User manual

Table Of Contents
Zynq-7000 AP SoC Technical Reference Manual www.xilinx.com 316
UG585 (v1.11) September 27, 2016
Chapter 10: DDR Memory Controller
across PVT is slightly less than 90 degrees, and will be automatically provided by Vivado Design Suite
for inclusion into the FSBL or other user code.
10.6.10 Alternatives to Automatic DRAM Training
If for some reason the automatic training is not successful, alternative calibration schemes can also
be used.
TIP: Training failures can be detected by performing a simple memory write-read-compare test. Since
training is done independently for each byte lane, the memory test should check each data byte
independently. In the event of training failure, two possible solutions are proposed here: a
semi-automatic and a manual training method. As the method gets more manual, the training time
increases. It is therefore recommended to follow this sequence:
1. Try automatic training, verify board measurement-driven initial values
2. If failed, try semi-automatic training
3. If failed, use manual training
Automatic Training
The standard training procedure is described above.
The estimated time for initialization and training is 1-2 ms.
Semi-Automatic Training
This method is useful when system/board delays are known, but PVT timing uncertainty causes the
automatic training to fail. Note that only two initial timing parameters are needed to enable
successful automatic training:
•Write DQS to CLK skew
The one-way board delay from Zynq to DRAM
These values are known in this case, but the PHY PVT variations modify these values in an additive
fashion. Therefore, given a nominal delay value T, the actual value might be in the range (T-delta,
T+delta), where delta is the maximum PVT variation.
The semi-automatic training method is performed as follows:
1. Divide the range (T-delta, T+delta) into n parts, and thus create (n+1) possible values for each of
the two delay parameters.
2. Perform (n+1)
2
automatic training procedures and follow each one with a memory test.
For example, for n=2, the three data points for each parameter are T-delta, T, and T+delta. Perform
nine automatic training procedures and observe the results. For n=4, perform 25 tests, etc.