New Approach for Estimating Accurate Statistical-Based DDR4 Margin

Blog Post created by KeysightEEsofEDA Employee on Mar 30, 2017

A statistical simulation technique has become popular for the design and analysis of high-speed signals. Especially where accurate prediction of random jitter is important, such as in the measurement of eye opening at ultra-low BERs. In this DesignCon 2017 paper published by my colleagues Hee-Soo Lee, Cindy Cui, Heidi Barnes, and Luis Boluna, you can learn about advantages of the statistical approach for the accurate prediction of random jitter at ultra-low BER, and the limitation of this approach due to SSN (simultaneous switching noise). This paper proposes a solution that extracts the mask correction factor from the voltage noise calculated from a transient simulation, then uses it for accurate prediction of eye height and eye width calculation in the statistical analysis. Measurement data is provided to validate the approach.


Crosstalk and Delta-I noise are significant noise sources for DDR4 designs and are known as simultaneous switching output noise (SSON), or SSN. For DDR4 systems (up to 3200 MT/s), the Inter-Symbol Interference (ISI) and Random Jitter (RJ) induce timing margin uncertainties, which cannot be ignored because the shrinking unit interval (UI). In order to take into account the RJ and ISI effects accurately, JEDEC introduced the new DQ receiver compliance mask at 1e-16 BER in the DDR4 specification.

The new DQ compliance specification requires an eye opening at an ultra-low BER level, 1e-16, which poses a new challenge to simulation-based design methodology. The traditional simulation approach was based on SPICE-like time domain simulation technologies.

As you may find from Figure 2, the eye shrinking induced by inter-symbol interference (ISI) and random jitter (RJ) is relatively small at a low data rate (800 Mb/s). However, the timing margin decreases by 9% UI (15ps) from 103 to 1016 bits because of ISI and RJ effects at 3200 Mb/s data rate system. This proves that time-domain simulation, even with several thousand bits, is far more inadequate to accurately predict the eye opening at 1e-16 BER level.


We can get the ultra-low BER contours at a fraction of the time required for SPICE-like time-domain simulation methods by using the statistical analysis method. The dilemma is that the statistical simulation has to be used for
calculation of the ultra-low BER contours but the Delta-I noise contribution for SSN is not taken into
consideration.To address this challenge, a practical and efficient SSN induced jitter and noise model extraction
method is proposed in this paper. The extracted jitter and noise values will be used to correct the
eye height and width calculation at a certain BER level as well as the JEDEC DQ compliance mask
to reflect the eye-margin correctly. This methodology improves the accuracy of DDR4 statistical simulation, by using the mask correction factor. The extraction process of mask correction factor is relatively simple and quick but still, delivers reasonable accuracy while overcoming the limitation of the statistical simulation approach with the SSN induced time variant Delta-I noise. The validated correlation between measured and simulated data as it is discussed in this paper proves that this methodology can be effectively used for DDR4 designs. 


If you want to learn more about a practical and efficient SSN induced jitter and noise model extraction
approach and the measurement process, download this 
DesignCon paper here: 

SIPro, PIPro, free trial