1064 with dual-die package (128Mbit) HyperRAM

[p_shep]

It's my favourite subject... again!

So we're having trouble sourcing enough of the Cypress S27KS0642GABHI02 for our project and have a couple of Winbond parts to try:

W956D8MBYA5I (on the 'tested' list in AN12239)

W957D8MFYA5I (16Mbyte dual-die package)

Now the 'tested' winbond one is failing a memtest after a few mins when using the same settings as the original Cypress. I guess a timing issue somewhere? Funnily enough it fails immediately at lower clocks rates.

On the DDP Winbond, the memtest fails at the 8Mb boundary, but it appears that the memory after can be read - at least with the memory viewer. Is there a way to get the FlexSPI driver to cope with the dual-die arrangement? If the size is set to just 8Mbytes, the memtest seems to be working fine with the Cypress settings.

[p_shep]

Just found this:

https://community.nxp.com/t5/i-MX-RT/I-MX-RT1176-use-128Mb-Hyper-Ram-has-Problem-when-enable-prefetc...

So MPU has the area set as device, and I've created 2 linker sections to cover the area, but still not working

[p_shep · Accepted solution]

So setting:

config.ahbConfig.enableAHBPrefetch = false;

Will get it working. But that's surely going to be a BIG hit on performance?

[p_shep]

Yep, turning prefetch off has a dramatic affect on performance..

At 327MHz, my memtest goes from ~1140ms to 1884ms to complete.

So, ways to abort the prefetch at the 8Mb boundary?