i.MX RT1050 Hyperflash Vs QSPI Vs SDRAM

[martinsmith]

Hi, I'm starting a hardware design using the i.MX RT1050 and am scratching my head about what memory to use. I think the program I end up with will be too large to load into the internal RAM memory and run from there, so I believe the options I'm left with are:

1. XIP from QSPI - Easy and cheap, but the QSPI will have approx. 100 MB/s bandwidth? which will severely limit the performance of the i.MX RT1050
2. XIP from Hyperflash - More work to put onto a PCB (12 lines), but faster 300 - 400 MB/s bandwidth. Will still throttle back the performance of the processor but not by too much? Big problem with this is that NONE of the Hyperflash ICs seem to be available in any of the usual places (Mouser, Digikey, etc) and have massive lead-times (42 weeks, seriously!?!)
3. Add external SDRAM and QSPI to the PCB and load the program on boot up into the external SDRAM and run from there - Most work to put on a PCB and again will throttle back the processor performance (~400 MB's bandwidth?), but has been around for a while and cheaper and more available than Hyperflash

Has anyone got any opinions on this? Which way should I go and are there any other memory configurations I have missed?

Thanks

Martin 

[willischang]

Hi 

based on NXP 's application note there is a HyperFlash support list on it. pls see table 7. for HyperFlash Support List

Document Number: AN12107
Application Note
Rev. 0 , 12/2017

**************************************************

Besides the EVK onboard HyperFlash, the following Flashes are also support:
Table 7. HyperFlash supports list
Vendor
Flash
ISSI (Hyper Flash) IS26KS256
SPANSION (Hyper Flash) KS512SBPHI02
Macronix  MX25UM513
Micron MT35X
Adesto ATXP032

[martinsmith]

Hi,

Thanks for the reply. I did find that app note and having checked Mouser, Digikey and Farnell NONE of the flash memories on that list are available in ANY of the catalogues. So I would need to go to the manufacturers directly. I only want <100 a year so I don't imagine I would be high on their list of customers?

I really wanted something that I can source from 2 or 3 different places to guarantee that I can get hold of them in the future.

[gordonmacnee]

Hi Martin,

These are new devices and are available as Engineering Samples but will be available from all of our distributors shortly. This will include, Arrow, EBV, Future Electronics and Ineltek through Europe. 

[Yuri]

Hello,

  When considering system performance issues, that are usually very application

dependent, use app notes AN120042 (Using the i.MXRT L1 Cache) and AN120077
(Using the i.MX RT FlexRAM).

  In particular, refer to section 6 (Conclusion) of AN120042.

Have a great day,

Yuri

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

Note: If this post answers your question, please click the Correct Answer

button. Thank you!

[ScottKerstein]

Hello Yuri,

AN120077 is a good discussion and as the app note states, "i.MXRT series takes advantage of the ARM CortexM7 core with 32K/32K L1 I/D-Cache. This delivers extremely high performance regardless the code is executed from on-chip RAM, external Flash or external memory."

(1) Is there an empirical data showing performance of XIP from QSPI Flash vs. XIP from HyperFlash?

(2) Is XIP from QSPI Flash predicable and repeatable?

(3) Can there be any hic-ups from this method causing performance issues vs. using traditional onboard TCM?

(4) Does NXP have a tool that can emulate performance?

Thanks in advance,

Scott

[Yuri]

Hello,

   I am afraid we do not have public performance estimations, at least right now: as usually it is very application

dependent. 

Regards,

Yuri.

[gordonmacnee]

Just to add to the comment above (and yes - I do work for Adesto...). Have a look through Youtube posts from Embedded Artists regarding their i.MX1050. Anders talks of performance with Adesto's ATXP032 EcoXiP and is fitting it as standard to their board. This gives them a high performance from a low power budget. We have since improved the timing and have better benchmark results but the videos remain a good guide.

iMXRT 1052 OEM Board - Introduction - YouTube 

iMX RT1052 OEM Board - OctalSPI Flash - YouTube  

[martinsmith]

Thanks Gordon for you reply. Looking at the youtube video you mentioned, Anders (at 3:39) mentions the performance of the EcoXiP vs running from internal memory. The performance using EcoXiP is only 33% of running from internal memory. That seems a huge amount of performance to give up? 

I know you will say that the performance is still very good (and it is), but wouldn't it be better to use a smaller cheaper processor with 33% of the performance of the RT1050, and run from internal flash on that processor?

[gideonintrater]

Hi Martin,  please refer to http://www.adestotech.com/wp-content/uploads/2-pp-Overview_EcoXiP_11_17-1.pdf . The performance of the RT1050 with the Adesto EcoXiP is 1912 coremark at 5% miss rate.  That's roughly 66% of the max performance of this CPU when running from on-chip memory (3000 Coremark).  The 1912 is equivalent to what you will get out of a 400MHz Coretex M7-based CPU if you can find one.  Note that any other external memory would reduce the performance of the CPU relative to what you could get when running from on-chip.

[willischang]

Hi Giddy,

Yes, I have logged on last night and posted something Flash support list information about Eco-XiP they call HyperFlash.

Thanks

Willis

[edmckernan]

Did this Martin smith contact you

Sent from my iPhone

[willischang]

Hi Ed.

Pls see the attached e-mail which Martin smith reply me last night.

Thanks

Willis

[edmckernan]

Martin

Adesto Technologies has an alternative to hyper flash that is called EcoXiP.  Currently it is sampling at 32Mb and we will have larger densities later this year.  The chip has an Octal interface operating up to 266MB/s.  Information and a data sheet is her: http://www.adestotech.com/products/ecoxip/ .  We have chips working on the NXP Eval board today.  If you provide an email to me with your location we can provide more technical assistance.  ed.mckernan@adestotech.com

[martinsmith]

Hi Ed,

Thanks for the reply. 

I'm not sure this is the solution? It is another form of Hyperflash and will probably have the same problems - single source, long lead times, etc. I'm guessing at 266MB/s it will slow the processor down also? (assuming 1 byte of code can be executed each clock cycle???).

What are the advantages of this over SDRAM and a cheap QSPI IC to hold the program?

Regards

Martin