an open-drain output

An open-drain output is a one-trick output: it can pull its pin down to ground, but it cannot push it up. Picture a person who can only ever open a drain valve to let water out of a tank, never pump water in — the tank fills only if something else feeds it. The pin's transistor connects it to ground when active and otherwise simply lets go, leaving the pin floating. To get a HIGH, you add an external pull-up resistor to whatever voltage you want.

Inside, the output is a single transistor — an N-channel MOSFET (open-drain) or an NPN transistor (the older open-collector) — whose drain or collector is the pin and whose source or emitter goes to ground. Turn the transistor on and it shorts the pin to ground: a solid LOW. Turn it off and the pin is connected to nothing internally; the pull-up resistor quietly raises it to the supply: a HIGH. Because the chip never actively drives high, you, the designer, choose the HIGH voltage by choosing which rail the pull-up connects to. That is exactly why open-drain is the natural way to interface between voltage domains and to build certain shared buses.

This output style buys three things a normal push-pull output cannot: a HIGH voltage you set yourself (handy for level shifting), the ability to switch a small load or LED returning to a higher voltage, and the wired-OR connection where many outputs share one line. The honest cost is speed and power: the pull-up resistor and the wire's capacitance form an RC that slows the rising edge, and while the output is LOW it continuously wastes current through the resistor. I2C, interrupt lines, and reset lines all use open-drain for these reasons, accepting the speed penalty in exchange for the sharing it enables.