Measuring Rbulk of transistors

From circuitbending to homebrew stompboxes & synths, keep the DIY spirit alive!

Moderators: lisa, luketeaford, Kent, Joe.

ixtern
Common Wiggler
Posts: 184
Joined: Mon Jan 22, 2018 4:34 am

Re: Measuring Rbulk of transistors

Post by ixtern » Wed Jan 22, 2020 5:46 pm

To better understand this Giacoletto (or Kulke-Miller) method, I've drawn BJT model to see what's going on.
BJT_model.png
You do not have the required permissions to view the files attached to this post.

User avatar
devinw1
Super Deluxe Wiggler
Posts: 1159
Joined: Tue Aug 07, 2018 11:20 am
Location: Portland, OR
Contact:

Re: Measuring Rbulk of transistors

Post by devinw1 » Fri Jan 24, 2020 12:06 am

I measured some more PNP matched pairs using my original Renee Schmitz setup for use on my TZ0. The results are below. Unfortunately Mouser had the best one of the lot sorted as TSOP-6 but it's really TSSOP-6 (smaller package), but I was able to solder it onto a TSOP-6 pad for testing. After I work with this stuff a little, it becomes less scary. Anyway, the BCM53DSF looked like it was gonna be a winner, and it did do quite well in linearity, but it had horrible settling time/thermal issues like what I saw with non monolithic pairs. The next best one was the PMBT4403YSX which did very well. I'll post in my other thread about that though.

To explain the chart: HFE is the average HFE of Q1 and Q2. Delta HFE is the difference between HFE of Q1 and Q2. Same with delta Vbe. I believe all these pairs specify a Vbe match of 2mV or less.
matchedpairs1.png
You do not have the required permissions to view the files attached to this post.

User avatar
cygmu
Super Deluxe Wiggler
Posts: 1423
Joined: Mon Feb 15, 2016 3:33 pm
Location: UK

Re: Measuring Rbulk of transistors

Post by cygmu » Fri Jan 24, 2020 7:30 am

That PMBT4403YSX seems to do really well, and the results in your TZO thread are great!

I checked the datasheet for that part [ https://assets.nexperia.com/documents/d ... 4403YS.pdf ] and couldn't see anything about matching or offset or anything like that, but evidently the one you measured has pretty good matching between the two transistors.

Is that likely to be luck, or a consequence of the manufacturing process that we can reasonably rely on, at least for DIY builds?

User avatar
devinw1
Super Deluxe Wiggler
Posts: 1159
Joined: Tue Aug 07, 2018 11:20 am
Location: Portland, OR
Contact:

Re: Measuring Rbulk of transistors

Post by devinw1 » Fri Jan 24, 2020 11:11 am

cygmu wrote:
Fri Jan 24, 2020 7:30 am
That PMBT4403YSX seems to do really well, and the results in your TZO thread are great!

I checked the datasheet for that part [ https://assets.nexperia.com/documents/d ... 4403YS.pdf ] and couldn't see anything about matching or offset or anything like that, but evidently the one you measured has pretty good matching between the two transistors.

Is that likely to be luck, or a consequence of the manufacturing process that we can reasonably rely on, at least for DIY builds?
You're right... I was looking at that datasheet too and realized that was one of the part numbers I rolled the dice on because it's billed as a switching transistor pair and not explicitly a matched Vbe pair. I think any of these modern pairs made on the same die probably have pretty tight process control and we benefit from that, but yes I will have to see if I just got a good one or if they really do have good Vbe matching across the board.

I've got 2 more of them I will measure for now and see. Sample of 3 is better than 1. :D

User avatar
guest
Super Deluxe Wiggler
Posts: 3815
Joined: Mon Aug 19, 2013 11:49 am

Re: Measuring Rbulk of transistors

Post by guest » Fri Jan 24, 2020 11:24 am

thats an interesting point you bring up: are they on the same die? most of the matched pairs are not. they are 2 seperate dice that are sorted by machines and then placed in the same package as something close. an unmatched pair might not bother with this, and therefore keep the transistors on the same die, giving better thermal characteristics. ive had good luck cracking the top off with a pair of diagonal cutters and just looking to see the configuration.
openmusiclabs.com

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1291
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Re: Measuring Rbulk of transistors

Post by neil.johnson » Fri Jan 24, 2020 11:59 am

When I discussed this issue with Philips/NXP years ago (I think it was about the BCM847) the response back was that while the two transistors were separate die (not same die) they were taken from the same area of the wafer after slicing so they are likely to be pretty closely matched electrically.

Being separate die the thermal tracking won't be perfect, but better than two discrete TO-92 epoxied together.

Neil
Modular stuff :: http://www.njohnson.co.uk

User avatar
devinw1
Super Deluxe Wiggler
Posts: 1159
Joined: Tue Aug 07, 2018 11:20 am
Location: Portland, OR
Contact:

Re: Measuring Rbulk of transistors

Post by devinw1 » Fri Jan 24, 2020 12:21 pm

Oh, well I stand corrected! Very interesting. I had thought that most matched pairs were made on the same die.

I definitely like the idea of destructively testing some units to see how they are actually made. :D

Synthiq
Wiggling with Experience
Posts: 471
Joined: Mon Feb 06, 2017 3:10 pm
Location: California

Re: Measuring Rbulk of transistors

Post by Synthiq » Fri Jan 24, 2020 4:15 pm

Mostl monolitic transistor arrays I have seen have a substrate connection that must be properly biased for correct function. One exception is the THAT 300 series with electrically isolated transistors but even they have a substrate connection that can be grounded for minimum capacitive coupling between the separate transistors. So it is a pretty safe bet that two transistors in a 6 pin package are two separate dies.

User avatar
guest
Super Deluxe Wiggler
Posts: 3815
Joined: Mon Aug 19, 2013 11:49 am

Re: Measuring Rbulk of transistors

Post by guest » Fri Jan 24, 2020 4:23 pm

is there a reason this is only on quads and quints? for example, the LM194, MAT02, SSM2212, etc dont have substrate connections and they are on a single die:

https://zeptobars.com/en/read/MAT02-low ... og-Devices
openmusiclabs.com

ixtern
Common Wiggler
Posts: 184
Joined: Mon Jan 22, 2018 4:34 am

Re: Measuring Rbulk of transistors

Post by ixtern » Fri Jan 24, 2020 5:17 pm

guest wrote:
Fri Jan 24, 2020 4:23 pm
is there a reason this is only on quads and quints? for example, the LM194, MAT02, SSM2212, etc dont have substrate connections and they are on a single die:
https://zeptobars.com/en/read/MAT02-low ... og-Devices
I think the answer may be this one (in second sentence):
valgamaa wrote:
Wed Oct 02, 2013 12:45 pm
The transistors in the THAT chips are fully isolated from each other, which is why they can put NPN and PNP devices together on the same die. This means that the substrate isn't doped, so there is no parasitic PN junction to reverse-bias, unlike the LM3046.
The short answer is that the substrate pins can be ignored.
viewtopic.php?t=94286

User avatar
guest
Super Deluxe Wiggler
Posts: 3815
Joined: Mon Aug 19, 2013 11:49 am

Re: Measuring Rbulk of transistors

Post by guest » Sat Jan 25, 2020 3:24 pm

ok, heres what im using to test with right now:

Image

the stability is a bit related to the transistors in use and how they are layed out on the protoboard, so the opamps/capacitors might need some tweaking. as it stands, its good to ~1kHz sine waves before the compensation starts to effect the accuracy (<1%). here are my latest numbers:

MAT14 - 0.48
2N3904 - 0.9 -> 1.3
THAT300 - 1.1
2SC1583 - 1.3
PMP4201 - 2.25
LM3046 - 2.9
MMPQ2222 - 1.0

the 2N3904 varied a lot depending upon how i placed them in the protoboard, and which pairs i selected (the latter being more critical). im suspecting that the matching of Vbe matters a fair bit, and the fact that they arent on the same die giving different temperatures is critical. the MAT14 numbers match really well with the datasheet, and the rest of the numbers match my data from using them in exponential converters, so i am very inclined to trust this setup as the best way to see how a pair will work as an expo converter (because it basically is an expo converter). the MMPQ2222 is a quad i had laying around that has seperate dice inside. it was absolutely horrible.

[EDIT] fixed the MMPQ2222 results - i had the collector and emitter swapped at first. it seems decent now.
Last edited by guest on Mon Jan 27, 2020 3:35 pm, edited 1 time in total.
openmusiclabs.com

User avatar
guest
Super Deluxe Wiggler
Posts: 3815
Joined: Mon Aug 19, 2013 11:49 am

Re: Measuring Rbulk of transistors

Post by guest » Mon Jan 27, 2020 3:34 pm

i messed up the MMPQ2222 test, and got a more reasonable result of 1ohm. ive edited the data above to reflect this.
openmusiclabs.com

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1291
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Re: Measuring Rbulk of transistors

Post by neil.johnson » Mon Jan 27, 2020 4:54 pm

The ZTX851 and ZTX951 have very low emitter resistance, so might we worth a try?

Neil
Modular stuff :: http://www.njohnson.co.uk

ixtern
Common Wiggler
Posts: 184
Joined: Mon Jan 22, 2018 4:34 am

Re: Measuring Rbulk of transistors

Post by ixtern » Tue Jan 28, 2020 3:00 am

neil.johnson wrote:
Mon Jan 27, 2020 4:54 pm
The ZTX851 and ZTX951 have very low emitter resistance, so might we worth a try?

Neil
Indeed, they are very interesting transistors, known for lowest base spreading resistance (what means low input noise).
Don't know about their emitter resistances, may be small also.
In the low-noise BJTs comparison (table 8.1a) in the "Art of Electronics" (P.Horowitz, W.Hill, 3rd ed.) they are winners.

But they are ..
- single transistors only,
- high current devices (up to 5A) with rather high lekeage currents (Icb0, Ice0).

So I doubt if they are suitable for temp-compensated expo converters.

User avatar
devinw1
Super Deluxe Wiggler
Posts: 1159
Joined: Tue Aug 07, 2018 11:20 am
Location: Portland, OR
Contact:

Re: Measuring Rbulk of transistors

Post by devinw1 » Tue Jan 28, 2020 11:32 am

Yeah that 5Amps pretty much means that they will have pretty crap (likely) linearity down in the region we will be using them (<1mA), right?

User avatar
devinw1
Super Deluxe Wiggler
Posts: 1159
Joined: Tue Aug 07, 2018 11:20 am
Location: Portland, OR
Contact:

Re: Measuring Rbulk of transistors

Post by devinw1 » Tue Jan 28, 2020 2:25 pm

Somewhat related side topic: Has anyone ever tried playing with the TL441 log amp for anything? It's like $6 bucks but seems like it could be useful for expo converters somehow as it basically functions based on the exponential behavior of BJTs and has matched pairs inside? I could be way off, but was just curious as I came across this chip a while back and was intrigued.

User avatar
guest
Super Deluxe Wiggler
Posts: 3815
Joined: Mon Aug 19, 2013 11:49 am

Re: Measuring Rbulk of transistors

Post by guest » Tue Jan 28, 2020 3:02 pm

i have never used the TL441, but i have puzzled over it before. i think i decided that for 6$ i would get way better performance from a MAT14. the output currents are though cascode buffers, which will have the added base current, giving 1% or worse error. the tail current of the diffpairs are not directly controllable, making it hard to setup an accurate exponential converter. there might be some fun other things to do with it, or it could be wrapped in a bunch of other circuitry to make it work as a decent expo converter. the thing i liked about it, was that it had 2 seperate circuits, so the other half could be used for temperature compensation.

i have used other log amps as expo converters, and wrote up those results in my paper. i have learned a lot since then, and i think i could get the ADI parts to perform a lot better these days.
openmusiclabs.com

User avatar
devinw1
Super Deluxe Wiggler
Posts: 1159
Joined: Tue Aug 07, 2018 11:20 am
Location: Portland, OR
Contact:

Re: Measuring Rbulk of transistors

Post by devinw1 » Tue Jan 28, 2020 4:45 pm

Yes... I was puzzled by it too, which typically makes me intrigued :D. What does this thing DO exactly, I thought.

I did just find this today https://www.analog.com/en/analog-dialog ... ained.html

which was helpful. It now makes sense to me that many of these are more or less designed to output the log of the *envelope* of a signal.

Anyway, like you said perhaps some interesting things could be done with it, but probably not worth adding even more to the pile to make a good expo converter when they are $6 a pop. And the DIP version only has like 40 in stock at Mouser which to me smells suspiciously like something that might be obsolted soon.


BTW, just started reading your paper on Expos on openmusiclabs. It's very good and in depth! :hail:

User avatar
guest
Super Deluxe Wiggler
Posts: 3815
Joined: Mon Aug 19, 2013 11:49 am

Re: Measuring Rbulk of transistors

Post by guest » Tue Jan 28, 2020 5:49 pm

the TL441 is one of their older generation. they have a lot of newer versions. there are two main kinds of log amps. this one is a "piecemeal" variety. it pieces together a series of pseudo-log responses that are accurate over a small range. by putting 8 of these in a line, you get enough range to have a useful device. unfortunately, they are not accurate enough for music. there is another variety that is made like a standard expo converter. most of these have built in temperature compensation, which would make them ideal expo converters. but, they are kind of expensive, and they are tough to stabilize. they cost more than putting an AS3340 down, so probably not worth it. the TL441 could be used to make a nice sinewave shaper ala the barrie gilbert method ive used. but its over twice the cost of doing it discrete.

so what finally got you to read the expo paper? have you read the other two papers i wrote?
openmusiclabs.com

User avatar
devinw1
Super Deluxe Wiggler
Posts: 1159
Joined: Tue Aug 07, 2018 11:20 am
Location: Portland, OR
Contact:

Re: Measuring Rbulk of transistors

Post by devinw1 » Tue Jan 28, 2020 9:46 pm

Gotcha. I think I accidentally came upon the TL441 while searching for something different and didn't realize there were a lot of other models that do the same thing that are much newer.

As for the paper... not sure, just had some free time finally and happened to be looking at expos a lot lately! Thanks for all that work. I have skimmed over your OTA paper too but need to give that an in depth read for sure.

A question on the expo side; I understand the the different ideal tempcos for using a divider vs putting the tempco in the op amp feedback and it's perfectly clear but what I don't get is that for a divider scenerio, I can adjust the ratio of the divider to get the "watered down" tempco to be ideal (~3356), but then now the divider doesn't necessarily provide the right scaling that I need. For a standard (56k/1k) ratio, the ideal original tempco is roughly 3500, which AFAIK only the KRLs are. These have the downside of being rather gigantic and also often out of stock at various outlets. Every other type of PTC or Platinum RTD that I've come across is a higher tempco (3850 or more). So how would one even use these? Scale the divider to get the right tempco and then adjust the gain of the preceeding op amp? Or would this create additional error?

User avatar
guest
Super Deluxe Wiggler
Posts: 3815
Joined: Mon Aug 19, 2013 11:49 am

Re: Measuring Rbulk of transistors

Post by guest » Wed Jan 29, 2020 5:35 am

all good questions. one thing i realized after writing that paper, is that the temperature the tempco is specified at is different for the PTS series and the rest of the thermistors. the PTS is specified at 0C, whereas the thermistors are at 25C, and the TCR you want is equal to 1/T_specified. so for 25C thats 1/298K = 3356ppm, but for 0C its 1/273K = 3663ppm/C. so the actual difference from ideal is not nearly as bad with the 3850 PTS series.

when it comes to "watering down" the tempco, there are a number of ways of doing this. putting a resistor in series, a resistor in parallel, or all sorts of combinations of both. the method which effects the linearity the least is to put a resistor in series, which means you would ideally minimize the effect of any other resistances. in the case of the voltage divider, that top resistor (from the vantage point of the base of the transistor) is actually in parallel with the thermistor. so we need to make that resistor as large as possible so it doesnt effect the tempco as much, and then put a small series resistor between the thermistor and ground. figures 42/43 in the paper are results from using this method with a 10k divider resistor, 100ohm PTS and 4.5ohm series resistor.

but, as you point out, you do not get the right V/octave with larger divider resistors. to overcome this, i added more gain to the CV summing stage to increase the voltage seen by the divider. if this is not an option for a circuit, then the linearity will suffer a small amount, but its probably not going to be a big issue, as other factors will dominate (like how well the thermistor tracks the die temperature).
openmusiclabs.com

User avatar
devinw1
Super Deluxe Wiggler
Posts: 1159
Joined: Tue Aug 07, 2018 11:20 am
Location: Portland, OR
Contact:

Re: Measuring Rbulk of transistors

Post by devinw1 » Wed Jan 29, 2020 11:30 am

Aha! i hadn't gotten to the addendum yet! :doh: Thanks for the answers, I was thinking already along the lines of something like the series resistor already, so this reaffirms that and makes sense.

I like the look of these RTDs with the 0805 package, and Mouser has them too! :tu:

I'm doing the calculations based on a 100 ppm/C tempco of the series resistor (Yageo Metal Film 1% MFR series) and coming up with 27k/500 ohm and a 63.5 ohm series R. I'm wondering why it's different from your calculation in the paper (you got 17 Ohms). I first found the weighted average tempco of the 500 ohm + Series combo and put that into TCRE=(R2/(R1+R2))*TCR formula.

User avatar
guest
Super Deluxe Wiggler
Posts: 3815
Joined: Mon Aug 19, 2013 11:49 am

Re: Measuring Rbulk of transistors

Post by guest » Wed Jan 29, 2020 12:46 pm

i got my numbers from doing simulations with the equations given in the datasheet for the resistance versus temperature. the general equation from earlier in the paper can be used to get a decent approximation. but remember that the tempco for the PTS series is at 0C, so you are aiming for 3663ppm, not 3356ppm. also, the 100ppm of the thin film resistors is +/-100ppm, so its best to assume 0ppm on those. you can get 50ppm for about the same price.
openmusiclabs.com

User avatar
devinw1
Super Deluxe Wiggler
Posts: 1159
Joined: Tue Aug 07, 2018 11:20 am
Location: Portland, OR
Contact:

Re: Measuring Rbulk of transistors

Post by devinw1 » Wed Jan 29, 2020 1:46 pm

Hmm, but if I use the datasheet equation: RT = R0 x (1 + A x T + B x T2) (listed as valid from 0 to 150C) and use 500 as Ro (given at 0C), then the tempco at room temerature (22C) I get is:

3896 ppm/C

If I use 24C like you did, I get 3896. Where are you getting the 3663?

User avatar
guest
Super Deluxe Wiggler
Posts: 3815
Joined: Mon Aug 19, 2013 11:49 am

Re: Measuring Rbulk of transistors

Post by guest » Wed Jan 29, 2020 2:26 pm

the tempco is the change in resistance with temperature, divided by the resistance. so if we want to calculate this from the equation, we take the first derivative of the resistance with temperature:

d/dT (R0(1+AT+BT^2) ) = R0(A+2BT)

we then divide by R25 to get the tempco at that point

R25 = R0(1+A*25+B*25^2) = 548.67

TCR25 = (A+2B*25)/(1+A*25+B*25^2) = 3535ppm/C

and, since (A+2B*25) ~= A, TCR25 ~= TCR0*R0/R25

im betting that all those 3500ppm/C thermistors out there are platinum like this one.
openmusiclabs.com

Post Reply

Return to “Music Tech DIY”