john_ellis said:
SiCED produces 1200V SiC FETs of the static induction type. They are used to cascode a MOSFET, as I recall. Being SiC, they would probably withstand a pretty amazing junction temperature.
Anyway, as for characteristics, SITs are essentially vertical JFETs with a buried gate that usually takes the form of a grid or slits, making them pretty close to a triode in structure. It also makes them behave somewhat differently to a classic JFET, as you effectively have a huge number of smaller channels, rather than a single larger one. The earliest examples are the Sony VFETs, and those were pretty linear, but the technology has advanced since then.
You can vary the characteristics pretty much any way you like, with some compromises.
Breakdown voltage is tuned by altering the resistivity or the length of the path, with SiC being able to sustain a significantly higher voltage per µm than Si.
Vgs is tuned by the crossection of the channel, and can be tuned over a very wide range. Lovoltech holds patents for the grid style gate layout, which they have used to produce JFETs that have a Vgs of zero volts, or even positive Vgs (like for MOSFETs), although there is a limit to how high you can go before you get gate current, and having a turnoff above this point doesn't make sense.
Forward transconductance and max current is largely determined by the area of the channel divided by its length, IIRC, although I'd guess the doping level has a significant impact as well.
Reverse transconductance, and hence a significant part of the transfer curve, is determined by the chanel length. A short channel will be more dominated by the Vds, giving a triode-like characteristic, while a long channel will be more controlled by the Vgs, giving a pentode-like characteristic. The proposed commercial devices are made to have a mixed characteristic, transitioning from triode to pentode mode over their operating range, which is presumably beneficial when switching.
Input capacitance is primarily controlled by the surface area of the gate, again IIRC, as is gate leakage. Noise is similar.
These devices also, at least in their SiC incarnations, handle high power and high frequencies with low noise. And they are available, although I wouldn't expect SiCED to produce anything interesting for audio. You might get some of the FET fabs to do something similar, using vertical slit gates (those aren't patented), but expect prices to be exorbitant.
Bonsai said:
Howdy. Glad to see an idea I posted has gotten someone experimenting
Jcx’s point on stability when bootstrapping the opamp supplies from the output stage instead of the VAS was a good and perfectly valid one, but it can be avoided by making an opamp whose gain stages are not influenced by the power supply voltages. A differential VAS for the opamp and not miller compensation has been working for me. My experimental prototype currently consists of a low power (50W 8R) bipolar class AB stage linerarised by a discreet opamp built as detailed above, whose supply rails are bootstrapped from the output stage, not the VAS. I built it this way because it was simpler to whack together as a proto-type.
I can’t meaningfully report on it’s THD performance as yet because I haven’t got a distortion analyser with an oscillator that can deliver 30V peak (it’s just a unity gain output stage after all), but it’s perfectly stable, even into highly capacitive loads. I’ll have to build a complete amplifier around the opamp-linearised output stage and plot it’s THD performance with and without the opamp linearisation engaged to make comparisons. I intend to do this with my 512W rms class A+ design, which I was hoping to have running by now. Unfortunately another rather nutty project (amongst a few others) is keeping me busy in the meanwhile......
http://homepages.picknowl.com.au/glenk/ctm2k.htm
For future updates on my progress with the amplifier, go here:
http://homepages.picknowl.com.au/glenk/512WClassA.htm
Cheers,
Glen
Nixie said:
Unfortunately, unlike for HiFi audio, low blower/fan noise isn't as much an issue for RF transmitters. I have never endeavoured to make one quiet, other than to change worn motor bearings when the rattling gets on my nerves.
Maybe you could locate the amplifier in a room other than the listening room and control it remotely. That’s what we do with big RF transmitters – they’re always installed outside of the studio.
Cheers,
Glen
Nixie said:
Sure, you could probably use two devices in a cascode configuration. But considering the minimum order quantity to get anyone to even talk to you, I don't see it as realistic except in a literal interpretation of the oft-touted "cost no object" expression.
Of course, if you really want to burn money, you might have success talking to Northrop Grumman... I think they still have a SiC direct-write process.
I do see that SiCED has developed MOSFETs with 3kV blocking voltage, although their VJFET/SIT offering is still "only" 1200V. Still optimized for switching, though, I think, still expensive to manufacture, and presumably more intended as a technology teaser than anything for consumers to start employing. Also, N-channel only.
I'd suggest IGBT, cascoded MOSFETs or somesuch instead.
Edit: It seems SiCED now lists 3.5kV VJFETs (Idss=3A, Rds(on)=1.25R), along with 1.7kV 25A schottky fast/soft recovery diodes and 4.5kV 2A regular diodes.
p-channel MOSFET frequency response anomaly
Awhile back John Curl and Nelson Pass mentioned an anomaly that was observed in the performance of the p-channel IRFP9240 HEXFET. The anomaly takes the form of a frequency response deviation in the a.c. transconductance of the device that occurs in the frequency range between 100 Hz and 10 kHz. Specifically, if the relative a.c. transconductance at 100 Hz and below is 0 dB, the observed transconductance decreases to about -4 dB at 10 kHz. The midpoint of the transconductance transition (where it is down about 2 dB) lies in the vicinity of 500-1000 Hz.
I got very interested in this anomaly and made some detailed measurements to see it for myself. I was indeed able to confirm the phenomenon and also confirmed that the phenomenon does not appear in the n-channel devices.
My measurements were made in a conventional test to measure a.c. transconductance of the device. The source was connected directly to ground. In a typical round of measurements, the drain was connected to a -17.5V power supply through a load resistance of 7.5 ohms. The device was operated at 1.0 Amp. The drain-source voltage was 10V. The a.c. signal from the source was a.c. coupled to the gate of the device, and was set to 26 mV rms.
I observed the phenomenon as a function of drain current, drain-source voltage, load resistance and signal level. Based on my experimental data I have made the following observations:
1) The effect is independent of signal level. This suggests that it is not a thermal effect.
2) The effect decreases with a decrease in drain load resistance RL from 14 ohms to 1 ohm, suggesting that the effect increases with increased common-source voltage gain. Some data points for the effect as a function of gain under many different conditions are as follows:
Voltage gain Delta dB
36.5 3.2
18.3 4.1
7.4 1.6
3.7 0.7
2.7 1.0
0.5 0.2
“Delta dB” is the amount by which the gain is down at 10 kHz relative to 100 Hz.
(this data may suggest that the effect will be nearly absent in a source follower, which would be good news)
3) The effect increases with current (hence gm)
Delta dB’s are 5X higher at 1A vs 0.1A
4) Use of low load resistance changes the results at 100 Hz by almost nothing, but changes results at 10 kHz by a lot. This suggests that the effect depresses gm at high frequencies (as opposed to enhancing gm at low frequencies). It further suggests that perhaps the channel conductance is being modulated by a capacitive electric field originating at the drain.
5) The effect decreases with increased Vds. Decreasing Vds has no effect on the gm at 100 Hz, but causes a reduction in gm at 10 kHz. Note that decreased Vds also increases capacitance from drain to gate and source, so this tends to support the observations in (4).
6) Changes in input impedance below 1000 ohms have virtually no effect on the phenomenon, suggesting that this is NOT a direct Miller effect to the gate.
I have been in email contact with the application engineering people at IR, but so far, surprisingly, they don’t have an explanation for this behavior.
Cheers,
Bob
Awhile back John Curl and Nelson Pass mentioned an anomaly that was observed in the performance of the p-channel IRFP9240 HEXFET. The anomaly takes the form of a frequency response deviation in the a.c. transconductance of the device that occurs in the frequency range between 100 Hz and 10 kHz. Specifically, if the relative a.c. transconductance at 100 Hz and below is 0 dB, the observed transconductance decreases to about -4 dB at 10 kHz. The midpoint of the transconductance transition (where it is down about 2 dB) lies in the vicinity of 500-1000 Hz.
I got very interested in this anomaly and made some detailed measurements to see it for myself. I was indeed able to confirm the phenomenon and also confirmed that the phenomenon does not appear in the n-channel devices.
My measurements were made in a conventional test to measure a.c. transconductance of the device. The source was connected directly to ground. In a typical round of measurements, the drain was connected to a -17.5V power supply through a load resistance of 7.5 ohms. The device was operated at 1.0 Amp. The drain-source voltage was 10V. The a.c. signal from the source was a.c. coupled to the gate of the device, and was set to 26 mV rms.
I observed the phenomenon as a function of drain current, drain-source voltage, load resistance and signal level. Based on my experimental data I have made the following observations:
1) The effect is independent of signal level. This suggests that it is not a thermal effect.
2) The effect decreases with a decrease in drain load resistance RL from 14 ohms to 1 ohm, suggesting that the effect increases with increased common-source voltage gain. Some data points for the effect as a function of gain under many different conditions are as follows:
Voltage gain Delta dB
36.5 3.2
18.3 4.1
7.4 1.6
3.7 0.7
2.7 1.0
0.5 0.2
“Delta dB” is the amount by which the gain is down at 10 kHz relative to 100 Hz.
(this data may suggest that the effect will be nearly absent in a source follower, which would be good news)
3) The effect increases with current (hence gm)
Delta dB’s are 5X higher at 1A vs 0.1A
4) Use of low load resistance changes the results at 100 Hz by almost nothing, but changes results at 10 kHz by a lot. This suggests that the effect depresses gm at high frequencies (as opposed to enhancing gm at low frequencies). It further suggests that perhaps the channel conductance is being modulated by a capacitive electric field originating at the drain.
5) The effect decreases with increased Vds. Decreasing Vds has no effect on the gm at 100 Hz, but causes a reduction in gm at 10 kHz. Note that decreased Vds also increases capacitance from drain to gate and source, so this tends to support the observations in (4).
6) Changes in input impedance below 1000 ohms have virtually no effect on the phenomenon, suggesting that this is NOT a direct Miller effect to the gate.
I have been in email contact with the application engineering people at IR, but so far, surprisingly, they don’t have an explanation for this behavior.
Cheers,
Bob
It would seem that you have pretty much duplicated my results.
As far as I know, this has only been seen on parts from IR, and
not from other manufacturers, even for the same part #.
As to its effect in complementary Source followers, it is observable
as a slight increase in distortion in the mid-band, but we are
still talking less than .01% open loop.
I will be interested in what IR has to say - I recall Charles Hansen
reported that they gave him an explanation, but he couldn't
remember it. (not that my memory is any better).
As far as I know, this has only been seen on parts from IR, and
not from other manufacturers, even for the same part #.
As to its effect in complementary Source followers, it is observable
as a slight increase in distortion in the mid-band, but we are
still talking less than .01% open loop.
I will be interested in what IR has to say - I recall Charles Hansen
reported that they gave him an explanation, but he couldn't
remember it. (not that my memory is any better).
Nelson Pass said:It would seem that you have pretty much duplicated my results.
As far as I know, this has only been seen on parts from IR, and
not from other manufacturers, even for the same part #.
As to its effect in complementary Source followers, it is observable
as a slight increase in distortion in the mid-band, but we are
still talking less than .01% open loop.
I will be interested in what IR has to say - I recall Charles Hansen
reported that they gave him an explanation, but he couldn't
remember it. (not that my memory is any better).
Thanks, Nelson. So far the only speculation that someone from IR has made is that it may have to do with the distributed RC transmission line formed by the gate lines. However, I don't think that the numbers work out to place this effect in the audio band. I'll keep you posted.
Bob