There are some interesting things about the limitations in amplifying devices.
Vacuum tubes are first-order limited by capacitance and internal resistance. For instance, a 12AX7 with 62kohm Rp and 5pF (or whatever) output capacitance is limited to a cutoff frequency (which is not the same as GBW) of 513kHz. (That would be grounded grid; in common cathode, miller effect drops this by about a decade.)
But we have a trick for dealing with capacitance: inductance. We can cancel the capacitive reactance at a single frequency and extend the operating frequency arbitrarily!
The real limiting factor for vacuum tubes arises from physics and construction. Grids are electrodes on wires, so they have series inductance and parallel capacitance; they make excellent resonators. This effect cannot be controlled externally, so it's a fundamental limitation for the tube in question. The other effect is transit time and electron bunching, which arises from the time it takes electrons to go from cathode to anode. In a typical beam tetrode like 6V6, this is on the order of 50ns (IIRC), so you can estimate that, for frequencies in the vicinity of 20MHz, you are going to start seeing problems from this effect. In practice, most conventional tubes function up to 60MHz, with reduced ratings (presumably because the tube doesn't seem to be "switching" on and off fully anymore) at higher frequencies. In the ARRL handbooks, I've seen 6BQ5s used for 2-meter (144MHz) ham radio equipment.
The peculiar thing about vacuum tubes is, even a fairly poor tyoe will work well into the VHF band. This is something which cannot be done with bipolar transistors!
Of the other types of amplifiers, we have:
JFETs: the most similar to tubes. As far as I know, these exhibit transit time effects as well, where the transport is resistive instead of ballistic, because electrons are flowing through silicon, not vacuum. Other than that, the physics are almost identical. Even a poor JFET (e.g., 2N3819) is usable well into the UHF range; many are advertised for operation up to 900MHz, and these are crummy TO-92 packaged transistors.
There is a special category of JFETs which we owe to modern semiconductor progress. GaAs FETs, PHEMTs and GaN FETs have insanely high transconductance, smallRds(on), and ridiculously low capacitance; they are capable of hitting DC to 1GHz without any special attention. (It goes without saying, you do NOT use these in audio unless you also want to build a cellphone jammer!) Because these are special-purpose devices (radar, cell phones, satellites, etc.), it's reasonable to assume they are very finely patterned to gain every possible advantage. After all, if you're going to buy that pricey GaN wafer, you'd better get every ounce of bandwidth out of it you can manage.
MOSFETs: the second most similar, MOSFETs suffer notably from gate spreading resistance and miller effect. Newer generation switching transistors are excellent at high voltage, have a somewhat shielded construction minimizing miller effect (not quite on par with the shielding in a pentode, but much better relative to a triode). This allows switching edges on the order of 20ns, maximizing power supply efficiency.
Your average MOSFET is limited by gate capacitance and spreading resistance, at around 10MHz. (Supposedly, grounded-gate MOSFETs are prone to oscillation in the 30MHz band.) They can indeed be constructed with finer features, lower gate resistance (aluminum metallization versus polysilicon gate, etc.) and better shielding, reaching the microwave band easily (2.4GHz MOSFETs). Of course, very, very small MOSFETs are used for digital circuits up to -- whatever clock rate they're up to now.
BJTs: these are peculiar over all devices because of stored charge effects. Even if capacitance is nulled, you won't get much more than advertised bandwidth, because the charge simply can't move that fast, in and out of the base.
For a given BJT, there isn't anything you can do about its frequency response (aside from tedious amplifiers like the "fT doubler" or distributed amp). The transistor really does set the bandwidth of your circuit.
HBTs are made from fancy mixed semiconductors. I don't know the physics of stored charge in them; evidently it's not much of an issue as HBTs have been demonstrated in the near-THz.
IGBTs: these are a cross between MOS and BJT. You might drive the MOS part fast enough to produce useful RF, but the BJT part will be dragging behind, so the whole transistor never fully cuts off, resulting in very low efficiency. For such an application, you would be better served by removing the offending BJT element and using a MOSFET directly. Commercial IGBTs aren't very useful over 100kHz.
SCRs: yes, I shall also include SCRs in this list. They have the same limitations as BJTs, being minority-carrier devices. Two major drawbacks limit their use at high frequency: *ALL* the stored ("base") charge must be removed before it will stay off. There is no direct current path to all electrodes, so this charge cannot be forcibly removed, as it can from a BJT. It must dissipate by recombination only. This typically takes t_q = 40 microseconds, even for fast ("inverter grade") SCRs. The other limitation is critical dV/dt, which is the SCR's equivalent of miller effect. In practice, SCRs are not useful above 10kHz, and typically used in audio or line frequency inverters (400Hz, 60Hz, etc.). I suppose you could build one with RF transistors and get a t_q in the nanoseconds, but you might as well use both transistors directly in a class C amp and have an easier time driving it. SCRs are kind of a pain, limiting their use to only the highest voltage and power applications, where their ease of manufacture and raw power density are unsurpassed.
Conclusions:
("Ordinary" means an RC coupled, wideband amplifier stage; frequency limits are given assuming ideal tuned conditions.)
- Most tubes can be used at fairly impressive frequencies, well beyond their ordinary operating frequencies; RF tubes have been used for GHz
- JFETs are traditionally used for UHF; PHEMTs reach into the THz
- MOSFETs are fast, but depend on internal construction; generic MOSFETs will do single MHz, RF MOS will do GHz+
- BJTs depend strongly on construction and materials; general purpose signal transistors are typically ~100MHz, cheezy power transistors can get below 1MHz, RF BJTs extend into the 100s MHz, and HBTs are approaching THz
- IGBTs and SCRs are excellent for switching, but are limited to medium and low frequency use, respectively.
Tim
Vacuum tubes are first-order limited by capacitance and internal resistance. For instance, a 12AX7 with 62kohm Rp and 5pF (or whatever) output capacitance is limited to a cutoff frequency (which is not the same as GBW) of 513kHz. (That would be grounded grid; in common cathode, miller effect drops this by about a decade.)
But we have a trick for dealing with capacitance: inductance. We can cancel the capacitive reactance at a single frequency and extend the operating frequency arbitrarily!
The real limiting factor for vacuum tubes arises from physics and construction. Grids are electrodes on wires, so they have series inductance and parallel capacitance; they make excellent resonators. This effect cannot be controlled externally, so it's a fundamental limitation for the tube in question. The other effect is transit time and electron bunching, which arises from the time it takes electrons to go from cathode to anode. In a typical beam tetrode like 6V6, this is on the order of 50ns (IIRC), so you can estimate that, for frequencies in the vicinity of 20MHz, you are going to start seeing problems from this effect. In practice, most conventional tubes function up to 60MHz, with reduced ratings (presumably because the tube doesn't seem to be "switching" on and off fully anymore) at higher frequencies. In the ARRL handbooks, I've seen 6BQ5s used for 2-meter (144MHz) ham radio equipment.
The peculiar thing about vacuum tubes is, even a fairly poor tyoe will work well into the VHF band. This is something which cannot be done with bipolar transistors!
Of the other types of amplifiers, we have:
JFETs: the most similar to tubes. As far as I know, these exhibit transit time effects as well, where the transport is resistive instead of ballistic, because electrons are flowing through silicon, not vacuum. Other than that, the physics are almost identical. Even a poor JFET (e.g., 2N3819) is usable well into the UHF range; many are advertised for operation up to 900MHz, and these are crummy TO-92 packaged transistors.
There is a special category of JFETs which we owe to modern semiconductor progress. GaAs FETs, PHEMTs and GaN FETs have insanely high transconductance, smallRds(on), and ridiculously low capacitance; they are capable of hitting DC to 1GHz without any special attention. (It goes without saying, you do NOT use these in audio unless you also want to build a cellphone jammer!) Because these are special-purpose devices (radar, cell phones, satellites, etc.), it's reasonable to assume they are very finely patterned to gain every possible advantage. After all, if you're going to buy that pricey GaN wafer, you'd better get every ounce of bandwidth out of it you can manage.
MOSFETs: the second most similar, MOSFETs suffer notably from gate spreading resistance and miller effect. Newer generation switching transistors are excellent at high voltage, have a somewhat shielded construction minimizing miller effect (not quite on par with the shielding in a pentode, but much better relative to a triode). This allows switching edges on the order of 20ns, maximizing power supply efficiency.
Your average MOSFET is limited by gate capacitance and spreading resistance, at around 10MHz. (Supposedly, grounded-gate MOSFETs are prone to oscillation in the 30MHz band.) They can indeed be constructed with finer features, lower gate resistance (aluminum metallization versus polysilicon gate, etc.) and better shielding, reaching the microwave band easily (2.4GHz MOSFETs). Of course, very, very small MOSFETs are used for digital circuits up to -- whatever clock rate they're up to now.
BJTs: these are peculiar over all devices because of stored charge effects. Even if capacitance is nulled, you won't get much more than advertised bandwidth, because the charge simply can't move that fast, in and out of the base.
For a given BJT, there isn't anything you can do about its frequency response (aside from tedious amplifiers like the "fT doubler" or distributed amp). The transistor really does set the bandwidth of your circuit.
HBTs are made from fancy mixed semiconductors. I don't know the physics of stored charge in them; evidently it's not much of an issue as HBTs have been demonstrated in the near-THz.
IGBTs: these are a cross between MOS and BJT. You might drive the MOS part fast enough to produce useful RF, but the BJT part will be dragging behind, so the whole transistor never fully cuts off, resulting in very low efficiency. For such an application, you would be better served by removing the offending BJT element and using a MOSFET directly. Commercial IGBTs aren't very useful over 100kHz.
SCRs: yes, I shall also include SCRs in this list. They have the same limitations as BJTs, being minority-carrier devices. Two major drawbacks limit their use at high frequency: *ALL* the stored ("base") charge must be removed before it will stay off. There is no direct current path to all electrodes, so this charge cannot be forcibly removed, as it can from a BJT. It must dissipate by recombination only. This typically takes t_q = 40 microseconds, even for fast ("inverter grade") SCRs. The other limitation is critical dV/dt, which is the SCR's equivalent of miller effect. In practice, SCRs are not useful above 10kHz, and typically used in audio or line frequency inverters (400Hz, 60Hz, etc.). I suppose you could build one with RF transistors and get a t_q in the nanoseconds, but you might as well use both transistors directly in a class C amp and have an easier time driving it. SCRs are kind of a pain, limiting their use to only the highest voltage and power applications, where their ease of manufacture and raw power density are unsurpassed.
Conclusions:
("Ordinary" means an RC coupled, wideband amplifier stage; frequency limits are given assuming ideal tuned conditions.)
- Most tubes can be used at fairly impressive frequencies, well beyond their ordinary operating frequencies; RF tubes have been used for GHz
- JFETs are traditionally used for UHF; PHEMTs reach into the THz
- MOSFETs are fast, but depend on internal construction; generic MOSFETs will do single MHz, RF MOS will do GHz+
- BJTs depend strongly on construction and materials; general purpose signal transistors are typically ~100MHz, cheezy power transistors can get below 1MHz, RF BJTs extend into the 100s MHz, and HBTs are approaching THz
- IGBTs and SCRs are excellent for switching, but are limited to medium and low frequency use, respectively.
Tim
That s it.
Just a little typo : RF bjts are used of course in common
emitter mode not EF...
Worth adding that they have a multi emitter design..
As an insight, a schematic of what ressemble a real bjt.
Of course, a real equivalent schematic is far more complexe.
Not really
As Schematic said, charge effects are dominant for bipolars.
Let us make a crude dimensional analyzis:
If "a" is the basic measurement of length, then the capacitance grows as a function of a²/a=a.
The current carrying capacity of the junction area grows as a².
This means that a 2A junction has in fact less capacitance than two paralleled 1A ones.
Of course, things are more complicated than that, but it shows capacitance is not a basic limiting factor for larger structures: time-constants will remain more or less the same, and more current will be available to drive larger capacitances.
On the other hand, transit time and charge storage effects (linked to volume) do increase with larger junctions.
It is possible to avoid the charge effects during turn on with BJT's and SCR's but the effort involved hardly make it practical and they operate as switches when driven in this manner.
By illuminating the wafer with Infra-red at just over a nanometre wavelength it is possible to switch on the entire device in the time it takes for the light to travel across the wafer. The problem is that to turn on the entire device and not just a few points it requires a lot of photons so a pulsed laser with appropriate pulse shaping and an optical amplifier is used. The turn off time is the same as any saturated BJT or SCR, good for pulsed power applications with loads of large wafers in parallel and massive di/dt requirements but not much else.
The photons switch on the device just like electrons but they are far easier to get into the base region with the silicon being transparent to the wavelength mentioned.
By illuminating the wafer with Infra-red at just over a nanometre wavelength it is possible to switch on the entire device in the time it takes for the light to travel across the wafer. The problem is that to turn on the entire device and not just a few points it requires a lot of photons so a pulsed laser with appropriate pulse shaping and an optical amplifier is used. The turn off time is the same as any saturated BJT or SCR, good for pulsed power applications with loads of large wafers in parallel and massive di/dt requirements but not much else.
The photons switch on the device just like electrons but they are far easier to get into the base region with the silicon being transparent to the wavelength mentioned.
Now this gets more and more interesting. If I would compare say a large geometry power bjt with, say, 10 parallel medium power bjt's. I mean comparing re: bandwidth.
The parallel medium powers would have 10x the individual capacitance etc, but how would the storage effects compare? If the parallel med powers combined would have the same die area as the single high power bjt, would the storage effects be the same? Or should we consider the parallel med power bjt's as a single high power bjt but with additional 'connections' to 10 different parts of the die? If so, would that mean that the storage effects (and bw limitations) in the parallel med powers are less than in the single high power?
jan didden
Uauuu, let's make a pulsed laser modulated power amplifier up to 1THz/-3dB.
I'm starting to cut open power transistor cases ...
Last edited:
No, at a comparable current density, the storage effect of one device is identical to that of n devices paralleled (in relative terms).
Yes, and in higher performance power transistors, the geometry attempts to mimic this.
Elvee, I don't understand your reply.
I said:
"The parallel medium powers would have 10x the individual capacitance etc, but how would the storage effects compare? If the parallel med powers combined would have the same die area as the single high power bjt, would the storage effects be the same?"
You answered:
Did you mean 'Yes'? The question was: 10 x parallel medium power with combined die area same as a single high power, both cases same current density (I assume that is A/cm^2). Correct?
jan didden
I said:
"The parallel medium powers would have 10x the individual capacitance etc, but how would the storage effects compare? If the parallel med powers combined would have the same die area as the single high power bjt, would the storage effects be the same?"
You answered:
Did you mean 'Yes'? The question was: 10 x parallel medium power with combined die area same as a single high power, both cases same current density (I assume that is A/cm^2). Correct?
jan didden
Err, we're talking about high frequency limits, which for all devices considered (i.e., not counting ancient germanium BJTs!) is well beyond the audio band! If that makes it an academic question, then why post in an academic thread?
To be precise, it's only correct if the large power transistor was designed, essentially, to be ten of the medium transistors in parallel. If, instead, the pattern is basically a magnified version (like comparing 2N3055 to 2N2222), it will be slower.
This is not true. Even if intrinsic capacitance were zero, inductance, transit time or charge storage effects will ultimately limit the bandwidth. This was my point in comparing "wideband" to "tuned" amplifier frequency ranges. I think of vacuum tubes as the most ideal case, because their "wideband" cutoff frequency is fairly low (single digit MHz), yet their tuned frequency range extends significantly beyond this (into the UHF band).
Tim
Tim,
of course that`s not true, initially, my intention was to add "almost".
Since all parameters depend on each other, at a given technical level, a certain parameter can be elevated, but not without worsening some of the other ones. I think the ultimate limit is technological, it´s rather a question of what can be reliably manufactured, generally, manufacturing technology is of decisive importance. I recommend Chinese fakes.
of course that`s not true, initially, my intention was to add "almost".
Since all parameters depend on each other, at a given technical level, a certain parameter can be elevated, but not without worsening some of the other ones. I think the ultimate limit is technological, it´s rather a question of what can be reliably manufactured, generally, manufacturing technology is of decisive importance. I recommend Chinese fakes.
Reckless use of high bandwidth transistors is a good way to turn an audio amplifier into a cellphone jammer. Needless to say, your average 20MHz scope will never tell you something is wrong!
Tim
Yes. Thank you. I agree that very carefull design is mandatory (obligatory ?) Some small non inductive resistors (smc) can help, circuit layout is critical. But can sound great !
Bias at 100mA might be what is necessary to achieve low-impedance drive (with minimal emulated drive inductance) into proceeding stages. Otherwise, your drive becomes current-drive by necessity and it becomes necessary to use weird reactive networks to produce exactly the base current needed before feedback can correct. Am I correct?
It would work well for a class A headphone amp, at least.
- keantoken
It would work well for a class A headphone amp, at least.
- keantoken
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.