DesignCon 2012: “Closing eye” panel discusses crosstalk

DesignCon 2012 opened on Monday, January 30 with “Case of the closing eye” panel discussion. Representatives of Agilent Technologies, LeCroy, and Tektronix joined “Engineers who actually do something,” said Ransom Stephens, for this annual gathering. The lively discussion filled the large room at the Santa Clara Convention Center.

Click here to listen to the entire discussion (mp3)

This year’s discussion focused on the crosstalk problems that engineers face when designing and testing serial links at speeds of 25 Gpbs and higher and what engineers will face in the near future. Links with 10 lanes, 10 Gbps each and four lanes, 25-29 Gbps each are here and growing in deployments opening the door for more interference from crosstalk. “They’re nothing but trouble,” said Stephens.

LeCroy’s Marty Miller, described by Stephens as “One of the great jitter theorists,”opened with a discussion on crosstalk, saying that crosstalk is indirectly related to jitter because its causes variances in signal amplitude, which, along with jitter, result in eye-diagram closures. “Crosstalk is nothing more than interference in the amplitude domain,” he noted. “It’s really a study of noise.”

Pavel Zivny of Tektronix followed Miller. “The problem with crosstalk measurements,” he said, “is that you need the entire system running to produce the interference.” That makes the measurement difficult. He also noted that crosstalk has a certain amount of random noise. “A few years ago, we couldn’t separate crosstalk noise from random jitter, but now we can.” When making jitter measurements, he noted that BERTs don’t make mistakes, but jitter analysis tools often do.

Eric Kvamme of LSI described the company’s test chips engineers use to characterize SerDes devices. The test chips have more pins than production parts so engineers can have complete access to nodes in the chip. Unfortunately, designers build systems with many more SerDes devices that LSI engineers can put in a test chip. He also noted that measurement systems in the lab use high-quality, now-noise cables and connectors, but they don’t represent the connections seen on a board (see photo of slide). Thus, a board can have hundreds of SerDes paths. “You have to understand the product environment.”

Altera’s “Dean of jitter,” Mike Peng Li, spoke about how equalization can solve the problem of the closed eye and how it can’t. He described equalization as “medicine” for fixing closed eyes. Equalizers are essentially high-pass filters that boost the high frequencies in a bit stream because the higher the frequency, the more loss will occur as a signal travels through cables, connectors and PCB traces.

Mark Marlett of Analog Bits explained SerDes standards are adapting to crosstalk problems. By “standards,” Marlett was referring to mathematical models of jitter and crosstalk. He emphaized the need for simple, consistent models in simulation software and in measurement equipment.

Agilent’s Greg LeCheminant pulled it all together with a historical perspective. His point was that with speeds of 30 Gbps, clocks run at 15 GHz with higher harmonics. Thus, modeling and measurement techniques used by microwave engineers are essential for digital engineers.