DDR3 calibration issue on custom iMX6Solo board

取消
显示结果 
显示  仅  | 搜索替代 
您的意思是: 

DDR3 calibration issue on custom iMX6Solo board

8,651 次查看
johnson_dana
Contributor II

# markmiddleton

Hi Mark, The failure we're seeing mimics thread https://community.nxp.com/thread/506442  .
However all the required voltages are present in our case. We're using both 24 MHz and 32.768 kHz external crystals. The clocks look clean at around 1.1 Vpp. We're using two 4 Gbit DDR3 chips from Alliance operating on 1.35V. Along with the NXP iMX6 Solo processor we are using the NXP PMIC MMPF0200F6AEP. The DRAM_RESET pin is high (1.35 V). SDCKE0 and SDCKE1 are both low. DDR_VREF is 0.68 V (comes from PMIC).

Here's the test tool window:

NXP DDR TEST WINDOW.JPG

Below is the text output of the tool after downloading the .inc script and then running the calibration:

============================================
        DDR Stress Test (3.0.0)
        Build: Dec 14 2018, 14:12:28
        NXP Semiconductors.
============================================

============================================
        Chip ID
CHIP ID = i.MX6 Solo/DualLite (0x61)
Internal Revision = TO1.4
============================================

============================================
        Boot Configuration
SRC_SBMR1(0x020d8004) = 0x0100d072
SRC_SBMR2(0x020d801c) = 0x01000001
============================================

ARM Clock set to 800MHz

============================================
        DDR configuration
BOOT_CFG3[5-4]: 0x00, Single DDR channel.
DDR type is DDR3
Data width: 32, bank num: 8
Row size: 15, col size: 10
Chip select CSD0 is used
Density per chip select: 1024MB
============================================

Current Temperature: 49
============================================

DDR Freq: 297 MHz

ddr_mr1=0x00000004
Start write leveling calibration...
running Write level HW calibration
  MPWLHWERR register read out for factory diagnostics:
  MPWLHWERR PHY0 = 0x00001e1e


HW WL cal status: no suitable delay value found for byte 2

HW WL cal status: no suitable delay value found for byte 3
Write leveling calibration completed but failed, the following results were found:
    MMDC_MPWLDECTRL0 ch0 (0x021b080c) = 0x000D0013
    MMDC_MPWLDECTRL1 ch0 (0x021b0810) = 0x001F001F
Write DQS delay result:
   Write DQS0 delay: 19/256 CK
   Write DQS1 delay: 13/256 CK
   Write DQS2 delay: 31/256 CK
   Write DQS3 delay: 31/256 CK


Error: failed during write leveling calibration

Since we can't get by the "simple" leveling routine, there must be something obvious that we are missing.

标签 (2)
标记 (1)
9 回复数

8,028 次查看
TheAdmiral
NXP Employee
NXP Employee

Hi Dana,

First, please do not run the DDR calibration routine much below 333 MHz.

I would recommend a setting of 350 MHz to start out with.

Please let the DDR Density setting remain at default.

Finally, set the Write DQS and Read DQS values at 0x40 for all four channels and write leveling at 0x1F and run the Stress Test a couple of times without calibration. Please report the results. Maybe there is some clue in the failing results that might help us. It looks like lanes #0 and #1 are working fine, but the other two are having problems.

Cheers,

Mark

8,028 次查看
johnson_dana
Contributor II

Hello Mark aka The Admiral,

Thanks for the advice. After discussion with other engineers there is some confusion. First the channels you are referring to are the byte lanes, correct? Do I need to write 0x1F to each byte 0x1F1F1F1F or just 0x001F001F for the write leveling? An example of the .inc file has the following:

=============

// Calibration setup.

//=============================================================================

setmem /32 0x021b0800 = 0xA1390003 // DDR_PHY_P0_MPZQHWCTRL, enable both one-time & periodic HW ZQ calibration.

// For target board, may need to run write leveling calibration to fine tune these settings.

setmem /32 0x021b080c = 0x00000000 //my comment 0x1F1F1F1F or 0x001F001F?

setmem /32 0x021b0810 = 0x00000000 //my comment 0x1F1F1F1F or 0x001F001F?

//setmem /32 0x021b480c = 0x00000000 //my comment 0x1F1F1F1F or 0x001F001F?

//setmem /32 0x021b4810 = 0x00000000 //my comment 0x1F1F1F1F or 0x001F001F ?

////Read DQS Gating calibration

setmem /32 0x021b083c = 0x00000000 // MPDGCTRL0 PHY0 //my comment what do we write here?

setmem /32 0x021b0840 = 0x00000000 // MPDGCTRL1 PHY0 //my comment ditto?

//setmem /32 0x021b483c = 0x00000000 // MPDGCTRL0 PHY1 //my comment ditto?

//setmem /32 0x021b4840 = 0x00000000 // MPDGCTRL1 PHY1 //my comment ditto?

//Read calibration

setmem /32 0x021b0848 = 0x40404040 // MPRDDLCTL PHY0 // my comment each byte lane gets x40? This is the from the original .inc file

//setmem /32 0x021b4848 = 0x40404040 // MPRDDLCTL PHY1

//Write calibration

setmem /32 0x021b0850 = 0x40404040 // MPWRDLCTL PHY0

//setmem /32 0x021b4850 = 0x40404040 // MPWRDLCTL PHY1

Help!

Dana C. Johnson, VP PI&D

ORBCOMM, Inc. Suite 300

22970 Indian Creek Drive

Dulles, VA 20166

Office: 703-433-6456 Cell#1: 201-220-4613 Cell#2: 703-728-5939

www.orbcomm.com<http://www.orbcomm.com>; johnson.dana@orbcomm.com<mailto:johnson.dana@orbcomm.com>

0 项奖励
回复

8,028 次查看
TheAdmiral
NXP Employee
NXP Employee

Hi Dana,

Sorry for the confusion. Those are correct, except the Write Level calibration line should look like (0x001F001F)

Just would like to see the results with default calibration values.

Cheers,

Mark

0 项奖励
回复

8,028 次查看
johnson_dana
Contributor II

Hi Mark,

I've attached the .inc file we used along with the stress test output. Definitely not encouraging!

Best Regards, Dana

0 项奖励
回复

8,028 次查看
johnson_dana
Contributor II

Hi Mark,

 We've been trying different things with the tool and we ran across this: change the bus width down from 32 bits to 16 bits. Include your previous recommendations. Select iMX6DL in the DDR test tool. Download our new .inc file (attached) that reflect the 16 bit bus width. Let the tool pick the ARM clock speed as well as the memory density. See below:

pastedImage_1.png

Run the calibration...it succeeded at 400 MHz every time but some times it fails at 350 MHz...strange! See attached log file. Ran the stress test at 400 MHz. Passed!!! See second log file.

Our memory is setup as two 4 Gbit chips organized as 256M x 16 giving us a total of 8 Gbits or 1 GByte. The first chip uses SDQS0 and SDQS1 for data strobes. The second chip uses SDQS2 and SDQS3 for the data strobes. DQM0 and DQM1 are used for the first chip and DQM2, DQM3 are used for the second chip. We noticed that a SOLOX design that uses the same memory organization as our board uses SDQS0 and SDQS2 for the first chip and SDQS1 and SDQS3 for the second chip. We didn't think that would matter but it is the only difference we have noticed.

0 项奖励
回复

8,028 次查看
johnson_dana
Contributor II

Here's some additional information on our design:

1. We used one chip select CS0 for both DDR3 memory chips.

2. The low word chip uses CLK0 along with data strobes DQS0 and DQS1. It uses SDCKE0.

3. The upper word chip uses CLK1 along with data strobes DQS2 and DQS3. It uses SDCKE1.

0 项奖励
回复

8,028 次查看
Yuri
NXP Employee
NXP Employee

Hello,

 

  Please check Your PCB design,  using  section 1.1 (Schematic and Layout Design Rules)

of the DDR3 Porting Guide. In  particular, there is an Excel page named “MX6 DRAM Bus Length Check”

in “HW Design Checking List  for i.Mx6”. Designer can use it for layout self-checking.  

Freescale i.MX6 DRAM Port Application Guide-DDR3 

 

https://community.nxp.com/docs/DOC-93819 

Have a great day,

Yuri.

 

-------------------------------------------------------------------------------

Note:

- If this post answers your question, please click the "Mark Correct" button. Thank you!

- We are following threads for 7 weeks after the last post, later replies are ignored

Please open a new thread and refer to the closed one, if you have a related question at a later point in time.

0 项奖励
回复

8,035 次查看
johnson_dana
Contributor II

“I was surprised to see that the HW Guide design rules are NOT consistent with the HW Design Checklist Excel spreadsheet. If you put our trace lengths into the spreadsheet then there are many “failures” (cell turns RED). The checklist says <= 25 mil is the tolerance while the design guide says +/- 25 mils. I’ve attached the spreadsheet for the ORBCOMM board.

 

Here’s what our PCB designer said:

 

“I designed the trace lengths to follow the “Routing by the same length” rules from the design guide. This was the simpler method of building the rules and the fastest way to route.

If you check the values in the table, I have routed the main clocks to ~1.62”, the DSQ clocks to 1.617”, and all of the remaining signals between 1.595” and 1.645” (+/-25mils of the main clocks). This should meet the rules from that simplified rule set.

 

 

What guidelines should we be following?

0 项奖励
回复

8,037 次查看
johnson_dana
Contributor II

DDR ROUTING RULES.jpg

0 项奖励
回复