How to choose crystal speed?

You will find answers if you read the AN2867 carefully and try to ensure the stability of the oscillator with common high-frequency crystals and the required capacitors (including board parasitics) and over the whole temperature range.

It is simply cheaper and easier to realise if you use low-frequency crystals such as 4...8MHz for HSE.

Regards

/Peter

my 2 cts : dont use a crystal, take crystal oscillator, frequ. medium, as reference boards use, maybe 8 or 12 or 25 MHz, and select the cheapest, thats available from different sources.

and set "bypass" in Cube for oszillator .

and put 80 or 100 ohm in clock connection, to avoid reflections and new trouble. (my way...)

The first question is whether the device needs HSE at all. Many devices can run from HSI/MSI just fine.

It sure will run, but if you want somewhat decent timing accuracy, it's sure not the way to go.

The specs are quite poor compared to a standard crystal setup...

The question is whether the specific device needs that accuracy. For example, SPI, I2C and other synchronous interfaces doesn't need crystal level accuracy. Ethernet needs +/-50 ppm, which obviously needs a crystal. USB needs +/-0,25 %, but your MCU can do it with MSI synced with LSE or HSI48 with clock recovery based on USB SOF sequences.

Usb host would command external precise hse, then for usb device, msi would relax the constrains. Ethernet would probably require 25mhz which could be buffered by external transceiver fed to the mcu. For the hse value, clive gave some clues. Msi has some sweet frequencies. Otherwise, check flash wait states vs mhz and if low power is a concern on non msi mcu, running with pll bypass at high frequency enables instant prescaling gearing down and up.

If possible, I use no crystal. If I need a crystal for accurate audio generation/analysis, I just use the 32KHz crystal (which can be used to trigger DMA etc. up to 64KHz), as long as the 32KHz is providing an accurate drum beat, I don't care if the main processor is running at an imprecise frequency. I only resort to a higher frequency crystal when MHz clock frequency accuracy is required eg. generating a video signal.

> use XTAL, not those pesky resonators, a proper XTAL, and you will live longer and happier

As I learned quite recently on this very forum, recent MEMS oscillators (which are emphatically not quartz crystals) are low-power and have excellent characteristics. They cost more (1.15$ vs. 0.3$) but their performance is competitive with crystals (and can exceed it in terms of power consumption).

Also, for more generalized applications which employ crystals, you may learn about the trade-offs between low and high-frequency crystals here :face_with_tears_of_joy:

Update: One argument I haven't seen above is that although the PLLs can generate many different frequencies from your reference frequency, they are limited to specific ratios. If you need a specific frequency, the PLL may be able to achieve it based on crystal frequency and not with another. This can be important for interfaces with specific frequency requirements, or when you want to generate several specific frequencies by dividing the generated PLL clock. You have to get it right so you can get all the frequencies you need from one reference frequency. This SE may be useful.

Helium???

One more consideration. "Power consumption in oscillators is essentially a linear function of frequency".

Example From STM32G4 datasheet:

If you're working to minimize power consumption, an extra 200uA can be very substantial.