Two wires, but not I²C

What goes down the wires

Everything the display is ever told is a byte, and every byte goes out the same way: a start, eight bits, an acknowledge from the chip, a stop.

Press through it once. The parts worth noticing:

Both wires idle high on their own. Your board never drives them up. The library either pulls a wire down to 0 V or lets go of it, by switching the pin to an input, and the two 10 kΩ resistors on the board pull it back up. I²C uses the same trick, and for the same reason: two pins can never fight over a wire neither of them drives high. What the wire rises to is the display's VCC, which is the whole of 3.3 V or 5 V.

The bits go out least significant first. I²C sends the most significant bit first. This does not. It matters the moment you start writing segment patterns by hand: bit 0 of the byte is segment A, and it is also the first bit on the wire.

The chip answers. After the eighth bit the display pulls DATA down itself and holds it through a ninth clock, to say it heard you. That amber pulse is the only thing on the wire your board did not put there.

The part that is missing

There is no address. Nowhere in any of those bytes is there anything that says which chip should pay attention.

I²C opens every exchange with an address precisely so that a dozen devices can share two wires. The TM1637 has nothing of the kind — the datasheet says so in as many words, that the protocol "is not equal to I2C bus protocol totally because there is no slave address". Every chip that can hear the wires acts on everything it hears.

Three consequences, all of them practical:

• An I²C scan finds nothing. Not a fault. There is nothing to find.
• It cannot share your I²C bus. SDA and SCL belong to devices that understand addressing; this display would corrupt their traffic and be corrupted by it.
• A second display needs its own DATA pin. This is the one that shapes real projects, and more than one display is entirely about it.

What it costs in time

The library waits 100 µs at every step of every edge. One update (four digits, seven bytes, three start–stop pairs) makes 208 of those waits, about 21 ms doing nothing but waiting.

For a clock, a counter or a thermometer that is invisible. It starts to matter when six displays are being refreshed in a loop, or when 21 ms of blocking upsets something else in the sketch. The fix is the constructor's third argument, which is the delay in microseconds:

TM1637Display display(2, 3, 10);   // ten times faster, still far inside the chip's limit

The chip's datasheet allows a clock up to 500 kHz, so there is a great deal of room. Long wires are the reason to leave the default alone: the same resistors that hold the wires high also make them slow to rise, and every extra length of wire slows them further.