That said, very few circuits have attempted to use active MOSFETs as clipping devices.
More often than not, the gate is shorted to the source so that the channel can never turn on and only conducts when reverse biased through the body diode, which is just a diode and won’t receive further attention.
One exception seems to be the popular Zendrive, which instead connects the gate to the drain, so that the channel can eventually turn on; at least in those schematics that don’t mistakenly redraw it with the gate again connected to the source. In this configuration, VGS is equal to VDS and the FET will ride on the “knee” of the characteristic curves, before the linear region transitions in the saturation region.
One issue with that is the high voltage required to turn on the MOSFET, which doesn’t pair well with the limited swing of most op-amps at 9v (about 3V peak), if you also add the voltage drop from the series Schottky diodes. The op-amp will run out of headroom before the MOSFET has a chance to work.
Or will it? We might be used to think about MOSFETs turning “on” with about 4-5V of gate voltage when used as switches, the current involved in feedback clipping is tiny (input voltage/grounded feedback resistor), and voltages just above Vth are already enough to fully conduct. This means a G-to-D MOSFET will clip at about 1.6-1.8V by itself, but some series diode must always be added to prevent reverse bias (unless that’s what you’re going for).
These considerations are what brought me to my new applications of active MOSFET limiting. I haven’t found a similar use elsewhere, other than it being a simple application of basic theory.
Indeed, due to the exponential curve of the diode and the quadratic curve of the MOS diode, that’s exactly what happens: MOSFETs start with a steeper slope than diodes at low currents, before being overtaken. We can see that for MOSFETs to be softer, they need to go above 1mA. We can calculate that this won’t happen even with diodes clamping the output of the op-amp, unless the series resistor is as small as 1kΩ and the op-amp output is already hitting its rails adding another layer of saturation. Does this mean that MOSFET clipping, even when done right, is a lie? Softer is better, right?
I’m not here to discuss the merits, or lack thereof, of MOSFET clipping or to argue with the taste of their users. I usually assume people know what they like, even if they might not know why. People seem to like MOSFET clipping, even just as an idea, and I’m here to explore possibilities.
What deviates from the usual formula are the MOSFET clippers and the associated circuitry. This represents a practical application of the “biased gate”: as in the G-to-D, the gate gets the full signal voltage, ensured by the large cap and bias resistor; on top of that though, the gate has applied a small positive DC voltage with respect to the reference.
To ensure the bias voltage is mostly independent from the supply, since a tenth of a V matters, it is derived by the same reference voltage the source refers to (with a diode drop from the series forward-biased body diode) with two series diodes, which lift it by about 0.9V at these currents. The small rise of the reference voltage, by few hundred mV, caused by this is small enough that it can be ignored and won’t cause premature clipping, and even the next standard value (12k) for the divider is too much to compensate for it.
The resulting circuit will soft clip at about 1.4V peak, 0.9V from the MOSFET and 0.5V from the body diode; but one of the cool aspects of this technique is that the threshold can be adjusted at will by changing the bias voltage: it’s perfectly possible to make it larger, smaller, or even smaller than a diode drop if the MOSFETs are configured in parallel instead of series. I’ve chosen the series configuration here because it’s more reliable: the body diode in series with the active MOS makes sure there’s no reverse bias, as it would be the case for parallel with not enough bias (classic body diode clipping); since that means ensuring a low enough threshold for the parallel case, it’s just as easy with parameter spread to end up with a MOS conducting at all time for the fanciest voltage follower of all.
If you’re curious about the name, it comes from green being the color of overdrive and turtles being the most beloved animals in Elden Ring (which means you can also call the circuit Green Dog).
Here’s a suggested application: a variable threshold, stereo, unity gain limiter intended for line level signals. The Schottky diodes set the minimum threshold achievable, while the maximum is actually higher than Vth because the potentiometer allows to bias the gates negative. The bias voltage is derived by a Zener diode so it’s not supply dependent, and varies between 3V and 6V, which works well with the typical Vth; attention should be paid to matching between the two channels and trimming possibly allowed to account for large Vth variations. Another possible variation is to amplify the gate voltage with respect to the drain: this reduces the threshold the same way as biasing does, but also makes the transition more abrupt. I didn’t explore this further because it also requires extra components.
By the same reasoning, attenuating the gate signal should mean the conduction threshold is apparently higher, and the transition possibly smoother. I didn’t include this technique in the Green Turtle because it would mean biasing closer to conduction to begin with, to keep the threshold low enough, with all the risks of biasing open because of parameter spread; I also wasn’t able to get it to work as desired in simulation.