Removal of Cable and Connector Dispersion in Time-Domain Waveform Measurements on 40Gb Integrated Circuits (slide presentation only)

Transcription

1 Jan Verspecht bvba Gertrudeveld Steenhuffel Belgium web: Removal of Cable and Connector Dispersion in Time-Domain Waveform Measurements on 40Gb Integrated Circuits (slide presentation only) Jonathan Scott, Babak Behnia, Marc Vanden Bossche, Alex Cognata, Jan Verspecht, Frans Verbeyst, Mark Thorn, Daniel R. Scherrer Used with Permission

2 Removal of Cable and Connector Dispersion (& Jitter) in Time-Domain Waveform Measurements on 40Gb Integrated Circuits Jonathan Scott, Babak Behnia, Jan Verspecht, Marc Vanden Bossche, Alex Cognata, Frans Verbeyst, Mark Thorn, and Daniel R. Scherrer

3 Measurement Problem Cables & connectors ruin fidelity Jitter blurs edges, corrupts averages Instrument bandwidth filters waveform June, 2002 MTT-S Page 2

4 Large-Signal Network Analyzer (LSNA) Hardware similar to a NA» + power meter» + phase reference generator Functions like a scope» Time-domain display» Arbitrary stimulus waveform Calibrates like NA Developed for precision measurement for nonlinear, active device characterization & modeling June, 2002 MTT-S Page 3

5 LSNA Constraints Prototype: 600MHz--50GHz» Soon 50MHz--50GHz» 50MHz-110GHz possible Known signal period Progress to wide release depends upon the company perceiving a market... Calibrate at (anticipate) all frequencies Similar to using sampling scope with directional coupler June, 2002 MTT-S Page 4

6 Calibration Normal 2-port NA calibration first (SOLT,LRM,etc)» Now know ratios of travelling waves at reference planes Power calibration (power ref plane) Power calibration» If on wafer, require extra 1-port cal at a coaxial connector» e.g.: On-wafer thru + coaxial SOL cal at port-2 termination» Now know absolute magnitudes of travelling waves Phase calibration» Substitute comb for sinewave (typically at power cal plane) June, 2002 MTT-S Page 5

7 Reference Comb Generator Phase of tones known Calibrated via nose-tonose sampler calibration Adequate level at all desired frequencies NIST/NPL traceability possible June, 2002 MTT-S Page 6

8 Measurement Setup IC on wafer Bias T, wafer probe & cables required Scope can attach only at arrow ( ) June, 2002 MTT-S Page 7

9 LSNA Hardware Resembles 8510 NA» slightly more complicated test set Power Meter Reference (Comb) Generator Ref Gen Power Meter Signal Gen Downconverter mw Test Set Comb source Power sensor ADC & PC June, 2002 MTT-S Page 8

10 Scope Connection View Requires» Attenuation (pad)» Bias T» Cable (distance, flexibility)» Wafer Probe Worse if scope Remote head is not small-» 4x6x12 (loaf of bread) vs...» 1x2x4 (cigarette packet) June, 2002 MTT-S Page 9

11 Measurement System Risetime 7ps (onwafer!)» Compare 12ps for 50GHz scope Some Gibbs phenomenon No (random) jitter» with NO averaging penalty No slow tail» cable response corrected 40Gb Data Amp at 1.25GB/s June, 2002 MTT-S Page 10

12 Bitstream Measurement Scope/LSNA comparison highlights difference June, 2002 MTT-S Page 11

13 Eye Measurement June, 2002 MTT-S Page 12

14 Eye Measurement Using LSNA:» Transition asymmetry visible» See true overshoot levels» Ripple uncovered» Pattern-dependent jitter (ISI) visible once random jitter reduced Simulation agrees closely with LSNA data June, 2002 MTT-S Page 13

15 Measured vs Simulated (ADS) out_eye time, psec June, 2002 MTT-S Page 14

16 Comparison with latest scopes Faster, smaller sampler modules released 2002» Dispersion Artifact? >80GHz Bw at connector Size allows less cable Dispersion still the limit on wafer! June, 2002 MTT-S Page 15

17 Conclusion LSNA versus scope:» No cable & connection distortion» No instrument response uncertainty» No instrument jitter penalty Precision» UI complicated (product immature)» LF response gap (less serious than it looks) Effort» Expensive (vector, 2-port, traceable) Readily extended in frequency 110GHz possible (t R <3.5ps) June, 2002 MTT-S Page 16

18 Did you say 40GB/s? 40GB/s signals are sinewaves when seen with 50GHz bandwidth--- not interesting ICs are intended for 40GB/s, but tested at lower rates No 40GB source 1 year ago June, 2002 MTT-S Page 17

19 What Happens to Random Jitter? Averaging convolution by jitter PDF (*PDF) Scope averaging fi degrade response (x FT(PDF)) Trigger (instrument LO) jitter is calibrated out Pattern-dependent jitter (ISI) remains---not random Comprehensive discussion in full paper submission June, 2002 MTT-S Page 18