• State
  • Locked Locked
  • Replies 3 replies
  • Subscribers 98 subscribers
  • Views 436 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Sitara AM3894 and DDR3 memory interfacing

Nicholas C Lee
Prodigy 225 points
Other Parts Discussed in Thread: AM3894, AM5K2E04

I am starting so spec a Linux mini-PC design using the AM3894 for which I would prefer to use standard socketed 2GByte SO-DIMM DDR3 memory.

Please can you tell me if the AM3894 can be interfaced to commercial SO-DIMM DDR3 laptop memory?

e.g. http://download.micron.com/pdf/datasheets/modules/ddr3/jsf16c256x64h.pdf

Please let me know if there are any limitations regarding how Linux would see the 2GByte address space. Thank you.

And/Or If there is some fundamental memory interface limitation of the AM3894 that prevents me from using a socketed 2GByte SO-DIMM DDR3 memory, then please can you deescribe a 2GByte-total discrete memory chips solution that I could use. i.e. memory chip part number, and how many of those chips to make 2GByte. (4 chips max please)

Thank you

 

  • 0
    TI__Guru**** 393215 points
    Hi Nicholas,
     
    I don't think this will be possible. The SO-DIMM is 64-bit, while the AM3894 has two 32-bit EMIF's. There is no accomodation on the SO-DIMM for the second set of A[13..0] and BA[2...0] signals, not to mention other differences in memory organisation.
     
    For a 4-chip solution you will need something like MT41J256M16 (256M x 16 DDR3), two chips per EMIF. For more details on DDR3 design please refer to AM3894 Datasheet Section 8.3.2.

  • 0 in reply to Biser Gatchev-XID
    Prodigy 225 points

    Dear Biser,

    Thank you for the prompt and helpful reply.

    The 32-bit memory interface does rather limit the AM3894 for our application.

    I have looked at the price of the MT41J256M16 and they were over £25 each (at 1K units), so to solder a 2GB memory on our board would cost over £100, which is not acceptable. For comparison, an equivalent SO-DIMM can be bought for £8.50.

    By using 8 of the 8-bit wide MT41J256M8 chips instead, the cost could be reduced (to about £30) at the expense of board area. This is under consideration, but it is hardly ideal. It would actually be cheaper to use only 32-bits of the 64-bit bus from a 4GB SO-DIMM, but our end users would then be confused as to why they couldn't access the whole 4GB.

    Are there any known issues or limitations regarding Linux accessing the whole 2GB memory space?

    Regards,

    Nicholas Lee

    PS: I have just noticed the new Keystone am5k2e04 processor has the 64-bit DDR3 EMIF we are really looking for. This sounds exciting, even if it is just vapour-ware at the moment. If I knew (roughly) what they were going to cost and when European distributors are likely to get any stock, then I could decide whether to start designing for that chip instead.

  • 0 in reply to Nicholas C Lee
    TI__Guru**** 393215 points
    Hi Nicholas,
     
    Here is a link that will help you on the Linux Memory Map: http://processors.wiki.ti.com/index.php/EZSDK_Memory_Map