• State Resolved
  • Locked Locked
  • Replies 7 replies
  • Subscribers 31 subscribers
  • Views 567 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

AM2434: TSN Features & 1000 Mbps Ethernet

Tyler
Prodigy 150 points
Part Number: AM2434
Other Parts Discussed in Thread: AM2431, AM2634,

Hi,

I want to be able to sync the clocks of two devices using IEEE TSN or IEEE 1588 standard with a jitter of less than 100 nanoseconds. I see that the AM243x series of MCUs supports TSN but I am concerned about the speed of these microcontrollers at 1 Gbps speeds.

Will the AM2434 be fast enough to achieve a jitter of less than 100 ns at 1 Gbps Ethernet speeds? What about the AM2431 or 2432?

I also see that many of TI's Sitara MPUs support TSN and are much faster than their MCU counterparts. I was looking at the AM652x and AM641x MPUs. I was interested in using the MCU over the MPUs because of the reduced complexity (utilizing TI-Clang & CCS vs. Linux) and the reduced FCCSP (ALX) package size (293 pin) of the AM243x MCUs.

For my application requirements: Gigabit Ethernet (1000 Mbps) speeds and TSN synchronization with a jitter of less than 100 ns, what would you suggest I use? The MCU or MPU listed above? Or does anyone have a different suggestion? Please note smaller and lower cost are better. My only other requirement for the processor/controller would be a couple of SPI & UART ports. I do not need PCIe, etc. I will be utilizing an external 1000BaseT1 PHY to convert to the required MDI.

  • 0
    TI__Mastermind 35002 points

    Hi Tyler,

    Apologies for the delayed response, due some external events we could not look into it, we have started to look into the issue, allow us to get back by early next week.

  • 0 in reply to Nilabh Anand
    TI__Genius 12076 points

    Hi Tyler,

    On AM243x device, it is possible to have a jitter < 100ns. The current TSN PTP stack sync jitter settles around 40-50ns. 

    A few questions would help me understand and cater better

    1. What is your throughput requirement for the application?

    2. Do you require LwIP Stack or just layer-2 functionality will suffice?

    3. The 1 Gbps you mentioned above is for Rx and Tx throughput or switching traffic?

    Also, You can check out the AM263x product range if it fits your requirements. (https://www.ti.com/product/AM2634)

    Regards,
    Shaunak

  • 0 in reply to Shaunak Deshpande
    Prodigy 150 points

    1. Gigabit speed, 1000 Mbps

    2. I am not 100% sure. I want two of the units to be able to perform a time sync over TSN and communicate at 1000 Mbps. They will not be on a network of devices and will only be talking to each other. When connected, the PHYs on each end will link up (1 master, 1 slave).

    3. This would be for Rx & Tx throughput

    Do both the AM263x and AM243x have software support for Ethernet and TSN?

  • +1 in reply to Tyler
    TI__Genius 12076 points

    Hi Tyler,

    Thanks for sharing the details,

    Both AM243x and AM263x support Ethernet and TSN. On both the devices we support LwIP Stack as well in case you need other general networking capablitiles at layer-4.

    You can check out the following list of examples that are available out-of-box for ethernet and networking:

    AM243x: https://software-dl.ti.com/mcu-plus-sdk/esd/AM243X/latest/exports/docs/api_guide_am243x/EXAMPLES_NETWORKING.html

    AM263x: https://software-dl.ti.com/mcu-plus-sdk/esd/AM263X/latest/exports/docs/api_guide_am263x/EXAMPLES_NETWORKING.html

    Note: The out of box examples are not the most optimized from performance perspective. They are created with the intent of having decent memory footprint and performance. Based on your end usecase, ethernet frame sizes you can modify the out of box examples to perform accordingly, generally the out-of-box examples are modified to have high throughput for large frame sizes with less application memory footprint.

  • 0 in reply to Shaunak Deshpande
    Prodigy 150 points

    Hi Shaunak & Nilabh,

    This is very helpful, thank you both.

  • 0 in reply to Shaunak Deshpande
    Prodigy 150 points

    Hi Shaunak,

    I have another follow up question after reading about the example software you linked above, specifically for the AM263x. I have read elsewhere on the E2E forums that the Enet software examples can use up to 1.4 MByte of MCU RAM while performing TSN sync, etc. There was talk that this would be reduced to a few hundred KB at some point. Do you know if this was ever accomplished?

    I ask because I would like to use a custom design of the LP-AM263 but I see that there is no external SRAM interface available for the AM2634 so I would only have the 2 MBytes of on chip RAM available to use. Would this be enough memory for the Enet examples found in SDK version 09.01 to accomplish both TSN while maintaining gigabit TX/RX speeds? Again, this would be the only two tasks the MCU would be doing.

    If not, would it be advisable to use the AM2434 ALV footprint such as on this EVM board, which has access to a DDR4 memory chip? 

  • 0 in reply to Tyler
    TI__Genius 12076 points

    Hi Tyler,

    Let's first talk about the performance. Below data is on AM263x-LP from MCU_PLUS_SDK 09.01.00.41

    • Below attached is the memory footprint data in release mode for the TSN application with LwIP stack (CPSW):
      The OCRAM consumption is 927,776‬B (927KB)

       
    • Below attached is the memory footprint data in release mode for the TSN application without LwIP stack (CPSW)
      The OCRAM consumption is 696,328 (696KB)
    • If you choose the AM243x ALV, you have the DDR and the footprint will be different:
      gPTP example on CPSW With LWIP stack:


    Next, talking about the ethernet performance, I'm attaching below the ethernet performance data published by our networking team. The performance depends on a lot of factors such as your packet size, your line rate, requirement of LwIP (TCP/IP Networking stack in case you want other general networking capabilities at Layer4), if the use case is just at layer-2 etc. Similarly this can affect your memory footprint as well. You can try out experimenting the same based on your end-application needs.

    We do have some examples in our SDK (both AM243x, AM263x) you can try out.

    AM243x:

    - Layer 2 example: software-dl.ti.com/.../EXAMPLES_ENET_LAYER2_CPSW.html

    - GPTP CPSW example: software-dl.ti.com/.../EXAMPLES_ENET_CPSW_TSN_GPTP_TR.html

    - GPTP CPSW example with LwIP Stack: software-dl.ti.com/.../EXAMPLES_ENET_CPSW_TSN_LWIP_GPTP.html

    AM263x:

    - Layer 2 example: software-dl.ti.com/.../EXAMPLES_ENET_LAYER2_CPSW.html

    - GPTP CPSW example: software-dl.ti.com/.../EXAMPLES_ENET_CPSW_TSN_GPTP_TR.html

    - GPTP CPSW example with LwIP Stack: https://software-dl.ti.com/mcu-plus-sdk/esd/AM263X/latest/exports/docs/api_guide_am263x/EXAMPLES_ENET_CPSW_TSN_LWIP_GPTP.html

    Performance data:

    We do support a bandwidth of 1Gbps on Rx/Tx but again, the actual throughput depends on the application. You can refer the below:

    AM243x CPSW performance data: software-dl.ti.com/.../DATASHEET_AM243X_EVM.html

    AM26x CPSW performance data: software-dl.ti.com/.../enetlld_performance.html

    Note: The out of box examples do not showcase the best performance. There will be a performance and memory tradeoff. Based on your packet sizes and performance requirement, memory footprint can vary.

    Regards,

    Shaunak