The NJL transistor and diode are on the same die. They are very close related in thermal behavior.
If the diode is an ultrafast MUR one, then NJL transistor is made with the same technology which is based on platinum diffusion, to create fast recovery spots in the depletion region.
Very intresting!
If the diode is an ultrafast MUR one, then NJL transistor is made with the same technology which is based on platinum diffusion, to create fast recovery spots in the depletion region.
Very intresting!
roender said:
I think we are having a problem with translation. The die is the silicon wafer. They are made in a large circle (usually between 4" and 12") and each wafer is cut into the individual die that is the semiconductor itself. The wafer will yield hundreds, and sometimes thousands, of parts.
The die of a power transistor is then normally soldered to a copper leadframe. This is the exposed metal backing you see that allows for heat transfer. Typically one of the leads is an integral part of the leadframe, while the other two leads are attached to wirebond (aluminum wires) that are soldered to the appropriate points on the die. On a power xJL1302/3281, it is the center lead (collector) that is formed as part of the copper leadframe. (That is why it is so much harder to solder this lead, as it has a built-in heatsink.)
In the ThermalTrak parts, the diode is NOT part of the transistor die. Instead a separate diode is attached to the copper leadframe. The diode die is NOT soldered to the leadframe, as then there would be a short-circuit between the collector of the transistor and one of the leads of the diode. The designer told me how the diode was attached to the leadframe, but I have forgotten. Sorry.
They are great parts. Buy them, use them, enjoy them! On Semi should be rewarded for making the best audio output transistors in the world.
Charles Hansen said:
Thanks, Charles.
Having been an IC designer, I just jumped too quickly in assuming it was a monolithic approach. That was pretty dumb of me, given other realities like the one you mentioned, and probably some others having to do with device fabrication itself.
Sooooo, this is disappointing. The ThermalTrak devices are still a great idea, but they are not nearly as great as I had naively presumed.
It being made of two different die brings up a whole host of questions, none of which is a total show-stopper, but all of which require thought and accomodation. I'm sure you've addressed them.
For example, we can't depend on the close matching that a monolithic approach would have guaranteed. Depending on how close and consistent OnSemi's two processes are, the relative matching between the diode and the BJT could range from quite good to quite terrible. So much for their assertion in their app note that one could dispense with the bias trim pot (I never cared that much about that "advantage", anyway).
Another one is the mismatching of the thermal voltage coefficient, which has been mentioned.
Yet another is the greater thermal lag between the BJT and the diode. This may not matter much in the real world, but is not the ideal we would seek. I was doing some measurements of thermal time constants this morning using a ThermalTrak device, with pulsed dissipation in the BJT and monitoring via the diode. This now explains some of the slightly longer-than-expected observed thermal time constants that I saw (still in the very low seconds, though).
Thanks again,
Bob
Charles Hansen said:
Hi Charles,
What you have said here fits my understanding completely (now that I've been reminded to think about how devices are made).
They are great parts. My only quibble is that OnSemi probably could have done a better job of clearly indicating the structure (separate die) in the spec sheet and the app note.
Cheers,
Bob
Bob,
I think I have mentioned Sanken's stab at thermal tracking before. As far as I can see from the limited info (which at any rate is more than I could find on the Toshiba thermal traks) they ARE of monolytic construction. Plus, Sanken went apparently out of their way to match the tempco of the diodes to the transistors (darlingtons), evidenced by the particular diode types which are used in the N- and P-types.
But, I'm not an IC designer; what's your view of that, would they be 'better' in the tracking dept. than the Toshiba's?
Jan Didden
I think I have mentioned Sanken's stab at thermal tracking before. As far as I can see from the limited info (which at any rate is more than I could find on the Toshiba thermal traks) they ARE of monolytic construction. Plus, Sanken went apparently out of their way to match the tempco of the diodes to the transistors (darlingtons), evidenced by the particular diode types which are used in the N- and P-types.
But, I'm not an IC designer; what's your view of that, would they be 'better' in the tracking dept. than the Toshiba's?
Jan Didden
Bob Cordell said:
Disagree. Take what is arguably already the best output device available, add in thermal sensors on the leadframe, and the result is a GREAT product.
Bob Cordell said:
I already told everyone that the app note was wrong. But like I said before, if it had been correct, then all the audio designers would be out of work and everyone would be building the exact same design....
Bob Cordell said:
Yes, this one is a bit trickier to solve. I'll give you a hint: You won't solve it with equations or SPICE. You'll have to get your hands dirty and build some output stages and run some experiments.
Or just take a shortcut like McIntosh apparently did. Just run the sensor to microprocessor and fiddle with the program until it does what you want it to do. That's fine as long as you can live with an RFI generator in the middle of your analog audio circuitry. I can't.
Bob Cordell said:
Again, I disagree that this is a problem in the real world. When the output device is mounted to a heatsink, the diode mounted on the copper leadframe is going to track the temperature variations plenty fast to keep everything just where you want it.
If you tried to go any faster than that, you would end up creating distortion as the sensor would be changing the bias from instant-to-instant during a low-frequency musical note.
And as far as reliability goes, the only thing that happens any faster than that are the "hot spot" formations that (I believe) are more commonly referred to as "secondary breakdown". The solution to that problem is to use more devices or lower your rails (easy with a bridged output stage) or both.
janneman said:
I am 99.9% sure that they are NOT monolithic. From the brochure:
"The temperature compensation diodes are mounted on one chip and placed in the center of the chip to detect temperature rises directly."
This does NOT sound like monolithic construction. Plus the diodes in one device are Schottkys. I don't know that you can make a monolithic part with a BJT and a Schottky. Plus it explicitly says that the builit-in emitter resistor is a thick-film device, which CANNOT be monolithic.
But the biggest problem with these devices (in my opinion) is that you cannot make the proper output topology. As noted in other threads, the "T-Circuit" developed by Locanthi and used by Leach is the best BJT output stage by a large margin (in my opinion also). The Sanken parts can only be used as a double (not a triple), plus the drivers are running class AB and not class A like the T-Circuit.
So as much as I wanted to use them, I never did. (A few British companies have done so in their less-expensive amps.) That's why it was a god-send when On Semi released their ThermalTrak devices. These things are practically perfect!
I looked at these dvices in some detail. Some (constructive) feedback to On would be to improve the characterisation data in the data sheet as follows:-
1. Diode Vbe spread (need this for a range of diode currents)
2. Matching between diode and transistor Vbe
3. Temp co-eff absolute spread for the sense diode
4. Same again for transistor base emitter
5. Some charactersisation data of the diode and transistor operating as a system - i.e. dynamic tracking performance at two or three diffferent power levels.
This is about 1 man month work for a charaterisation engineer, so On need to decide whether it s worth it or not - a lot of the data can be pulled from the test records in the assembly centre.
I don't know that the higher coupling capacitance in a monolithic solution would be a show stopper - the problem would just have to be tackled in a different way (i.e. addressed with some engineering around the layout and circuit design).
For reference, an integrated tremperature sensor on a power chip of circa 25mm square (mosfet based power device with integrated protection) can operate at about 1KHz from overtemp sense to shutdown, cool, switch on again and back up to over temp (can't remember the trip and reset temp values - I think it was 20 or 30 degrees). Device was free standing for this test.
1. Diode Vbe spread (need this for a range of diode currents)
2. Matching between diode and transistor Vbe
3. Temp co-eff absolute spread for the sense diode
4. Same again for transistor base emitter
5. Some charactersisation data of the diode and transistor operating as a system - i.e. dynamic tracking performance at two or three diffferent power levels.
This is about 1 man month work for a charaterisation engineer, so On need to decide whether it s worth it or not - a lot of the data can be pulled from the test records in the assembly centre.
I don't know that the higher coupling capacitance in a monolithic solution would be a show stopper - the problem would just have to be tackled in a different way (i.e. addressed with some engineering around the layout and circuit design).
For reference, an integrated tremperature sensor on a power chip of circa 25mm square (mosfet based power device with integrated protection) can operate at about 1KHz from overtemp sense to shutdown, cool, switch on again and back up to over temp (can't remember the trip and reset temp values - I think it was 20 or 30 degrees). Device was free standing for this test.
Hi Bonsai,
I don't agree. We already accept variable parameters with outputs and diodes for sensing temp. Why would we now demand more from a much improved product??? There will be some sort of bias compensation adjustment. This is unless device to device variables are close enough to not worry about. So build some stuff.
Look at the price of these and look at how close these transistors are to each other for measured parameters. Not 5 years ago I was very depressed over unit to unit variations and now On Semi can compete with Japanese Semiconductor manufacturing.
-Chris
I don't agree. We already accept variable parameters with outputs and diodes for sensing temp. Why would we now demand more from a much improved product??? There will be some sort of bias compensation adjustment. This is unless device to device variables are close enough to not worry about. So build some stuff.
Look at the price of these and look at how close these transistors are to each other for measured parameters. Not 5 years ago I was very depressed over unit to unit variations and now On Semi can compete with Japanese Semiconductor manufacturing.
-Chris
Bonsai said:
Chris already covered the first three items -- you just need to have a bias adjustment in the amp (like every other amp in the world). This actually covers item #4 as well.
As far as item #5, there is no use to do this. Every amp uses a different arrangement for driving the output devices, both circuit topology-wise and physical arrangement-wise. So it doesn't really matter what the tempco of the diode is, you are going to have to add additional compensation elsewhere. What and how you add it is up to you.
Or just buy the Sanken devices. They are all set up to work as a perfectly matched system (although you still need a bias adjustment pot!). The only problem is that you HAVE to use their circuit, whether you like it or not. I don't like their circuit, so I don't use their parts. The ThermalTrak parts let me use any circuit I want. I just have to be smart enough to figure out how to make the thermal feedback work properly also.
As I said in my post, I was providing some constructuve feedback on the devices.
The issue is not about a cost adder on the devices - you only need to characterize them once as part of the transistor development cycle. This will take an engineer some weeks to do and to pull the data together and update the data sheet. This is an opportunity cost to the semiconductor manufacturer. The amortization or recovery cost adder on a high volume part for this activity is peanuts and should not affect it.
The motivation for characterising these devices more fully would be to help designers get to a working design more quickly, and to ascertain what the likely production spreads in an amp would be. Earlier on in this thread (and indeed in the On data sheet), the point is made about th e fact that trimless designs are possible. Well, to cater for this, you need to know what the productions spreads will be and to ensure your design can cater for them - hence characterisation data on the semiconductor devices concerned.
The issue is not about a cost adder on the devices - you only need to characterize them once as part of the transistor development cycle. This will take an engineer some weeks to do and to pull the data together and update the data sheet. This is an opportunity cost to the semiconductor manufacturer. The amortization or recovery cost adder on a high volume part for this activity is peanuts and should not affect it.
The motivation for characterising these devices more fully would be to help designers get to a working design more quickly, and to ascertain what the likely production spreads in an amp would be. Earlier on in this thread (and indeed in the On data sheet), the point is made about th e fact that trimless designs are possible. Well, to cater for this, you need to know what the productions spreads will be and to ensure your design can cater for them - hence characterisation data on the semiconductor devices concerned.
Charles, How about posting a GOOD schematic using Thermal Trac devices for the not so initiated as my self. Showing us how to incorporate them into Leach's amp circuit would be nice. We would have a T circuit and Thermal Trac's. Is that a viable option. Thanks for the long detailed thread. Tad
Charles Hansen said:
Charles, your thinking here is very one-dimensional. Maybe YOU would not make good use of the added engineering information, but you should not speak for other engineers. Most of them would rather not waste their time dunking these things in ice baths to find out what should have been on the spec sheet in the first place.
Bob
A key link for a successful bias compensation design
http://www.national.com/rap/Story/vbe.html
http://www.national.com/rap/Story/vbe.html
Hi Bob,
Your comment also applies to what I said, and I can see your point. But I must ask, when does an engineer ever blindly trust the data given? It's a direction, that's all. I will bet that the price would go up on the parts and it would not be any use to the better designers.
To engineer anything, you must know your parts. That helps define the better designers, wouldn't you say? Also, designing with the idea the there is a parameter spread also insures that newer replacement parts can be used once this first series has been superseded. So being lazy may preclude future service when required.
-Chris
Your comment also applies to what I said, and I can see your point. But I must ask, when does an engineer ever blindly trust the data given? It's a direction, that's all. I will bet that the price would go up on the parts and it would not be any use to the better designers.
To engineer anything, you must know your parts. That helps define the better designers, wouldn't you say? Also, designing with the idea the there is a parameter spread also insures that newer replacement parts can be used once this first series has been superseded. So being lazy may preclude future service when required.
-Chris
anatech said:
Hi Chris,
Designers should not blindly trust the data given, but when the data is not given in the first place, one may not even know what to trust to any degree.
Engineers like to know the spread, and they like to know what they can depend on and what they cannot depend on. It is all about risk management.
Putting extra information on a data sheet does not necessarily increase the cost of the part. Information on a data sheet can take many forms from mere typicals and characterization of what is there all the way up to performance grades based on selected product. The latter, of course, will increase the cost.
If you know the typical behavior for something, that is a start, and it is helpful, even if it is not guaranteed. Never throw the baby out with the bathwater - mine as much usefulness from the information as you can.
If something is not specified, yet it could be depended upon with some confidence, then there may be a lost opportunity to take advantage of that in some circuit configuration. At the same time, if something cannot be depended upon, it is also easy to communicate that with a spec as well - with just a big tolerance.
For example, OnSemi could have said, if you run 100 mA through the BJT and 25 mA through the diode, at room temperature the typical difference in junction drop will be 0 mV, but that will have a tolerance of +/- 30 mV. At least you bracket the behavior.
The ThermalTrak parts are indeed wonderful, and I like them just as much as Charles. I just think that if OnSemi had done a little more in the spec sheet (even those things that don't add to the cost of the part) that would have been helpful. That's all. Maybe I'm being too much of a stickler for specification. In my day job I often will have a fifty-page spec sheet for a 50-cent IC. Maybe I'm just spoiled. Even if OnSemi had just noted on their spec sheet that the thing was made of a separate BJT and a certain kind of diode, that would have been helpful.
Knowing the consistency of the BJT parts among same-sex and opposite sex, and the implied precision of OnSemi's overall process, I would not be surprized if there was fairly good consistency in the matching of certain characteristics between the diode and the BJT that one might be able to take advantage of in some circuit architectures. Maybe not. But without a spec, we can probably forget about it.
I also would not be surprized if there were designers out there who would be willing to pay an extra 50 cents per transistor to have matching between the diode and the BJT be within some range guaranteed by test. Never underestimate the creativity of a designer when you can give him something he can hang his hat on.
Cheers,
Bob
Hi Bob,
I do agree with you on these points, but I am fearful of a cost increase added and justified by some extra data with wide tolerances. I make it a point to test groups of devices from different lots before I have any confidence in them.
I must say that On Semi's process appears to be under control. I'm very happy with the transistor part but I haven't checked the diode section.
-Chris
I do agree with you on these points, but I am fearful of a cost increase added and justified by some extra data with wide tolerances. I make it a point to test groups of devices from different lots before I have any confidence in them.
I must say that On Semi's process appears to be under control. I'm very happy with the transistor part but I haven't checked the diode section.
-Chris
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.