For a client, I am using a LTM4638 on a 4"x4" PCB. The planes driven by this regulator are seeing 330Mhz impulse noise events once at each switching event. Amplitudes are sometimes upward of 200mV on a 1.2V plane. It has enough energy to get through a buffer powered by this plane to the digital output of that buffer and cause digital failures.
Figuring is was bad design practice on the PCB, I've spent a lot of time there (layout, caps, ...)
I just got an ADI eval board (DC2665B-B) with the same part on it and IT SHOWS THE SAME RINGING. It changes top to bottom. Measurement method: 3GHz Agilent diff probe into 33Ghz scope. Also did a soldered micro-coax to a cap-point and ran that straight into 50Ω input on scope. It's always there.
Do all fully integrated VRs like this have this issue? Are there some references for this model on how to keep that from getting to the rest of the board?
I know multiple methods to attempt to not let the affect the rest of the board (e.g. separate planes, maybe some FB, etc...) but I'd like to have a high confidence next design.
Thoughts?
A couple more bits of data while checking the eval board:
* bursts happened EVERY time
* burst aptitudes are consistent on every cycle (but different based on where on board it is measured)
* same result when load was both 0 and 750mA
Another key question: Do the engineers at ADI see the same on their eval board?
added more detail
[edited by: mtntrx at 3:49 PM (GMT -4) on 17 Oct 2024]