| Samuel Jayaraj |
What are typical voltages and resistance values involved in matching Lateral Mosfets like 2sk1058 and 2sj162? Could someone post a schematic if possible?
Thanks! |
|
|
| Charles Hansen |
One relatively easy way is to connect the drain to the supply, the gate to ground, and the source to a current source that has the other end grounded. Then just measure the voltage across the current source to find Vgs.
On the one hand you want to come reasonably close to the actual operating conditions of the transistor, which will typically be 50 Vds at 100 mA or so. However, this would dissipate 5 watts in the device which would destroy it unless heatsinked. So instead pick something smaller. I would suggest 10 Vds and 25 mA. This gives a total device dissipation of 250 mW, which will be fine.
You can make a current source from a 3-pin regulator if you read the app notes.
Don't try to match the P-channel devices to the N-channel devices. Just match each polarity to itself. |
|
|
| Charles Hansen |
| Also, I think there is an article on matching (vertical) MOSFETs on the passdiy.com site, but it seems to be down at the moment. :( |
|
|
| lucpes |
| Did not test this, so an 'approval' reply is needed: |
|
|
| jam |
Charles,
What is your prefered method of matching bipolar devices?
Regards,
Jam |
|
|
| Charles Hansen |
The method lucpes posted is really only suitable for vertical devices. With this approach you will only apply Vgs (threshold) across the device. This is OK for vertical devices as you will have 3 or 4 volts Vds. But for lateral devices you will only have a few hundred millivolts, which is kind of sketchy.
However if you don't want to build the current source as I proposed in my previous post, you could just use a resistor as lucpes proposed. Please note that it should replace the current source in my posting and be connected between source and ground.
Also, please be aware that the pinouts in lucpes' diagram are for vertical MOSFETs. The lateral MOSFETs have the source and drain pins interchanged. |
|
|
| Charles Hansen |
| The PassDIY site is back up now. However their matching article uses the same method as in lucpes' post and is therefore only suitable for vertical devices. |
|
|
| jackinnj |
| quote: | Originally posted by Charles Hansen
.... However, this would dissipate 5 watts in the device which would destroy it unless heatsinked. So instead pick something smaller. I would suggest 10 Vds and 25 mA. This gives a total device dissipation of 250 mW, which will be fine. |
I'm pretty sure that IRF uses a 20uS pulse -- I haven't gotten around to making a programmable pulse generator to automate the process, i.e. to ramp the 20uS pulses in voltage to measure transconductance etc.
I think (correct me if I'm wrong) they (IRF, not Hitachi) also characterize the threshold at 100mA.
the thought of a programmable pulse generator is percolating somewhere between the front and back burners right now... |
|
|
| Nelson Pass |
| If you're only matching, I beleive it will work fine with both types. |
|
|
| Charles Hansen |
Sorry, my previous instructions were not correct. They would work for a depletion mode JFET, but not for an enhancement mode MOSFET.
You will need to add a second supply as shown in the attached schematic to test lateral MOSFETs. You can either use a true current source, or just substitute a resistor. I would recommend +/- 10 volts for the supplies. Then for the 2SK1058/2SJ162 parts I would recommend either a 25 mA current source or a 350 ohm resistor. These values are not critical and can easily be varied +/- 50% or more. |
|
|
| Christer |
| quote: | Originally posted by Charles Hansen
Don't try to match the P-channel devices to the N-channel devices. Just match each polarity to itself. |
I have always got the impression that it is recommended to
match the N and P devices and now you recommend against
this. Could you please elaborate a bit on why you make this
recommendation? This has immediate practical consequences
which I have failed to get satisfactory answers to on previous
occasions. |
|
|
| Charles Hansen |
There are a couple of reasons for this. The first is a practical one, in that it is almost impossible to match Vgs across both polarities. I suppose that you could do it by throwing away 60% to 80% of the parts you buy.
The next reason is that the N-channel parts have a lower output conductance than the P-channel parts. This is especially true for the Toshiba brand MOSFETs (as opposed to the Magnatec clones which are *much* better in this regard). This means that whatever reading you get will depend slightly on Vds for the N-channel parts, but significantly for the P-channel parts. So you could only match them for one specific Vds condition.
But guess what? When you are running them as followers in an audio amp, Vds is constantly changing!
Besides, about the only thing you are losing is a slight DC offset by not matching them. This is easily compensated for elsewhere in the circuit. The bottom line is that there is no point whatsoever to try and match across polarities in the output stage. |
|
|
| Christer |
Charles,
Thankyou for the explanation. Yes the practical problem of
matching devices is what has worried me, especially if one
ends up with a bunch N devices from the same batch and a
number of P devices from the same batch. I have tried to get
opinions on this problem several times without getting much
wiser. It would simplify
things a lot if one doesn't have to worry about matching.
Your point that matching won't result in equivalent behaviour
is a good point and I guess it applies also to BJTs for instance.
A counterargument might be that one shouldn't match for a
certain parameter to be equal, but aim for a deliberate mismatch
that gives the best overall compromise, but it is probably difficult
to find the optimal compromise.
I don't really know about the MOSFETs, but I did some simulations
about the effect of mismatching BJTs in class A OPS. I found that
mismatching NPN and PNP devices resulted in increased
even order distorsion, but almost unaltered odd-order distorsion.
Of ocurse, those are only simulation results and have to be
taken as such.
BTW, I had planned on using the Exicon clones of the Hitachi
K162 and J1058 devices. They seem to be the cheapest
alternative and have slightly better spec's than the Hitachis.
Do you have any experience of them? The Magnatecs are also
much cheaper than the Hitachis, of course and seem rather
easy to get. |
|
|
| Charles Hansen |
| The Magnatecs are quite a bit better than the Hitachi parts. The Hitachi N-channel parts are fine, but the P-channel parts are very "triode-like" in their characteristic curves (high output conductance). This lowers their PSRR and makes them a poorer match for the N-channel parts. Also the zero tempco point differs by about 25% between polarities on the Hitachi parts but is virtually identical for the Magnatecs. |
|
|
| Charles Hansen |
As far as the mis-matching of opposite polarity's impact on distortion, this really only applies to mismatches of transconductance. But MOSFETs are pretty uniform in this regard, so this is not really a problem. Where they differ is in Vgs (threshold), which as I said only has an effect on DC offset. That is easily corrected elsewhere in the circuit.
The main reason to match output devices is just to make several separate (same polarity) transistors act as one big composite device. You need to match them so that they are sharing the work more-or-less equally. |
|
|
| Christer |
| quote: | Originally posted by Charles Hansen
The Magnatecs are quite a bit better than the Hitachi parts. The Hitachi N-channel parts are fine, but the P-channel parts are very "triode-like" in their characteristic curves (high output conductance). This lowers their PSRR and makes them a poorer match for the N-channel parts. Also the zero tempco point differs by about 25% between polarities on the Hitachi parts but is virtually identical for the Magnatecs. |
OK, well the Hitachis are not my first choice anyway. Although I
can buy them off-the-shelf locally, they are about three times
as expensive as ordering Exicons or Magnatecs from the UK.
I take it you haven't tried the Exicons then? I have seen some
people report they sound slightly better than the Hitachis. |
|
|
| Charles Hansen |
| Exicon = Magnatec |
|
|
| Christer |
| quote: | Originally posted by Charles Hansen
Exicon = Magnatec |
No, not at all, unless they sell the same devices undet two
different brand names and with different device numbers.
As I understand it Magnatec is owned by Semelab and sell
Magnatec lateral MOSFETs (as well as other things). They also
sell other brands, but not Exicon. It is not clear if the devices
are manufactured by Semelab of Magnatec themselves.
Exicon is a brand name owned by Profusion, who in addition
also sell a lot of other brands including semelab products, but
not the Magnatex LMOSFETs. It is unclear also here who
manufactures the Exicon devices.
http://www.magnatec-uk.co.uk/
http://www.semelab.co.uk/
http://www.exicon.com/
http://www.profusionplc.com/
It could, perhaps, be that both Magnatec and Profusion are
just distributors and for some reason buy identical devices
from Semelab, but branded by their own name. |
|
|
| Charles Hansen |
| Exicon = Magnatec |
|
|
| Christer |
OK Charles,
You seem so certain so I guess I'll have to believe you. There is
nothing in any those links I posted that reveals any such
connections, but there is nothing that contradicts it either.
So what we have then are two different brand names using
two different sets of device numbers for exactly the same
physical devices (either die or whole package)? |
|
|
| tcpip |
A bit of a diversion, but are you all sure that device matching is needed for OPS? I had once asked Randy (Slone) about this, and he said that he's never matched OPS devices. He said that the differences in hfe and other parameters from device to device would be taken care of by the negative feedback. Of course, it's possible that amps designed by others may not have enough NFB to achieve this.
What are your views on this? I don't know enough about amp design to have views of my own. All I know is that when I build a Randy Slone amp, I won't bother to match the OPS devices. |
|
|
| Charles Hansen |
| I guess it just goes to show that you can be wrong, even if you write a book about it. |
|
|
| jackinnj |
| quote: | Originally posted by Charles Hansen
I guess it just goes to show that you can be wrong, even if you write a book about it. |
two books now -- |
|
|
| tcpip |
| quote: | Originally posted by jackinnj
two books now -- |
As I said, I don't know enough about amp design to have my own opinions. All I know is that unless he's lying, he's getting phenomenal measured performance, and amp builders on this forum who have built his OptiMOS (see "sam9" and "5th element" postings elsewhere) are raving about the lovely sound, and all this without matching OPS devices. I guess Randy Slone may not know how to build amps, but whatever he knows will probably be good enough for me. :D |
|
|
| Charles Hansen |
Obviously his amps work, so that's not the issue. But if the output stage FETs are not matched, then they are not all doing their "fair" share of the work. So maybe some are doing 50% of their share and some are doing 150%. Or maybe worse than that.
I suppose that the amps would even work if you just used a single pair of output devices. But there is a difference between just "working" and offering truly high performance.
To my way of thinking, it is false economy to save on the extra "work" or "cost" of matching the output devices. Think about how much time you spend on your project. Is it really worth it to save $10 or $20 plus an hour or so of time to not match the parts?
Nelson Pass says this on the PassDIY website:
"In the case of a power output stage, we often want to
parallel power devices so as to share the current and heat
dissipation, as in Figure 3a. Here it is desirable that the
devices be identical so that the current is shared equally.
If the Vgs of each device is different, some device will
find itself doing the disproportionate share of the work
while some other device might not be turned on at all."
I'll put my money with Nelson any day of the week. |
|
|
| millwood |
| quote: | Originally posted by Charles Hansen
But if the output stage FETs are not matched, then they are not all doing their "fair" share of the work. So maybe some are doing 50% of their share and some are doing 150%. Or maybe worse than that. |
maybe it is indeed worse than that. But can you hear it?
I thought the best test is for some of us to intentionally mis-match output transistors and see if we can hear that inferior sound from mis-matched transistors.
If we cannot, what benefits do you get from matching transistors?
for the record, I haven't matched my transistors for a long time. |
|
|
| slowhands |
| quote: | Originally posted by tcpip
.....Randy (Slone) ...said that he's never matched OPS devices. He said that the differences in hfe and other parameters from device to device would be taken care of by the negative feedback. Of course, it's possible that amps designed by others may not have enough NFB to achieve this. .... |
It depends on how much you want to equalize current between devices. One school of thought might be to use large emitter resistors and not select parts, because the emitter resistors will encourage current sharing. This is practical with bipolars, not with mosfets. MOSFETs vary so widely in Vgs(on) and transconductance that matching is probably wise, since no practical source resistor will equalize currents. The good news is that once you have laterals matched, they share well and you don't need source resistors at all (but I use 0.1 ohm source resistors to simplify measuring bias currents, lazy me).
I do wonder if matching of MOSFETs might not be needed in one case. In the CF configuration, if you use multiple dedicated drivers, one per output device, and tie the driver emitter to the drain and drain resistor, the driver will compensate for wide variations in MOSFET parameters. It costs money to match the MOSFETs, and individual drivers are cheap, so would be a win. Wherever possible I try to design out the need for fussy adjustments and matching.
I have not been able to do a stable MOSFET CF output stage (Grr.... I'll get it if it kills me!!!), so I have not tried this but I want to. 100% local negative feedback is a wonderful thing, if it can be made stable. |
|
|
| Christer |
| quote: | Originally posted by Charles Hansen
Obviously his amps work, so that's not the issue. But if the output stage FETs are not matched, then they are not all doing their "fair" share of the work. So maybe some are doing 50% of their share and some are doing 150%. Or maybe worse than that.
I suppose that the amps would even work if you just used a single pair of output devices. But there is a difference between just "working" and offering truly high performance.
To my way of thinking, it is false economy to save on the extra "work" or "cost" of matching the output devices. Think about how much time you spend on your project. Is it really worth it to save $10 or $20 plus an hour or so of time to not match the parts?
|
I suppose you are now talking only about matching parallelled
N devices with each other and parallelled P devices with each
other, since previously in this thread you recommended
against matching complementary LMOSFETs? While I see your
previous point about matching complementary devices I have
a suspicion that the mismatching between the complementaries
may still be a bigger problem than mismatching between
parallelled devices. Some people, including Borbely and certain
people on this forum, do acutally sell matched complementary
pairs of Hitachi L-MOSFETs. Why would Borbely do that unless
he thinks it matters?
Of course, this does not in any way provide an argument
against matching parallell devices, which seems a sensible
thing to do.
Edit: The above was mainly considering distorsion etc. Of course,
matching parallell devices or otherwise guaranteeing a fair
current sharing is necessary to meet the intended power and
current handling spec. |
|
|
| lumanauw |
What is the Vgs of this K1058 pairs? Is the Vgs for Nch and Pch differs alot like mosfets or similiar like bipolars?
I'm looking for device for VAS. Bipolars have too much gain, Mosfets and fets have lower gain, but their Vgs is not predictable as bipolars. Has anyone try this lateral mosfet for VAS? |
|
|
| tcpip |
| quote: | Originally posted by slowhands
I do wonder if matching of MOSFETs might not be needed in one case. In the CF configuration, if you use multiple dedicated drivers, one per output device, and tie the driver emitter to the drain and drain resistor, the driver will compensate for wide variations in MOSFET parameters. It costs money to match the MOSFETs, and individual drivers are cheap, so would be a win. Wherever possible I try to design out the need for fussy adjustments and matching. |
This is very interesting. I have not read about this idea anywhere else. Let me know if you have any other details or ideas along these lines, just so that I can understand this a bit better. |
|
|
| audionline |
Can you design a best and simpleness schematic for K214/J77 matching Please?
Thank you very mach /... |
|
|
| Charles Hansen |
| quote: | Originally posted by Christer
Some people, including Borbely and certain
people on this forum, do acutally sell matched complementary
pairs of Hitachi L-MOSFETs. Why would Borbely do that unless
he thinks it matters?
Edit: The above was mainly considering distorsion etc. Of course,
matching parallell devices or otherwise guaranteeing a fair
current sharing is necessary to meet the intended power and
current handling spec. |
You would have to ask Borbely why he does that, because it makes no sense to me.
In my experience, the transconductance of the MOSFETs is very consistent from part-to-part. What varies is the Vgs (threshold). If the transconductance is mismatched it will create distortion (essentially only a small amount of 2nd harmonic). But that is not an issue because the parts don't vary in this regard. If Vgs (threshold) is mismatched it will create a small DC offset. That is trivially compensated for elsewhere. |
|
|
| Charles Hansen |
| quote: | Originally posted by lumanauw
What is the Vgs of this K1058 pairs? Is the Vgs for Nch and Pch differs alot like mosfets or similiar like bipolars? |
Typical Vgs (threshold) is between 0.4 V and 0.8 V. It is fairly random for both N-channel and P-channel. |
|
|
| Charles Hansen |
| quote: | Originally posted by audionline
Can you design a best and simpleness schematic for K214/J77 matching Please? |
Just use the previous schematic I posted. You will want to reduce the operating current. If you are using a current source, try 10 mA or so. If you are using a resistor, try increasing the value to 1000 ohms or so.
(Normally there is no need to match these smaller parts. If you are planning to parallel them, you would be better off to use the bigger parts instead.) |
|
|
| audionline |
OK...I will try it see the effect..
Thank you very mach .Good luck everyday ... |
|
|
| sam9 |
First a generalized observation about buying BJTs that seems to apply to both small ignal devices and output devices: When I buy a large quantity, 20 for example, the odds are that the difference between the highest and lowest measurement is less than when I buy just two. Some devices have lot numbers printed on them; getting out my magnifying lens and squinting at them confirms the following scenario.
This is counterintuitive at first glance but I think there is a reason that I can infer from being around stockroom operations from time to time. When a large quantity order is being pulled from stock, pre-packaged batched are just grabbed off the shelf and these are batches more likely to come from a single production run - maybe even the same wafer if you are lucky. When you order one or two, the pieces are pulled from a "left-overs" or "odds and ends" bin which have an inherently lower uniformity. I've seen a similar practice with a variety of products --that it is independant of the nature of the product. It applies equally to nuts&bolts, donuts, window frames and semi-conductors.
I see no reason why the phenomonon should be limited to BJTs. Slone probably buys his L-MOSFETS in larger batches than most DIYer's and is more likely to get devices made in the same run. Add to this that he seems to be using Magnatec devices. Since they make a "double die" package where two dies are package in parrallel without separate gate stoppers, drain or source resistors or any way to add such, it is likely that by necessity they have achieved a manufacturing uniformity greater than "average". Note they have a "quad-die" product under development which places even greater demands on their process. It is not likely to be just be post process sorting because that gets way too expensive, especially if you are trying to do a four way match. Thus I suspect that there is a high probability that Slone is correct with regard to the stock of components HE buys but it mostly likely does not apply to the small number of units you or I might buy.
[An slightly analogous aside: D. Self claims to have parralleled BJTs without RE resistors. He probably did. I tried it and could have use the result to roast marshmallows!! Possibly similar to the above situation]
Gratuitous plug: You can get matched sets (2 each) of 2SK1058 and 2SJ162 fom www.tech-diy.com. The price asked looks quite reasonable. |
|
|
| jackinnj |
| quote: | Originally posted by Charles Hansen
Typical Vgs (threshold) is between 0.4 V and 0.8 V. It is fairly random for both N-channel and P-channel. |
From my experienc the distribution of devices is pretty "normal". You do get some real outliers where the Vgs(t) is 1.4 or 1.5, but very few. It doesn't matter whether you're matching devices from Hitachi, International Rectifier, On-Semi or ST -- I want to be careful that the definition of "random" actually means statistically random, and exhibitiing the probabilities one would expect in a sample chosen at random. from a properly defined popoulation.
As the device warms up (as any semiconductor warms up) things begin to change. Moreover, the ambient temperature will have an effect.
But how does it sound? |
|
|
| Nelson Pass |
| quote: | Originally posted by sam9
[An slightly analogous aside: D. Self claims to have parralleled BJTs without RE resistors. He probably did. I tried it and could have use the result to roast marshmallows!! Possibly similar to the above situation] |
I did that with Thermistors in an CRC power supply once. It
worked on one amplifier, and never again. :cool: |
|
|
| hieroglyphics |
I've been searching the forum and read both articles about matching on the passdiy site, but I'm still somewhat unsure of a couple things. I set up the circuit on the "How to: Matching Devices" page using an R1 of 62 ohms. I'm testing IRFP044's: the datasheet says that its a Hexfet--whether that classifies as "lateral" or not, I'm not sure. I'm mainly wondering if I need to measure the Vgs immediately upon connecting the circuit or if I need to wait until the temperature and Vgs stabilizes?
Many thanks! |
|
|
| Nelson Pass |
First off an IRFP044 is a vertical Mosfet.
Second, you want to test the Mosfet under the conditions
of linear usage, so you set the current (i.e. the resistor in
series with the supply) to the appropriate value, considering
the 3-4 volt drop of the Vgs. For example a 4 volt typical
Vgs means that a 12 volt supply will want an 8 ohm resistor
for a 1 amp test.
None of this is critical, but it's nice to be in the ball park. We
are looking to match devices, and so the important thing is
consistency in testing more than anything else. The Mosfets
should start out at the same temperature, and you should
take readings after a consistent number of seconds (1 - 2- 3)
unless you are heat sinking the devices, in which case you are
playing a more sophisticated game. :cool: |
|
|
| hieroglyphics |
Ahhh, thank you kindly! I had gone through and tested all 32 of them and recorded the immediate value but I'll go back and use that 3 second delay. When I've got the time, perhaps I'll get sophisticated--but that doesn't really suit my personality :crazy:
Cheers! |
|
|
| Charles Hansen |
| I don't think you need to re-test them. I think what Nelson was saying was to be consistent. So first, make sure that all the devices start at the same (presumably room) temperature. Then, measure the devices at a consistent time interval. So it doesn't matter if it is 1/2 second, or 1 second, or 2 seconds, or 100 seconds. As long as all the devices were measured in the same way, then they can be matched to each other. |
|
|
| larryg |
| When testing any device for the purpose of matching whatever is desired, don't you have to establish first the tolerance +/- of the test it self?:att'n: |
|
|
| jackinnj |
| quote: | Originally posted by larryg
When testing any device for the purpose of matching whatever is desired, don't you have to establish first the tolerance +/- of the test it self?:att'n: |
Sounds like my quantitative organic chemistry prof.
+/- 10 or 20 millivolts on a lateral MOSFET is 0.4 to 0.8% at a I(d) of 1 amp., but 1.3% to 2.6% at 100ma.
The breakpoints for acuracy tolerances (let me see if I remember this correctly from grad school 30 years ago) are 30%, 3%, 0.3% etc. -- a couple decades ago I could have derived the formulas, from the logarithms, std deviations, mean-variance, etc...sorry, I have killed too many brain cells in the intervening years.
i should hook up a datalogger and graph the changes in I(d) as the device warms up, mbe something to do later this afternoon. |
|
|
| Charles Hansen |
The original poster is trying to build an amplifier for himself. He's not trying to write a scientific paper for a peer-reviewed journal.
So, no, he does not have to establish the accuracy of the test.
Anything he does will be better than randomly selecting parts. The more care he takes, the better his results will be. That is true of anything in life. That's why some things are better than others. |
|
|
| larryg |
| Yes I agree, more or less!:dodgy: |
|
|
| Nelson Pass |
| quote: | Originally posted by hieroglyphics
I had gone through and tested all 32 of them and recorded the immediate value but I'll go back and use that 3 second delay. |
Actually, it takes my voltmeter 3 seconds to make up its mind.
:cool: |
|
|
| tcpip |
I am getting one stubborn doubt about this issue. There have been two types of matching discussed here:- Matching between one N-channel device and its complementary P-channel device. I don't know whether this is necessary, but if you say it is, then it is.
- matching between multiple N-channel devices in cases when you have them in parallel. (Ditto, multiple P-channel devices.) This is where I have my doubt....
MOSFETs have negative tempco. If one MOSFET conducts more current than its parallel neighbour, it gets hotter, which means it automatically begins to conduct less. That's why you can use multiple parallel power-MOSFETs without drain resistors, but you can't take this liberty with power-BJT. Doesn't this mean that the original reason to match these devices is gone... the devices will themselves avoid the situation where one MOSFET is doing 150% of the work and the other only 50%?
I had this niggling doubt, then I went back and checked in the "Art of Electronics" this morning, and now my doubt is one big question which refuses to go away. So... is matching of multiple parallel power MOSFETs needed at all? |
|
|
| Charles Hansen |
I'm not sure who you're asking this question of, but here is my answer:
1) No, actually the opposite. According to me, there is no need to match between opposite polarities.
2) No, again. Matching is very important here. First of all, MOSFETs have a positive tempco at low currents and a negative tempco at high currents. What is "high" and what is low"? Depends on the device. Vertical MOSFETs (ie, just about anything besides a Hitachi or Magnatec) have a zero tempco point at many (ie, 5 - 20) amps. This means that in actual use, the vertical MOSFET will always have a positive tempco.
But even with a lateral MOSFET that has a negative tempco above 100 mA or so, your argument doesn't hold water. For the negative tempco to start to equalize things, there needs to be a temperature mismatch between devices. The only way there will be a temperature mismatch is because one device is doing more work.
To see the actual effect that would occur, just look at the data sheet. On the 2SK1058 operating at 500mA (a few watts into a loudspeaker), you can see that a mismatch of Vgs (threshold) of 0.15 V is equivalent to a 50 C temperature difference!
So don't be stubborn, just use a little common sense (and look at the data sheets). |
|
|
| tcpip |
| quote: | Originally posted by Charles Hansen
I'm not sure who you're asking this question of... |
No one in particular, and thanks for your answer.
| quote: | | 1) No, actually the opposite. According to me, there is no need to match between opposite polarities. |
Cool.
| quote: | | 2) No, again. Matching is very important here. First of all, MOSFETs have a positive tempco at low currents and a negative tempco at high currents. What is "high" and what is low"? Depends on the device. Vertical MOSFETs (ie, just about anything besides a Hitachi or Magnatec) have a zero tempco point at many (ie, 5 - 20) amps. This means that in actual use, the vertical MOSFET will always have a positive tempco. |
But is it true that parallel MOSFET devices can be used without any drain resistors, unlike power BJTs? I'm trying to take the concepts one step at a time... |
|
|
| Charles Hansen |
| quote: | Originally posted by tcpip
But is it true that parallel MOSFET devices can be used without any drain resistors, unlike power BJTs? |
Any device with a negative tempco can be paralleled without resistors (you meant "source", not "drain") IF you match them carefully enough.
The key parameter for any output device in this situation is its transconductance (X amps of output for Y volts of input) under the actual operating conditions. If the devices are matched, then you don't need source resistors.
You could even parallel bipolars in the same way IF you made a composite device that had a negative tempco. In other words, if each output pair had it's own temperature compensation, then you wouldn't need emitter resistors (although you would still want to match the devices). |
|
|
| Christer |
Charles has several times said that he does not think it is
important or even good to match N-channel vs. P-channel
devices. This is actually very good news in my view, since it
simplifies the process of buying devices, and settles a question
I have brought up on several occasion before in other threads,
without getting satisfactory answers. My concern was that if
buying quantities at once, one may probably end up with
devices from the same batches (one N batch and one P batch)
so the the N devices are likely to be reasonably matched and
the P devices to be reasonably matches, but on the other hand
it would considerably decrease the chance of finding complementary
matches. The alternative then would be to buy just a few devices
at a time, spread out over a lengthish time period, which would
increase the statistical spread, thus increasing the chance of
finding complementary matches, but decrease the chance of
finding matchin N devices and matchin P devices.
So far, it seems nobody in this thread has really argued against
Charles' point of view on this, so should we take that as an
indication that all the people with experience on this agree
with him? |
|
|
| jackinnj |
| within one tube of IRF or Hitachi devices you get one "mean" and one std dev, the next tube may have another "mean" -- |
|
|
| GRollins |
The one thing that I would add is that different device types may behave differently while being tested. I bought a large batch of IRF644s and had a devil of a time getting them matched. The first minute or two the digits were whizzing by so quickly that I couldn't even read them, much less count on them to be stable. I ended up waiting about ten minutes to get a decent reading.
On the other hand, the IRFP044s I bought to build the Aleph-X prototype settled down within seconds. Pure bliss compared to the '644s.
My operating assumption is that it's a thermal phenomenon. The '644 is a TO-220, whereas the '044 is a TO-247. Why it should make such a remarkable difference, I don't know.
At any rate, yes, match your N devices to each other and your P devices to each other. Let Self, Slone, et. al. go their own way.
Grey |
|
|
| tcpip |
| quote: | Originally posted by Charles Hansen
Any device with a negative tempco can be paralleled without resistors (you meant "source", not "drain") IF you match them carefully enough. |
Okay, so there seems to be a disagreement between you and "The Art of Electronics". They say power MOSFETs can be freely paralleled without any source resistors (sorry about the drain vs source goof-up). They don't mention any need for matching. In fact they make it clear that it's the negative tempco which allows this to be done, thus (I presume) making the matching unnecessary.
I have the book at home.... I can quote giving page number if anyone wants. I found this comment mentioned in two places, Chapters 3 and 6.
Now I'm in a jam. Who do I believe? :) |
|
|
| millwood |
| quote: | Originally posted by tcpip
They say power MOSFETs can be freely paralleled without any source resistors |
that is if they have negative temp co. Not all mosfets exhibit negative temp co, especially at all current range. some of the irf devices (irf540) for example exhibit positive temp co until Ids > 15amp. |
|
|
| Charles Hansen |
| You should believe the book. After all, they wouldn't print it if it weren't true. |
|
|
| GRollins |
H&H may not be looking at the kind of heavy biasing that audio folks take for granted, particularly in class A operation. They're more likely to use parallel MOSFETs in a power supply regulator or something of that nature.
Two points:
--One is that this is DIY. If you don't want to match devices, no one is going to make you. There's no such thing as a DIY police force that will come arrest you for using unmatched devices. Music will come out, and you will be happy that you built something that plays music.
--The other is that you might mount a set of unmatched devices with really, really small Source resistors. Something like .01 ohms. Run the amp. Measure the voltage drop across the resistors. Since they're so small in value, they won't have much effect on the operation of the circuit; they're just there to give you a measuring point. Calculate the current differences between the output devices. If you feel that the current differences aren't worth worrying about, then remove the .01 ohm resistors and carry on. If you find that one device is running, say, 40% higher current than the one next to it (note that this means it's running 40% more heat dissipation), then you might want to consider increasing the resistance of the Source resistors...or using matched devices.
The choice is yours...not mine, Charles's, or anyone else's. The ball's in your court.
Grey |
|
|
| slowhands |
| quote: | Originally posted by tcpip
This is very interesting. I have not read about this idea anywhere else. Let me know if you have any other details or ideas along these lines, just so that I can understand this a bit better. |
There is not much to understand. It's an expensive nuisance to match MOSFETs, but it's necessary in all the designs I've seen if you want reasonable current sharing.
For the case only of the CF configuration, there is a way not to have to do this. Rather than use a single driver for multiple parallel CF outputs, use instead multiple drivers, one per MOSFET device. It sounds very brute force, but consider that the drivers will do the matching, using the inherent 100% negative feedback in the CF configuration. And the drivers could be very cheap TO-92 devices, since each drives only a single MOSFET.
Assuming we have precisely matched source resistors, each driver will set Vgs to the "right" value to force equal current through the source resistor, so the device currents must match within the tolerances of the source resistors and driver Vbe's. I would use 1% source resistors; driver Vbe spread is not great.
Again, this is trading modest higher parts cost of the extra drivers for inventory and labor cost reduction due to not having to match output parts during manufacturing or field repair. For a DIY builder, it means you use all the output devices you have, not just the ones that match closely.
So, it's brute force and simple but has some significant benefits. |
|
|
| tcpip |
Dear all,
Thanks for the answers.
| quote: | Originally posted by millwood
that is if they have negative temp co. |
Yes, this seems to be something the book didn't cover. I guess I'll now have to graduate to datasheets. :)
| quote: | Originally posted by Charles Hansen
You should believe the book. After all, they wouldn't print it if it weren't true. |
Thank you.
| quote: | Originally posted by GRollins
H&H may not be looking at the kind of heavy biasing that audio folks take for granted, particularly in class A operation. They're more likely to use parallel MOSFETs in a power supply regulator or something of that nature. |
Yes, this is my suspicion too. The book seems to be discussing only on-off binary applications like motor switching, not linear applications, when they discuss power MOSFETs.
| quote: | | --One is that this is DIY. If you don't want to match devices, no one is going to make you. There's no such thing as a DIY police force that will come arrest you for using unmatched devices. |
I already knew this. :D My reason for posting was because I was getting confusing inputs from different sources. By treating opinions expressed here as being on par with opinions in a classic text, I thought I was showing you all how important I considered your opinions.. I am not certain some of you understood this. Anyway, let's focus on MOSFETs, not misunderstandings.
| quote: | | --The other is that you might mount a set of unmatched devices with really, really small Source resistors. Something like .01 ohms. Run the amp. Measure the voltage drop across the resistors. Since they're so small in value, they won't have much effect on the operation of the circuit; they're just there to give you a measuring point. Calculate the current differences between the output devices. If you feel that the current differences aren't worth worrying about, then remove the .01 ohm resistors and carry on. If you find that one device is running, say, 40% higher current than the one next to it (note that this means it's running 40% more heat dissipation), then you might want to consider increasing the resistance of the Source resistors...or using matched devices. |
Yes, this is interesting. I think I'll try this when I build my first MOSFET amp. Where can I get 2W 1% 0.01Ohm resistors?
| quote: | | The choice is yours...not mine, Charles's, or anyone else's. The ball's in your court. |
Yes, I know. I'd think this is the norm in DIY. However, when many DIYers get together, it's quite normal for one DIYer to ask questions to others to help clear his confusion. I was under the impression that that's what I was doing, but it appears that I've tested the patience of some members on this thread severely. I can't figure out why.
I find an opinion expressed in a very highly-regarded textbook to be something I can't dismiss lightly. That was the reason for my re-visiting the issue. If some of you feel that such sources can be dismissed, maybe you could have said so, and explained why. That would have helped me look at opinions from that book in a new way, and would have helped me learn faster. For instance, I really appreciate your comment about the context of the book's comments (switching) may be different from that of audio. That's the kind of pointers a learner needs.
The success and usefulness of something like diyaudio depends on- all of us carefully differentiating genuine questions/confusion from "argument for the same of argument"
- being patient with those of us who know less than we do
- giving pointers to help others learn, instead of making remarks which don't help. I've had to receive remarks like: "So don't be stubborn, just use a little common sense..." Etc. Most of us are already trying to use whatever common sense we have; let's just take that on trust.
Have fun. :D |
|
|
| Christer |
| quote: | Originally posted by jackinnj
within one tube of IRF or Hitachi devices you get one "mean" and one std dev, the next tube may have another "mean" -- |
Sure, but that still means it should be easier to find matches
between devices from the same tube than between the same
amount of randomly bought devices, I supppose. |
|
|
| sam9 |
"Rather than use a single driver for multiple parallel CF outputs, use instead multiple drivers, one per MOSFET device. It sounds very brute force, . . . "
This is somewhat analogous the the idea, "Power Slice", Linear Technology promotes for its LT1166. You could build a bunch of output module complete with their own heatsink then daisy chan as many as you need for a devised output. Of cource one goal is to sell as many LT1166's as possible. |
|
|
| GRollins |
tcpip,
My patience is intact, although I can't speak for others. It's just that it's difficult to resolve things like this easily. H&H is as close to an "audio textbook" as we're likely to have in the near future (I am specifically not including Self and Slone--others may feel differently--for what it's worth, I'd take Self over Slone any day of the week). Nelson Pass has produced a series of projects that have instructive value in addition to being working circuits that you can actually listen to. I, myself, get more ideas from reading patents, although the writing is sometimes rather opaque. Lacking a definitive, dedicated "audio" book, you have to learn on your own.
Note the number of posts you see from EEs who find that their courses did not cover some of the things that are considered fundamental in audio. We're off in a very small corner of the electronics field with very little overlap between our area of interest and any other part of electronics, with the arguable exception of radio.
If it was easy, it wouldn't be so much fun as a hobby. Cookbook recipies don't interest me much. I want to roam around a little.
As far as .01 resistors go, you'll have to see what's available. If all else fails, use .1 ohm. You should be able to find those. I used to have a store that stocked a lot of NTE stuff locally. They went out of business last summer, so I'm reduced to mail order now. If my perception of the world is correct, things will get worse before they get better. The world economy will be variable in the near future, poor after that. Go ahead and think mail order if you can't get something locally. You won't need 1% parts. When dealing with 5% or greater parts, I measure the resistance and write on the side of the piece with an indelible marker. Easy to do if your're using a white ceramic resistor. Don't worry about inductance for my test--the bias at idle (i.e. DC) will answer your question. Buy cheap resistors, measure them, write it on the side, and go.
Try not to let the people who sneer at you keep you from trying things. It's a pain the the (*ahem*) sometimes, but there are people who get their jollies from criticizing others. Makes 'em feel important. Just keep plugging and you'll get there.
Incidentally, I'm not saying that H&H is a bad book. It's just that we're in a niche market with our own odd requirements. By all means, keep reading the book, but remember that some things may have to be seen from a different point of view.
Might I suggest the word experience instead of the phrase common sense? As you gain experience, it sometimes begins to seem like common sense, but you have to keep in mind that none of us were born knowing the difference between MOSFETS, tubes, and bipolars. We had to learn. Some people take that learning process for granted. I had to fight too hard to learn what I know. Hopefully the path will be somewhat easier for others.
Grey |
|
|
| sam9 |
To expand on the post by GRollins, it becomes obvious (to me anyway) after reading this form and a couple of books that there is more than one way to build a good amplifier. However, you have to deduce that from information, often contradictory, provided by people who believe they have found the one true path.
Most important build SOMETHING! You won't really learn what works until you (try to) power it up. I don't think there is any source, book or forum poster, that is 100% right about all things audio, but excluding the obvious snakeoil salesmen nearly all have some gem of informatiom. Some have more gems than others. |
|
|
| tcpip |
| quote: | Originally posted by GRollins
My patience is intact, although I can't speak for others. It's just that it's difficult to resolve things like this easily. |
| quote: | Originally posted by sam9
Most important build SOMETHING! |
Thanks, both of you. Guys like you make the time I spend here worth the while. I hope I'll be able to pay it forward some day, in some way, if not pay it back. :) |
|
|
| GRollins |
I stand with one foot in the tube camp and the other in the solid state camp, so I can agree that there are more ways than one to build a good circuit. People who insist that there's only one way and it's their way are best ignored. They may be right some of the time, but it's not always worth the trouble to sort through the bluster in search of useful information.
I certainly don't claim to know everything about electronics. If I did I'd get bored and move on to something else where I could learn things. I'm only truly happy when I'm learning something.
Pay forward sounds good to me. It shouldn't be so confounded hard to learn this stuff, but we're reduced to picking up crumbs here and there and it takes literally years to get even the rudiments of the audio trade under control.
The irony is that reproduced music touches nearly everyone in the world sooner or later. Yes, there are aboriginal tribes in Australia and the Amazonian rain forest who may not have even a boom box, but expressed as a percentage, the vast majority of humanity uses some sort of electronic device to listen to music. Given that large a potential customer base and high demand (home, car, and personal portables, not to mention pro gear for concerts), audio designers should be more in demand. You'd think that universities would be churning out scores of them every time graduation rolled around. The reality is that there are very few people involved. A few mega corporations such as Sony and Pioneer control the lion's share of the market. They sell a large number of units of a limited number of models and their design teams are comparatively small. Then there's the sad fact that they design to a price point instead of designing for performance. The high end designers try to take up the slack, but their product is expensive. If you have sufficient discretionary income you can have a very fine system indeed.
The rest of us have to learn...and do...on our own.
Grey |
|
|
| tcpip |
| quote: | Originally posted by GRollins
The irony is that reproduced music touches nearly everyone in the world sooner or later. |
But then there are different ways to experience music. The way that a lot of my friends listen to music does not require all that realistic a reproduction. So I guess the demand for discerning audio system designers will always be low, don't you think?
| quote: | The high end designers try to take up the slack, but their product is expensive. If you have sufficient discretionary income you can have a very fine system indeed.
The rest of us have to learn...and do...on our own. |
So true. :) I came into diyaudio because I have more taste than money. But now that I'm in, I feel that I'd stick to diy audio even if I had lots of money. There are things which commercial designers simply don't attempt, purely for pragmatic reasons. For instance, how many active multi-amped speakers exist commercially? And this in spite of most people firmly believing that it's easier to build good line-level xo than high-current ones. |
|
|
| millwood |
| quote: | Originally posted by GRollins
Given that large a potential customer base and high demand (home, car, and personal portables, not to mention pro gear for concerts), audio designers should be more in demand. |
that's if everthing else is held constant. Unfortunate, economy of scale and fierce competition will drive out specialty designers, or anyone who failed to expand. That's why we had so many designers / audio companies back then than we do now.
Plus, technology has advanced to such a point that inexpensive boxes today can compete in performance with expensive boxes back then and now.
| quote: | Originally posted by GRollins
You'd think that universities would be churning out scores of them every time graduation rolled around. The reality is that there are very few people involved.
Grey |
Because I think we are approaching the limit of human hearing, with even the consumer grade boxes. I am one of those who don't believe in 0.0000007% THD, or egyptian maple boxes or golden feet, magic sound stages, 'positively gray', or 'shortest signal path', to name a few. I think most of us, if put to an objective test, will not be able to detect 99% of the things we claim to be able to detect.
The fastly dwindling audio designer circle is a testiment to that, in my view. |
|
|
| tcpip |
| quote: | Originally posted by millwood
Because I think we are approaching the limit of human hearing, with even the consumer grade boxes. I am one of those who don't believe in 0.0000007% THD, or egyptian maple boxes or golden feet, magic sound stages, 'positively gray', or 'shortest signal path', to name a few. I think most of us, if put to an objective test, will not be able to detect 99% of the things we claim to be able to detect. |
I wonder about this, though.
I too am the kind of person you describe. I bought Van den Hull cables and couldn't hear any difference with my home-grown cables. I can't hear a difference between an NE5532, a TL082, and an OPA2134. And in terms of approach, I find Randy Slone and Doug Self's approach the one that makes sense. (As far as I have seen, they believe in measurements, comparison testing, listening tests, etc, but are unwilling to entertain things which do not have an explanation in theory, e.g. the directionality of audio cables.)
So, to that extent, I am the kind of person you describe: I can't hear many of the differences that many others on this forum seem to.
However, the difference between a simple home-grown pair of speakers and the ones that come with combo systems (which 99.99% of users seem to buy for their homes today) is so dramatic that I (and all my non-diy non-audiophile friends) can hear it. Thus it seems that there are major improvements possible.to the sound of modern music systems. Why don't we see more of it happening in commercial audio systems?
I actually have a theory about why not. :D I believe that there's such a thing as "adequate quality" in free markets. This means that the market will not always opt for a better product, even if the price increase is affordable. The market often seems to have a concensus about a "good enough" level, and the bulk of the demand comes at that level, not higher. I've seen it in modems. Good modems could hold connections much longer with India's noisy telephone lines, but today no one buys anything other than the cheapest ones, even though noisy lines are common even now. And I'm seeing it in music systems. |
|
|
| sam9 |
Amplifiers are one thing, speakers are another. Even if a perfect speaker were definable, it will sound different in different rooms or different locations in the same room.
An additional thing to think about is that as one ages, hearing declines. Think about what that means as far as harmonics are concerned. If you calculate THD with a limit on bandwidth, yiou start with HD beyond 10k mmeaning little and each year the cut-off gets lower and lower. THD is probably more important because it is and indirect indicator of IM which may be less convenirnt to measure.
Other things I've found in commercial amps that could be avoided include buzzing transformers, off-on thumps, ground loops, ripple contamination the signal and inability to drive Magnepans (low sensitivity and loww impedance) with out suffereing premature death. So there are still some thing you can go for in diy amps that add to the quaity of listening even if THD hads been puhed below audibility.
One other thing about THD and THD+N. It is cummulative. Your amp in itself may well have performance beyond the audible, but is is not beyond the realm of possibility that it could add just enough to signal it takes as input to cross the threshold of audibilty. If I were designing an amp I would want it's noise and distortion far enough below the audible that it could not be guilty of being the "final straw". |
|
|
| jackinnj |
| quote: | Originally posted by sam9
... THD is probably more important because it is and indirect indicator of IM which may be less convenirnt to measure. |
Perhaps you didn't see the IM gadget in AudioXpress a few months back. There were a couple things I could do with the design to simplify it for broader appeal, but I don't have the time.
The thread is drifting away from L-MOSFETs |
|
|
| kasra |
Charles,
When using your method in post #6, from which points does one measure the VGS?
Is this method appropriate for 2SK1530& 2SJ201?
Thank you! |
|
|
| AndrewT |
Hi,
be careful you don't mix up GDS(V & 201) and GSD(L).
Change your source polarity to suit either N or P type. The diagram is actually the same.
The difference between lateral and vertical types does not affect the measurement technique, only the voltage reading you get as a result.
Your 201s (Y grade) will have a Vgs voltage closer to a vertical type. O grade have a lower Vgs nearer Lateral type.
When carrying out the testing try to keep the Drain current constant. Either use a current limiter based on a 3pin regulator or use the adjustable current limit in a regulated PSU.
I use a plugboard for the connections. The fets don't fit into the holes but on mine can just touch the top of the recepticles.
I tested at 100mA (+- some unknown error) and found that 1058s matched very well out of the tube. The parallel quads measure very close when installed with 0r22 source resistors.
However the Pchannel 162s had a wider spread of Vgs. One out of the sixteen tested was significantly different. When installed the source current is about 70% higher than it's three partners. 60mA to 65mA vs 105mA for the low Vgs Fet. Target for each was 75mA. The measured Vgs @100mA was within 20mV from highest to lowest for any group of four except on the errant group where one Pch was 40mV low:- 1260mV, 1300mV, 1300mV, 1310mV. My heatsinks get too hot at Iq=400mA so had to settle for Iq=300mA at +-69V=41W dissipation.
Hope the testing procedure and background help. |
|
|
| viktor1986 |
| Here is schematic for matching mosfet's.With this schematic I matched IRFP240&9240.Ugs for IRFP240 was 3.85V(100mA) and 4.65V(3A).Ugs for IRFP9240 was 3.73V(100mA) and 4.63V(3A) |
|
|
| AndrewT |
Hi Victor,
how did you measure drain current? |
|
|
| viktor1986 |
| quote: | Originally posted by AndrewT
Hi Victor,
how did you measure drain current? |
On picture you will see +48V.It's wrong,i used +17V.I put +A on +17V and -A on drain and I measure current... |
|
|
| AndrewT |
Hi Victor.
How did you measure the drain current? |
|
|
| mzzj |
| quote: | Originally posted by tcpip
Okay, so there seems to be a disagreement between you and "The Art of Electronics". They say power MOSFETs can be freely paralleled without any source resistors (sorry about the drain vs source goof-up). They don't mention any need for matching. In fact they make it clear that it's the negative tempco which allows this to be done, thus (I presume) making the matching unnecessary.
I have the book at home.... I can quote giving page number if anyone wants. I found this comment mentioned in two places, Chapters 3 and 6.
Now I'm in a jam. Who do I believe? :) |
As already stated earlier mosfets have negative tempco only above certain current level. Mostly when books claim that mosfets can be paralled just like that without matching they are talking about SMPS-stuff where mosfets are operated at high currents switching on or off but not in linear mode.
For linear use like audio amplifier normal v-mosfets never operate at sufficiently high current levels to have negative tempco. So matching and current balancing resistors are good to have. |
|
|
|
