Does a baudrate in serial communication (USB or RS232) need to be exact?
The UARTs typically used in RS232-type serial systems work by sampling the data line somewhere mid bit according to a division of the predefined baud rate base frequency clock. As such, if the sent data and receiver are not on the same frequency the "sample-point" will wander closer to the edge of the bit frame on successive bits.
With a normal UART the bit length for a byte is 10 or 11 bits. 1 Start, 8 data, and 1 or 2 stop bits. Half a bit of wander on the 10th bit translates into 0.5/10 = 5% error.
However, in reality your tolerance is less than that because you also need to add in the latency of you base frequency period which will add in an offset from the leading edge of the start bit. The higher your base frequency, the less effect that has.
As for using a 555 timer for this purpose, I would not recommend it unless you plan on having a manual adjustment in the 555 circuit.
A USART on the other hand uses a more complex control method that attempts to synchronize the transmission to the received data. This can be through using a data pattern that has an embedded clock, by using a passed clock, or by some form of phase locking to the received data edges. (Though arguably the latter is really pseudo-synchronous.)
The timings have to be accurate enough that they don't drift apart before the protocol resyncronises.
UART serial resyncronises on each byte and a byte is around 10 bits (8 bits of data plus start and stop). We assume that our UART targets the middle of each bit. If everything is is perfect and only one end is inaccurate that allows for approximately 5% difference between the two ends of the link.
However:
- Both ends of the link may be inaccurate relative to nominal. In the worst case one end may be under nominal while the other is over nominal.
- There may be systematic inaccuraries. For example most UARTS have a limited range of baud rate generator settings.
- Your UART is based on a master clock with a finite speed. Even if the clocking and baud rate generator are perfect this will result in it not hitting the exact middle of each bit.
The bottom line is that 1% error on your clock is almost certainly fine. 5% error is almost certainly a problem. Between those two figures it may or may not be a problem depending on the overall picture.
That is a tall order for a RC oscilator. Lets say your R has a 1% tolerance and your C has a 2% tolerance. That gives about a 3% tolerance for the RC network's time constant and that is before you think about any errors introduced by your driver chip.
So the bottom line is you should be looking at a crystal or ceramic resonator.
As for USB I don't have the experiance to analyse from first principles but https://www.silabs.com/community/interface/knowledge-base.entry.html/2004/03/15/usb_clock_tolerance-gVai says 1.5% for low speed and 0.25% for full speed.