About master and slave design of EMIF on arria 10

[lambert_yu]

Hi all,

     I face problems about master and slave design of EMIF's IPs on arria 10:

       for this design case :

       1) each emif_usr_clk signals from emif_master and emif_slave is always the same?

       2) the amm_* signals of EMIF's controller is timinged to emif_usr_clk?

       3) Could I can only use the emif_usr_clk from emif_master to timing  the amm_* signals of emif_master and emif_slave?

       4) If above problems is correct, I found that amm_rdvalid signals from emif_master and emif_slave is not synchronous and have big timing skew between each other from scope.

       5) from the example simulation result, I found that EMIF's controller can receive up to 11 requests (write/read or both), I want to that the burstcounter of read request is up to 2^7? And if I send one this read request, I can still send up to 10 write request, right?

    software : quartus pro 16.1

    fpga type : arria 10

       Could someone help me about above problems?

Brs,

Lambert

[AdzimZM_Intel]

Hi Lambert,

I'm Adzim. Thanks for using the Intel Community.

May I know which memory type that you used?

Thanks,

Adzim

[lambert_yu]

Hi Adzim,

   I am using DDR4-2666 ( MT40A512M16LY-075), thanks.

Brs,

Lambert

[AdzimZM_Intel]

Hi Lambert,

The core clock network sharing will necessary the PLL reference clock sharing.

One of the interfaces will be a master and the other will be slaves.

The master and slaves are used the same reference clock and the output clocks signals are also same.

So you are right for your questions.

But I'm still working on to provide the answer for the question 4.

For the burstcount, it is limited to width of 1 - 11.

The minimum burst is 2^1 and the maximum burst is 2^11.

So for your design, you can create the burstcount of 2^7.

And by the way, the maximum burst length for Arria 10 external memory interface IP is 127 which is equivalence to 2^11.

You can refer to the Avalon Interface Specification document in the link below:

https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/manual/mnl_avalon_spec.pdf

Thanks,

Adzim

[lambert_yu]

Hi Adzim,

     I don't do the thing that compares the phase of the emif_usr_clk from the ddr4 master and ddr4 slave, just because I got your answer.  About the problem4, I use one async_fifo to store the ddr4 commands which generated through timinged to emif_usr_clk of ddr4 master for the ddr4 slave, but I still want to get one real reason about this problem just because the efficiency issues so that I must wait that the slowest one finishes the current burst operation before next burst operation(specially wr2rd or rd2wr switch operation).  

     Besides, for the avalon interface of the ddr4 controller, which is better for  instancing one avalon source IP or use my own logic to emunite one avalon source IP? Now I chose later, but I feel it's not good though I satisfy the timing requirement for the avalon interface.

Brs,

Lambert

[AdzimZM_Intel]

Hi Lambert,

Switching between read and write operation can reduce the efficiency of your controller because its needs to issue the precharge and activate commands for the operation and access the bank and row.

By the way, there are methods to improve the efficiency of your controller such as Auto-Precharge Commands, Additive Latency, Calibration and so on.

You can refer to the Arria 10 EMIF User Guide in the link below on Chapter 13.

https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/ug/ug-20115.pdf

But its depends on your design controller to handle the traffics of the operations.

Thanks,

Adzim

[lambert_yu]

Hi Adzim,

      Is there any information for the timing skew of avl_master_ready and avl_slave_ready signals based on master and slave design of EMIF ?

Brs,

Lambert

[AdzimZM_Intel]

Hi Lambert,

Do you have the timing report regarding to this?

Thanks,

Adzim

[AdzimZM_Intel]

Hi Lambert,

Do you have any update on my last feedback?

Regards,

Adzim

[lambert_yu]

Hi Adzim,

     Just because the time gap got from the scope between avl_ready_s and avl_ready_m, the time gap got from the scope between avl_read_valid_s and avl_read_valid_m are all not one a fix value, so I can not privide the timing skew.

Brs,

Lambert

[AdzimZM_Intel]

Hi Lambert,

I put the case to closure since you already found the solution for your issue.

Feel free to file a new case if you encounter a new problem.

Regards,

Adzim

[AdzimZM_Intel]

Hi Lambert,

I put the case to closure since you already found the solution for your issue.

Feel free to file a new case if you encounter a new problem.

Regards,

Adzim