Like it says on the tin.
Been looking at all kinds of transistors from bog standard BJT to RF HEMT. From my understanding, configurations like darlington, sziklai (can never remember the spelling of that), and composit amplifiers allow you to mix and match properties of different semiconductors in order to get better overall functionality. Is it worth looking into this further, or should I just stick with one set of transistors? If it is worth it, I would really appreciate some part recommendations. While I understand there's a few gold standard parts (Toshiba transistors for example) I'm sure there are a lot of others out there that may not be as prevalent because they are not used typically for audio, but can still perform as well if not better.
Been looking at all kinds of transistors from bog standard BJT to RF HEMT. From my understanding, configurations like darlington, sziklai (can never remember the spelling of that), and composit amplifiers allow you to mix and match properties of different semiconductors in order to get better overall functionality. Is it worth looking into this further, or should I just stick with one set of transistors? If it is worth it, I would really appreciate some part recommendations. While I understand there's a few gold standard parts (Toshiba transistors for example) I'm sure there are a lot of others out there that may not be as prevalent because they are not used typically for audio, but can still perform as well if not better.
If you truly knew... you'd depletion your Nelson against your Hex self, and know it's already been done... in at least one example.
Lateral MOSFETs have great linearity and thermal coefficients making them great for audio outputs, but they have availability issues. As time passes, analog amplifiers are becoming a niche market so switch mode MOSFETs dominate the market. The best amplifiers are a matter of topology and design and not the choice of devices. Vendors boosting the use of exotic devices are probably not the best engineers.
I mean, I figure it's not the most original idea (and go figure @Nelson Pass has already done that. Seems to be a trend) but it's not like there's a "how to" or a "hall of fame" for reference lol
Input is all op amp. No getting around that they beat classical semiconductors on small signal, but also no getting around that they aren't known for high voltage or current. I've also noticed laterals are becoming rarer and rarer. Seems Exicon and Toshiba are the heads of industry there. I don't necessarily agree about not using exotic devices though. Sometimes you just gotta try something ya know? That and I would like to think there's at least SOME sort of engineering rationale for picking odder things.
Edit: today I learned there are diamond fets that have twice the band gap of SiC
Laterals are less linear than BJTs (due to low gm), their advantages are thermal stability and freedom from secondary breakdown - they don't need thermal compensation of the bias point, which is always tricky to get right for BJT output stages. Most output stage failures are secondary breakdown from what I can gather, so lateral MOSFET amps have a reputation for robustness as a result.
MOSFETs have always been dominated by devices designed for switching, its not a new thing. Incidentally lateral MOSFETs also have a niche as RF amplifiers, where they are called LDMOS.
So would it be safe to say then that, in terms of audio and the companies that mass produce product, that choosing mosfets wasn't so much a design improvement decision as much as it was just switching to a more common technology?
Profusion still sell lateral mosfets.
The To3 parts are being run down but the TO247 versions are still in production.
Some Exicon TO3 parts are no longer available, the rest are EOL, but TO247 and TO264 (double-die) devices are all active. Some of the still available TO3 parts are double-die ones as they presumably sell slower.
The current gain is of MOSFETs is enormous compared to BJTs, 'almost' (some do it!) allowing direct drive with JFETs. At least with a push-pull source follower output, the parasitic Cdg and Csg looks like it charges one gate and symmetrically discharges the other, drawing almost no current from VAS. So most of the VAS overhead is only really needed to overpower the mismatch when the devices being heavily modulated with voltage and power swings.
I think the TO-3 vs TO-247 issue is likely to do with thermal 'speed', so even though the metal looks technically superior to the plastic version, the plastic wins if the design is "too simple" and doesn't account for (or prevent) rapid thermal cycling.
A Sziklai stage that has its voltage clamped by a cascode could be an interesting idea. The Amnesis and Peufeu's headphone amp go with this.
I think the TO-3 vs TO-247 issue is likely to do with thermal 'speed', so even though the metal looks technically superior to the plastic version, the plastic wins if the design is "too simple" and doesn't account for (or prevent) rapid thermal cycling.
A Sziklai stage that has its voltage clamped by a cascode could be an interesting idea. The Amnesis and Peufeu's headphone amp go with this.
Huh? The charging currents sum to double that of one device, so the VAS needs enough quiscent current to handle that.
Ya, I noticed that too. Yes, the driver charges the N-channel while discharging the P-channel BUT the N-channel gate is the negative side of Cdg while the P-channel gate is the positive side of it's Cdg, so the two gate currents ADD, not cancel. The two Cdg are like two capacitors in parallel, and the fact that the other sides of these caps have a different DC voltage is irrelevant. The Cdg's are dominant over Cgs's because the ~2Volt gate-source charge delta is trivial compared to the ~50V "Miller" charge for the output slew.
Maybe you're right -- I'll double-check later on a simulator. Thinking out loud here, there are some non-obvious aspects to it. Looking at the IRFP240 as an example, the capacitances have a negative slope in relation to Vds. Although 'opening' the N-channel requires charging, Vds falls at a faster rate than Cgd increases.
This by itself suggests that if the gate is somewhat charged and then switched to floating, external influences to the channel, such as controlling the voltage would produce inverse changes to the current. Pretty neat.
Vgs is pretty constant, so Cgs contributes very little load.
Say that the N-ch on top is charged higher, pulling the output voltage up, the increased P-ch Vds should automatically reduce its current, even if the VAS has very low current.
This by itself suggests that if the gate is somewhat charged and then switched to floating, external influences to the channel, such as controlling the voltage would produce inverse changes to the current. Pretty neat.
Vgs is pretty constant, so Cgs contributes very little load.
Say that the N-ch on top is charged higher, pulling the output voltage up, the increased P-ch Vds should automatically reduce its current, even if the VAS has very low current.
Wow look at all this great discussion! Thanks for all the responses 🙂
Since posting this I actually picked back up a class D design that I was working on before. I realized I just need to put the modulating FETs on a lower rail and I should be able to use it. There is however a big problem: DSD at high levels of modulation. Although the FETs I'm using are very good (sub nano charge times!) that doesn't change that I'm still going to run into the miller effect and possible ringing of the FETs themselves. It actually brings me back to cascode. I've been looking into a lot of documentation at it seems to be a bit split on the efficacy of cascode at high switching speeds. Some integrators use one fet to drive the other in compliment, while others put something like a regulator or an LDO/UVLO on the FET not being directly driven. Would a cascode be good for the output FETs? (Actually I would love to put this in sim, but I'm running into a problem with utilizing the LMG1210 in spice so right now all I got is theory lol)
Since posting this I actually picked back up a class D design that I was working on before. I realized I just need to put the modulating FETs on a lower rail and I should be able to use it. There is however a big problem: DSD at high levels of modulation. Although the FETs I'm using are very good (sub nano charge times!) that doesn't change that I'm still going to run into the miller effect and possible ringing of the FETs themselves. It actually brings me back to cascode. I've been looking into a lot of documentation at it seems to be a bit split on the efficacy of cascode at high switching speeds. Some integrators use one fet to drive the other in compliment, while others put something like a regulator or an LDO/UVLO on the FET not being directly driven. Would a cascode be good for the output FETs? (Actually I would love to put this in sim, but I'm running into a problem with utilizing the LMG1210 in spice so right now all I got is theory lol)