1G/10G PHY single reference clock

[jsenesi]

I have been asked to implement the 1G/10Gbe PHY design on Arria 10 based hardware that only provides a 644.53125 MHz reference oscillator. Our typical implementations include the 125 MHz reference for 1Gbe.

I have been able to build the design with two cascaded fPLLs producing the 125 and 625MHz references required for the 1GbE operational mode.

10GbE works just fine. 1Gbe mode does not work. When I try to switch to 1GbE the calibration sequence never completes; signal 'rc_busy' remains active..

I am monitoring signals from the reset controller. On a 1G reconfig that always fail the status ends up: rc_busy = ‘1’ , rx_is_lockedtodata_out = ‘0’, pll_locked = ‘1’, tx_analogreset = ‘0’, rx_analogreset= ‘0’, tx_digitalreset=’0’, rx_digitalreset=’1’, tx_cal_busy=’0’, rx_cal_busy=’0’

I'm looking for debug suggestions and/or implementation ideas (that don't use a 125MHz oscillator).

Thanks.

[Deshi_Intel]

HI,

Pls see my comment below

1. Regarding 1G/10G PHY IP clocking clarification
   • A10 PHY user guide doc (page 168, figure 75) show you clocking explanation where it uses  fPLL/CMU PLL for 1G PMA, ATX PLL/CMU PLL for 10G PMA and another fPLL for PCS clocking 
   • https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/hb/arria-10/ug_arria10_xcvr_phy.pdf
   • Sorry but it's not clear to me how you implement the clocking design at your side. Can you explain exactly how does it differ from figure 75
2. Regarding 1G operation stuck in calibration process issue
   • I am not sure whether your 1G/10G PHY IP is connected and configured correctly or not.
   • Why not you generate a working 1G/10G PHY IP design example from "A10 low latency Ethernet 10G MAC IP" and then cross check with your own design ?
   • https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/ug/ug-20016.pdf     (refer to chapter 3)

Thanks.

Regards,

dlim

[jsenesi]

Hi,

I've attached an illustration of my clocking. The notable difference is the lack of a 125MHz reference clock and my using the 644.53125 with cascaded fPLLs to generate 1GbE references.

I will look at the reference design and see if there are any other differences I should be concerned about.

Thanks,

Joe

[Deshi_Intel]

Hi Joe,

Thanks for sharing your design clocking diagram.

Yup, the difference is mainly on 1G clocking side. 

• Cascading fPLL is not recommended due to higher jitter affecting the clock network performance.
• Also, I am not sure how you connect 1G CDR  (rx_cdr_refclk_1g) in your design. Based on reference design, this should be using same 125MHz clock source to clock both 1G Tx_PLL and 1G Rx_CDR

Thanks.

Regards,

dlim  

[jsenesi · Accepted solution]

Hi dlim,

I have found a solution working in the lab. For some reason generating the 125MHz from 644MHz with an IOPll seems to work, while using an fPLL has not.

I have attached a new clock diagram illustrating.

Regards,

Joe