diagram is often used at frequency of interest when socket is combined with rest of circuit: 6 GB/sec Bit-error rate (BER) Silicon Valley Test Conference 2013 8 South East Asia Test Symposium 8
Silicon Valley Test Conference 2013 Socket vs. PCB Performance of socket is intertwined with performance of PCB. Example: High speed circuit, 12 GB/s, cases 1 (blue),4 (green) and 8 (red) Design File: GNDConfig.ffs HyperLynx LineSim v9.0 U1_B01.19 generic TDR Net001 TL1 J2 Via1b_HFSSDesign1.s2p Po... TL2 J1 7_S2POut_1.s2p Port1 Port2 TL5 R10_B00 DR41D_B0... Net004 Silicon Valley Test Conference 2013 9 South East Asia Test Symposium 9
Silicon Valley Test Conference 2013 Load board Real world load board parameters are set up in Mentor HyperLynx simulator in order to generate a representative signal path model Silicon Valley Test Conference 2013 10 South East Asia Test Symposium 10
Complete signal path with PCB Socket: Socket is modeled as a connector between device-under-test (DUT) and load board Socket is modeled as a connector between device-under-test (DUT) and load board Silicon Valley Test Conference 2013 11 South East Asia Test Symposium 11
Complete signal circuit with PCB Device-under-test (DUT) is represented with its actual properties Load board Socket Socket block can be either model output from 3D simulator or actual measurement results Silicon Valley Test Conference 2013 12 South East Asia Test Symposium 12
Real grounding, coupling, DUT@50 Ohms Insertion loss S21 (f) FEXT,NEXT (f) Device-under-test (DUT) is assumed to have 50 Ohm termination Silicon Valley Test Conference 2013 13 South East Asia Test Symposium 13
Real grounding, coupling, real DUT with L,C,R Insertion loss S21 (f) and FEXT,NEXT (f) Device-under-test (DUT) is represented with its actual properties Silicon Valley Test Conference 2013 14 South East Asia Test Symposium 14
Ground position dependence Cases: Ground near a signal pair, ground far from pair ground between pair – 2 different types of socket Z = 35 Ohms - Z = 49 Ohms Silicon Valley Test Conference 2013 15 South East Asia Test Symposium 15
Silicon Valley Test Conference 2013 Far ground 3 GB/s signal through one connection and NEXT plus FEXT -> Eye worse for near 50 Ohm socket but crosstalk lower NEXT/FEXT Z = 35 Ohms - Z = 49 Ohms Silicon Valley Test Conference 2013 16 South East Asia Test Symposium 16
Silicon Valley Test Conference 2013 Near ground 3 GB/s signal through one connection and NEXT plus FEXT -> Eye worse for near 50 Ohm socket but crosstalk lower NEXT/FEXT Z = 35 Ohms - Z = 49 Ohms Silicon Valley Test Conference 2013 17 South East Asia Test Symposium 17
Ground between signals 3 GB/s signal through one connection and NEXT plus FEXT -> Eye worse for near 50 Ohm socket / crosstalk very comparable NEXT/FEXT Z = 35 Ohms - Z = 49 Ohms Silicon Valley Test Conference 2013 18 South East Asia Test Symposium 18
50 Ohm socket comparison ground between signal pins Two (near 50 Ohm) sockets with comparable (good) – 1dB points -> Ground between pins results in very comparable performance NEXT/FEXT Z = 49 Ohms - Z = 49 Ohms Silicon Valley Test Conference 2013 19 South East Asia Test Symposium 19
50 Ohm socket comparison ground adjacent to signal pins Two (near 50 Ohm) sockets with comparable (good) – 1dB points -> Ground adjacent to pins no longer sufficient to avoid Xtalk NEXT/FEXT Z = 49 Ohms - Z = 49 Ohms Silicon Valley Test Conference 2013 20 South East Asia Test Symposium 20
Silicon Valley Test Conference 2013 Conclusion Socket design, analysis and measurement for SI is straightforward ----- but ----- Socket optimization for a given application can be case specific Overall system analysis is mandatory as frequencies increase Silicon Valley Test Conference 2013 21 South East Asia Test Symposium 21