A diode does one deceptively simple thing: it passes current one way and blocks the other. Everything interesting in audio comes from how it makes that transition, not that it makes it. The turn-on is not a switch; it is a smooth exponential curve with a soft corner (the knee) at a threshold voltage Vf. Bend a signal against that corner and you get distortion whose entire character is set by the shape of one small piece of a curve.
This page is mostly about that: clipping, overdrive, fuzz, and how diode choice sculpts the sound. But the same junction shows up all over the signal chain, so the general physics comes first and the audio-only tricks build on it.
the arrow points the way conventional current flows anode (+) ──────────►|────────── cathode (−) PN real part: ──[ 1N4148 █]── ▲ band = cathode current flows anode → cathode only, and only once V across it > Vf
Dope one side of a silicon crystal with extra electrons (N) and the other with holes (P). At the boundary they diffuse across and cancel, leaving a thin depletion region with a built-in field. Forward bias collapses that field and current pours across; reverse bias widens it and current stops. The math is the Shockley diode equation, and every audio parameter is one term of it.
I = Is · ( e^(V / (n·Vt)) − 1 ) Is saturation (leakage) current ─ how leaky in reverse; sets noise floor V voltage across the diode n ideality factor (~1-2) ─ how SOFT the knee is Vt thermal voltage ≈ 25.85 mV @ 300K (Vt = kT/q, rises with temperature) because it is exponential, current changes ~10× for every ~60 mV (n=1), so "Vf" is not a hard number -- it is just where the curve gets steep enough to matter
| Property | Symbol | What it does to audio |
|---|---|---|
| Forward voltage | Vf | Sets the clip threshold. Lower Vf clips earlier and quieter; higher Vf passes more signal before clipping and is louder. |
| Knee softness | n | Sharp knee = hard, buzzy, high-order harmonics. Gradual knee = soft, warm, low-order harmonics. |
| Junction cap. | Cj | Acts as a small parallel capacitor = a lowpass. High Cj rolls off treble; low Cj keeps the top end clear. |
| Reverse leakage | Ir | High leakage (germanium) adds hiss and can shift DC bias; low leakage stays quiet when not clipping. |
Push a diode backwards hard enough and it breaks down and conducts anyway. In a rectifier that is failure; in a Zener it is the whole point (a fixed reference voltage). Two other reverse-bias effects earn their own audio uses: the depletion region behaves as a voltage-variable capacitor (a varactor), and the time it takes to stop conducting when the voltage flips (trr, reverse recovery) is what makes slow rectifiers "snappy" and fast ones clean.
Vt scales with absolute temperature and Is roughly doubles every 10°C. Germanium, with its already-large leakage, drifts audibly with heat. This is the real reason vintage germanium fuzz "behaves differently on a cold stage" -- it is not mojo, it is Is(T).
Same junction, different materials and geometry, wildly different Vf and knee. For clipping, the choice of type is the tone control before any resistor value.
| Type | Vf (typical) | Knee | Audio character | Examples |
|---|---|---|---|---|
| Silicon signal | 0.6-0.7 V | fairly sharp | hard, punchy, bright; the default | 1N4148 / 1N914 |
| Silicon rectifier | 0.7-1.0 V | sharp | power supplies; higher Vf, slower | 1N4001-4007 |
| Germanium | 0.2-0.3 V | soft | warm, smooth, compressed, "vintage"; leaky + temp-drifty | 1N34A, OA90 |
| Schottky | 0.15-0.45 V | soft-ish | very low threshold, very fast, low Cj; asymmetric pairings | 1N5817, BAT54 |
| LED | 1.5-3.4 V | fairly soft | loud, dynamic, "open" -- high threshold = more headroom before clip | red ~1.8, grn ~2.1, blue ~3.0 |
| Zener | set in reverse | hard | back-to-back = symmetric hard clip at a chosen higher voltage | 3V3, 5V1 zeners |
| Diode-connected Q / body diode | ~0.6 V | varies | a transistor wired as a diode, or a MOSFET body diode; used for unusual asymmetric curves | 2N3904 B-C, MOSFET |
An LED is a diode you can also see. As a clipper its high Vf gives loud, dynamic distortion (the "Turbo Rat" trick). But an LED glued to a light-dependent resistor is a vactrol, and a vactrol is the heart of a Buchla-style lowpass gate -- the LED's brightness (its forward current) squeezes the LDR and controls a filter/VCA in one soft, organic "bongo" pluck. So the same part sits in two very different audio stories: as a junction it clips, as an emitter it gates.
You can characterize a clipping diode with a multimeter and a bench supply -- no fancy rig. Two measurements tell you almost everything that matters for tone: the forward voltage at a real current, and whether a batch is matched.
# most DMMs have a diode-test mode (the ►| symbol) -- # it pushes ~1 mA through and reads the voltage drop red probe → anode, black probe → cathode (the band) reading = Vf at ~1 mA # 0.55-0.65 Si · 0.15-0.30 Ge · OL if reversed/open reversed probes → OL # good: it blocks the other way OL both ways → open (dead) ~0 both ways → shorted (dead)
Vf = Vsupply − (I · R), read across the diode.
Two "identical" 1N34A germaniums can differ by 50-100 mV and a lot in leakage. For a symmetric clipper you often want the up-diode and down-diode close, so the two halves of the waveform clip the same. For a ring modulator (four diodes) matching is critical -- mismatch bleeds carrier into the output. Sort a bag: measure Vf (and, for germanium, reverse leakage) and group them.
sorting a bag of germaniums -- measure two things, group by both Vf @ 1 mA Ir (reverse leak, µA) use 0.28 V low matched pair for symmetric clip 0.24 V high asymmetric side / character 0.31 V low ring-mod quad candidate for asymmetry you WANT a mismatch -- pick it on purpose, don't fight it
The relevant datasheet parameters, if you have the sheet: Vf (with its test current), Ir (reverse leakage), Cj or Ct (junction capacitance), trr (reverse recovery, for rectifiers), and If/Vrrm ratings. These live in partinfo for the common audio diodes.
Clipping is the act of limiting a signal's amplitude. A diode clipper does it by clamping the voltage: once the signal tries to exceed the diode's threshold, the diode conducts and holds the node near ±Vf. The corners it lops off are new harmonics -- that is the distortion.
Think of a clipper as a memoryless map y = f(x): for each input voltage, one output voltage. The shape of that map is the sound.
Two circuits dominate guitar/audio distortion, and they sound different for a structural reason, not just a diode reason. Where you put the diodes decides hard vs soft as much as which diodes you pick.
A series resistor feeds a node with anti-parallel diodes to ground. Below Vf the diodes are open and the signal passes; above it they conduct and clamp the node hard to ±Vf. Because the clamp is abrupt, this is the hard-clipping topology (ProCo Rat, DOD 250, MXR Distortion+).
R (sets how hard it slams the diodes) in ○──────[ R ]──────┬────────○ out │ ┌───────┼───────┐ │ │ │ ──►|── │ ──|◄── anti-parallel pair │ │ │ (one conducts each half) └───────┼───────┘ ═╪═ GND clamps the node to ±Vf · add diodes in SERIES per leg to raise the ceiling (n·Vf) mix a different diode on one leg → asymmetric
Put the diodes in the op-amp's feedback path, in parallel with the feedback resistor. At low levels the amp has full gain; as the output rises past Vf the diodes start conducting and progressively reduce the gain. The limiting is gradual, so this is the soft-clipping topology (Tube Screamer, Klon, Bluesbreaker). The output is the clean input plus a smoothly limited overdrive.
┌────────[ Rf ]────────┐ │ ┌───►|───┐ │ diodes across the ├─────┤ ├───────┤ feedback resistor: │ └───|◄───┘ │ the gain folds down │ │ past ±Vf in ○─[Ri]───┴──┤−╲ │ │ ╲ │ │ ►───────────────┴───○ out │ ╱ ref ○───────────┤+╱ softer because gain is reduced, not slammed · the dry signal is always present underneath
Less common but worth knowing: a diode in series with the signal chops off the part of the wave below its threshold (used in older fuzz and in envelope/rectifier stages). Adding a bias voltage to a shunt clipper shifts where on the wave the clip happens, giving gated or "sputtery" textures (velcro fuzz). Stacking N diodes in series in each leg multiplies the threshold to roughly N·Vf.
| Move | Effect |
|---|---|
| More diodes in series per leg | Higher threshold → louder, more headroom, less compression |
| Different diode on one leg | Asymmetric clip → even harmonics + DC offset |
| Smaller series R (shunt) | Harder slam, more aggressive clip |
| Small cap across the diodes | Rolls off fizzy high harmonics (a built-in treble tame) |
| Socket the diodes | Swap Si/Ge/LED by ear -- the classic "clipping options" mod |
Put the physics and the topology together and you can predict a sound before you build it. This is the map from the four junction properties to what you hear.
PROPERTY → WHAT YOU HEAR Vf (threshold) → loudness & headroom low Vf = early, quiet, squishy high Vf = late, loud, open knee (n, softness) → harmonic ORDER sharp = harsh/buzzy (high-order) soft = warm/round (low-order) symmetry → harmonic FLAVOUR symmetric = odd (hollow) asymmetric = +even (2nd, musical) junction cap Cj → treble high Cj = darker top end reverse leakage Ir → noise / bias high Ir = hiss, DC drift (Ge) topology → hard vs soft shunt-to-gnd = hard across-feedback = soft
| Pedal | Hard/Soft | Diodes | Placement | Character |
|---|---|---|---|---|
| Tube Screamer | soft | 2× 1N4148 | feedback | smooth, mid-hump, low gain |
| ProCo Rat | hard | 2× Si | shunt to gnd | aggressive, buzzy, cutting |
| MXR Distortion+ | soft-ish | 2× Ge 1N34A | shunt to gnd | warm, compressed |
| DOD 250 | soft-ish | 2× Si | shunt to gnd | brighter cousin of the + |
| Big Muff Pi | soft ×2 | 1N4148 | 2 feedback stages | huge sustain, very compressed |
| Klon Centaur | soft | 2× Ge 1N34A | feedback + clean blend | transparent, "still your amp" |
| Turbo Rat | hard | 2× LED | shunt to gnd | loud, open, more headroom |
Values and exact placements vary between revisions and clones; treat this as the classic configuration, not gospel for every unit.
The clipper is one job. The same junction does a half-dozen others in audio gear, and they are worth knowing because they show up when you open any real box.
Rectifiers turn AC into DC -- the bridge in every power supply (1N400x), and the "rectifier sag" that gives tube amps their bounce. Downstream of a signal, a diode + RC is an envelope follower: rectify the audio, smooth it, and you have its amplitude over time. That envelope is exactly what a compressor's sidechain, a noise gate, and an auto-wah run on -- which is why compressor attack/release measurement lives right next door to this.
envelope follower -- the guts of a comp/gate sidechain audio ○──►|──┬─────┬──────○ envelope (a slow DC that tracks level) rectify │ │ ═╪═ C █ R ← R·C sets the RELEASE; the diode+source R the ATTACK │ │ ═╪═ ═╪═ GND
A plain rectifier loses everything below Vf, useless for small signals. Put the diode inside an op-amp feedback loop and the amp's gain "divides out" the Vf, so it rectifies down to millivolts. This is how accurate meters, level detectors, and clean envelope followers are built.
| Role | How | Where in audio |
|---|---|---|
| Protection / clamp | diodes to the rails catch overvoltage/ESD; back-to-back across an input | op-amp inputs, jacks, phantom-power blocking |
| Ring modulator | 4 matched diodes in a ring + transformers multiply two signals | bells, robots, classic sci-fi (Dalek) tones |
| Varactor (varicap) | reverse-bias Cj becomes a voltage-controlled capacitor | VCOs, tuning, some analog chorus/vibrato |
| Voltage reference | Zener breakdown holds a fixed voltage | bias rails, regulators, clamp thresholds |
| Vactrol (LED + LDR) | LED light drives a resistor; not a junction effect but the LED's other life | lowpass gates, opto compressors, tremolo |
Because a diode clipper is (mostly) memoryless, it is one of the cleanest things to model in software: a single transfer curve y = f(x). That is also why it is the "hello world" of analog emulation -- the whole measure-model-null loop fits on one circuit.
WHITE BOX solve Shockley for the node -- physically true, needs the schematic diode current = Is(e^(V/nVt) − 1), Kirchhoff at the clip node, solved per sample (Newton-Raphson for the implicit loop) GRAY BOX hypothesize tanh-ish curve + a pre/post filter, FIT to measurements f(x) = threshold · tanh(x / threshold) ← Vf sets the threshold, the softer the real knee, the better tanh already fits BLACK BOX train a tiny neural net on input/output audio pairs (NAM-style)
The catch is the same one every non-linearity has: a sharp curve makes harmonics without limit, and everything above half the sample rate folds back as aliasing -- the inharmonic "digital fizz". Hard clipping is the worst case. The fixes are oversampling (run the curve at 4-16× rate, then filter down) or antiderivative anti-aliasing (ADAA). SPICE is the useful bridge here: simulate the real schematic to get the true curve, then fit the cheap runtime model to it and null-test the difference.
tanh, mind the aliasing.
Where to go deeper, and the parts in the offline reference.
# the common audio diodes, datasheet params on hand partinfo 1n4148 # silicon signal -- the default clipper partinfo 1n34a # germanium -- warm, soft, leaky partinfo 1n5817 # schottky -- low Vf, fast partinfo search diode # the whole family in the db