In my audio journey, I've have come upon the output stage. I have two questions:
1) Is there a discrete semiconductor that doesn't have a lot of variance in production? (I'm assuming there's no 1% tolerant transistor or the like)
2) When it comes to output stage, while chip amps do simplify design, does the cost proposition of them outweigh the performance of a high specced discrete amplifier?
1) Is there a discrete semiconductor that doesn't have a lot of variance in production? (I'm assuming there's no 1% tolerant transistor or the like)
2) When it comes to output stage, while chip amps do simplify design, does the cost proposition of them outweigh the performance of a high specced discrete amplifier?
1) Typically, a semi vendor makes a bunch of parts and sorts them by voltage, gain etc. As a result, many cheaper parts are actually surplus better parts. Besides variations like gain/beta, transistors are very thermal sensitive so circuits must be arranged so that the exact gain is not critical.
2)Chip amps can do things like integrated thermal and output protection better than most discrete designs, but the inability to make best quality PNP transistors on a N-type chip means that they have a performance limit. However, that limit today is very good so ~nobody cares about it.
Note that chip amps are generally a limited power capability. Some people have extended them in various ways, but if you want a big amp, the advantages of using chip-amps go away. We now have some very good class-D chip-amps, which are taking over the commercial market. Class A, and AB amps, tube amps and vinyl may remain in niche markets, but most of the world is going digital.
2)Chip amps can do things like integrated thermal and output protection better than most discrete designs, but the inability to make best quality PNP transistors on a N-type chip means that they have a performance limit. However, that limit today is very good so ~nobody cares about it.
Note that chip amps are generally a limited power capability. Some people have extended them in various ways, but if you want a big amp, the advantages of using chip-amps go away. We now have some very good class-D chip-amps, which are taking over the commercial market. Class A, and AB amps, tube amps and vinyl may remain in niche markets, but most of the world is going digital.
Oh OK so it's like a silicon lottery for processors and the "binning" they do.
The one I'm probably going to use is the TPA3255. I figure if I'm doing DIY I may as well use the best if it's only a few bucks more than what I would spend on a typical hand of FETs.
That said, while the performance numbers are VERY good, I just figured I would ask just in case if I was leaving anything in the table if I were to go with that as opposed to a discrete solution. (Still trying to decide if I want to go analog or digital, but I think the question is a good one to ask none the less).
That’s not quite true. The process they use to make IC power amplifier chips still has crap PNPs. The process used to create all these 1 GHz fT 1000v/us op amps is not a power process. And still relatively expensive. But really nice PNPs.
But you can’t TELL these clowns that. They insist it’s better to stack these chips in series-parallel. That works with individual transistors, right? But when the entire block is a high gain amplifier instead of an individual transistor, they are simply not as well behaved in terms of voltage/current sharing as it is in a large bank of emitter followers with individual emitter resistors. They start fighting one another.
BKr0n - Designing with bipolar transistors requires making the circuit tolerant of large variations in Hfe, Is, and temperature. There is no getting around the kT/q in the Ebers-Moll equation. Emitter degeneration and feedback go a long way.
Chip amps are cost-effective at moderate power levels. Beyond that, discrete transistors are required.
Ed
Chip amps are cost-effective at moderate power levels. Beyond that, discrete transistors are required.
Ed
No point for a single device as it would involve throwing away most of the devices, matched pairs are a thing though (you don't have to throw away devices!). For IC's laser-etching is available to provide precision.
When you realize the thickness of a gate or base in a FET or transistor in atoms, and note that its all constructed by etching and diffusion, you'll realize why precise repeatability is very hard to achieve.
Why are PNPs so much more difficult than NPNs to make?
Interesting you should say that. There's a couple of application notes on the web page for the TPA3255 about this
https://www.ti.com/lit/an/slaa787/s...=https%3A%2F%2Fwww.ti.com%2Fproduct%2FTPA3255
https://www.ti.com/lit/an/slaa788a/...=https%3A%2F%2Fwww.ti.com%2Fproduct%2FTPA3255
I'm familiar with feedback, but not so much on emitter degeneration. Could you expand on this please?
By that rationale, does that mean the ones you find in an IC pack are more precise than ones that are strictly discrete?
PNP are probably no harder to make than NPN, but holes are 3 times less mobile than electrons (in silicon), so all else being equal PNP have significantly lower performance (holes are the majority carriers in PNP). This means there isn't symmetry in properties without compensating some other way.
Also on an IC you need one extra layer for one of the polarities, since the substrate is doped uniformly (its the wafer originally), so if the substrate is P-type PNP have to be PNPN, and NPN are NPN, in order to isolate from each other the bottom layer must be opposite to the substrate. Also the diffusion properties of the different dopant atoms will be different, affecting the ability to create these layers at various depths - although ion-implantation can be used for more control, its more steps, more cost, another process variable that has to be strictly controlled.
In short its complicated.
Within one IC you can get good matching, because its the same part of the same wafer. Many wafers processed have test circuits dotted around it specifically for measuring process characteristics for detecting manufacturing process changes over time and across the wafer.
Also on an IC you need one extra layer for one of the polarities, since the substrate is doped uniformly (its the wafer originally), so if the substrate is P-type PNP have to be PNPN, and NPN are NPN, in order to isolate from each other the bottom layer must be opposite to the substrate. Also the diffusion properties of the different dopant atoms will be different, affecting the ability to create these layers at various depths - although ion-implantation can be used for more control, its more steps, more cost, another process variable that has to be strictly controlled.
In short its complicated.
Within one IC you can get good matching, because its the same part of the same wafer. Many wafers processed have test circuits dotted around it specifically for measuring process characteristics for detecting manufacturing process changes over time and across the wafer.
Allow for up to 20% in variance with discretes when designing a solid state amplifier - or worse.
Consider dc-, ac- and thermal-stability, noise, open loop and closed loop issues also, to name a few.
It's not like programming and fixing bugs, those things. It is good or crap, only some minor tweaking.
Consider dc-, ac- and thermal-stability, noise, open loop and closed loop issues also, to name a few.
It's not like programming and fixing bugs, those things. It is good or crap, only some minor tweaking.
Ok so its more about manufacturing costs. That said, does that mean things like GaNs are inherently more precision made? I see a lot more of those in IC packs as opposed to discrete semiconductors (I would assume also because of the decreased footprint you would want for them). Additionally, including what @steveu said, do the manufacturers specify where the higher grade versions of each semiconductor go? Or is that more just about digging through the product stack?
Wow... that's quite the variance. How the hell are you supposed to design around that? lol. Can you order all the parts be matched from certain places?
There is no problem making PNP transistors on a N-type substrate, but a chip-amp lives on one "monolithic" chip, a P-type substrate that makes great NPN transistors, not great PNP transistors. PNP transistors on a P-type substrate are "lateral" transistors.
https://eng.libretexts.org/Bookshelves/Electrical_Engineering/Electronics/Operational_Amplifiers:_Theory_and_Practice_(Roberge)/10:_Integrated-Circuit_Operational_Amplifiers/10.02:_FABRICATION
https://eng.libretexts.org/Bookshelves/Electrical_Engineering/Electronics/Operational_Amplifiers:_Theory_and_Practice_(Roberge)/10:_Integrated-Circuit_Operational_Amplifiers/10.02:_FABRICATION
That's the real challange, and the reality to deal with.
Most semi's have this tolerance, and when designing for industrial products it's a ruling condition. Adjustment pots are cheap as parts, but extremely expensive to turn to its proper setting. Now, design & build an outstanding electronic device, idiot-proof, everlasting. The real thing.
Ngl this is a beautiful explanation for all the various high power anything I've had to calibrate over the years lol. Actually really does explain a lot. Also explains why there's so much literature out there on specifically discrete amplifier design.
Monolithic processes are better about getting them matched to one another than discretes, since the physical variances that cause the electrical ones are pretty uniform on the same DIE. One die to the next, you get more variance. One wafer to the next, the variation is the same as two discretes bought at different times, from different places. But they can be more uniform than 50 or even 10 years ago - IF they control the process as tight as they can. Making old products to old spec, they let things go a bit before tightening the screws. Yield is everything and cost not much further behind in terms of priorities in the fab.
The ADI (Analog Devices) process that makes high speed PNPs alongside their NPN counterparts is expensive and just not what’s used to make LM3886’s and 7800 series regulators. It’s not your grandfather’s op amp. There is special isolation steps needed, and almost can be considered non-monolithic by comparison.
The PNPs still aren’t as good as the NPNs but it’s more like what you can get with discretes.
The ADI (Analog Devices) process that makes high speed PNPs alongside their NPN counterparts is expensive and just not what’s used to make LM3886’s and 7800 series regulators. It’s not your grandfather’s op amp. There is special isolation steps needed, and almost can be considered non-monolithic by comparison.
The PNPs still aren’t as good as the NPNs but it’s more like what you can get with discretes.
Emitter degeneration is done by putting a resistor in series with the emitter. The resistor provides local feedback.
Ed
The advantage of separate discretes is the better noise performance, the noise vectors are never identical.
I try to design with 10% resistors in mind, the circuit should be able cope with that.
Also, small local feedback circuits (just more then ED) with 20% resistors are doable.
I try to design with 10% resistors in mind, the circuit should be able cope with that.
Also, small local feedback circuits (just more then ED) with 20% resistors are doable.