# Decoupling Capacitors: Why Every Chip Wants 100nF

> A 100nF decoupling capacitor beside a chip's power pins acts as a tiny local battery, absorbing current spikes that otherwise cause resets and mystery bugs.

[HTML version](https://soldr.ai/blog/decoupling-capacitors-why-every-chip-wants-100nf)

23 August 2026 · Bharat Raj · [originally published on Compoden](https://compoden.com/blogs/guides/decoupling-capacitors-why-every-chip-wants-100nf)

**A decoupling capacitor is a small capacitor (almost always 100nF ceramic) placed right next to a chip's power pins, acting as a tiny local battery that supplies the chip's sudden bursts of current so the supply voltage at the chip never dips.** Digital chips do not draw current smoothly. Every time internal transistors switch, the chip gulps a spike of current for a few nanoseconds. Wires and breadboard traces have inductance, which means they cannot deliver current that fast, so without a capacitor the voltage at the chip's pin sags for an instant. Those sags cause resets, corrupted sensor readings, and crashes that look exactly like software bugs. One 100nF capacitor across VCC and GND, placed close to the chip, absorbs the problem almost entirely.

## Why chips need bursts of current
Inside a microcontroller, millions of transistors flip between states on every clock tick. Each flip charges or discharges a tiny internal capacitance, and that takes a pulse of current. Averaged out, an ATmega328 might draw 15mA. But zoomed in to nanosecond scale, the demand looks like silence punctuated by sharp spikes many times larger than the average.
Your power supply, whether a USB port or a battery, could deliver that current easily. The problem is the path. Every centimetre of wire has a little inductance, and inductance is the electrical property that resists changes in current. Slow changes pass through fine; nanosecond spikes get choked. So at the exact moment the chip demands current, the wire says "give me a moment", and the voltage at the chip's power pin briefly collapses. Engineers call this supply droop or rail noise.

## What the capacitor actually does
A capacitor stores charge and can release it almost instantly. Placed directly across the chip's power and ground pins, it sits there charged to the supply voltage. When the chip spikes, the capacitor delivers the charge from millimetres away, faster than the supply ever could through its long leads. Between spikes, the supply quietly refills the capacitor. The name says it well: it decouples the chip's frantic short-term demands from the lazy long-distance supply. You will also hear the term bypass capacitor, which is the same thing viewed differently: it gives high-frequency noise a bypass route straight to ground instead of travelling through your circuit.

## Why 100nF specifically?
It is a sweet spot, not a magic number. The capacitor needs to be large enough to hold meaningful charge, but small ceramic capacitors respond faster than big electrolytics because they have less internal inductance of their own. 100nF ceramic handles the frequency range where digital switching noise lives, it is cheap, and it is small. That combination made it the universal default, to the point that datasheets simply assume it. For chips with several power pins, the rule is one 100nF per power pin, each as close to its pin as you can manage.
You will often see a larger companion, 10µF to 100µF electrolytic, added per board. That one handles slower, bigger events, like a motor starting or an LED strip lighting up. The two work as a team: the electrolytic is the neighbourhood water tank, the 100nF is the bottle on your desk.

## Worked example: a bare ATmega328 on a breadboard
Say you have prototyped on an Arduino Uno R3 and now want to run the bare chip on a 400-point breadboard. The Uno's designers already placed decoupling capacitors on the board, which is why the Uno feels so dependable. Your breadboard version has none, and it will misbehave until you add them.
Place the ATmega328 across the centre channel. Connect 5V and GND. Now place one 100nF ceramic capacitor with one leg in the same 5-hole row as pin 7 (VCC) and the other leg in the row for pin 8 (GND), spanning the channel or hopping rows as needed. Add a second 100nF between AVCC (pin 20) and GND, because the analog section has its own power pin. Keep the capacitor legs short; every extra centimetre of Dupont jumper wire between capacitor and chip adds inductance and weakens the effect. Finally, one 10µF electrolytic across the power rails (stripe to GND) steadies the whole board.

## Where this bites you
The classic story: a project works perfectly until you add a servo or a relay, then the microcontroller starts resetting at random. An SG90 servo yanks big gulps of current from the rail every time it moves. The rail dips, the chip's brown-out detector fires, and the board reboots. Beginners hunt through their code for days, because the crash correlates with "when the motor moves", which looks like a software event.
The signature of missing decoupling is randomness: works on Monday, fails on Tuesday, works with the multimeter probe touching the rail (the probe adds capacitance!), fails when you remove it. Analog readings that jump around, I2C sensors that occasionally drop off the bus, SPI transfers that corrupt one byte in a thousand: all classic symptoms. Before blaming code, check that every chip has its 100nF and that motors have their own bulk capacitor, or better, their own supply.

## FAQ

### Does the orientation of a decoupling capacitor matter?
Ceramic capacitors (the small yellowish discs or rectangles, often marked 104 for 100nF) have no polarity; either leg can go to either rail. Electrolytic capacitors (the small cans) are polarised: the striped leg must go to ground. Reversing an electrolytic can destroy it, sometimes with a pop.

### Can I use 1µF or 47nF instead if that is what I have?
Usually, yes. Anything from 47nF to 1µF ceramic will decouple a hobby microcontroller adequately. The habit that matters is placement: close to the power pin, short legs. A perfect capacitor 10cm away does far less than a mediocre one at 5mm.

### My circuit works fine without any capacitors. Should I still add them?
Yes. "Works on my desk today" is the most fragile kind of working. Noise problems scale with wire length, load current, and bad luck. The capacitor costs almost nothing and removes an entire category of future mystery bugs, which is why every commercial board is covered in them.
If you would rather not think about which pins need what, Compoden's AI build assistant Soldr can plan the wiring for your project with the boring-but-vital parts included.