jneutron said:
Hey, there's an echo in here..
dinna I jusss say that???
Anywhoo..
I modelled the chip, they give .625 C per watt as junction to case.
I put in typical die and copper thickness, to get that number requires about a 130 by 130 chip.
With 50 watts input and a 40 degree heatsink, the numbers are:
Junction 103 C
top of baseplate 80 C
So at 50 watts, I'd have the diodes 40C over heatsink but 23C under junction...2/3rds of the thermal excursion..
Course, that 80 C is at the silicon/copper interface..the diode could easily be at 50% excursion...
Anybody know the actual die dimensions?
Cheers, John
John
That’s interesting.
I just have to do some thinking.
BTW: Good to see the thread back on Trak.
Stinius
jneutron said:
Hey, there's an echo in here..
dinna I jusss say that???
Anywhoo..
I modelled the chip, they give .625 C per watt as junction to case.
I put in typical die and copper thickness, to get that number requires about a 130 by 130 chip.
With 50 watts input and a 40 degree heatsink, the numbers are:
Junction 103 C
top of baseplate 80 C
So at 50 watts, I'd have the diodes 40C over heatsink but 23C under junction...2/3rds of the thermal excursion..
Course, that 80 C is at the silicon/copper interface..the diode could easily be at 50% excursion...
Anybody know the actual die dimensions?
Cheers, John
What do you use for thermal modeling? Sounds like fun (and useful!) stuff
I guess 130x130 is in mils? I just had a look at the pictures at Rod Elliot's site of fakes and originals and a MJ(L)21193 chip looks like about 5.6x5.6mm = 220x220 mils. I'd expect the MJL3281 to be about the same size as same power rating is claimed. (high voltage SOA is different though)
There is also pictures of similar size Sanken transistors, one picture has text that says die is 5x5mm. Also I've read somewhere on the forum that Toshiba 2SC5200/2SA1943 use 5x5mm dies. So it seems to be in the 200x200 mil range for this size of transistor.
Lots of fun pictures:
http://sound.westhost.com/counterfeit.htm
I particulary like the picture of a 2N3055 besides a MJ21193... There is some difference in die size there. Also, there is a picture of a large plastic transistor fake with a die that's even smaller than the 2N3055!
A question I have:
About how big is the thermal capacity of a typical 5x5mm chip by itself? (that is, "capacitance" in an electro-thermal model)
syn08 said:
I may know a thing or two
Conceptually, the same equivalent schematic you got from Motorola applies for the NJL; all decent power devices (be it in TO3, TO247 or TO218, fakes not included) are mounted using what is called a "copper pre-form". Now, the values for the elements are generally not exactly known and no semiconductor manufacturer is going to oficially reveal and guarantee such parameters. What you got is most likely the result of some calculations, simulations and estimates for TO3.
I may crack a new NJL device, but last time I've checked the diode is not mounted using a copper preform, (this would connect the diode substrate to the collector; also copper preforms are expensive in terms of mass production) but using insulating epoxy. Which makes the thermal series resistance much larger than the transistor silicon-preform stack. Therefore, it is safe to assume there is a significant temeperature gradient between the transistor chip and the diode chip. How large, it is hard to say, only some calculations could be made... My estimate (based on other devices I worked with) is 10-15%
Look at these NJL devices as the best of the worlds in terms of temperature tracking; the Sanken devices are a little bit more elaborated but they are essentially not that much better. The ultimate device would have to be built with on chip, dielectrically insulated, diode but I certainly doubt this will ever be in production due to the economics. A separate diode chip is almost free, while an integrated, dielectrically insulated diode will most likely double the cost of the transistor itself. Add to this the lack of flexibility (OnSemi is selling the same transistor chips but without the diode, in the three pin TO218, as the MJL series) to understand why we are probably never going to get the goodies we are hoping for. All we can do is to understand how these devices are working and hence how to use them to our best. In this sense, what John is doing is very good stuff.
Douglas,
What about the following estimate? (after your test #6)
Lets estimate a safe power dissipation of 10 W ( normally 8 w) in the two transistors so 5W per transistor
Let's take a 1°C/W thermal resistance between junction and case
Let's suppose you have a good finger and heatsink is 40°C
Then Junction temperature = 40°C + 5 = 45°C
Let suppose a diode temp at half way as suggested by JNeutron.
Tdiode 42.5°C
Let's have a diode temco at 9.2mA of 1.6mV/°C
Let suppose a transistor Vbe tempco on the high side: 2.5mV/°C
Delta Vbias required = ( 45-25) x 2.5
Delta Vbias furnished = (42.5-25) x Tc
Tc required is Tc= 1.15 x 2.5 = 2.88
Mutiplication factor required 2.88/1.6 =1.8
If we repeat this with junction transistor=50°C and junction diode = 45°C
Multiplication factor required: 1.25 x 2.5/1.6 = 1.9
This is close to your requirement to double the Tempco.
Is this correct?
JPV
I just ran characteristic curves for JPV's 3xxx transistors. He's got a really well matched set there. The curves are virtually identical for all 4 NPN's and all 4 PNP's respectively.
All of his PNPs have a Hfe of between 90 and 110 for collector currents from 100 mA to 10A.
The NPN's had an Hfe from 100 to 120 with Ic from 100 mA to 10A. Sample "N1" was slightly out of family at lower Ic with a Hfe of 140.
JPV, I took photos of the curve tracer plots for you... 3 for each unit, 24 in all. I'm ready to send the units to you...
If anyone else is interested in seeing the traces, please email me privately and I'll send a ZIP file.
-------------------------------------------
My 4xxx units aren't quite so well matched, but exhibited almost twice the Hfe of the 3xxx in all cases. The NPNs and PNPs ran from 160 to 210 for Hfe. No photos for these.
Cheers,
John
All of his PNPs have a Hfe of between 90 and 110 for collector currents from 100 mA to 10A.
The NPN's had an Hfe from 100 to 120 with Ic from 100 mA to 10A. Sample "N1" was slightly out of family at lower Ic with a Hfe of 140.
JPV, I took photos of the curve tracer plots for you... 3 for each unit, 24 in all. I'm ready to send the units to you...
If anyone else is interested in seeing the traces, please email me privately and I'll send a ZIP file.
-------------------------------------------
My 4xxx units aren't quite so well matched, but exhibited almost twice the Hfe of the 3xxx in all cases. The NPNs and PNPs ran from 160 to 210 for Hfe. No photos for these.
Cheers,
John
ostripper said:An externally hosted image should be here but it was not working when we last tested it.
That is a copper preform (on semi -njw0281) t-trak has same
die but with integrated diode. There seems to be a very thin
layer of grey conductive epoxy under silicon, molded HIGH
impact thermoplastic on top..
Have smashed blown sanken OP devices w/internal diode
comp. ,same construction..
Thanks for that. Can I just check I'm seeing right? I assume I'm looking at the metal plate, and the underside of the part in the photo is the metal face you usually see. The chip is the white square, but what is the yellow area?
What I would really like to know is what is between the junction and the plastic side. Does the chip sit in a little hollow in the plastic? Is that hollow full of air, or maybe silicon grease? How deep is the hollow?
Any info would be useful.
jneutron said:
Hey, there's an echo in here..
dinna I jusss say that???
Anywhoo..
I modelled the chip, they give .625 C per watt as junction to case.
I put in typical die and copper thickness, to get that number requires about a 130 by 130 chip.
With 50 watts input and a 40 degree heatsink, the numbers are:
Junction 103 C
top of baseplate 80 C
So at 50 watts, I'd have the diodes 40C over heatsink but 23C under junction...2/3rds of the thermal excursion..
Course, that 80 C is at the silicon/copper interface..the diode could easily be at 50% excursion...
Anybody know the actual die dimensions?
Cheers, John
John, I reckon you've put your finger on it. I think we have to face the fact that the diode is nowhere near the junction temperature, and that is why I needed a multiplication factor of four to get correct compensation.
I must admit I'm a bit disappointed.
DouglasSelf said:
Correct. This is what I also mentioned in my early post http://www.diyaudio.com/forums/showthread.php?postid=1683583#post1683583 second paragraph.
DouglasSelf said:
John, I reckon you've put your finger on it. I think we have to face the fact that the diode is nowhere near the junction temperature, and that is why I needed a multiplication factor of four to get correct compensation.
I must admit I'm a bit disappointed.
Did you need a factor of two or a factor of 4 ??
The estimation I gave above shows for your experiment a factor of two but four ... then I made a mistake.
JPV
Doing some background reading last night I came upon this graph in "Bipolar Semiconductor Devices" by David Roulston, (see http://www.dself.dsl.pipex.com/ampins/library/library.htm) which gives insight into temperature drops actually inside a power transistor chip. It is the result of simulation rather than measurement. I believe quoting it here comes within "fair use".
In the diagram YS is the thickness of the silicon at 200um. YH is not defined, but at a thickness of 1400um (=1.4mm) it is presumably a metal header. As you can see, the maximum temp difference between junction and the top of the header is about 14 degC; between junction and the bottom of the header it is 41 degC. Diodes mounted on top of the header would apparently see almost exactly 2/3 of the junction-case temperature drop, just as predicted by jneutron in Post #379. However, as you move away from the junction sideways, the temp on top of the header drops rapidly towards the temp at the bottom of the header, so 2/3 is probably very optimistic. A lot depends on just where the diode is mounted.
In the diagram YS is the thickness of the silicon at 200um. YH is not defined, but at a thickness of 1400um (=1.4mm) it is presumably a metal header. As you can see, the maximum temp difference between junction and the top of the header is about 14 degC; between junction and the bottom of the header it is 41 degC. Diodes mounted on top of the header would apparently see almost exactly 2/3 of the junction-case temperature drop, just as predicted by jneutron in Post #379. However, as you move away from the junction sideways, the temp on top of the header drops rapidly towards the temp at the bottom of the header, so 2/3 is probably very optimistic. A lot depends on just where the diode is mounted.
Attachments
DouglasSelf said:I reckon you've put your finger on it. I think we have to face the fact that the diode is nowhere near the junction temperature, and that is why I needed a multiplication factor of four to get correct compensation.
I must admit I'm a bit disappointed.
Because of the large thermal mass of the heatsink and the different thermal paths of the ThermalTrak diode die vs. transistor die, will diodes glued on the back of a plastic package have equal or superior thermal tracking to ThermalTraks?
Perhaps everything is not so bad.
The thermal circuit is linear, this means that in steady state the delta T of the diode is proportional to the delta T of the transistor chip for a specific delta power input
There is only one source of heat: the transistor, it is then possible to correct the tempco by multiplying it with a constant.
One measurement is enough to estimate this correction and it looks like it is a factor of two to correct a diode tempco of 1.6mV/°C.
If we bias the diode at 0.1mA, then the temco is multiplied by 1.2 and the gain of the Vbe multiplier should be 1.6. Reasonnable.
The transient will remain a problem. But the copper specific heat is half the specific heat of silicon and if the diode is close to the chip, the copper path has perhaps a negligeable capacitance. The ratio of mass of the two chips is the driving parameter.
The transient mistraking will be only relevant at very low frequencies so the crossover transient distortion due to mistracking. This is a good reason to use multiamplifiers and electronic crossovers.
JPV
The thermal circuit is linear, this means that in steady state the delta T of the diode is proportional to the delta T of the transistor chip for a specific delta power input
There is only one source of heat: the transistor, it is then possible to correct the tempco by multiplying it with a constant.
One measurement is enough to estimate this correction and it looks like it is a factor of two to correct a diode tempco of 1.6mV/°C.
If we bias the diode at 0.1mA, then the temco is multiplied by 1.2 and the gain of the Vbe multiplier should be 1.6. Reasonnable.
The transient will remain a problem. But the copper specific heat is half the specific heat of silicon and if the diode is close to the chip, the copper path has perhaps a negligeable capacitance. The ratio of mass of the two chips is the driving parameter.
The transient mistraking will be only relevant at very low frequencies so the crossover transient distortion due to mistracking. This is a good reason to use multiamplifiers and electronic crossovers.
JPV
DouglasSelf said:
When you tried the circuit and found you needed the mulitplication of about 2, was this needed to get stable bias with heatsink temperature?
If it was, then it couldn't be because the diodes are at an intermediate temperature between junction and sink because when bias is constant (which it is when tempco is set right) temperature from junction to case will be the same independent of what the temperature of the sink is. A 10 degree increase of heatsink temperature will cause an increase of 10 degrees for both the diode and transistor junction.
Transients are another matter though, but if temperature coefficient is increased to track under transients it will be overcompensated for heatsink and case temperature changes due to thermal resistance of insulator. Putting drivers on the heatsink may or may not help with this, increasing back the bias when heatsink temperature increases. Won't compensate for the thermal resistance and drop over the insulator though.
DouglasSelf said:
I understand that. They are not tracking perfectly. My question is related to the amount of misstracking.
Is it correct to say that you had to multiply the delta T of the 2 diodes by two to compensate the delta T of the two ouput transistors
Was it two or four? I am talking about compensating the delta's.
What we have to do is creating a Vbias equivalent to the Vbe's of the ouput transistors and we can get there without having the diodes at the same temperature as the ouput transistors if everything is proportionnal and if we know the correction factor.
Am I correct?
Cheers
JPV
JPV said:
The tempco of the diodes was multiplied by about four times to get approximately correct compensation. Looking back at the data, the actual factor was somewhat below four.
Yes, if all the temperatures were proportional then you could simply set the tempco of the bias generator to a figure that would give perfect compensation. But this only applies in a static situation. Given that a decent-sized heatsink can take an hour to get within one degree C of the final temperature, when are you going to have that situation? Surely the whole point of the TTraks is that they promise better compensation under dynamic conditions.
For those who want to go further than emitter resistors and match these units, here's Hfe data for all of my 4xxx ThermalTrak transistors (I received 5 more NJL4281 today) to give you an idea of how many you'd need to get to be able to find a matched set.
NJL4281: (sample: Hfe @ Ib=1mA)
Sample 1: 205
Sample 2: 150
Sample 3: 160
Sample 4: 215
Sample 5: 190
Sample 6: 140
Sample 7: 130
Sample 8: 140
Sample 9: 210
Sample 10: 220
NJL4302 (sample: Hfe @ Ib=1mA)
Sample 1: 180
Sample 2: 175
Sample 3: 180
Sample 4: 180
Sample 5: 180
Looks like I didn't get as lucky with matching for my NPN's as JPV did with all of his units. All of my PNPs were from the same lot. Samples 1-5 of my NPNs were from one lot, samples 6-10 from another...
John
NJL4281: (sample: Hfe @ Ib=1mA)
Sample 1: 205
Sample 2: 150
Sample 3: 160
Sample 4: 215
Sample 5: 190
Sample 6: 140
Sample 7: 130
Sample 8: 140
Sample 9: 210
Sample 10: 220
NJL4302 (sample: Hfe @ Ib=1mA)
Sample 1: 180
Sample 2: 175
Sample 3: 180
Sample 4: 180
Sample 5: 180
Looks like I didn't get as lucky with matching for my NPN's as JPV did with all of his units. All of my PNPs were from the same lot. Samples 1-5 of my NPNs were from one lot, samples 6-10 from another...
John
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.