Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: I repeat my Request
Just what I thought. You should save your bluster and bragging until you actually build and measure the real thing. Hopefully it will not oscillate.
Bob
G.Kleinschmidt said:
Just what I thought. You should save your bluster and bragging until you actually build and measure the real thing. Hopefully it will not oscillate.
Bob
MOSFET Cgs
Based on some earlier discussions, I was under the impression that the gate-source capacitance, Cgs, of a MOSFET would rise as the forward gate voltage was increased, particularly as the device passed threshold and entered conduction.
We know that the capacitances are often a function of voltages, and indeed the spec sheet shows how Cgs and other capacitances vary as a function of Vds. Unfortunately, I have not seen curves of Cgs as a function of Vgs or of Id.
So this weekend I decided to measure the Cgs of an IRFP240 for myself. I biased it up with an ac gate circuit impedance of 8.2K and measured the gate voltage as a function of frequency. In particular, I measured the frequency at which the ac gate voltage was down 20 dB from its LF value. The device was held at a Vds of 10V, and had a load resistor of 0.24 ohms (this minimized Miller multiplication of Cgd while allowing me to measure the ac current).
I measured the device from Vgs = 0 up to Vgs = 4.5V, where Id had risen to 1.5A.
With Id = 0, Cgs was 1267 pF, virtually independent of Vgs.
As the device turned on, Cgs increased to only 1471 pF at Id = 1.5 A. This is only a 16% increase.
Needless to say, I was surprized that there was not a significant increase in Cgs as the device turns on when properly measured and the effects of Cgd and any Miller multiplication thereof are taken out.
Cheers,
Bob
Based on some earlier discussions, I was under the impression that the gate-source capacitance, Cgs, of a MOSFET would rise as the forward gate voltage was increased, particularly as the device passed threshold and entered conduction.
We know that the capacitances are often a function of voltages, and indeed the spec sheet shows how Cgs and other capacitances vary as a function of Vds. Unfortunately, I have not seen curves of Cgs as a function of Vgs or of Id.
So this weekend I decided to measure the Cgs of an IRFP240 for myself. I biased it up with an ac gate circuit impedance of 8.2K and measured the gate voltage as a function of frequency. In particular, I measured the frequency at which the ac gate voltage was down 20 dB from its LF value. The device was held at a Vds of 10V, and had a load resistor of 0.24 ohms (this minimized Miller multiplication of Cgd while allowing me to measure the ac current).
I measured the device from Vgs = 0 up to Vgs = 4.5V, where Id had risen to 1.5A.
With Id = 0, Cgs was 1267 pF, virtually independent of Vgs.
As the device turned on, Cgs increased to only 1471 pF at Id = 1.5 A. This is only a 16% increase.
Needless to say, I was surprized that there was not a significant increase in Cgs as the device turns on when properly measured and the effects of Cgd and any Miller multiplication thereof are taken out.
Cheers,
Bob
Re: MOSFET Cgs
Thank you for this data. It seems to me quite in line with what I've measured earlier for IRL530N.
http://www.diyaudio.com/forums/showthread.php?postid=1193842#post1193842
The relative increase could vary greatly with Vd and the MOSFET type. In my measurement the increase was around 100% for Vd=5V and only 50% for Vd=12V for IRL530N. It would be interesting to see if this dependancy would have a similar character for IRFP240. In a way this kind of measurement may help to find MOSFETs better suitable for audio use.
Cheers
Alex
Bob Cordell said:
Thank you for this data. It seems to me quite in line with what I've measured earlier for IRL530N.
http://www.diyaudio.com/forums/showthread.php?postid=1193842#post1193842
The relative increase could vary greatly with Vd and the MOSFET type. In my measurement the increase was around 100% for Vd=5V and only 50% for Vd=12V for IRL530N. It would be interesting to see if this dependancy would have a similar character for IRFP240. In a way this kind of measurement may help to find MOSFETs better suitable for audio use.
Cheers
Alex
Re: Re: MOSFET Cgs
Hi Alex,
Good point.
I'll give it a shot at Vds = 5V and Vds = 20V.
Cheers,
Bob
x-pro said:
Hi Alex,
Good point.
I'll give it a shot at Vds = 5V and Vds = 20V.
Cheers,
Bob
Glen, I am not super-sure, but I think you are bootstrapping the input to raise the input impedance? If so, I have never used this technique.
My approach is different, but it is the one that I gave to R.G. Meyer to use in a low noise design that we were working on. For a note on R.G. Meyer, see message above.
My approach is different, but it is the one that I gave to R.G. Meyer to use in a low noise design that we were working on. For a note on R.G. Meyer, see message above.
estuart said:
The basis should be the size of the original output signal. On that
scale the difference between .001% and .00001% is very small.
😎
There is a game being played here that is beneath the standard set by the generally high quality discussion. Certain of you believe that you can skirt the limit of acceptable behavior and avoid any consequences. This will no longer work. Although we enforce our policies on all threads, this thread is supposed to be a discussion on a professional level so abuse is even more glaring.- and there is actually more of it.
We are taking action against posters here that poison the tone of these threads by constant sour, aggressive, belittling or condescending attitudes. We can't punish for attitude? -just watch us. We have already had members leave involuntarily or voluntarily because they aren't willing to change their attitudes. We have sent members to the sin bin and more will follow if there is ANY personal comment made.
Words such as: ignorant, stupid, incompetant, fool, bogus, will result in bin time. We have tried to allow you guys some freedom but you have abused that freedom in these technical threads more than any others.
estuart is in the bin for 3 days due to his disrespectful comments to John Curl.
john curl said:
John expressed some concern about my earlier post on BJT and MOSFET interectrode capacitances. Although I certainly addressed the bootstrapping issue in that post, I did not put any numbers on it. I assume that this was what John was unhappy with. Here I'll put some numbers on it for the IRFP240 and the 2SC3264.
Let's assume that each device is operating at 1 amp and is driving a 4 ohm load (all by itself, for simplicity).
Gm of the IRFP is about 3 S at this current. Gm is about 10 S for the 2SC3264 at this current (ideally, it would be 40S, but this is not an ideal device).
For the IRFP240, we have a source follower gain of 0.924, so the bootstrapping factor for the capacitance is 0.076. The Cgs is about 1250 pF, so its effective value after bootstrapping is only 95 pF. Note that this is smaller than the value of Cgd.
For the 2SC3264, we have a source follower gain (to the emitter, before any external ballast resistance) of 0.976, for a bootstrapping factor of 0.024. Cbe at 1 amp is 0.126 uF, so we end up with an effective value after bootstrapping of about 3024 pF. The greater amount of bootstrapping in the BJT is not enough to overcome the much larger starting amount of capacitance.
If we instead consider the numbers at a typical bias current of 100 mA for the BJT and 150 mA for the MOSFET, we obtain the following effective Cbe/Cgs capacitances after bootstrapping:
IRFP240: 250 pF
2SC3264: 2183 pF
Even though its effective input capacitance has increased due to its lower transconductance, the MOSFET still has a significantly lower Cbe/Cgd capacitance after bootstrapping.
In a typical push-pull arrangement, both the P and N devices contribute to total stage transconductance, so in that case the effective Cbe/Cgs input ca[pacitance of each of the p and n devices will be about half.
Cheers,
Bob
The stuff that rings my bell during day hours is usually modelled and tested at a 1/100th scale, probably because they're often a thousand feet in reality.
Always saves a lot of money in the end, but still some that skip it.
Something as innovative as a modulated rails Class A amp should have a place of its own, even if the signal split stage doesn't enable it to go higher than 50 double-u on a life size scale.
Other than that i rather get the Sanken numbers import tax free across the border from profu in Germany for AUS 2.85 each, i'll bet you a skippy i can do it for 1K.
Always saves a lot of money in the end, but still some that skip it.
Something as innovative as a modulated rails Class A amp should have a place of its own, even if the signal split stage doesn't enable it to go higher than 50 double-u on a life size scale.
Other than that i rather get the Sanken numbers import tax free across the border from profu in Germany for AUS 2.85 each, i'll bet you a skippy i can do it for 1K.
john curl said:
Hi John,
I estimated Cbe from the data sheet information, mainly based on the ft vs Ic curve. I outlined it in Post #1740, but here are some more details.
If you look at ft at several different values of Ic, you can deduce the two components of the Cbe capacitance, the one due to the junction capacitance and the one that is proportional to Ic. The hybrid pi model relation, Cpi = gm/Wt is the key relationship. Wt is the radian transition frequency (Wt = 6.28 ft) and gm is the intrinsic transconductance.
Here is some of the data for the 2SC3264:
20 mA; ft = 5 MHz; gm = 0.8S; Cpi = 0.025 uF
100 mA; ft = 17 MHz; gm = 4S; Cpi = 0.037 uF
300 mA; ft = 31 MHz; gm = 12S; Cpi = 0.062 uF
1000 mA; ft = 50.5 MHz; gm = 40S; Cpi = 0.126 uF
The junction capacitance component of Cpi is estimated to be 0.023 uF, the rest is the component that is proportional to Ic.
There is one caveat in evaluating this device. The extrinsic transconductance at currents like 1A and above is substantially less than the intrinsic transconductance. Indeed, at 1A, the extrinsic transconductance is only about 10S. Not sure what the source of this non-ideality is in this device, whether it is base spreading resistance or emitter resistance, or something else.
In my earlier post where I took into account the bootstrapping effect on Cbe, I used the extrinsic gm, since that is what the outside world sees. However, depending on device details that I am not sure of, one might argue that the bootstrapping calculation should use the instrinsic transconductance, which would increase the degree of bootstrapping for the BJT.
Cheers,
Bob
jacco vermeulen said:The stuff that rings my bell during day hours is usually modelled and tested at a 1/100th scale, probably because they're often a thousand feet in reality.
Always saves a lot of money in the end, but still some that skip it.
Something as innovative as a modulated rails Class A amp should have a place of its own, even if the signal split stage doesn't enable it to go higher than 50 double-u on a life size scale.
Other than that i rather get the Sanken numbers import tax free across the border from profu in Germany for AUS 2.85 each, i'll bet you a skippy i can do it for 1K.
Hi jacco,
I agree. Doing tests and measurements on a scaled basis is often very valuable and important in engineering. However, it is incumbent upon the person making the claims based on such modeling to explain the assumptions, limitations and caveats of the model so that others can make up their own mind as to whether the scale model is a valid substantiation of the claim. The bolder the claim, the more substantiation that the model is appropriate is needed.
Cheers,
Bob
Bob Cordell said:
Hi Bob,
if I remember correctly, these devices use series nichrome resistors in every emitter in a multi-emitter construction. Perhaps this is what limits transconductance for large currents.
Cheers
Alex
x-pro said:
That could very well be true. If that is the source of the limited gm, it is good news for two reasons: first, it is probably linear; second, it would then be pretty much like an extension of the external emitter ballast resistor, meaning that the amount of bootstrapping of Cbe might be more than the amount that I calculated earlier (I think - this is really at the edge of my expertise in terms of device operation).
Cheers,
Bob
Nelson Pass said:
Agreed, seems the oscillation theory is a good one especially considering the protoboard construction method. I prefer point to point wiring over a solid ground plane for prototype work like this, dead bug, etc.
Pete B.
Why power transistor inter-electrode capacitances are important
In an earlier post I compared the inter-electrode capacitances of some typical MOSFETs with some typical RET bipolars, looking at both Cgd/Ccb and Cgs/Cbe. The conclusion I drew was that MOSFETs are typically equal or less than the bipolars in the Cgd/Ccb component and often a lot less in the Cgs/Cbe component, even when typical source-follower/emitter follower bootstrapping effects are taken into account. The main point was to illustrate that, while the input capacitance of MOSFETs is often cited as a concern, it is certainly not worse than that of BJTs.
Now let's take an example and see why these capacitances can be so important, particularly in a large-signal situation. We'll focus on the amount of drive current that the driver must supply to a number of paralleled output devices under conditions of large voltage slew rates and large current slew rates into low impedances.
As an example, let's say we want to have a slew rate of 100 V/us into a 2-ohm load. Let's further say we are building a nice big amplifier like John Curl's JC-1, with nine pairs of paralleled RET output devices. John, forgive me for using this as an example, but it is a great design to look at as an example. This may also serve to illustrate to others one reason why you chose to use MOSFETs to drive the output bipolars. Forgive me if I make some wrong assumptions.
We'll look at the PNP side of the output stage and assume a 2SA1295 PNP device. This is the worst case side, both in terms of the Ccb of 500 pF and the ft of 35 MHz.
Since there are a total of 18 output devices driving a 2 ohm load while in the Class-A region, we can look at just one transistor as if it were driving a 36 ohm load. We'll assume that we are looking at the needed driver currents as we are passing through the Class-A region. Assume that each device pair is biased at 133 mA at idle.
re = 1/gm at idle is 0.19 ohms, so bootstrapping G = 0.99475;
Cbe at idle is about 0.043 uF, so effective Cbe = 227 pF.
Ccb = 500 pF, so total effective input capacitance is 727 pF per device.
With nine devices on each side, we have an input capacitance to be driven by the driver of 6543 pF. At a slew rate of 100 V/us, this corresponds to an input current to the group of PNP output transistors of 650 mA. You can see that things can get ugly pretty fast under these extreme conditions.
Of particular note, and perhaps surprisingly, the Ccb capacitance of the nine paralleled devices dominates the problem. This is one of the prices we pay for paralleling a lot of devices.
Note also that we still have to drive the total amount of ccb capacitance of 4500 pF even when we are driving a much lighter load of 8 ohms.
The power that must be dissipated in the driver with 90V rails will be pretty significant (assuming a Class-A driver that stays in Class-A even under these conditions).
Now let's assume that we built the output stage with nine paralleled pairs of IRFP240/9240 MOSFETs instead.
We'll assume Cgd = 300 pf and Cgs = 1250 pF; The total effective Cin for the nine devices is very strongly dominated by the Cgd in this case, and comes out to be about 2900 pF. This compares very favorably with the 6500 pF in the BJT case.
The bottom line is that, especially in amplifiers with a good number of paralleled output devices, one must really allow for quite a bit of available high-frequency dynamic drive current from the drivers.
Cheers,
Bob
In an earlier post I compared the inter-electrode capacitances of some typical MOSFETs with some typical RET bipolars, looking at both Cgd/Ccb and Cgs/Cbe. The conclusion I drew was that MOSFETs are typically equal or less than the bipolars in the Cgd/Ccb component and often a lot less in the Cgs/Cbe component, even when typical source-follower/emitter follower bootstrapping effects are taken into account. The main point was to illustrate that, while the input capacitance of MOSFETs is often cited as a concern, it is certainly not worse than that of BJTs.
Now let's take an example and see why these capacitances can be so important, particularly in a large-signal situation. We'll focus on the amount of drive current that the driver must supply to a number of paralleled output devices under conditions of large voltage slew rates and large current slew rates into low impedances.
As an example, let's say we want to have a slew rate of 100 V/us into a 2-ohm load. Let's further say we are building a nice big amplifier like John Curl's JC-1, with nine pairs of paralleled RET output devices. John, forgive me for using this as an example, but it is a great design to look at as an example. This may also serve to illustrate to others one reason why you chose to use MOSFETs to drive the output bipolars. Forgive me if I make some wrong assumptions.
We'll look at the PNP side of the output stage and assume a 2SA1295 PNP device. This is the worst case side, both in terms of the Ccb of 500 pF and the ft of 35 MHz.
Since there are a total of 18 output devices driving a 2 ohm load while in the Class-A region, we can look at just one transistor as if it were driving a 36 ohm load. We'll assume that we are looking at the needed driver currents as we are passing through the Class-A region. Assume that each device pair is biased at 133 mA at idle.
re = 1/gm at idle is 0.19 ohms, so bootstrapping G = 0.99475;
Cbe at idle is about 0.043 uF, so effective Cbe = 227 pF.
Ccb = 500 pF, so total effective input capacitance is 727 pF per device.
With nine devices on each side, we have an input capacitance to be driven by the driver of 6543 pF. At a slew rate of 100 V/us, this corresponds to an input current to the group of PNP output transistors of 650 mA. You can see that things can get ugly pretty fast under these extreme conditions.
Of particular note, and perhaps surprisingly, the Ccb capacitance of the nine paralleled devices dominates the problem. This is one of the prices we pay for paralleling a lot of devices.
Note also that we still have to drive the total amount of ccb capacitance of 4500 pF even when we are driving a much lighter load of 8 ohms.
The power that must be dissipated in the driver with 90V rails will be pretty significant (assuming a Class-A driver that stays in Class-A even under these conditions).
Now let's assume that we built the output stage with nine paralleled pairs of IRFP240/9240 MOSFETs instead.
We'll assume Cgd = 300 pf and Cgs = 1250 pF; The total effective Cin for the nine devices is very strongly dominated by the Cgd in this case, and comes out to be about 2900 pF. This compares very favorably with the 6500 pF in the BJT case.
The bottom line is that, especially in amplifiers with a good number of paralleled output devices, one must really allow for quite a bit of available high-frequency dynamic drive current from the drivers.
Cheers,
Bob
Well, I guess that the complementary power fet 12A drive stage composed of a pair of a 2SJ201 and the 2SK1530 was not a bad investment, afterall! ;-)