Pull-ups, debouncing, protection

YOU WILL BE ABLE TO

Explain why a floating input is undefined and size a pull-up.
Design an RC debounce and state its limits.
Describe standard input protection against overvoltage.
Assemble a checklist for a robust interface.

ME-5 established that a digital input recognises two voltage bands with a forbidden zone between them. A pin connected to nothing sits nowhere in particular — it floats, picks up whatever charge and interference is nearby, and reads unpredictably, sometimes oscillating and drawing extra current as its input stage lingers in the forbidden region. The cure is one resistor that defines the idle level.

A pull-up resistor ties the input to the supply, so the pin idles high; a switch to ground pulls it low when pressed. A pull-down does the mirror image. The value is a compromise: small enough that leakage and interference cannot shift the pin, large enough that the current wasted while the switch is closed is acceptable. Ten kilohms is the default answer, drawing $330$ µA at $3.3$ V while pressed.

I_{pull} = \frac{V_{supply}}{R_{pull}}

current wasted while the switch is closed

Now bounce. A mechanical switch's contacts do not close once; they strike, separate, and strike again for a few milliseconds. A microcontroller polling every microsecond sees dozens of presses. CA-5's capstone designed the hardware cure — an RC whose time constant outlasts the chatter — and this is where it belongs.

A complete button interface: R_pu defines the idle level, the RC outlasts the contact bounce.

Two cautions on the RC. First, its output crosses the input threshold slowly, and ME-5 L4 explained the consequence: a slowly-moving signal near a threshold invites both chatter and metastability. A proper design follows the RC with a Schmitt-trigger input — which most microcontroller pins have — or with a comparator given hysteresis (ME-4 L4). Second, software debouncing is a legitimate alternative and often the better one: sample the pin, and accept a change only after it has been stable for several reads. Hardware costs two components; software costs a few lines and no board area. Use hardware where the signal is also an interrupt source or where the noise is electrical rather than mechanical.

Finally, protection. A pin connected to the outside world can meet voltages it was not designed for — static from a fingertip, a miswired supply, a sensor's fault condition, an inductive spike from a nearby motor. The standard defences are cheap and stack well: a series resistor of a few hundred ohms to a few kilohms limits fault current into the pin; clamp diodes to the supply and ground shunt anything outside the rails (most microcontrollers include weak ones internally, which the series resistor makes effective); and a Zener or TVS device holds a hard ceiling, which is ME-1 L4's Zener clamp in its natural application.

Hazard	Symptom	Standard cure
Floating input	Random or oscillating reads	Pull-up or pull-down resistor
Contact bounce	One press counted many times	RC plus Schmitt input, or software debounce
Inductive kickback	Switching device fails, often on first use	Flyback diode across the coil
Overvoltage on an input	Damaged pin, unstable chip	Series resistor plus clamp diodes or Zener
Excess load current	Hot part, sagging rail, brown-out	MOSFET switch sized for stall, adequate supply
Asynchronous input	Rare unreproducible faults	Two-flip-flop synchroniser (ME-5 L4)

The standard interface hazards and their standard cures — a checklist worth applying to every input and output.

WORKED EXAMPLE

A button interface, completely specified

Pull-up: $10$ kΩ external (or the internal one), giving $330$ µA while pressed.
Debounce: contacts bounce about $2$ ms, so target $\tau = 10$ ms. With $C_d = 100$ nF, $R_d = \tau/C = 100$ kΩ.
Check the interaction the CA-5 capstone flagged: $R_d$ is in series into the pin, and the pin draws essentially no current, so it does not disturb the level — but $R_d$ and $R_{pu}$ do form a divider with respect to leakage, so keep $R_d$ well below the pin's leakage-equivalent resistance (megohms). $100$ kΩ is fine.
Feed a Schmitt-trigger input so the slow edge does not chatter.
Protection: the series $100$ kΩ already limits fault current handsomely, so no additional resistor is needed for this input.

R_{pu} = 10

kΩ,

R_d = 100

kΩ,

C_d = 100

nF, Schmitt input.

WORKED EXAMPLE

Why the checklist exists

Every row of the table above describes a failure that works fine on a bench and fails in the field — which is the most expensive class of defect there is.
Floating inputs work until someone touches the board. Missing flyback diodes work until the relay switches off. Missing synchronisers work until the timing happens to align.
All six cures together cost a handful of passive components and no design time once memorised.
This is the difference between a circuit that functions and a product that ships.

The cures are cheap; the failures are intermittent, which makes them dear.

PRACTICE

A $4.7$ kΩ pull-up is used on a $3.3$ V I2C bus. How much current flows while a device holds the line low?

Design an RC debounce for contacts that bounce for $5$ ms, using a $220$ nF capacitor.

Name the two distinct problems a slowly-rising debounced signal can cause at a digital input, and the cure for each.

Why does a series resistor make a microcontroller's internal clamp diodes far more effective?