rf generator circuit
An RF generator circuit is the power-conversion stage that turns direct current from a supply rail into a high-frequency wave of controlled amplitude and frequency, and it is the physical machine behind every radio-frequency plasma tool, induction heater, and industrial source. In a semiconductor fabrication facility the generator drives the plasma that etches, deposits, and strips films, so the circuit must deliver hundreds or thousands of watts into a load that changes as the plasma ignites and drifts. The circuit is not a single amplifier but a chain: a low-power oscillator or frequency source, a driver that raises the signal to a usable level, a high-efficiency power stage that lifts it to the rated output, and a matching network that transforms the impedance of the plasma load so the final stage sees the impedance it was designed around. Because the load moves, the circuit also carries a sensing and control loop that reads reflected power and adjusts the tuning so the delivered power stays at its setpoint, and it is this combination of power conversion and feedback that makes a modern generator a small closed-loop system rather than a bare amplifier.
**An RF generator circuit is a chain of stages, not a single block.** The signal path begins in an oscillator or a phase-locked loop that produces a stable tone at the operating frequency, then passes through a buffer and driver amplifier that raises it from milliwatts to a level the final stage can accept, and finally through the power stage that delivers the rated output. Each stage has a different job: the source sets frequency and stability, the driver supplies the gain and voltage swing, and the final stage handles the current and heat. In a plasma generator the operating frequency is fixed by regulation, so the source is locked to a crystal reference and the control loop adjusts amplitude and impedance rather than frequency.
**The final stage is built for efficiency, not for linearity.** Because the generator spends its working life delivering essentially full power into a resonant load, the power stage is run as a switch-mode amplifier rather than a linear one. A switch-mode stage turns the transistor fully on and fully off, so the transistor rarely carries current and voltage at the same time, which is what allows it to reach an efficiency near 90 percent instead of the 60 percent of a linear class-AB stage. The higher efficiency matters at kilowatt power levels, where the difference between 90 and 60 percent is hundreds of watts that would otherwise be heat, requiring bigger heatsinks, more cooling, and a larger enclosure. This is why industrial generators run switch-mode stages and reserve linear stages for low-power or amplitude-modulated applications.
**The output stage sees the load through a matching network, and the two must agree.** The transistor inside the power stage is designed to deliver its rated power into a specific low impedance, typically a few ohms, while the plasma chamber and its cable present a nominal 50 ohm load. A matching network bridges the gap by transforming the 50 ohm load so the transistor sees its design impedance, and the most common form is a two-element L-network. For a stage with a 3 ohm design impedance driving a 50 ohm load, the network has a loaded quality factor of 3.96, a series element of 11.9 ohms that can be built from a 139.4 nanobenry inductor, and a parallel element of 12.6 ohms that can be built from a 929.1 picofarad capacitor. The network is deliberately narrow-band, so it filters the switching harmonics of the square-wave drive and leaves a clean sine at the output.
**The control loop is what keeps the generator safe when the load refuses to stay put.** Because a plasma changes its impedance as it ignites and drifts, the generator cannot simply be turned on and left to run. A directional coupler on the output samples the forward wave and the reflected wave, a control circuit turns that ratio into a standing wave ratio and a reflected-power figure, and the loop then acts to keep the delivered power at its setpoint. When the reflected power rises toward a limit the loop folds the forward power back or retunes the match, and it is this closed loop that protects the switch-mode stage from the very real possibility that a drifting plasma reflects the full output back into the transistors.
**The matching network is the part that has to move as the plasma drifts.** When a plasma ignites, its impedance changes from nearly an open circuit to a low, lossy value, and it keeps drifting as the process runs, so the matching network cannot be fixed. A matching network in a generator is therefore tunable, usually a variable capacitor and inductor adjusted by motors or by fixed elements switched in and out, and the control loop steers it to minimize reflected power in real time. This is the same matching problem solved on a Smith chart, but here it is automated: the generator measures the forward and reflected wave, computes the impedance, and rotates the tuning elements to move the operating point back toward the center. The speed of this adjustment is what separates a modern frequency-agile generator, which can retune in milliseconds, from an older one that moved heavy vacuum capacitors over the course of a second.
```flowchart
flowchart TD
A[Crystal oscillator / PLL at 13.56 MHz] --> B[Buffer and driver amplifier]
B --> C[Switch-mode power stage (Class-D/E, ~90% eff)]
C --> D[L-match network: 3 ohm to 50 ohm]
D --> E[Directional coupler samples forward and reflected wave]
E --> F[Control loop computes delivered power and VSWR]
F --> G{Reflected power below limit?}
G -- yes --> H[Hold setpoint, deliver to plasma]
G -- no --> I[Adjust tunable match elements / fold back power]
I --> F
```
The table below connects the amplifier class to its efficiency and to what a kilowatt generator must supply on the rail, and it turns the L-match elements into the component values that an engineer would order. All figures assume the 13.56 MHz plasma band.
| Amplifier class | Efficiency | DC in for 1000 W out | L-match Q (3 to 50 Ω) | Series X (Ω) | Parallel X (Ω) |
|---|---|---|---|---|---|
| Class-AB | 60% | 1667 W | 3.96 | 11.9 | 12.6 |
| Class-D/E | 90% | 1111 W | 3.96 | 11.9 | 12.6 |
The arithmetic that sizes the matching network is compact, and it is the same L-match mathematics used across radio-frequency design. A two-element network transforms a low device impedance to a high load impedance with a quality factor set by the ratio of the two, and the two reactances follow directly.
$$Q = \sqrt{\frac{R_{load}}{R_{device}} - 1}, \qquad X_{series} = Q \cdot R_{device}, \qquad X_{parallel} = \frac{R_{load}}{Q}$$
For a 3 ohm device optimum and a 50 ohm load, the quality factor is the square root of fifty over three minus one, which is 3.96, and the series reactance is 3.96 times three, which is 11.9 ohms. A series element of that reactance at 13.56 MHz is an inductor of 139.4 nanobenries, and the parallel reactance of 12.6 ohms is a capacitor of 929.1 picofarads.
$$L_{series} = \frac{X_{series}}{\omega}, \qquad C_{parallel} = \frac{1}{\omega \, X_{parallel}}, \qquad \omega = 2\pi f$$
The efficiency that decides the heat load is the ratio of radio-frequency output to the direct-current input, and a switch-mode stage holds it high because the transistor is either fully on or fully off.
$$\eta = \frac{P_{RF}}{P_{DC}}, \qquad P_{DC} = \frac{P_{RF}}{\eta}$$
A generator delivering 1000 W at 90 percent efficiency draws 1111 W from its rail, while a 60 percent stage draws 1667 W, and the 556 W difference is heat that the enclosure must remove. The power that a mismatch reflects follows the reflection coefficient, so at a standing wave ratio of 2 to 1 the reflected coefficient is 0.333, and 11.1 percent of the forward power, 111 W of a 1000 W forward wave, comes back toward the final stage, leaving 889 W delivered. Because a switch-mode stage is not designed to absorb that reflected wave, the control loop watches the directional coupler and folds the power back or retunes before the return power exceeds what the transistors can survive.
The components that make up a modern generator are the everyday parts of the power electronics industry. The final stage is almost always a laterally diffused metal-oxide-semiconductor transistor, an LDMOS device built to switch at high voltage and high frequency, run from a rail that can be as high as 50 V, with a design impedance near 3 ohms. The tuning network uses vacuum or air-variable capacitors that can swing their value by a large ratio under a control signal, and the directional coupler that samples the forward and reflected wave is a short transmission-line section with a pair of coupled lines. The whole assembly is built to sit next to the plasma chamber, where the output is connected by a short run of 50 ohm coaxial cable; at 13.56 MHz the wavelength is 22.12 meters, so a quarter-wave section is 5.53 meters, and the physical network is a small fraction of that length, which is why a lumped L-match is practical at this frequency rather than a transmission-line transformer.
The instruments and suppliers that support generator design are the same names found across radio-frequency power work. Keysight and Rohde & Schwarz analyzers characterize the matching network and the load, and vector network analyzers plot the impedance of the chamber so the tuning range of the network can be set to cover it. Anritsu and Bird instruments measure forward and reflected power in the field, and Belden and Times Microwave supply the 50 ohm cable and SMA and N-type connectors that carry the output. A semiconductor tool maker sizes the L-match, chooses the LDMOS stage, and selects a tuning range wide enough to follow the plasma, and the numbers that matter are the ones this treatment derives: the 3.96 quality factor, the 11.9 ohm series element, the 12.6 ohm parallel element, and the 90 percent efficiency that keeps a kilowatt generator from melting its own enclosure.
The numbers that make the circuit concrete are easy to remember once they are tied to hardware. A 1000 W generator at 90 percent efficiency draws 1111 W from the rail and rejects only 111 W as heat, while the same generator built in class-AB would draw 1667 W and reject 667 W, a difference that decides the difference between an air-cooled and a liquid-cooled chassis. At a standing wave ratio of 2 to 1, the same generator sees 111 W of reflected power out of its 1000 W forward wave, and at a ratio of 3 to 1 that jumps to 250 W reflected with only 750 W delivered. The L-match that transforms the 3 ohm device to the 50 ohm load uses a 139.4 nanobenry series inductor and a 929.1 picofarad parallel capacitor, and when the plasma drifts the tunable element swings to keep the reflected power under the protection limit, which on a 50 V LDMOS rail is set so the return wave does not push the device voltage past its rating. Across a 915 MHz induction band the same L-match mathematics applies, but the wavelength of 328 millimeters makes the network smaller and the component values correspondingly tighter, so a lumped network is replaced by printed or distributed elements.
Read the RF generator circuit through a *power-conversion* lens rather than a *signal* lens: the circuit exists to move hundreds of watts from a direct-current rail into a plasma with the least possible waste, and every element in it is sized by efficiency, impedance transformation, and the heat that a mismatch can dump into the final stage. An engineer who reads the generator as an amplifier that merely makes a signal is missing the real constraint, which is that the transistor must be protected from the power it reflects when the plasma drifts. The professional habit is to read the amplifier class for the efficiency, to read the L-match quality factor for the bandwidth and the component values, and to know that the 90 percent efficiency, the 3.96 quality factor, the 11.9 ohm series element, and the 111 W reflected at a 2 to 1 standing wave ratio are not separate facts but one circuit viewed through the physics of power conversion.