Vrefint and Vts very noisy with SAI active on G4

[unterberg_rtb]

We are using an STM32G491 with TDM audio output active on SAI. When activating the 4 MHz bit-clock on SAI1_SCK_A, Vref and temperature measurement become very noisy, while externally connected ADC inputs are not affected.

See attached graph for difference in Vref measurement with SAI enabled and SAI disabled. Vref is used as input, it is externally connected to 3.3V. This graph has been recorded on a STM32G4 eval board, bit we see the same behavior on our own boards. ADC clock is 6,25 MHz (100 MHz with prescaler 16), and the Vref sample time is set to 47.5 cycles (makes 7,6 us sample time).

Is this a known issue? I could not find anything about this in the G4 errata sheets; however, the L4 errata sheet lists a similar issue.

[LCE]

Any capacitors at the Vref input, any cap at the ADC input?

For the ADC inputs, the PCB layout on the eval boards is often terrible, especially on the Nucleos, where the lines go across the whole board to get to some connectors.

[unterberg_rtb]

On our board we have 1nF on each active ADC input pin, and 100nF connected between VREF+ and VSSA, very close to the CPU pins.

Sample values from external ADC pins are very stable, so I assume VREF+ is stable. Only the internal VREFINT ADC channel is noisy. Can this be caused by internal interference, similar to what is described in STM32L431xx
errata sheet es0320 Section 2.5.3 "Spurious temperature measurement due to spike noise"?

[LCE]

Ah, so it's mostly the internal ref, that's not too bad.

> "Spurious temperature measurement due to spike noise"

That sounds reasonable.

I always use the internal reference as a sanity check, and I found that it is not the most accurate voltage (as the datasheets say), and almost always a little bit noisy.