Mixing ThermalTrak with non-ThermalTrak?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I am working with a friend to develop a high-power amplifier using several pairs of the On-Semi MJL32181A/MJL1302A output devices. I would like to take advantage of the internal diode in the ThermalTrak (NJL) versions of these devices, but I don’t want to specify the more expensive NJL for all the transistors when I need only one or two pairs for sampling the temperature for bias stability purposes. In large quantities the ThermalTrak devices cost almost 50% more. So we would prefer to specify the MJL for the majority of the devices which are not required for sampling temperature.

My assumption is that the transistors have identical dice in the NJL and MJL versions, the NJL only adding a MUR-120 diode to the internal header. The specs look the same, including the thermal resistance of junction-to-case. Therefore, my assumption is that a pair (or two as might be needed for Vbe drop) of NJL devices, which share output duties with adjacent MJL devices on the same heatsink should serve to provide bias stability as well as if all devices were the costlier NJL devices.

I will measure and test this combination, but I thought I would inquire here to see if anyone has actual experience mixing NJLs and MJLs together and using the smaller number of the NJLs to sample temperature for all. Does the added diode on the header cause the transistor which shares the header to perform differently in terms of thermal time constant or thermal resistance, or are there any pitfalls I have not considered?

Forgive me if this has been addressed before but I searched and didn’t see anyone mentioning mixing these devices.
 
If it is any consolation, when I check Bob Cordell's ltspice models, they are identical. there will be lot to lot variation however, unless you do some selecting/testing. I think Bob talked about it in the book, I'd have to find the page. Suggest to ltspice your new design, it can only help. Post your design.
 
Yes, I have been simulating this circuit with SPICE. I have several models for the 3281/1302 but I am using models that I got from Bob preferentially.

As I said, I really don’t expect any significant differences due to the added diode, but I thought I’d ask if anyone has had experience mixing MJLs and NJLs and relying on the NJL’s diodes to set bias for both device types. If there is a difference it might show up as different thermal time constants or resistances that would not be modeled in a standard SPICE simulation.
 
Harman Kardon do absolutely the same in their HK990 amplifier, using 5 pairs of output devices, where only one pair is Thermaltrak:

Sajti
 

Attachments

  • HK_out.png
    HK_out.png
    45.6 KB · Views: 364
There is no specification about how the diode tracks in a thermaltrack.
No time constant, no thermal resistance; no thermal capacitance.

The diode and the transistor are in the same case, on separate dies.
You get about the same as if the diode were glued onto the case.

You’re right that these parameters are not specified (that I have seen).

However, as I understand it, both devices’ dice are not just in the same case, but bonded next to each other on the same copper header. The header is electrically connected to the collector of the BJT while the diode is insulated. This thermal “intimacy” shortens the thermal time response by at least an order of magnitude and maybe more than two orders of magnitude, compared to the heatsink’s response. This makes the composite device useful for safer bias spreading. Bob Cordell describes the device well in his excellent book.
 
You’re right that these parameters are not specified (that I have seen).

However, as I understand it, both devices’ dice are not just in the same case, but bonded next to each other on the same copper header. The header is electrically connected to the collector of the BJT while the diode is insulated. This thermal “intimacy” shortens the thermal time response by at least an order of magnitude and maybe more than two orders of magnitude, compared to the heatsink’s response. This makes the composite device useful for safer bias spreading. Bob Cordell describes the device well in his excellent book.

Hi Brian,

There is no fundamental problem mixing NJL and MJL devices. The power transistor die are identical, and many commercial amplifiers that use ThermalTrak devices use one NJL and the rest MJL. Depending on your concern about beta and Vbe matching, just be aware that the die in the NJL and MJL parts you have most likely cam from different wafers, just as they would for MJL devices bought at different times.

If you are using Triple output stages, beta matching is usually not as important, as there is plenty of current gain. However, if you are using relatively large base stopper resistors and relatively small emitter resistors, current hogging among paralleled devices can be a concern of beta among the paralleled devices is significantly different. If you use 2.2 ohm base stoppers and 0.22 ohm emitter resistors, this should not be a problem.

Vbe matching is pretty good these days due to overall process uniformity. Nevertheless, it does not hurt to check Vbe between the MJL and NJL devices and see that they are not too far different, perhaps within 10mV of each other.

Finally, although there is no evidence of a difference in the spec sheets, it is conceivable that the NJL parts could have a lower junction-to-case thermal resistance if there was something about the construction that required a larger structure inside to make room for the diode on the header. I doubt this is the case.

Cheers,
Bob
 
There is no specification about how the diode tracks in a thermaltrack.
No time constant, no thermal resistance; no thermal capacitance.

The diode and the transistor are in the same case, on separate dies.
You get about the same as if the diode were glued onto the case.

Unfortunately, this is way wrong. There are comparison curves in my book that show the difference.

Cheers,
Bob
 
Bob,

Thanks for your reassurance about mixing MJLs with a few NJLs.

Your point about the dice coming from different batches is well noted.

The possibility that the header is larger in the NJLs or that the BJT is positioned closer to the edge of the header to accommodate the diode was the genesis of my original question.

Nevertheless I feel comfortable in proceeding with mixing the devices, and measuring to be sure.
 
You’re right that these parameters are not specified (that I have seen).
This thermal “intimacy” shortens the thermal time response by at least an order of magnitude and maybe more than two orders of magnitude, compared to the heatsink’s response. This makes the composite device useful for safer bias spreading.
How do you know ?
Anyway, it is your money.... Buy transistors 50% more expensive, for a diode inside that doesn't do anything specified better than a diode glued on the transistor case.
 
AX tech editor
Joined 2002
Paid Member
Hi Brian,

There is no fundamental problem mixing NJL and MJL devices. The power transistor die are identical, and many commercial amplifiers that use ThermalTrak devices use one NJL and the rest MJL. Depending on your concern about beta and Vbe matching, just be aware that the die in the NJL and MJL parts you have most likely cam from different wafers, just as they would for MJL devices bought at different times.

If you are using Triple output stages, beta matching is usually not as important, as there is plenty of current gain. However, if you are using relatively large base stopper resistors and relatively small emitter resistors, current hogging among paralleled devices can be a concern of beta among the paralleled devices is significantly different. If you use 2.2 ohm base stoppers and 0.22 ohm emitter resistors, this should not be a problem.

Vbe matching is pretty good these days due to overall process uniformity. Nevertheless, it does not hurt to check Vbe between the MJL and NJL devices and see that they are not too far different, perhaps within 10mV of each other.

Finally, although there is no evidence of a difference in the spec sheets, it is conceivable that the NJL parts could have a lower junction-to-case thermal resistance if there was something about the construction that required a larger structure inside to make room for the diode on the header. I doubt this is the case.

Cheers,
Bob

I wonder if it is absolutely necessary to have a pair ThermalTraks. Would a single device not work as well, with a suitably dimensioned bias circuit?
Assuming all devices on the same heatsink.

Jan
 
I wonder if it is absolutely necessary to have a pair ThermalTraks. Would a single device not work as well, with a suitably dimensioned bias circuit?
Assuming all devices on the same heatsink.

Jan

Good question. In other words could you double the drop across a single ThermalTrak diode to provide the sensing needed for a complementary pair? Sounds reasonable.
 
How do you know ?
Anyway, it is your money.... Buy transistors 50% more expensive, for a diode inside that doesn't do anything specified better than a diode glued on the transistor case.

Page 309 of Bob’s book (original version) shows measured tracking diode response times for a ThermalTrak diode versus two other common mounting options. The ThermalTrak diode responds much faster.

The 50% greater cost for the ThermalTrak device versus the regular BJT-only device is high, but if you only need one or two among many output devices in a big amp, then it’s reasonable. Hence my original question.
 
I wonder if it is absolutely necessary to have a pair ThermalTraks. Would a single device not work as well, with a suitably dimensioned bias circuit?
Assuming all devices on the same heatsink.

Jan

Hi Jan,

In principle that works, i.e., just make one NPN the ThermalTrak and use only one diode for thermal feedback. There is a caveat, however. Depending on the details of the bias spreader and how picky you are, using only one ThermalTrak will give a thermal feedback voltage that is potentially rich in second harmonic at low frequencies. Using one NPN ThermalTrack and one PNP ThermalTrak tends to cancel out this dynamic effect. I could be wrong, but I seem to recall that the thermal time constant from BJT die to the diode is fairly fast, maybe as little as 20 ms. I just don't have my book in front of me now. This consideration may be in the "everything matters" category for highest quality audio performance.

Cheers,
Bob
 
How do you know ?
Anyway, it is your money.... Buy transistors 50% more expensive, for a diode inside that doesn't do anything specified better than a diode glued on the transistor case.

I disagree with your assertion, but I would be happy to reconsider my position if you can answer yes to at least twof of the following five questions:

1. Have you measured the thermal characteristics of a ThermalTrak device?

2. Have you built and measured an amplifier that uses ThermalTrak devices?

3. have you read Chapter 14 of my book, published in 2010?

4. Have you read Chapter 22 of Doug Self's book, published in 2013?

5. Have you read a credible article by someone else that contradicts my findings and those of Doug Self?

Cheers,
Bob
 
I disagree with your assertion, but I would be happy to reconsider my position if you can answer yes to at least twof of the following five questions:

1. Have you measured the thermal characteristics of a ThermalTrak device?

2. Have you built and measured an amplifier that uses ThermalTrak devices?

3. have you read Chapter 14 of my book, published in 2010?

4. Have you read Chapter 22 of Doug Self's book, published in 2013?

5. Have you read a credible article by someone else that contradicts my findings and those of Doug Self?

Cheers,
Bob
1 & 2 No, I do not need ThemalTrack devices.
3 & 4 No, I do not need these books.
My first audio power amplifier was built in 1967 a modest 2 x 10W rms using germanium power transistors AD149. Perfectly stable, I did not wait for ThermalTracks nor the said books.

My best regards.
 
Sanken employee? :)

Lol, unlikely as sanken itself has the STD03N, and STD03P which use the same diode temperature sensing principle ;-)

BTW I am curious as to if you considered the sanken's as they work quite well and measure well too. For example Jeff Rowland's stereo model 625 has 12 sanken STD03 power transistors for the output stage per channel and is able to dish out 600 watts into 4 ohms.

https://i.nextmedia.com.au/Assets/2..._685_s2_power_amplifier_review_test_lores.pdf
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.