I read that one of the disadvantages of the 555 timer is that it varies in timing accuracy depending on temperature.
How much can accuracy vary, and why does this happen? Is temperature the only thing that affects timing accuracy?
It seems the 555 is generally regarded as being inaccurate to some degree, but often the how and why is not explored.
I had a very quick skim over the 555 datasheet, but nothing jumped out to me (though I’m sure it’s probably in there somewhere).
I had a very quick skim over the 555 datasheet, but nothing jumped out to me (though I’m sure it’s probably in there somewhere).
It depends whether you are implementing a monostable or astable: -
Plus the drift from the timing capacitor (that can be the most significant drift of all).
This Wiki answer covers it from all angles. Have a look when you have time for a more comprehensive bit of reading :)
This has been rehashed in many places, but a 555 is just not a thing to use if you care about stability, parameter spread, etc. It's perfectly fine for applications where none of that matters, but in those a tiny MCU might provide way more functionality for the same or lower (!) cost.
If you want to go old-skool analog, and show that you really dig the 2nd- and 3rd-order effects of temperature on parts, then you can design interesting compensation circuits that will take care of both the thermal drifts inside the chip, as well as thermal drifts of the external parts - with the capacitor being the worse the less you pay for it. It'll be pretty analog art, and you could trim such circuits to perfection if you have a small environmental chamber, but none of that is practical or even necessary for something you'd like to make any money on (unless it's sold as unique art, where nothing needs to make sense, cost-wise).
There are many other ways to get the timing functions you need, without even knowing about 555. And almost universally, they'll be better. A modern 6- or 8-pin MCU with a built-in oscillator is almost always more stable, has a narrower frequency tolerance, and way more versatile in timing tasks. If you don't care much for an MCU, then TimerBlox are purpose-designed parts with superior specs to almost anything that a 555 can do.
For longer timing intervals, a decades-old watch IC, TI's AC5948N/AC5954N was much more accurate than a 555, although a bit clumsier to use :)
If you need 10% ish accuracy (that could be trimmed and coddled to get to 1-2% ish) then a 555 is fine, especially if you need high voltage capability or high output current capability. Many, many millions are still sold industrially for simple applications such as novelty neon sign power supplies. The 555 itself may change by 1% over the temperature range, a film timing capacitor will be similar, and a timing resistor will add a bit more. There are also effects from noise on the power supply, bypassing of the control voltage and so on. If you make a CMOS 555 oscillator in the kHz range with an NP0 capacitor and a precision timing resistor, it will be pretty good in terms of accuracy and stability with a quiet well-regulated and bypassed power supply. Even if the power supply and external parts are perfect the 555 itself typically drifts 150 ppm/°C, so for a 0-70 °C range that's about a 1% change and the initial accuracy is typically +/-1%. Other oscillator types mentioned below typically operate at much higher frequencies and divide down, so variations in supply, etc. tend to average out, whereas a 10% blip in the supply voltage can truncate a pulse from a 555 a similar amount of time.
If you need a bit better accuracy then a small MPU with built-in calibrated RC oscillator can get 1-2% unadjusted accuracy with no external components. An oscillator based on a ceramic resonators could be sub-1% in initial accuracy. There are also "silicon oscillator" timing chips with a pre-programmed output frequency.
If you need timekeeping accuracy in an oscillator then quartz crystals and oscillators (including those built into MCUs and real-time clock chips) based on them are a good choice. The worst ones tend to be better than 50 ppm (0.005%) and you can get accuracy and stability down into the ppb range with temperature compensated and ovenized oscillators. A watch that loses or gains 2 seconds per day requires an accuracy of about 20 ppm (a 555 oscillator is maybe 1000 times too inaccurate for that application).
Finally, if none of that is good enough, you can even ̶b̶u̶y̶ order atomic clocks as components.
The first two items can be 5-10 cents each in quantity, up to a starting price of about US$5000 for an atomic clock module, so a range of 10,000:1 in cost for an accuracy range of roughly 1,000,000,000:1.
