How could we improve flashing time?

[ppalant]

Hello,

We are using NRF5340 in our product and running automated hardware testing. As part of this process, we need to program the chip twice: once with a dedicated testing firmware, and a second time with the actual production firmware.

One issue for us is that the whole flashing process takes about 55 seconds, which means almostj 2 minutes per device tested is spent on flashing.
We would like to reduce flashing time if possible and would appreciate any pointer on how to do so. 

Here are some more details:

• Tools used:
  • nrfjprog 10.24.2
  • Segger JLink V9
• Commands used to flash:
  1. nrfjprog -f NRF53 --coprocessor CP_NETWORK --recover
  2. nrfjprog -f NRF53 --coprocessor CP_APPLICATION --recover 
  3. nrfjprog -f NRF53 --config ./tools/helper_scripts/flash_scripts/config.toml --erasepage 0x10000000-0x103FFFFF
  4. nrfjprog -f NRF53 --coprocessor CP_NETWORK --program build/ipc_radio/zephyr/zephyr.hex --config ./tools/helper_scripts/flash_scripts/config.toml --chiperase --verify
  5. nrfjprog -f NRF53 --coprocessor CP_APPLICATION --program build/merged.hex --config ./tools/helper_scripts/flash_scripts/config.toml --sectorerase --verify
• Execution times for those commands:
  1. Recover CP_NETWORK: 1.4 s
  2. Recover CP_APPLICATION: 1.4 s
  3. Erase page: 16 s
  4. Program CP_NETWORK: 2.1 s
  5. Program CP_APPLICATION: 31.9 s
• Firmware file sizes:
  • 2.2 MB for APP coprocessor
  • 428 KB for NET coprocessor
• Config toml file:

  • [log]
        # level - A spdlog log level enumerator name. Case insensitive.
        # pattern - A spdlog log formatting pattern.
        # file - Target log file, will be written to in append mode. If file is a directory, or it's missing a file extension, "/log.log" will be appended to the log file path.
        [log.dll]
            # nrfjprogdll logger configuration
            level = "TRACE"
            pattern = "%v"
        [log.file]
            # nrfjprog executable file log sink configuration
            file = "log.log"
            level = "TRACE"
            pattern = "[%Y-%b-%d %H:%M:%S] [%5l] %v"
        [log.terminal]
            # nrfjprog executable terminal log sink configuration
            level = "ERROR"
            pattern = "[%5l] %v"
        
    [rtt]
        # If true, NRFJPROG_rtt_stop invalidates the current RTT control block. 
        # To restart RTT communication the target device must reinitialize the RTT control block before NRFJPROG_rtt_start is called.
        invalidate_control_block_on_stop = true
    
        # Set the period between rtt API calls in the async rtt worker threads. 
        # Shorter periods may improve performance, but can overwhelm the J-Link debug probe.
        async_read_period_ms = 10
        async_write_period_ms = 10
    
    [jlink]
        # If true, J-Link debug probe fw will be silently updated on NRFJPROG_connect_to_emu_with*().
        auto_update_fw=false
    
    [target]
        # Define the desired family of devices to use by default by nrfjprog.exe when -f or --family arguments are not used. 
        # Family identifiers are case insensitive, no suffixes, for example NRF51, nRF52, Nrf53 are all valid. 
        # If set to AUTO an automatic family detection will be performed. The autodetect family operation will take time, so it is advised to use correct family if known.
        family = "AUTO"
    
        # Define the desired clockspeed in kHz you want nrfjprog.exe use by default when -c or --clockspeed arguments are not used.
        clockspeed = 11999
    
    [approtect]
        # Configurations for devices with hardened ap-protection.
    
        # Should nrfjprog write UICR APPROTECT registers after erasing?
        # Certain devices perform this write in hardware, in this case this configuration does nothing.
        write_uicr_approtect = true
        # Should nrfjprog verify UICR APPROTECT registers?
        # On certain devices UICR->APPROTECT is written automatically to some reset value != 0xFFFFFFFFF
        # on erase. On these devices it's not possible to directly compare this NVM register to image file contents.
        verify_uicr_approtect = false
        # Should nrfjprog cli write an unlock image to NVM memory after --recover?
        # Certain devices require this image to stay unlocked.
        install_unlock_image = true
    
    # nrfjprog QSPI configuration.
    [qspi]
        # Define the capacity of the non-volatile memory device in bytes. Set to 0 if no external memory device is present in your board.
        mem_size = 0x400000
    
        # Define the desired ReadMode. Valid options are FASTREAD, READ2O, READ2IO, READ4O and READ4IO
        read_mode = "READ4O"
    
        # Define the desired WriteMode. Valid options are PP, PP2O, PP4O and PP4IO
        write_mode = "PP4O"
    
        # Define the desired AddressMode. Valid options are BIT24 and BIT32
        address_mode = "BIT24"
    
        # Define the desired Frequency. Valid options are M2, M4, M8, M16 and M32
        frequency = "M2"
    
        # Define the desired SPI mode. Valid options are MODE0 and MODE3
        spi_mode = "MODE0"
    
    
        # Define SPI interface timing. Valid options are in the range of 0 to 7.
        # This argument is only used for devices where the dll function NRFJPROG_qspi_set_rx_delay() is supported.
        rx_delay = 2
    
        # Define the Index of the Write In Progress (WIP) bit in the status register. Valid options are in the range of 0 to 7.
        wip_index = 0
    
        # Define page size for commands. Valid sizes are PAGE256 and PAGE512.
        page_program_size = "PAGE256"
        
        # If retention is enabled, device RAM contents will be read and buffered during QSPI driver initialization.
        # The buffered data will be written back to the device when unitializing the driver, restoring the original device RAM state.
        # Enabled: retain_ram = true, Disabled: retain_ram = false
        retain_ram = true
    
        # Define the assigned pins for the QSPI peripheral. Valid options are those existing in your device
        # For nRF53, QSPI pins are not configurable and these values are ignored.    
        [qspi.sck]
            # Define the desired SckDelay. Valid options are in the range 0 to 255
            delay = 0x80
            pin = 21
            port = 0
    
        [qspi.csn]
            pin = 18
            port = 0
        [qspi.dio0]
            pin = 23
            port = 0
        [qspi.dio1]
            pin = 22
            port = 0
        [qspi.dio2]
            pin = 10
            port = 0
        [qspi.dio3]
            pin = 9
            port = 0
    
        [qspi.custom]
            # Define the desired IO level for DIO2 and DIO3 during a custom instruction. Valid options are LEVEL_HIGH and LEVEL_LOW
            io2_level = "LEVEL_LOW"
            io3_level = "LEVEL_HIGH"
    
            # Custom instructions to send to the external memory after initialization. Format is instruction code plus data to send in between optional brakets. 
            # These instructions will be executed each time the qspi peripheral is initiated by nrfjprog. 
            # To improve execution speed on consecutive interations with QSPI, you can run nrfjprog once with custom initialization, and then comment out the lines below.
            # Numbers can be given in decimal, hex (starting with either 0x or 0X) and binary (starting with either 0b or 0B) formats. 
            # The custom instructions will be executed in the order found. 
            # This example includes two commands, first a WREN (WRite ENable) and then a WRSR (WRite Satus Register) enabling the Quad Operation
            # mode for the MX25L6433F flash memory.
            instructions =  [
                {command=0x06, data=[]},
                {command=0x01, data=[0x40, 0, 0x2]},
                {command=0x06, data=[]},
                {command=0xB7, data=[]}
            ]
    
    [system]
        # If true, the underlying jlinkarm_nrf_worker process will ignore SIGINT, SIGTERM, SIGABRT, SIGILL and SIGSEGV signals.
        # It is strongly recommended to keep this as false for nrfjprog exe.
        # This feature is mainly inteded for use with the pynrfjprog API, where you sometimes want to manually handle a ctrl-c event without the signal
        # propagating to the jlinkarm_nrf_worker process.
        worker_ignore_signals = false
    

[Einar Thorsrud]

Hi,

I see that the two steps that take the most time is step 3 where you erase the external flash, and step 5, where I assume most of the time is spent programming the external flash. I see on line 71 in the config file that you have set the QSPI speed to 2M.  Can you try increasing this?

[ppalant]

Hi, thank you for your input. 
I set QSPI speed to 32M, but it makes no difference to total flashing time. 

However, we noticed that we can gain 20 seconds on CP_APPLICATION programming if we use --chiperase instead of --sectoreerase. 

[Einar Thorsrud]

Hi,

Yes, I did not spot that there was a secotor erase there. On the nRF, chiperase and recover  is much faster than sectore erase.

Do you actually write to the QSPI flash when you program the application or not? If you do not, the QSPI frequency will not matter.