TITLE The Challenges of Measuring PAM4 Signals Panelists: Doug Burns, SiSoft Stephen Mueller, Teledyne LeCroy Luis Boluña, Keysight Technologies Mark Guenther, Tektronix Image Jose Moreira, Advantest Martin Rowe, UBM (Moderator) January 21, 2016
The Challenges of Measuring PAM4 Signals Douglas Burns Signal Integrity Software Inc. (SiSoft)
PAM4 PAM4 vs NRZ An old Story PAM4 Features 2 bits of data per UI (4 discrete logic levels) Gray Code data encoding may be used May require Forward Error Correction (FEC) Multiple codes available, needs validation with Hardware PAM4 Measurements: Measurement problem not really changed Except we are working with 3 smaller eyes, not 1 eye. Eye height at 3 separate measurement levels Vertical Bathtubs Eye Width for 3 eyes Horizontal bathtubs
Shows Eye Height as a function of probability Easier to understand than a probability color chart Great alternate view from Eye Diagram Vertical Bathtubs
Shows Eye Width as a function of probability Easy to see offsets and Eye width variations Each Eye can have its own sampling time Great alternate view from Eye Diagram Horizontal Bathtubs
Encoding Goal: reduce the number of bit errors when there is a symbol error Symbols contain 2 bits of data Gray Code Popular Code style can be programmable
Noise is More Important than Jitter Stephen Mueller, Teledyne LeCroy
How Jitter Closes Eyes Rj = 0 ps Rj = 1.6 ps Rj = 2.4 ps Rj = 3.2 ps 12 Gbaud NRZ 6 Gbaud PAM4 Rj = 3.2 ps Rj = 10.0 ps Rj = 4.0 ps Comparison showing an NRZ and PAM4 signals at the same bit rate and how the eyes close with increasing amounts of Rj applied. The PAM4 signal shows little effect from the jitter relative to NRZ.
How Noise Closes Eyes Rn = 0 mv Rn = 6 mv Rn = 12 mv Rn = 18 mv Rn = 24 mv 12 Gbaud NRZ 6 Gbaud PAM4 Rn = 32 mv Rn = 32 mv Comparison showing an NRZ and PAM4 signals at the same bit rate and how the eyes close with increasing amounts of Rn applied. The effect on the NRZ and PAM4 signals are similar.
PAM4 vs NRZ Slew Rates
NRZ Slew Rate = 4.95 V/ns Small increase in noise causes small increase in horizontal eye closure PAM4 Slew Rate = 1.41 V/ns Small increase in noise causes large increase in horizontal eye closure Δt ΔV Δt ΔV
PAM4 vs NRZ, 8 mv Rn, 0 ps Rj Slew NRZ = 4.95 nv/s Rn = 8 mv Rj = 8 / 4.95 ps = 1.62 ps NRZ Slew PAM4 = 1.61 nv/s Rn = 8 mv Rj = 8 / 1.61 ps = 4.97 ps PAM4 Due to low slew rate, PAM4 is relatively insensitive to jitter but very sensitive to noise!
The Challenges of Measuring PAM-4 Signals Luis Boluña, Keysight Technologies We need to understand PAM-4 more deeply than we do now.
Challenges of PAM-4 Impairments New Tx output topology may create new types of distortion Linear output stage Analog summing two level weighted NRZ streams to create PAM-4 Other new topology??? New Output measurements needed to quantify these new distortions New Rx Input circuits may be susceptible to new types of impairments Triple input level detectors Transition qualified phase detector in clock recovery Multi-level replacement for DFE Other new topologies??? Stressed input testing will need to emulate new impairments which cause bit errors
Generating and Measuring PAM-4 Impairments Impairments of levels and eye levels can be hard to emulate at speed. An Arbitrary Waveform Generator is suitable to provide such capabilities.
PAM-4 Validation PAM-4 Stimulus QSFP28 3 meter cable DUT Output of lossy channel with no EQ
PAM-4 Simulation to Measurement Correlation Tx with 3-tap FFE 12.5 Gbps (6.25 GBaud), PRBS7 Keysight ADS 2016.01 Channel Simulation PAM-4 at Rx Download Simulated Tx Waveform to AWG Tx AWG Tx with FFE AWG DUT Rx Sampling Oscilloscope Compare simulation with measurement RX:PAM-4 at DCA DUT:QSFP28 3m
Measured Validation Summary: Comparison inside of FlexDCA Channel Output Measured Simulated 2/3 Width 72.55ps 69.70ps 1/2 Width 74.70ps 67.90ps Simulated 0/1 Width 62.60ps 62.95ps 2/3 Height 86.2mV 87.6mV 1/2 Height 84.6mV 80.0mV 0/1 Height 106.2mV 103.0mV
Comments on FEC Oversimplified Diagrams (transcoding omitted) Past and current implementation of Forward Error Correction (FEC) have dealt with an independent FEC per lane. Future work is implementing striping packets of data across multiple lanes. Challenges arise when you stripe FEC across SerDes lanesmeaning all cumulative errors (random and burst) need to be accounted for across all lanes not just the one you are monitoring. (Frame Loss Ratio)
The Challenges of Measuring PAM4 Signals Mark Guenther, Tektrronix
PAM4 TECHNOLOGY MATURITY Dynamic Standards Environment Dynamic Tools Environment Design / Debug Goal Compliance Goal
NEW MEASUREMENTS & TECHNIQUES Measurements per Transition Type Relative Eye Position Measurements Clock Recovery Options Conventional? Selected Edges? Noise-Tolerant? IBIS Model? Spec-Compliant? Example: Rise & Fall Times Vertical Linearity Horizontal Offset
ADDITIONAL PAM4 CHALLENGES The Importance of Equalization and De-Embedding New Challenges to Equipment Fidelity With PAM4 s greatly reduced height for each eye, the noise performance (effective bits) of measurement hardware is at a premium
The Challenges of Measuring PAM4 Signals Jose Moreira, Advantest
Volume Production for PAM-4 Devices Expectation is that volume production is handled by loopback testing even for first generation devices (failure coverage and customer returns will determine if this is true) How will the PAM-4 production ramp be implemented (loopback or at-speed, bench/rack or ATE, plain wire loopback or parametric ATE loopback, golden device) AT-SPEED PATTERN NUMBER OF TESTED UNITS AT-SPEED TESTING AT-SPEED? LOOPBACK AT-SPEED LOOPBACK R&D/ LAB CHARACTERIZATION PRODUCTION RAMP VOLUME PRODUCTION TIME
Challenges for ATE ROI is not there for dedicated PAM-4 ATE pinelectronics cards. For at-speed testing the best approach would be an extension of current 28/32 Gbps NRZ solutions: Power combiner or DAC on the stimulus side. Single comparator with three independent functional tests to cover all PAM-4 levels. Main challenges are: Signal integrity Pattern (standard ATE PRBS engines cannot handle some PAM-4 patterns forcing the use of memory base patterns) test time Cost 32 Gbps at-speed NRZ ATE System
TITLE Thank you! Questions? Image If you d like a copy of these slides, contact Martin Rowe, martin.rowe@ubm.com or leave your business card at the end of the session.
TITLE Interested in contributing to EDN or EE Times? EDN: Technical articles, tutorials, how to. Image EE Times: News, opinion, tech trends, or quizzes. Contact martin.rowe@ubm.com