by Christian Filip, Mentor Graphics
Today’s electronic products are becoming increasingly dense and complex, making it ever more challenging to meet performance targets. High-speed design requires fast and accurate simulations from the design planning phase, throughout the design process, and in the post-layout stage, to ensure that products can be manufactured and work as intended in the field. Selecting the right tools to simulate analog and digital circuits for high-speed applications helps avoid costly re-spins and physical prototypes. There are several tools available in the market to check for BER, COM, EMI, SERDES, crosstalk analysis, signal integrity, and power integrity to ensure high-speed performance. Of particular interest is the difference between Channel Operating Margin (COM) and Bit Error Rate (BER) analysis for accurate and efficient high-speed design performance.
Since the IEEE 802.3bj 100 Gb/s Backplane Ethernet Standard has been officially approved, COM is becoming an important quality-evaluation method for SERDES links. COM was preceded by a long history of eye/BER evaluation techniques but they have never been officially standardized. Different tools and methods have been used to identify channel characterization response and evaluate the effect of ISI, jitter noise, and crosstalk. Overall, many decisions – especially those regarding channel characterization, waveform processing, data accumulation, and impairment handling, are left to the developers. Although many tools are available on the market, we can hardly find two that produce identical results and in some cases the differences can be significant.
Mentor Graphics has analyzed the computational procedure specified for COM and compared it to traditional statistical eye/BER analysis. The differences between these two approaches ranged from how they perform channel characterization, to how they consider Tx and Rx noise and apply termination, to the differences between numerical procedures employed to convert given jitter and crosstalk responses into the vertical distribution characterizing eye diagrams and BER. Depending on the channel, COM may potentially overestimate the effect of crosstalk and, depending on a number of factors, over- or underestimate the effect of transmit jitter, especially when the channel operates at the rate limits. A modification to the COM procedure eliminates these problems without considerable work.
One of the goals for COM is to minimize ambiguity in finding the signal/noise ratio, a measure closely related to BER. Mathematically, COM is equivalent to statistical analysis, but performs it only for one vertical cross-section of the eye diagram that corresponds to the “best” sampling time. Therefore, it is reasonable to compare COM with statistical eyes, rather than with bit-by-bit analysis which is often unable to provide sufficient sample size. Mentor Graphics and Wild River have written a white paper titled, “BER- and COM-Way of Channel-Compliance Evaluation: What Are the Sources of Differences?,” which just received the DesignCon Best Paper Award. This paper analyzes the computational procedure specified for COM and compares it to traditional statistical eye/BER analysis, including examples of false-positive and false-negative cases to explain the difference by certain features of the links that differently affect the BER and COM numbers.
Christian Filip is a Product Marketing Manager for High Speed Analysis Products at Mentor Graphics Boards Systems Division. Filip joined Mentor Graphics Corporation in 2014 as a product marketing manager for high-speed products. His interest includes high-speed serial link design, modeling and simulations of DDR memory interfaces as well as power integrity. Prior to joining Mentor Graphics, Cristian was a Senior Hardware Engineer specializing in signal and power integrity at General Dynamics Canada in Ottawa. Cristian holds a M. Eng. In Electronics and Telecommunications from the Polytechnic University, Timisoara, Romania and is member of Professional Engineers Ontario.
Leave a Reply
You must be logged in to post a comment.