Thyristors, triacs and transistor switches in sewing-machine speed controls. A companion to Speed Control of Universal Motors in Sewing Machines
Development Build v82, November 2026
About this paper. The text, analysis, figures and simulations were written by Claude, an AI model made by Anthropic, at the direction of Robert Drzyzgula, who set its scope and questions. It is the second companion to Speed Control of Universal Motors in Sewing Machines (cited here as “the speed-control paper”). The first companion, Inside the Motor, looks inside the motors. This one looks inside the switches that control them. The sources are listed in the last section; please report errors as issues on the project’s GitHub repository.
What it covers. The speed-control paper treats a phase-control pedal as a box that chops the line: it sets a firing angle, and the motor gets part of each half-cycle. This paper opens the box.
The short answers.
On the simulations. The firing curves are computed from a simple model of the timing network, not measured. It uses a 32 V breakover, 22 V left on the capacitor after firing, a 120 V line and a resistive load. The voltage left after firing is the least certain of these. The motor test rig is meant to measure the real board, and its results should replace these figures.
The silicon-controlled rectifier, or SCR, is the oldest of the power switches in this paper and the parent of the triac. “Thyristor” is the family name; in everyday use it means the SCR. Bell Labs worked out the four-layer PNPN device in the mid-1950s, and General Electric brought the SCR to market in 1957. Within a few years it had replaced the thyratron tube and the magnetic amplifier in motor drives, lamp dimmers and battery chargers. It is still the cheapest way to switch large currents from an AC line.
An SCR is a sandwich of four alternately doped layers: P, N, P, N. The outer P layer is the anode, the outer N layer the cathode, and the gate connects to the inner P layer next to the cathode.
The structure has three junctions, J1, J2 and J3. With the anode positive, J1 and J3 are forward biased, but J2 in the middle is reverse biased and blocks. With the anode negative, J1 and J3 block. So an SCR that has not been triggered blocks in both directions.
The key to the device is to see the four layers as two transistors that share their middle layers, as the figure shows: a PNP made of the first three layers and an NPN made of the last three. The PNP’s collector is the NPN’s base, and the NPN’s collector is the PNP’s base. Each transistor’s output current is the other’s input.
Call the two transistors’ common-base current gains (PNP) and (NPN), the fraction of each one’s emitter current that reaches its collector. Summing the currents at the junctions gives the anode current:
Here is the gate current and the terms are the two transistors’ small leakage currents. In silicon, is small at tiny currents and rises with current. With no gate current, the sum is well below 1 and only leakage flows: the SCR is off. A pulse of gate current raises the NPN’s current and so its . That sends more current into the PNP’s base, which raises its , and so on. When reaches 1, the denominator goes to zero: the loop gain is one, and the current is limited only by the external circuit. The SCR has switched on, and it stays on with no gate current at all.
Why should rise with current? At low current densities, much of the emitter current is lost to recombination, at defects in the junctions’ depletion regions and at the surface, before it can cross the base. That loss grows only slowly with current, so as the current rises, a larger share of it gets through. Designers use this. The emitters are shorted to the P layer next to them in places, so that small currents (leakage, or the displacement current of a fast voltage edge) bypass the emitter entirely and can’t start the loop. Only a real gate current, or a large enough fault, raises the gains to one.
The layers’ doping sets the voltages. The N1 layer next to the anode is thick and lightly doped, because the blocking junction’s depletion region spreads into it: the thicker and purer it is, the higher the voltage it holds off. It is also where the stored charge sits when the device conducts, which is why high-voltage thyristors are slower to turn off.
Two currents describe the latch:
The gate can start the regenerative loop but can’t stop it. A negative gate current large enough to steal the loop’s own current would turn it off, but for an ordinary SCR that is about the whole anode current. The gate turn-off thyristor (GTO) is built to make that practical; §7 comes back to it.
So an SCR is turned off by commutation: something outside it brings its current to zero. On an AC line, the line does that every half-cycle (“natural” or “line” commutation). After the current stops, the SCR also needs a short time with no forward voltage, its turn-off time , tens of microseconds for a phase-control part, to sweep the stored charge out of its middle layers. Apply forward voltage sooner and it turns back on. Early DC choppers and inverters had to force commutation with extra capacitors and SCRs, and the circuitry was elaborate; transistor switches retired them (§7).
Any current that drives the loop gain to one fires an SCR. The gate is the intended way, but there are others:
There is also a limit on how fast the current may rise once it fires, . Conduction starts in a small region near the gate and spreads across the chip over microseconds. A current that rises too fast crowds into that region and can burn it.
An SCR conducts in one direction only, so on AC it gives controlled rectification: DC whose average is set by the firing angle. The classic uses:
For a sewing-machine pedal, though, the useful step was putting two SCRs into one chip with one gate: the triac.
A triac is, in effect, two SCRs connected inverse-parallel and built into a single piece of silicon with a single gate. General Electric introduced it in the 1960s, for lamp dimmers and appliance controls. Because it conducts either way, its main terminals aren’t called anode and cathode. They are MT1, the terminal the gate is referenced to, and MT2. On a TO-220 package such as the BT136, MT2 is the metal tab, so the tab is live.
Inside, it is one chip with two SCRs folded together. The figure shows the layout. The thick, lightly doped N1 layer in the middle blocks the voltage in both directions, as in an SCR. Above and below it are P2 and P1. The N-type emitters are let into the outer faces at different places:
Each terminal’s metal touches both an N emitter and the P layer beside it. That is a deliberate short. It keeps the emitters from injecting until a real gate or load current flows, which is what sets the triac’s sensitivity and its dV/dt rating.
The triac behaves as described for the SCR: it latches, the gate can’t turn it off, and it turns off when the current through it falls below the holding current. So all a control circuit can choose is when to switch it on in each half-cycle. That moment is the firing angle , measured from the voltage zero. This is leading-edge (or forward-phase) control:
With a resistive load the current reaches zero with the voltage. A motor is inductive, so its current lags and the triac stays on a little past the voltage zero. Inside the Motor (§3 there) shows the motor current behind a triac in detail.
A triac can be fired with gate current of either sign, whichever way MT2 is biased. That gives four combinations, called quadrants:
| Quadrant | MT2, relative to MT1 | Gate current | BT136-600E: gate current to fire (typ / max) |
|---|---|---|---|
| I | positive | positive, into the gate | 2.5 / 10 mA |
| II | positive | negative, out of the gate | 4 / 10 mA |
| III | negative | negative | 5 / 10 mA |
| IV | negative | positive | 11 / 25 mA |
The four work by different routes through the structure, which is why they aren’t equally easy:
Circuits avoid quadrant IV where they can:
In an inverse-parallel pair, each SCR conducts for one half-cycle and then has the whole next half-cycle, 8 ms, to recover while the other one carries the current. In a triac the same silicon carries both directions. When the current falls to zero at the end of one half-cycle, the region that just conducted is still full of carriers. The voltage across the triac is about to rise in the opposite direction, and that direction belongs to the other half of the same chip. If the voltage rises fast enough, the leftover carriers act like gate current and the triac turns straight back on, now in the other direction.
So triac datasheets add two ratings:
With a resistive load the voltage rises gently from zero after the current stops, and neither rating matters. With a motor, the current reaches zero after the voltage, so the line is already tens of volts into the next half-cycle. That voltage appears across the triac within microseconds. This is the case the snubber in §5 is for.
A diac is a three-layer, two-terminal trigger, symmetrical, with no gate. In effect it is a transistor with no base connection, built with both junctions alike so that it works the same either way round.
With voltage across it, one junction is forward-biased and the other reverse-biased. Almost nothing flows until the reverse-biased junction reaches its avalanche voltage, about 32 V for the common DB3 (28–36 V). The avalanche current flows into the floating middle layer, where it acts as base current. The transistor’s gain rises with current, as an SCR’s does (§2), so the more current flows, the less voltage it takes to keep it flowing. That is a negative resistance: past breakover, the voltage falls back by at least 5 V (the “breakback”) as the current rises, until the device is conducting heavily at a lower voltage.
That snap is the point. A capacitor charged to 32 V dumps part of its charge through the diac into the triac’s gate as a short, sharp pulse of current, far more than the gate needs. Unlike an SCR, a diac doesn’t latch: it has no fourth layer, so when the capacitor’s current falls away, it stops conducting.
Why not feed the gate straight from the capacitor through a resistor? The gate fires at a fraction of a volt and a few milliamps, so it would fire on a slowly rising, poorly defined current that varies with temperature and from one triac to the next. The diac waits for a definite voltage, then delivers a definite pulse. The firing point is set by the RC network and the diac, not by the triac’s gate sensitivity.
The figure strips a phase-control pedal to its essentials, without the extra parts of a real board such as the donor board in §6.
The triac sits in series with the motor, as §3 describes. While it is off, it holds off the whole line and the motor gets nothing.
The timing network is VR, R and C, connected across the triac. While the triac is off, the network sees the whole line voltage. The motor in series drops almost nothing, since the network draws well under a milliamp. C charges through VR + R, and its voltage lags the line and is smaller than it. In steady state it is the line voltage divided by and delayed by :
Here is VR + R, and . With 106 kΩ and 0.1 µF, , so C swings to 41 V peak, lagging the line by 76°. Turning VR up both shrinks this voltage and delays it, so it reaches the diac’s breakover later in the half-cycle. Past a certain point its peak no longer reaches the breakover at all, and nothing fires. On 120 V, setting V gives . That is ms, about 137 kΩ with 0.1 µF. It is where a pedal’s motor starts.
The diac fires the triac when C reaches about 32 V, in quadrant I on the positive half-cycle and quadrant III on the negative (§3).
R, in series with VR, keeps the total resistance from reaching zero. It limits the current through the pot’s wiper and track at the full-speed end, and so sets the earliest firing. Even with VR at zero, C still has to reach 32 V while the line rises, so the circuit fires a little after the line passes 32 V: about 11° into the half-cycle, essentially full voltage.
Each half-cycle, C charges until it reaches the diac’s breakover. The diac then fires the triac, and the triac conducts until the motor current falls to zero. That happens a little after the voltage zero, because the motor is inductive. Delaying the firing angle reduces the share of each half-cycle applied to the motor. For a resistive load, the RMS output is:
The motor’s inductance stretches the conduction past the voltage zero, so the real curve differs, but the shape holds: a smooth fall from full voltage at to nothing at .
The bottom panel of the figure shows that the circuit has a memory. When the diac fires, it leaves C at about 22 V. With the triac on, the network sees only the triac’s volt or so across it, so C drains slowly through VR + R. It still holds about 15 V of the same sign when the line reverses. The next half-cycle has to pull it back through zero before charging it the other way. A capacitor that has not been firing starts each half-cycle further back: at about −30 V near the start threshold, the trough of the plain RC’s lagging sine.
That difference is why a simple pedal snaps on. The setting that starts the motor is not the one that stops it:
So the pedal needs a push to start, the motor jumps to a speed above where you wanted it, and then it will run slower than it would start. The jump happens near 125° whatever the pot’s value, because it comes from the capacitor’s memory. A different pot or trimmer only moves where in the pedal’s travel it happens (§6).
Two-stage networks reduce the jump. A second resistor and capacitor feed the timing capacitor from a node that is itself an RC delay of the line. After each firing, that second capacitor recharges the timing capacitor from a defined voltage, so the timing capacitor starts each half-cycle from nearly the same place whether or not the last one fired. The memory, and with it most of the hysteresis, is erased. Lamp-dimmer and tool-speed application notes have used this since the 1960s; better pedals do it, and cheap ones don’t.
Before the triac fires, the RC network charges from the voltage across the triac. That is the line voltage minus the small back-EMF the turning armature generates from residual field magnetism. A slowing motor makes less back-EMF, so the capacitor charges slightly faster and fires earlier, which speeds the motor back up. The effect is small in a simple pedal. Exploiting it deliberately, with the network arranged to sense the armature’s voltage, is the basis of the regulated designs in the speed-control paper (§7 there).
The simulation is in tools/make_triac_figures.py. Within
each half-cycle it solves the RC network exactly (a capacitor charged
through a resistor from a sine), and it carries the capacitor’s voltage
from one half-cycle to the next, so the start and stop thresholds come
out of the model rather than being put in. Its assumptions:
The snubber, Rs in series with Cs straight across the triac (the dashed box in §4’s figure), is there for the moment the triac turns off. §3 explained the problem: a fast-rising voltage across a triac can turn it on, either just after it has conducted (commutating dV/dt) or at any time (static dV/dt).
Commutation. With a resistive load, the current reaches zero together with the voltage, so the voltage across the triac grows gently from zero. A motor is inductive: its current lags. By the time the current has fallen to zero and the triac turns off, the line is already some way into the next half-cycle, at tens of volts of the opposite polarity. That voltage appears across the triac at once, limited only by stray capacitance: a step of many volts per microsecond. The triac, still full of charge carriers, turns straight back on and conducts the whole next half-cycle, whatever the timing network wanted. Control is lost, and the motor runs on at full speed or lurches.
Static dV/dt. Even a triac that has been off for some time can be fired by a fast enough edge, through its own internal capacitance. A spike on the line or the electrical noise from a motor’s brushes can do it.
Cs has to charge through Rs and the motor, so the voltage across the triac can rise no faster than that RC allows. The edge is slowed from a fraction of a microsecond to tens of microseconds. Rs limits the current when the triac fires and Cs dumps its charge into it, and it damps the ringing between Cs and the motor’s inductance. Typical values for a small motor are 0.1 µF (X2-rated, for use across the line) and a few hundred ohms to 1 kΩ. The donor board’s snubber is 1 kΩ and 0.1 µF.
A snubber has one cost: Cs passes current to the motor even with the triac off. At 60 Hz, 0.1 µF is about 27 kΩ, which lets about 4 mA through. That is enough to light a neon lamp but not to turn a motor. It does mean a snubbed control is never fully off; only a switch is. The VSM Base’s knee controller has one.
Some triacs are built to switch inductive loads off without help. ST calls these “Snubberless” (its BW and CW versions), and NXP and WeEn call theirs “Hi-Com”. They are the three-quadrant types of §3. ST’s datasheet rates the BTA08-600BW, chosen for the MCU design in the speed-control paper, to turn off a current falling at 7 A/ms with no snubber, and to withstand 1,000 V/µs while off. A sewing motor’s current falls far more slowly than that. At 1 A RMS, its 1.4 A peak falls at no more than A/ms. The price is a less sensitive gate: the BW version needs up to 50 mA to fire and fires only in quadrants I–III. The donor board’s BT136-600E is an ordinary sensitive-gate triac: it fires on 10 mA, its datasheet gives 50 V/µs static and no commutating figure, and the board carries a snubber.
The MCU design uses a snubberless triac but leaves room for a snubber. There are three reasons it might be needed:
So the plan is to build without a snubber, measure, and fit one if the triac turns on by itself or fails to latch.
The VSM Base’s first knee controller reuses an inexpensive foot-pedal board, the XWW KB3151C (75 × 35 mm, labeled 110–120 V, 1.0 A), with a knee lever turning a pot in place of the pedal. The speed-control paper (§11 there) describes the controller and its mechanism. This section looks at the board’s circuit.
What is on it. It is §4’s minimal circuit with a few additions:
It is a two-terminal series device. It has only two spade tabs and sits in series in one leg of the motor circuit, with no separate mains input or motor output. That works because, as §4 showed, the timing network needs only the voltage across the triac. In the original pedal, only this control loop enters the box, as a two-wire lead from a standard cordset, and one pedal fits many machines.
At the end of the slide pot’s travel there is an off zone, where the wiper runs off the track and nothing fires. That is the pedal at rest.
The board’s timing network is the one in §4’s figure, with one addition: the 2 MΩ trimmer RV2 is connected around the slide pot RV1 (200 kΩ). As traced, RV2 joins RV1’s pin 1, which is tied to MT2, to RV1’s pin-3 end. So current from MT2 can reach RV1’s wiper by two paths. One runs directly along RV1’s track from pin 1 to the wiper. The other runs through RV2 to pin 3, then back along the rest of the track to the wiper. The wiper feeds C2 through R2.
Which of RV1’s pins is the wiper is still to be checked with a meter. The algebra below doesn’t depend on it.
The parallel resistance. Let be RV1’s whole track (200 kΩ), the section from pin 1 to the wiper, and the rest. Let be RV2’s setting, 0 to 2 MΩ. The direct path is , and the other is . In parallel:
C2 (0.1 µF) charges through plus R2 (1 kΩ). Pressing the pedal moves the wiper from the pin-3 end toward pin 1, so falls from to 0. Three things follow from the formula:
Firing. §4’s two thresholds turn these resistances into speeds. With C2 = 0.1 µF on 120 V, the motor starts when the timing resistance falls below about 137 kΩ, and it settles at about 125° firing, roughly 40% of the line’s RMS voltage. Once running, it keeps firing until the resistance rises past about 158 kΩ. The next figure runs the board’s network through the simulation of §4, pressing the pedal from rest and then easing back. It plots motor voltage as a stand-in for speed: at light load, a universal motor’s speed rises with its voltage, though not in proportion.
| RV2 | at the start of travel | Dead travel | Voltage when it starts | Lowest voltage while running |
|---|---|---|---|---|
| 2 MΩ | 182 kΩ | 27% | 38% | 9%, at 15% of travel |
| 1 MΩ | 167 kΩ | 21% | 38% | 7%, at 7% of travel |
| about 430 kΩ | 137 kΩ | none | 39% | 39%, at the off zone |
| 200 kΩ | 100 kΩ | none | 66% | 66% |
| 100 kΩ | 67 kΩ | none | 83% | 80%, a quarter of the way in |
| 0 | 0 | none | 100% | 90%, halfway |
What a sewist notices. RV2 is the board’s minimum-speed adjustment, and turning it goes through four distinct behaviors.
What RV2 cannot do is lower the snap-on speed. In every setting with dead travel, the motor starts at about 125° firing, near 40% of line voltage. RV2 only moves where in the travel that happens. The jump comes from the timing capacitor’s memory (§4), so a gentler start needs a different network, such as §4’s two-stage RC, not a different setting.
The start threshold also varies from board to board. A DB3’s breakover can be anywhere from 28 to 36 V, which alone moves the threshold from about 157 kΩ down to 121 kΩ. C2’s tolerance and the line voltage move it further. So the 430 kΩ setting is a model’s figure, not a dial position. That is why RV2 is a trimmer, set by ear: back it off until the motor just starts as the pedal leaves the off zone.
In the knee controller, the slide pot is replaced by the Bourns pot, wired as a two-terminal resistor, with RV1’s pin-3 pad jumpered to the wiper pad. That puts RV2 directly across the pot’s working section, so , which always rises with . The off switch covers the dead travel up to the switch point, about 1” of knee travel. There the pot’s section is about 220 kΩ, and it falls to zero at full travel. With RV2 at 2 MΩ, the motor starts when falls to about 147 kΩ, a third of the way into the speed inch. At about 360 kΩ, which makes 220 kΩ ∥ RV2 = 137 kΩ, it starts as the switch closes. The same caveat applies: set it by ear.
A thyristor can only be switched on; the circuit has to switch it off. That suits AC phase control, where the line does the switching off for free, and little else. Everything that chops DC, which includes every brake, every PWM motor drive and every inverter, needs a switch its control terminal can turn off as well as on. Three families do that job: bipolar transistors, power MOSFETs and IGBTs.
A bipolar transistor conducts while current flows into its base, and stops when it doesn’t. That’s the turn-off a thyristor lacks, but it is expensive: a power transistor’s current gain is low, often 10 or 20, so switching 10 A takes most of an amp of continuous base drive. Darlington pairs raise the gain but add a volt or more of drop. Power bipolars ran the first transistor motor drives and switching supplies of the 1970s and 1980s. They have since been almost entirely replaced by the next two devices.
A MOSFET’s gate is insulated from the channel by a thin oxide, so it draws no steady current. It is voltage-driven: about 10 V on the gate (4–5 V for a “logic-level” part) turns the channel fully on, and 0 V turns it off. The gate is a capacitor, so switching it takes a burst of current to move the gate charge, and a gate driver supplies that. Power MOSFETs reached the market in the late 1970s.
Its character:
MOSFETs rule at low voltage. A cordless drill’s speed trigger is a MOSFET switching the battery onto the motor at a few kilohertz, which is the whine Inside the Motor measured (§5 there). So is the H-bridge on the Pololu G2 board in the DC-motor variant of the MCU controller.
The insulated-gate bipolar transistor, developed in the early 1980s (notably by B. Jayant Baliga at General Electric), combines the two. A MOSFET section at the gate drives a bipolar structure underneath, built on an extra P layer. When it conducts, that layer floods the thick, lightly doped region that blocks the high voltage with carriers, a process called conductivity modulation, and the region’s resistance collapses.
IGBTs rule at mains-derived voltages and kilowatts:
A switch that can turn off allows a second way to chop the line. Leading-edge (forward-phase) control, the only kind a triac can do, starts conduction late in each half-cycle. Trailing-edge (reverse-phase) control starts at the zero and cuts off part way through. It needs a MOSFET or IGBT pair, one for each direction, or one inside a diode bridge.
Which suits a load depends on what it does at the edge:
On DC the freewheeling path is easy: a diode across the motor, or the body diodes of an H-bridge. With it, a transistor switch can chop the supply at any frequency and duty cycle, which is pulse-width modulation. The motor’s inductance smooths the current between pulses. Inside the Motor shows the waveforms (§5 there).
| SCR | Triac | Power MOSFET | IGBT | |
|---|---|---|---|---|
| Turned off by | the circuit (current zero) | the circuit (current zero) | its gate | its gate |
| Conducts | one way | both ways | one way, plus body diode | one way |
| Gate drive | current pulse | current pulse, either sign | voltage, gate charge | voltage, gate charge |
| On-state loss | about 1–1.5 V drop | about 1–1.5 V drop | about 1.5–2 V drop | |
| Switching speed | line frequency | line frequency | tens of ns; up to hundreds of kHz | about 1 µs tail; up to tens of kHz |
| Typical home | controlled rectifiers, big AC switches | pedals, dimmers, appliance motors | battery tools, DC drives, power supplies | VFDs, servos, cooktops, EVs |
Every family in this paper turns up somewhere in the VSM Base’s controllers or in the controllers the speed-control paper compares.
The device physics in §2, §3 and §7, the two-transistor model of the
thyristor included, is standard textbook material, given here without
citation line by line; Mohan, Undeland and Robbins is the standard
engineering text. The firing simulation and the figures are generated by
tools/make_triac_figures.py in the project repository. The
KB3151C’s schematic was traced from photographs of the board; its KiCad
source is in hardware/triac-board/.