PAM4 challenges at DesignCon 2016
PAM4 modulation drew crowds from end-to-end at DesignCon 2016. It started on Tuesday morning, January 19 with a tutorial, "PAM4 Signaling for 56G Serial Link Applications," before moving to the afternoon Case of the Closing Eye panel.
The following day saw several technical papers and, as the exhibit hall opened, PAM4 appeared just about everywhere. On Thursday, January 21, DesignCon concluded with another panel, "Measurement Challenges of PAM4 Signals." The five panelists touched on some topics covered in other sessions and added points points not covered earlier. Panelists covered issues from simulation to bench measurements to production test.
Stephen Mueller of Teledyne LeCroy addresses the audience.
(Photo by Steve Taranovich)
Click on any image to download the panel slides (PDF)
Doug Burns of SiSoft opened the panel with a discussion about simulation. He explained how you'll need to use other horizontal and vertical bathtub curves to characterize the three PAM4 eyes. In the figure below, Burns rotated the bathtub curve 90º so it aligns with the eyes. The vertical curve shown below lets you see eye height as a function of BER (bit-error rate). The PAM4 signal's smaller eyes make it more susceptible to noise than the single NRZ eye.
Teledyne LeCroy's Stephen Mueller followed Burns with a comparison of PAM4 and NRZ signals. Using animated slides, he highlighted the differences on slope. In the figure below, Mueller shows how the PAM4 signal can have a slower dV/dt than NRZ when not transitioning the full signal amplitude. That is, when not transitioning from bits 00 to 11 or 11 to 00.
Luis Boluña of Keysight Technologies opened by saying "We need to better understand PAM4. For example, linearity is an issue now." The figure below shows the effect on nonlinearity on a PAM4 signal. Note the different eye heights.
Mark Guenther from Tektronix followed Boluña, starting with an update on Ethernet standards that are adopting PAM4: IEEE 802.3ba, 802.3bj, 802.3bm, and 802.3bs. He then explained jitter and clock-recovery techniques for PAM4. Guenther showed how a BER of 10E-6 changes how BER will be measured. "At 10E-6," he explained, "BER can be measured directly instead of having to be extrapolated as it is with 10E-12 end lower, they way we do with NRZ." Finally, Guenther showed how noise performance is important with PAM4 signals, as shown in the figure below.
Jose Moreira of Advantest closed the presentations with a topic not heard at any other PAM4 session: production test. "At this point," he said, "we're not sure if we'll test PAM4 devices in production using at-speed or loopback techniques, but the expectation is that we'll use loopback, even for early products."
Before opening the floor to questions, I stepped back and offered some of the choice words I'd heard for PAM4 over the three days of DesignCon 2016.
- Not cheap
- The wild, wild west
- Not viable for high-performance computing
- Why did IEEE pick PAM4? Don't know.
- I'm a digital guy. What's this analog signal?