stm32U585 + 2 x APS25608N OSPI - speed problems

[ccuebler]

Hai All!

I use 2 pieces of APS25608N PSRAM on OSPI bus with memory mapped mode. The MCU run at 1.8V, 160MHz system clock. Previously, I used APS12808N at 80MHz and 3.3V VDD without any issues.

Currently, the PSRAMs run at 50MHz, but need to speed up to 100 or 160MHz.
I thought the aps256's default settings should be good up to 100MHz, but they weren't.

My working settings at 50MHz was ( relevant source in attachment): 
MR0: Fix type latency, latency code 5, and full strength (25 ohms).
MR8: 32Byte Warp burst and without crossing.

OSPI settings are:
Dummy cycles:  Read = 8, Write = 4

  hospi->Init.FifoThreshold           = 4;
  hospi->Init.DualQuad                = HAL_OSPI_DUALQUAD_DISABLE;
  hospi->Init.MemoryType              = HAL_OSPI_MEMTYPE_APMEMORY;
  hospi->Init.DeviceSize              = 25;
  hospi->Init.ChipSelectHighTime      = 2;
  hospi->Init.FreeRunningClock        = HAL_OSPI_FREERUNCLK_DISABLE;
  hospi->Init.ClockMode               = HAL_OSPI_CLOCK_MODE_0;
  hospi->Init.WrapSize                = HAL_OSPI_WRAP_NOT_SUPPORTED;
  hospi->Init.ClockPrescaler          = 2;
  hospi->Init.SampleShifting          = HAL_OSPI_SAMPLE_SHIFTING_NONE; // None
  hospi->Init.DelayHoldQuarterCycle   = HAL_OSPI_DHQC_ENABLE;
  hospi->Init.ChipSelectBoundary      = 10;
  hospi->Init.DelayBlockBypass        = HAL_OSPI_DELAY_BLOCK_USED;
  hospi->Init.MaxTran                 = 0;
  hospi->Init.Refresh                 = 640;

So, I try to speed up to 80MHz:
The MR0 and MR8 register settings are same, and it written MRs to APS256 successfully (I read back and compared ).
For the best delay block settings, I ran the delay block fine tuner, but doesn't found any good option. 

I also tried to modify MR0 values ( increase / decrease the latency code, changed the drive strength, etc.), without any success.
I played with the refresh time used calculated values. I wrote a simple delay block tuner like function to search a good refresh time.
I measured the clock, chip select and data0 lines against with oscilloscope, and I saw only a little overshoot, but the timings was good. (fast signal rising and falling, and data lines stabilized on time)
I think, it's only a configuration or settings-related issue. 
At the end, I confused, what is the relationship between latency code and dummy cycle.

What is the best settings: How can I speed up to 100 or 160MHz the PSRAMs? 
Please check my attached code.

Thank you.

[KDJEM.1]

Hello @ccuebler ;

Could you please share clock and data waveforms for different frequencies: 50 MHz and 80 MHz?

Which voltage range are you using?

Please refer to STM32U585 datasheet and check the maximum OCTOSPI clock?

Thank you.

Kaouthar

[Sco]

I suggest you refer to the datasheet, and find the "power configuration" of your board to see the maximum frequency the STM32U585 can run.

if you are developping a new board, the next point maybe to check there's no hardware constrain for the layout of the board, Impedance matching, equal length...

[Alex - APMemory · Accepted solution]

Hi, 

Just one comment related to : "I use 2 pieces of APS25608N PSRAM on OSPI bus with memory mapped mode. The MCU run at 1.8V, 160MHz system clock. Previously, I used APS12808N at 80MHz and 3.3V VDD without any issues."

Both APS25608N and APS51208N are only 1.8V, up to 200Mhz (max frequency also related to MCU max memory frequency). If you have the choice you better using APS51208N (2pces of APS25608N already inside one package with one CE) rather than 2 pces of APS25608N on your PCB. You can also use APS512XXN-OB9-BG/APS256XXN-OB9-BG in OPI mode.

Thanks

Alex

[ccuebler]

Hi @Alex - APMemory !

Thank you! 
We try to find the problem, and our solution was the same with Your recommendation.
We using the APS51208N, and it working fine. 
I think the problem was with the tracing, because we can't reach higher speed than 50MHz.

Thanks
Krisztian.

[Alex - APMemory]

Good to hear, I guess 512Mb single package have only advantage (design, comptivness...) vs 2pcs.

Alex