Hi Rob,
hfe is one of the prime factors
I've found 2N3055's with hfe's from 60 - 200 all having the same batch number....
I went trough this batch of 100 and ended with 7 having the same hfe-reading.
Matching to within 1% is not too bad. Note the accuracy of the measuring devices, probably these limit the accuracy you can get.
I would suggest you buy (at least) 5 - 10 times the number you need, and you will get some sets. If you need to meet some specs like hfe > Y you;ll need more.
hfe is one of the prime factors
I've found 2N3055's with hfe's from 60 - 200 all having the same batch number....
I went trough this batch of 100 and ended with 7 having the same hfe-reading.
Matching to within 1% is not too bad. Note the accuracy of the measuring devices, probably these limit the accuracy you can get.
I would suggest you buy (at least) 5 - 10 times the number you need, and you will get some sets. If you need to meet some specs like hfe > Y you;ll need more.
I have considered exactly the same problem for a while, though
in my case it be rather expensive MOSFETs and JFETs and more
about finding matched complementary pairs, which is likely even
harder. So, well it is a bit different since the pairs need not be
matched to each other (or maybe, I'd need that too in some
cases). One could just approach it purely statistically, but since
all transistors of each type will most likely be from the same
production batch the statistics gets quite different. I don't
know, but I have assumed that devices from the same batch
are likely to have a smaller variation than the same number
of devices picked randomly. On the other hand, that would
mean that I may end up a bunch of similar N devices and
a number of similar P devices, but no matched pairs at all.
What is peoples experience and wisdom about this? Maybe
it is better to pay the considerably higher price of buying
matched devices from people who provde this service??
in my case it be rather expensive MOSFETs and JFETs and more
about finding matched complementary pairs, which is likely even
harder. So, well it is a bit different since the pairs need not be
matched to each other (or maybe, I'd need that too in some
cases). One could just approach it purely statistically, but since
all transistors of each type will most likely be from the same
production batch the statistics gets quite different. I don't
know, but I have assumed that devices from the same batch
are likely to have a smaller variation than the same number
of devices picked randomly. On the other hand, that would
mean that I may end up a bunch of similar N devices and
a number of similar P devices, but no matched pairs at all.
What is peoples experience and wisdom about this? Maybe
it is better to pay the considerably higher price of buying
matched devices from people who provde this service??
I have looked at this -- if using Excel make sure you have the Pro version with the statistics package -- I guess I am revealing the fact that I was a lab rat at one time.
for MOSFET's (like IRFP240 + IRFP9240) -- within a "tube" you will find pretty close Vgs(t) -- if you buy them loose it's not the case. I would say that if you have a couple hundred of P and N channel devices you can match complementary pairs to some tens of millivolts. The transconductances match up pretty closely when the Vgs(t)'s match. (I have done the regression analysis, and the "eyeball" analysis via scatter plotting or histogram.)
for small signal BJT's used as current mirror's -- sometimes in a bunch I will find a bimodal distribution of beta's -- and sometimes a batch of a few hundred will have one mean and a standard deviation of only a few percent.
Jack
for MOSFET's (like IRFP240 + IRFP9240) -- within a "tube" you will find pretty close Vgs(t) -- if you buy them loose it's not the case. I would say that if you have a couple hundred of P and N channel devices you can match complementary pairs to some tens of millivolts. The transconductances match up pretty closely when the Vgs(t)'s match. (I have done the regression analysis, and the "eyeball" analysis via scatter plotting or histogram.)
for small signal BJT's used as current mirror's -- sometimes in a bunch I will find a bimodal distribution of beta's -- and sometimes a batch of a few hundred will have one mean and a standard deviation of only a few percent.
Jack
For the MOSFETS (lateral ones) I was planning on buying them
directly from Exicon, so I would most certainly get devices from
the same batch, so maybe it isn't a very good idea, as I
suspected.
BTW, Jack, I know you sell matched pairs of the Hitachis, so now
you don't have to tell that yourself and get flamed for it.
directly from Exicon, so I would most certainly get devices from
the same batch, so maybe it isn't a very good idea, as I
suspected.
BTW, Jack, I know you sell matched pairs of the Hitachis, so now
you don't have to tell that yourself and get flamed for it.
I bet you dont sell a matched pair of socks like this Jack!!!hahaha
and the amp below has 12 fets per channel and that was after going through ALOT...but I was buying them by the thousands at the time..LOL
Mods feel free to delete this...my Irish humour gets carried away
DIRT®
and the amp below has 12 fets per channel and that was after going through ALOT...but I was buying them by the thousands at the time..LOL
Mods feel free to delete this...my Irish humour gets carried away
DIRT®
Attachments
I've found 2N3055's with hfe's from 60 - 200 all having the same batch number....
I went trough this batch of 100 and ended with 7 having the same hfe-reading.
-------------------------------------------------
If you want good match, go for Japanese transistors in the same gain group. Much easier to match, even dynamically and in complementary pairs.
US makers seem to sell the matched ones to trade customers first, then the remainder for retail.
I went trough this batch of 100 and ended with 7 having the same hfe-reading.
-------------------------------------------------
If you want good match, go for Japanese transistors in the same gain group. Much easier to match, even dynamically and in complementary pairs.
US makers seem to sell the matched ones to trade customers first, then the remainder for retail.
JOE DIRT® said:
Yes, well I was originally kind of hoping to get a decent number
of matched pairs out of the lot, but I guess I would have to get
a "hit rate" of at least 35-40% of the lot for it to break even.
However, the higher the number of devices the higher the
percentage of matched pairs should be, statistically.
Actually, my plan is to build a number of amplifier modules of
at least two topologies using these MOSFETs, and probably
some experimental variations, so I might even be interested in
an unmatched pair or two just for experimentation curiosity,
but I won't have a need for a lot of them.
jackinnj said:
Hi Jackinnj,
perhaps you sell matched quads of MJ15003's.?
As there are serious counterfeit problems with this device so far I've not found ppl selling matched pairs anymore.
So I'm about to order a tray of those and do the matching myself.
That 'll leave 92 devices not directly used by me although I could probably sell some to other dutch builders.
Buying a tray of 100 from avnet will end up cheaper than buying 40 unmatched devices here in the Netherlands. (shipping included)
With 100 I think I get a decent change of having 2 matched quads (although not necessarely equal)
Can of worms
Matching is a real can of worms.
So you buy a pair of matched transistors. What exactly does that mean? Perhaps the vendor specified that they were matched for hfe. But hfe varies slightly with Ic, so was their Ic the same as your proposed Ic? For power MOSFETs, perhaps they matched Vgs for Ic = 0. Just as well, really, because there is huge variation here. But what about gm matching? I went through my stock of power MOSFETs to find complementary P and N channel pairs. Out of about 30 transistors, I found two "pairs" that produced similar sets of output characteristic curves.
Worse, when I tested my stock of (probably 10) MJE340, I found two devices with reduced hfe (although still within specification) but poor hfe linearity with Ic. The others looked just like the transistors we were taught about.
Matching is a real can of worms.
So you buy a pair of matched transistors. What exactly does that mean? Perhaps the vendor specified that they were matched for hfe. But hfe varies slightly with Ic, so was their Ic the same as your proposed Ic? For power MOSFETs, perhaps they matched Vgs for Ic = 0. Just as well, really, because there is huge variation here. But what about gm matching? I went through my stock of power MOSFETs to find complementary P and N channel pairs. Out of about 30 transistors, I found two "pairs" that produced similar sets of output characteristic curves.
Worse, when I tested my stock of (probably 10) MJE340, I found two devices with reduced hfe (although still within specification) but poor hfe linearity with Ic. The others looked just like the transistors we were taught about.
The best way of matching is to use a curve tracer. Sadly, they are unbelievably expensive because there's a very limited market for them so the development costs have to be recouped over a few units.
What you could try doing is to set up a test rig that applies the collector voltage you will be using (in series with an ammeter), and apply a known current to the base via a constant current sink. Then, as you swap transistors in and out, you can determine hfe. Even better, you could adjust the CCS to produce the collector current you will use in circuit, and measure base current, then calculate hfe. This is all easier than it sounds.
The previous method will only give you hfe at one operating point, but it's better than nothing.
What you could try doing is to set up a test rig that applies the collector voltage you will be using (in series with an ammeter), and apply a known current to the base via a constant current sink. Then, as you swap transistors in and out, you can determine hfe. Even better, you could adjust the CCS to produce the collector current you will use in circuit, and measure base current, then calculate hfe. This is all easier than it sounds.
The previous method will only give you hfe at one operating point, but it's better than nothing.
Re: Can of worms
When buying matched pairs You should specify the aspect to match. Vgs at 1 A? hfe at 10mA?
there is the trade off between doing it your self, in which you can define all aspects to match, of buy commercial-of-the-shelf.
The problem with DIY is the number of devices which has to pass through your hands before you end up with a good enough match.
That's the reason I just bought a tray of 100 BJT in the US (in stead of 30 for the same bucks in the NL) to end up with 8 devices which will be used in 2 monoblocks.
They'll be selected under working conditions in two stages; first:
-hfe is determined at an Ic of 27% of the actual operating Ic.
This gives a distribution of hfe vs device count
When visualised this distribution probably will have "peaks"
"peaks" are tested for hfe at operating Ic from the lowest hfe onwards.
And bloody well I hope to end up with two sets of 4 with a hfe at operating condition within 1%.
btw I'll measure the hfe indirectly as:
- Ic will be measured as Volts over a known fixed resistor and
- Ib also as Volts over an fixed resistor.
and no resistor between emittor and +Vss
Knowing the range your devices should be in, you still need several values of resistors for Rc and Rb to get the most accurate hfe.
One advantage in my position is the large Ic at operating conditions.
Errors and in-accuracies from the measurement-instruments will be in the last digit-readout. Which is fine, as there are three digits left.
And the knowledge the matching
improves
Hi EC8010EC8010 said:
When buying matched pairs You should specify the aspect to match. Vgs at 1 A? hfe at 10mA?
there is the trade off between doing it your self, in which you can define all aspects to match, of buy commercial-of-the-shelf.
The problem with DIY is the number of devices which has to pass through your hands before you end up with a good enough match.
That's the reason I just bought a tray of 100 BJT in the US (in stead of 30 for the same bucks in the NL) to end up with 8 devices which will be used in 2 monoblocks.
They'll be selected under working conditions in two stages; first:
-hfe is determined at an Ic of 27% of the actual operating Ic.
This gives a distribution of hfe vs device count
When visualised this distribution probably will have "peaks"
"peaks" are tested for hfe at operating Ic from the lowest hfe onwards.
And bloody well I hope to end up with two sets of 4 with a hfe at operating condition within 1%.
btw I'll measure the hfe indirectly as:
- Ic will be measured as Volts over a known fixed resistor and
- Ib also as Volts over an fixed resistor.
and no resistor between emittor and +Vss
Knowing the range your devices should be in, you still need several values of resistors for Rc and Rb to get the most accurate hfe.
One advantage in my position is the large Ic at operating conditions.
Errors and in-accuracies from the measurement-instruments will be in the last digit-readout. Which is fine, as there are three digits left.
And the knowledge the matching
improves
EC8010 said:
There have been dozens of articles on "how to" build a curve tracer -- you just need a stack of old "Audio Amateur" or "Wireless World" magazines. The simplest of these use an oscilloscope and a couple blanking tricks to sweep in one quadrant --
Actually, I have been too lazy to finish up the article...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.