Skip to main content
Fabrizio Pizzolato
Fabrizio Pizzolato
Associate II
August 29, 2017
Question

STM32 resetting on first sdram access

  • August 29, 2017
  • 13 replies
  • 2775 views
Posted on August 29, 2017 at 10:09

Good morning, I have a STM32F429 based board, the MCU is interfacing a Micron SDRAM (MT48LC4M32B2). We successfully configured the sdram interface using the FMC driver. We use 1.4.3 version.

We had a first production round where about half of the boards have the following problem: on the very first access to the sdram the MCU hard resets. I could verify that a 1ms negative pulse is present on the MCU reset line. After this reset the application restarts and access to the SDRAM is executed with no problems. 

Note: by saying first access I mean the first time the memory is used. I always configure it (by using low level functions from the driver) without problems. At a later time (actually through an external command) I create a pointer to the first memory location in sdram and try to use it, and this is when the problem occurs.

Anybody here had similar problems? Any ideas on how to identify the problem?

thank you very much

    This topic has been closed for replies.

    13 replies

    waclawek.jan
    waclawek.jan
    Super User
    August 29, 2017
    Posted on August 29, 2017 at 10:37

    Inadequate power supply/grounding/decoupling, and memory having an increased consumption when it first time accesses the memory array?

    JW

    Fabrizio Pizzolato
    Fabrizio PizzolatoAuthor
    Associate II
    August 29, 2017
    Posted on August 29, 2017 at 10:48

    I monitored the MCU supply and it is stable during the reset pulse. Consumption, anyway, is about 50 mA and does not increase when the problem shows.

    Tesla DeLorean
    Tesla DeLorean
    Guru
    August 29, 2017
    Posted on August 29, 2017 at 12:15

    No defined mechanism to generate a reset from such an access. Do you have a watch dog timer? Pull up on NRST, other circuitry?

    Not much to work with here.

    Anyone one from ST review circuit or PCB design?

    Tips, Buy me a coffee, or three.. PayPal Venmo (See Profile) Up vote any posts that you find helpful, it shows what's working..
    Fabrizio Pizzolato
    Fabrizio PizzolatoAuthor
    Associate II
    August 29, 2017
    Posted on August 29, 2017 at 14:57

    No watch dog timer active and I removed a couple of lines that coupled electrically NRST to other arts of the circuit.

    So it seems indeed an electrical problem, unless anybody can suggest something else to verify.

    waclawek.jan
    waclawek.jan
    Super User
    August 29, 2017
    Posted on August 29, 2017 at 15:17

    Let me reiterate the power supply problem idea: what exactly is the VDDA/VDDA voltage with regard to VSS/VSSA? What is its primary source? Is this same source used to power the SDRAM? Have you observed it with an oscilloscope, both GND and probe pin directly on the mcu pins, synced to the critical moment?  Is PVD and/or BOR enabled; in the latter case what is the BOR level? What are the VOS settings? What is the clock and FLASH latency setting? Does the problem occur if you set the clock to some other source, e.g. HSI? Do you have some sort of (hard- or any other-) fault handlers in place?

    JW

    Tesla DeLorean
    Tesla DeLorean
    Guru
    August 29, 2017
    Posted on August 29, 2017 at 15:28

    Voltage and caps at VCAP1/2 pins would be things to double check. But honestly not sure why SDRAM should trigger.

    Would have built a SDRAM memory tester for prototypes, and saturate test those, and 'working' boards from batch.

    Tips, Buy me a coffee, or three.. PayPal Venmo (See Profile) Up vote any posts that you find helpful, it shows what's working..
    AVI-crak
    AVI-crak
    Senior
    August 29, 2017
    Posted on August 29, 2017 at 15:38

    Initialize the contact connections to the memory in two stages:

    1) address lines + control, reset sdram memory.

    2) data lines, ending settings and clearing memory.

    The MT48LCxxx may be in sleep mode immediately after the reset, this is an undefined state. But a quick reset can cause a current consumption pulse. The documentation specifies the time and sequence of initialization of the MT48LCxxx, you can not delay and you must not rush. If you put more ceramics - then you can act on the break.

    By the way, the sleep mode is a cool thing.

    Fabrizio Pizzolato
    Fabrizio PizzolatoAuthor
    Associate II
    August 29, 2017
    Posted on August 29, 2017 at 16:58

    I had some doubts on this one. I actually tried to follow step by step what's on the 'Initialization' chapter, with two problems.

    1. I have a 1ms tick on the system, so some of the instructions can't be accurately replied.

    2. also using FMC driver some commands are not available at the driver level: in stm32f4xx_ll_fmc.h see 'FMC_SDRAM_CMD...' defines.

    So for example when the datasheets states, for example 'Starting at some point during this 100μs period, bring CKE HIGH. Continuing at least through the end of this period, 1 or more COMMAND INHIBIT or NOP commands must be applied' I can't be sure on what to do.

    AVI-crak
    AVI-crak
    Senior
    August 29, 2017
    Posted on August 29, 2017 at 17:39

    For the sake of experiment, I collected all references to the initialization of sdram using hal. The minimum file size is more than a thousand lines - all related mentions of initialization in one place. The project was going and working successfully, but it is impossible to understand how it is going under the GCC.

    In such cases, it is better to use direct access to registers.

    http://en.radzio.dxp.pl/stm32f429idiscovery/sdram.html

     
    Terms & ConditionsAccessibility statement