I have a board with two LTM8073 "silent switcher µModule regulators" by Analog Devices/Linear Technology being used to provide two separate power supplies. This is a 10-layer dual-sided board that was put together by a very experienced PCB designer.
Both supplies receive 48V in and they provide 3.3V and 5.0V respectively. These power a Trenz SBC that has its own set of 1.0V, 1.5V, 1.8V supplies being fed from the 5.0V supply and thus taking most of the load. Otherwise most of the design and layout is nearly identical. The design is very close to the recommendations presented in table 1 of the data sheet, I only increased the switching frequency of the 3.3V supply slightly to 1MHz, well within the recommended range. But I am having a couple of odd issues with the 3.3V supply.
Although I doubt this will be useful, here is a bottom-side view of the board with only the top and bottom layers shown. The 3.3V supply is U19, the one at the top (the supplies are mounted on the top layer, but nearly all of their passives are in the bottom layer).
As an aside, I checked the traces that seem to snake and surround the supplies. These are either unconnected (share lines), outputs (Power Good), or low-impedance configuration lines (Sync).
The data sheet has an error in their recommendations for feedback resistance for a 3.3V supply. The design equations indicate 79.9kΩ while Table 1 says 80.6kΩ, I stupidly followed the table which later I corrected by adding a 5MΩ parallel resistor. This should result in 79.3kΩ and 3.32V.
However, with both supplies unloaded the 5V supply is at 5.006V while the 3.3V supply is at 3.23V. I have tested this on two different boards and both have nearly identical values.
I proceeded to modify the 5V supply (due to debugging issue #2) to provide 3.61V, and I obtained 3.61V almost exactly. This suggest there is some issue with the layout of the 3.3V supply that is not present in the layout of the 5V supply.
Where should I look?
When connecting the SBC (which should take most of the power from the 5V supply) I cannot get the module to power-up properly. The module goes through its initial power sequencing, starting its 1.0V, 1.5V, & 1.8V supplies all of which come from the 5V supply, but when it switches on the pass-switch from the 3.3V supply it causes it to droop enough for it to go out of regulation and drop its power-good signal. Which then restarts the power sequence again. This can go on indefinitely.
I measure the load on the 3.3V supply to rise to about 1A in 5µs (it's supposed to regulate in excess of 3A) before the supply drops its power-good signal. On the other side of the switch the voltage only increases by about 100mV. The Trenz 0702 SBC uses the TPS27082 power switch which provides soft-start capability and seems to be configured for about 300µs startup time.
By placing a temporary bridge across the TPS27082 switch, I can get the system to start and the board works as designed with a total load of ~4.5W (most of which is in the 5V supply). I have not yet measured the current consumption in the 3.3V supply, but it is sitting at ~3.16V.
The issue is clearly triggered by power-up inrush, but (1) the 3.3V supply should handle the inrush, and (2) this does not explain the relatively large error that it starts from. I looked at the additional capacitive loading I added to that side of the switch and it is less than 50µF in total.
I added extra output capacitance to the 3.3V supply output (total of 300uF), it increased the on period and the peak current but it still went out of regulation.
Calculating peak current and duration, we estimate that the surge can be caused by as little as 90uF load, this is well within acceptable ranges. With such capacitance and the set slew rate of the switch, peak surge current should be about 1.1A.
After a forced startup, although on steady state the 5V supply does take most of the load, it is much more constant @0.6A. During operation the 3.3V supply sees transient load changes of 1.4A (it does not lose regulation though).
Adding a resistive load of about 100mA to the 3.3V supply allows the board to start up. This leads me to suspect that somehow the supply is in some “slow mode,” and does not have time to react to the fast change in load.
Adding a 120ohm resistor across the power-up switch does not solve the problem.
Adding a 100pF capacitor from the feedback node to the output sort of solves the problem as in the system is likely to start up within a second, but it still hiccups multiple times until it can get out of that condition.
I should have inspected the layout more carefully before we sent it out, there are a few changes I would now make. The most notable one (and the possible cause of the lower regulation) is that the 1uF input capacitor for the 3.3V supply is about 2cm away, while the 5V one is 2mm away. These are on polygon pours connected to the same planes, but still an issue. Adding a capacitor (through a bottom via) slightly improves regulation, but it does not solve the problem.
Typical startup sequence, the purple trace is the 3.3V enable signal on the SBC, the yellow trace the 3.3V supply, and the blue trace the 3.3V supply after the switch on the SBC.
Any suggestions on where to start looking or how to fix it? I have ways to hack it, but I would like to know the actual root of the problem.
A 1A initial surge in a 3A supply should not cause it to lose regulation. I see no specifications in the datasheet for transient load increase from zero.