Inquiry Regarding Jetson Thor (T5000) Boot Failure

Dear Support Team,

We have developed a custom carrier board for the Jetson Thor (T5000) and are currently performing bring-up activities. However, we are experiencing a boot failure and have been investigating the power-on sequence.

The waveform below shows the issue we are observing.

When power is connected to the system, SV (3.3V) is applied immediately.

Our design follows the EVK schematic to ensure that VDDIN_PWR_BAD_N remains High after MV (5V) and HV (15V) have stabilized.

However, after VDDIN_PWR_BAD_N goes High, when we assert MODULE_POWER_ON after a delay of 100 ms, the power rails remain stable, but VDDIN_PWR_BAD_N toggles Low, as shown in the waveform.

If MODULE_POWER_ON is not asserted, VDDIN_PWR_BAD_N remains High.

For regular booting, pins J62 and G5 are connected to GND.

Could you please investigate this behavior and advise us on the possible cause and recommended solution?

Thank you for your support.

Hi,

If your circuit follows our reference carrier board design, is VIN_PWR_BAD_N connected to the VIN loss detection circuit?

If so, the signal going Low may be caused by a drop in the carrier board’s input voltage after MODULE_POWER_ON is asserted. Could you please measure the input voltage and check whether it drops at the same time as VIN_PWR_BAD_N goes Low?

Thank you,

Hello,

As shown in the waveform, the MV and HV voltages remain stable when VDDIN_PWR_BAD_N goes Low.

This signal is driven from the Carrier Board to the T5000 module.

Could you please confirm whether there are any other possible causes that could cause the T5000 module to pull this pin Low?

Thank you for your support.

Hi,

On the reference carrier board, VDDIN_PWR_BAD_N is connected to U69, which monitors the carrier board’s input voltage.

Does your carrier board use the same circuit configuration?
Could you please explain how VDDIN_PWR_BAD_N is connected and used on your carrier board?

Thank you,

Hello,

The circuit is configured as shown in the schematic below.

The MCU monitors the Power Good signal from the MV regulator and enables the HV regulator accordingly.

Since the HV voltage is only available after the MV voltage has been established, the power supplied to the THOR module is monitored by checking the HV regulator output.

The VDDIN_PWR_BAD_N signal is designed to be monitored by both the T5000 module and the MCU.

When VDDIN_PWR_BAD_N is High, the MCU drives MODULE_PWR_ON High to proceed with the power-on sequence.

Please review the above configuration and let us know your findings.

Thank you for your support.

Best regards,

Hi,

Thank you for sharing the schematic.

Could you please check the JT_VDDIN_PWR_BAD_N signal waveform with R1221 unpopulated?

Thank you,

Hello,

Please find the requested waveforms attached.

We removed R1221 and captured the signals separately at the Reset IC output (CH2) and the THOR input side (CH3).

CH3 initially goes High due to the internal pull-up within the module, which is referenced to the MV input. CH2 then goes High after the HV voltage is detected.

Furthermore, when the Module Power On signal is asserted, CH3 goes Low briefly before returning to High.

Please review the waveforms and let us know your findings.

Thank you for your support.

Hi,

Could you please check whether the J62 pin is tied to GND, as shown in our reference schematic?

As mentioned previously, for normal booting, J62 and G5 (BOOT_CHAIN_SEL) are connected to GND through 0-ohm resistors.

This issue appears to be the same as the one reported in the following NVIDIA Developer Forum thread:

The issue begins during MB1 boot, when an I2C timeout occurs while attempting to access the internal temperature sensor.

Could you please advise if there are any known causes or potential factors that could lead to this issue?

Additionally, could you please confirm whether there are any other pins that must be strapped to specific levels for normal booting, similar to J62 and G5 (BOOT_CHAIN_SEL)?

Any guidance on the potential root cause and the required strap pin configurations would be greatly appreciated.

Thank you for your support.

Hi,

Could you please share the waveform with the following points?

  • VDDIN_PWR_BAD_N
  • MODULE_POWER_ON
  • MODULE_SHDN_N (L52)
  • SYS_RESET_N (L60)

Thank you,

Please find attached the waveforms you requested.

Based on the waveforms, MODULE_SHDN_N does not appear to transition to a stable High state, seemingly in correlation with the drop in PWR_BAD_N.

Additionally, although it is not shown in the attached waveforms, CARRIER_POWER_ON was observed to be asserted by the module after PWR_BAD_N went Low but before it returned to High.

Could you please review the waveforms and advise on the possible cause of this behavior?

Thank you for your support.

Hi,

Thank you for sharing the waveforms.

Could you please confirm whether these waveforms were captured in Recover mode or during a normal boot? If they were captured during a normal boot, please also attach the debug UART log.
Also, could you please share the waveforms with the following points?

  • VDDIN_PWR_BAD_N
  • MODULE_POWER_ON
  • MODULE_SHDN_N
  • CARRIER_POWR_ON

The previously shared log was captured in Recovery Mode and appeared to show an I2C write failure to the thermal sensor inside the CVM via VM_I2C.

If the same error occurs during a normal boot in the debug log, could you please also capture the VM_I2C_SCK and DAT waveforms and check for any abnormalities?

Additionally, could you please confirm whether the 3.3V supply to pin L53 (SYS_VIN_SV) follows the same power supply as our reference design?

Thank you,

The waveforms attached to my previous message were captured during normal booting.

For this test, I connected a T5000 module, whose onboard flash had been successfully written, to a different type of carrier board.

However, the same error occurred during initrd execution in Recovery Mode, at the same stage as the error reported in the previous log:

THOR_UART_DEBUG_MSG_19102026.txt (7.7 KB)

[0000.660] E> I2C: Timeout while polling for transfer complete. Last value 0x00000002.
[0000.665] E> I2C: Could not write 0 bytes to slave: 0x0098 with repeat start false.

Additionally, please find attached the additional waveforms you requested.

Regarding VM_I2C, it would be helpful to examine its waveform for further investigation. However, the carrier board was not designed to use the module’s internal I2C port, and unfortunately, VM_I2C is not routed to an external connector. Therefore, we are unable to measure this signal directly.

Finally, the power applied to L53 (SYS_VIN_SV) is an always-on 3.3 V supply that is enabled by the Power Good signal from the always-on 5 V supply. We believe this power sequencing approach is consistent with the reference design.

For your review, I have also attached the schematics for the T5000 power circuitry and connector page.

T5000 Platform.pdf (606.3 KB)

Could you please review the additional waveforms and schematics and advise on any potential causes of the I2C timeout, as well as any further debugging steps you would recommend?

Thank you for your continued support.

Hi,

Thank you for sharing the customer carrier board schematic.

We noticed that THERMAL_ALERT_N (L56), which is pulled up to 3.3V on the CVM, is also pulled up to 1.8V on your carrier board.

Could you please review the notes in the Comment column of the pinmux sheet and check whether there are any other similar pulled up voltage mismatches?

Thank you,

You are correct regarding the issue you pointed out.

The pull-up power source for THERMAL_ALERT_N (L56) was incorrectly configured, and this issue has already been corrected.

I also reviewed the comments in the pinmux sheet again and confirmed that there are no additional pins that violate the specified requirements.

However, we have discovered something puzzling.

We previously developed carrier boards for JAO and JAOi. The current Jetson Thor carrier board was developed as a revision of those existing designs.

To investigate further, we installed the T5000 module on our existing JAO/JAOi carrier board. Surprisingly, no I2C timeout occurred, and the internal flash download completed successfully.

There are several important observations:

  • The THERMAL_ALERT_N pull-up issue was also present on the original JAO/JAOi carrier board, yet the T5000 operated successfully on that board.

  • BOOT_CHAIN_SEL is not connected to GND on the original JAO/JAOi carrier board.

  • Even on the original carrier board, VDDIN_PWR_BAD_N temporarily drops Low before returning High.

These observations make it difficult to explain the issue in terms of the pull-up configuration, power sequencing, or strap pin settings alone.

We are now in a very challenging situation because the T5000 works correctly on the original JAO/JAOi carrier board but fails on the revised Jetson Thor carrier board, despite the similarities in their designs.

Could you please help us investigate this issue further? Any suggestions regarding potential differences between the two carrier boards that could cause the MB1 I2C timeout, or any additional checks you would recommend, would be greatly appreciated.

We would sincerely appreciate your technical guidance in helping us identify the root cause.

Thank you for your continued support.

Hi,

Thank you for the detailed investigation.

Could you please compare the schematic differences between the original and revised carrier board (for Orin), particularly the SYS_VIN_SV supply circuit and the connections to VM_I2C?

Please also check whether there are any differences in the timing of the power sequence signals between the two boards.

Thank you,

Hi,

Please do not populate the following resisters.

R2103 (10kΩ)
R2112 (0Ω)
R2122 (47kΩ)
R2300 - R2305 (2.2kΩ)
R2308 - R2309 (2.2kΩ)
R2316 - R2319 (2.2kΩ)
R1215 (100kΩ)

Thank you,

Regarding your response to my first message:

Before receiving your response, we had already attempted to eliminate the circuit differences between the original JAO/JAOi carrier board and the revised carrier board by removing all connected series resistors, pull-up resistors, pull-down resistors, and other related components. We retained only the GND connections for J62 and G5.

However, the same error persisted.

As an additional observation, on the original Orin carrier board, G5 is connected to the Ethernet switch for RGMII functionality and is not connected to GND. Nevertheless, MB1 booting on the Thor module completed successfully on this board.

SYS_VIN_SV becomes active immediately when power is applied to the board. Both carrier boards use the same circuit configuration and components for this supply.

VM_I2C is not connected to the carrier board on either design and is therefore inaccessible for external measurement.

Regarding the boot sequence, both boards use MCUs of the same part number to control the power-up sequence. We also tested both MCUs with the firmware exchanged between the boards, but the results remained unchanged. Furthermore, we observed no significant differences in the waveforms captured using an oscilloscope.

Regarding your response to my second message:

We had already implemented some of the suggested resistor-removal measures as part of our efforts to eliminate circuit differences. We applied the remaining measures as well, but the same error persisted.

We have now exhausted several avenues of investigation, yet the root cause remains unclear.

Our development schedule is becoming extremely tight, so we urgently need to identify the next steps.

Based on the findings above, are there any additional areas we should investigate? In particular, could you recommend any specific checks or diagnostic procedures to help determine why MB1 encounters an I2C timeout on the revised carrier board but works correctly on the original JAO/JAOi carrier board?

Any further technical guidance or assistance in identifying the root cause would be greatly appreciated.

Thank you for your continued support.

As a last resort, we would like to re-verify the power sequencing timing from the beginning.

The Design Guide does not appear to specify strict minimum and maximum timing limits for all the relevant intervals. Therefore, we plan to measure the power sequencing timing on the NVIDIA EVK and apply the same timing values to our carrier board.

Could you please advise if there are any specific timing requirements or minimum/maximum durations that must be strictly observed during the power-on sequence?

Additionally, we would appreciate your confirmation regarding the required strap pin configurations.

We have generally followed Table 18-1 of the Design Guide. However, the recommended implementation for the RAM_CODE pins does not appear to be explicitly specified.

In our current design, the RAM_CODE pins are left unconnected (N.C.). Could you please confirm whether this configuration is acceptable or whether these pins require specific pull-up/pull-down configurations?

Similarly, BOOT_CHAIN0 and BOOT_CHAIN1 are currently left unconnected. Do these pins require any specific connections or defined logic levels for normal booting?

For the remaining items, we will apply the previously identified correction to the THERMAL_ALERT_N pull-up power source and remove unnecessary resistors as discussed.

Given the persistent boot failure despite our previous debugging efforts, we would greatly appreciate your confirmation of the required power sequencing timing and strap pin configurations before proceeding with further hardware modifications.

Thank you for your continued support.

Hi,

The RAM_CODE pins can be left unconnected if unused. BOOT_CHAIN0 and BOOT_CHAIN1 can also be left unconnected when BOOT_CHAIN_SEL is low.

Please also attach the debug UART log captured during normal boot, rather than during flashing.

Thank you,