You need to test, we're here to help.

You need to test, we're here to help.

11 January 2021

Four Measurement Best Practices

To start the New Year right, we’re going to talk about four measurement "best practices", which will help you get the most out of any oscilloscope you have. These are important when doing any type of measurement—and you can get a good start on them simply by asking yourself the four questions in the sidebar.

1. Anticipate the results

Those who are familiar with Dr. Eric Bogatin’s Rule #9 will know this one. Before you do any measurement, anticipate what you expect the result to be, because that is the most important way of identifying if there is a potential problem. 

04 January 2021

Decision Feedback Equalization in DDR

Figure 1. A transmitted rectangular pulse suffers
distortion by the time it reaches the receiver. 
Broadening and reflections from previous
transmitted bits add to the pulse response, 
creating inter-symbol interference.

High-speed serial links such as those used in DDR4 and DDR5 are subject to a variety of signal degradation challenges.  Insertion losses, frequency dependent attenuation and inter-symbol interference (ISI), as well as others, are among the most commonly encountered sources of signal degradation. 

Figure 1 shows how reflections can cause ISI on a rectangular pulse. When a rectangular pulse is transmitted, it suffers distortion which is apparent when it reaches the receiver.  It may be broadened due to group delay dispersion because different frequency components of the signal propagate along the signal path at differing velocities. In addition, there may be echo pulses, due to impedance mismatches in the channel.  These mismatches cause reflections that propagate back and forth over the channel and appear as these echoes where subsequent bits should be.

Generally, these signal losses can be compensated for using any of several equalization techniques. The commonly used equalization techniques are feed forward Equalization (FFE), Continuous Time Linear Equalizations (CTLE), and Decision Feedback Equalization (DFE).  FFE and CTLE basically filter the signal to increase the amplitude of high frequency signal components eliminating the frequency dependent attenuation of the signal path. 

Figure 2. Example of a 4-tap DFE equalizer.
DFE is a non-linear equalizer which quantizes the signals and feeds back discrete symbols, as shown in Figure 2.

The DQ signal goes through an analog gain adjustment stage and is applied to a summer.  The output of the summer goes to the first slicer, implemented as a flip flop, where the logical state (0 or 1) is determined, and the signal is delayed by one clock cycle.  The signal propagates through a total of four such slicers.  The output of each slicer is weighted by individual tap weights T1 – T4 and is then fed back to the summer where the weighted sum is subtracted from the DQ signal.  The feedback returns the scaled copies of the last four bits to the input.  This effectively removes ISI from the input. Since the signal is quantized and noise component is ignored in that process it is not propagated.  The clock is required to assure correct timing of the summation components.

DDR5 specifies the use of Decision Feedback Equalization or DFE.  DFE is used rather than FFE or CTLE because it provides equalization without increasing the noise level on the signal. FFE and CTLE basically boost the high frequency response of the channel to compensate for the channel’s frequency dependent attenuation.  The boost is the result of an analog filtering process. This has the effect of increasing noise and noise-like effects like crosstalk, as well.  The slicer stages in DFE quantizes the signal ignoring the noise voltage, and it does not propagate noise to the output of the equalizer. So, it provides high frequency boost without the noise.

In addition to the excellent noise performance, DFE is relatively inexpensive to implement.

The DFE equalizer has one possible negative characteristic: if one of the slicers makes an incorrect determination, then the output will see a burst of errors until the sequence is cleared.  This is a not common event, especially if the data sequence is random.

The DDR Debug Toolkit includes tools for setting up DFE. The weight of each tap can be entered manually, or the DFE can be “trained”.  The training process will automatically determine and enter the correct tap weights.

To learn more about DDR testing and our DDR solutions, watch the on-demand webinar, DDR4/5 & LPDDR4/5 Probing and Debug Solutions.

Also see:

Which Virtual Probing Method to Use?

Isolating DDR Read and Write Operations

Removing Reflections from DDR Signals Probed Mid-Bus