Actions
Forum Thread Details
  • State Verified Answer
  • +1 person also asked this people also asked this
  • Replies 6 replies
  • Subscribers 284 subscribers
  • Views 891 views
  • Users 0 members are here
Related

FSBL DDR4 timings

userid0

Hi,

I am currently trying to prepare a custom PMU firmware and FSBL. What would be extremely helpful are the DDR4 memory timings.

Is there a way to obtain these?

-- Alex

Parents
  • 0

    I noticed that the Micron part # MT53B512M32D2NP-062 is listed as a 1600 (800MHz clk) part, yet the settings in the BSP are for 1066 (533 MHz).  I want to know if the setting can be changed to 1600 for better performance and if the board layout can handle it. If so, then what are the settings for the other timing parameters for the memory controller.  Micron will not release the datasheet for the part w/o an NDA at this time so it's not easy find starting points.     

    Have you guys tested the part at the higher speed ?  I realize it draws more power, produces more heat, and the timings are tighter and it may not work at that speed for any of those reasons.   But, in some applications it woudl be nice to have the memory operate at higher speeds if it works.  So if you tested it and it worked, could you give the other timing parameters for the 1600 speed setting.

    Thanks

  • 0 in reply to Former Member

    Very good question!

    Check out https://www.xilinx.com/support/documentation/data_sheets/ds925-zynq-ultrascale-plus.pdf, Table 30, LPDDR4. The specific device on the Ultra96 is the ZU3EG device, -1 speed grade, in the SBVA484 package. As you can see in Table 30, the Max data rate that this device/package supports is 1066. We are using the 1600 Micron device because that is what is available, not because that is the rate we can support.

    If you are happy with what you can do on Ultra96 but want to experiment with higher performance, then you could look at something like the UltraZed to get more performance.

    Bryan

Reply Children
  • 0 in reply to bhfletcher

    Wow, didn't realize the SBVA484 package had such limits on it.  So weird that only that package does.  You would think that the die speed grade would be the limiting factor and not the package.   I guess that's what I get for only looking at part of the table when I was researching the part.

    Thanks for clearing that up.  If the single channel ADC product sells and customers want more channels in a system then I'll probably use the ZU7EV based board.  It has 64 bit memory and a lot more of it.