Why are power transistors (bjt's) so slow, compared to small-signal transistors? Small-signal devices have bandwidths into the 100's of MHz or even GHz. Medium power devices have lower bandwidths. High power transistors are slower yet. Is it a result of the construction to make them rugged to withstand higher powers, or is there another mechanism?
jan didden
jan didden
So, more doping lowers mobility. Does that mean power transistors have more doping than small-signal transistors?
And does electron mobility have a relation to device bandwidth?
jan didden
Hello
Substrat are not the right name, I've corrected it, I was meaning lateral thickness not vertical, like a thicker base-collector-emitor sandwitch.
Sorry for my limited english explanation, I'm french speaking.
Bye
Gaetan
Last edited:
Actually, it's not very self-explanatory...
RF transistors often have higher doping levels, which is necessary due to the smaller feature size or reduced carrier lifetime.
If doping were the only explanation, power transistors would work very well indeed, because they usually have fairly light doping. High voltage transistors (up to 1500Vcbo in HOTs, and more in industrial use) necessarily have very light doping, on the order of 10^14, or a P-I-N structure at the collector, just like a high voltage diode, and for the same reason: to minimize electric field strength, maximizing breakdown voltage.
However, this is not the case. The reason is structure. Most transistors are planar. This worked nicely for some of the first planar transistors, like 2N2222, which boasts a ~150MHz fT (though not part of the spec). It didn't work so well for the first planar power transistors, like 2N3055, which is a poor power transistor by any measure. They're basically the same thing, so why are they so different? A few reasons. One, if they are truely the same process, then the doping and depth will be equal, but the area much wider. A thin base has a large base-spreading resistance, so it takes a long time to drive the base. Generally, the emitter is broken up into strips, giving more base contact area. The finer this structure, the more emitter "fingers" and base edges are brought out, the faster it runs.
RF power transistors are little more than monolithic arrays of 2N3904s. They may be printed on a, say, ~2um feature size, or mechanically connected, where the emitters are built as islands rather than fingers; a bond wire per emitter is required! Obviously, bond wires have lower capacitance than metallization which overlays the base and collector.
Other reasons include voltage capacity, which is generally higher. (Yes, there are a few 200, 300, even 450V small-signal transistors.) hFE goes up with doping, so the light doping required to obtain the higher breakdown voltage necessarily reduces hFE. (Those 1500Vcbo transistors have a typical saturated hFE ~= 2.5!!)
Incidentially, there are a few medium power transistors with ridiculous bandwidth. These are made for high resolution CRT drivers, and include the 2SC3597. I don't recommend using one as VAS unless you damn well know what you're doing!
Tim
RF transistors often have higher doping levels, which is necessary due to the smaller feature size or reduced carrier lifetime.
If doping were the only explanation, power transistors would work very well indeed, because they usually have fairly light doping. High voltage transistors (up to 1500Vcbo in HOTs, and more in industrial use) necessarily have very light doping, on the order of 10^14, or a P-I-N structure at the collector, just like a high voltage diode, and for the same reason: to minimize electric field strength, maximizing breakdown voltage.
However, this is not the case. The reason is structure. Most transistors are planar. This worked nicely for some of the first planar transistors, like 2N2222, which boasts a ~150MHz fT (though not part of the spec). It didn't work so well for the first planar power transistors, like 2N3055, which is a poor power transistor by any measure. They're basically the same thing, so why are they so different? A few reasons. One, if they are truely the same process, then the doping and depth will be equal, but the area much wider. A thin base has a large base-spreading resistance, so it takes a long time to drive the base. Generally, the emitter is broken up into strips, giving more base contact area. The finer this structure, the more emitter "fingers" and base edges are brought out, the faster it runs.
RF power transistors are little more than monolithic arrays of 2N3904s. They may be printed on a, say, ~2um feature size, or mechanically connected, where the emitters are built as islands rather than fingers; a bond wire per emitter is required! Obviously, bond wires have lower capacitance than metallization which overlays the base and collector.
Other reasons include voltage capacity, which is generally higher. (Yes, there are a few 200, 300, even 450V small-signal transistors.) hFE goes up with doping, so the light doping required to obtain the higher breakdown voltage necessarily reduces hFE. (Those 1500Vcbo transistors have a typical saturated hFE ~= 2.5!!)
Incidentially, there are a few medium power transistors with ridiculous bandwidth. These are made for high resolution CRT drivers, and include the 2SC3597. I don't recommend using one as VAS unless you damn well know what you're doing!
Tim
Thanks Tim. So, if it's the structure, why don't we just adapt the structure to build high-speed power transistors? What's the problem?
jan didden
Mobility and size again: when holes and electrons have to travel 100µ instead of 10µ, they take ten times longer.
If you simply scale up structure, you increase the transit time.
If you create more involved structures, they are difficult to implement and more fragile.
High speed power transistors exist; they are called RF power transistors. They are expensive and fragile. They also tend to oscillate if not carefully tamed by people who know what they are doing. You would not want to make an audio amp with RF power transistors - even an RF engineer would not want to, and everyone else would not be able to.
The nearest we get is that some people use an RF transistor for VAS, as it has low feedback capacitance. The capacitance itself is not the issue, as we will probably need to increase it anyway, but device capacitance is very non-linear so creates distortion.
The nearest we get is that some people use an RF transistor for VAS, as it has low feedback capacitance. The capacitance itself is not the issue, as we will probably need to increase it anyway, but device capacitance is very non-linear so creates distortion.
How about thermal conditions?
Nelson Pass has addressed the idea of ideal biasing sweet spot. As I was measuring the case temperature of my output transistor the other day and seeing 54.4 deg C, and seeing that it is dissipating approx 15 watts steady, I observed that it seems to be well below the area of concern for overheating, but I wonder what running closer to the edge, or less so---would do to the sound?
Terry
Nelson Pass has addressed the idea of ideal biasing sweet spot. As I was measuring the case temperature of my output transistor the other day and seeing 54.4 deg C, and seeing that it is dissipating approx 15 watts steady, I observed that it seems to be well below the area of concern for overheating, but I wonder what running closer to the edge, or less so---would do to the sound?
Terry
With Ft ranging from 30 to 60 mhz , modern power bjts
are not exactly what would be called slow...
Propagation delay and carriers mobility are not the main culprits
of these "limited" gbw products.
Rather, the parasistic internal capacitances are the main factor
in speed.
Rf bjts are indeed fast, but remeber that they are driven using
a lc circuit that drive them at very low impedance to counteract
the slowing effect of the base related parasistic caps, and this
in an emitter follower circuit with emitter to ground....
Such a solution is evidently not transferable to decent audio amps..
are not exactly what would be called slow...
Propagation delay and carriers mobility are not the main culprits
of these "limited" gbw products.
Rather, the parasistic internal capacitances are the main factor
in speed.
Rf bjts are indeed fast, but remeber that they are driven using
a lc circuit that drive them at very low impedance to counteract
the slowing effect of the base related parasistic caps, and this
in an emitter follower circuit with emitter to ground....
Such a solution is evidently not transferable to decent audio amps..
So you say it's a matter of construction? The larger geometries necessary for larger Pd mean larger internal capacitances therefore less bandwidth?
On the Rf bjt's, you seem to be saying they are not really intrinsically fast but they are driven by a circuit that takes care of the charge/discharge requirements of the internal capacitances? Or did I get that wrong?
jan didden
im familiar with designs in the digital domain (chips, cpu's and memory hierarchies) and i assume its the same underlying reason in both world. there is a correlation betweeen physical size and speed. the larger the size the slower the speed due to signal travel times, capacitance between signal paths etc. and then there is a correlation between size and power, the bigger the power in a path the more issues with speed we get for the above mentioned reasons. finally there is the problem of yields in the factory fir a given process. with very small litographic dimensions and high speed and power yields goes down due to the process which drives up costs, if the yield is unacceptable at a given power or speed its often more economical for the manufacturer to use lower soecs and have more parts passing qa.
now i am a designer/architec and ive never really been in any of the layout teams. but ive hanged out with enough of them to pick up a thing or two and the above at least in the digital domain is what they keep on telling me. btw architecture rarely yields as much performance as physics, mastering shrinking the dimensions while keeping yields high gives much better performance in general than trying to work more architecture in a chip (this probably doesnt matter for transistors but its certainly true in cpu's in my experience).
now i am a designer/architec and ive never really been in any of the layout teams. but ive hanged out with enough of them to pick up a thing or two and the above at least in the digital domain is what they keep on telling me. btw architecture rarely yields as much performance as physics, mastering shrinking the dimensions while keeping yields high gives much better performance in general than trying to work more architecture in a chip (this probably doesnt matter for transistors but its certainly true in cpu's in my experience).
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.