|Everything Else Anything related to audio / video / electronics etc) BUT remember- we have many new forums where your thread may now fit! .... Parts, Equipment & Tools, Construction Tips, Software Tools......|
Please consider donating to help us continue to serve you.
Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
||Thread Tools||Search this Thread|
|10th May 2012, 10:30 PM||#1|
Measuring mosfets using the locky_z instrument
I love original data so I will post some here for everybody to scrutinize
The data represents 5 independent (done on different days) measurements of Vgs vs. Id for 10 different (2SK1530) devices under test (DUT(1-10)).
The raw data are here.
Looks pretty good to me
Note that the DUTs does not represent a random sample. They are two sets of devices (6+4) previously matched by zhoufang.
|11th May 2012, 10:25 PM||#4|
The purpose of these measurements are not to verify Zhou's precision, but obviously they may be used for this purpose
I bought the devises DUT 1-6 as "precision matched" and DUT 7-10 as N-P matched. Both types of matches offered by Zhou comes with the promise of 10 mA precision between the N (and P) devises. For the moment I will only look at the N channel devices.
Did Zhou deliver what he promised? Using the means of the 5 independent measurements I get the following ranges (which should be < 10 mA):
Set A (DUT 1-6): 15 mA
Set B (DUT 7-10): 48 mA
This is clearly not within specs........
Zhou measure "@ 500mA @ equilibrium" while I'm doing a pulse measurement (takes about 1.2 second) over a range of current using the locky_z curve tracer (LZCT). These are clearly not identical conditions so it is a bit early to claim that Zhou does not deliver good quality.
As I recorded traces with the LZCT I can readily have a look at the data @ 500 mA (same current as Zhou), but I cannot (or dont feel like) replicate the equilibrium condition without additional experiments.
Did Zhou deliver what he promised if measured in pulse mode at 500 mA?
Set A (DUT 1-6): 8.8 mA
Set B (DUT 7-10): 10.2 mA
I would say that in this case Zhou has indeed delivered what he promised, even if evaluated using very a different method
1) Sorting at operating conditions is (obviously) the best thing to do.
2) Sorting at any condition is (obviously) better than not sorting.
3) The accuracy of the locky+z devise is sufficient to rank devises already closely matched by other methods.
4) Zhou: get a curve tracer so that you can sort according to the clients desire!
Finally: can anybody recommend some kind of tight fitting terminal block where I can "plug" in the DUT without having to fix 3 screws? The time limiting factor is absolutely the mounting/unmounting of the DUT.
Graphic of @ 500 mA data attached and spread sheet data updated.
|26th May 2012, 10:35 PM||#5|
Join Date: Nov 2008
Thanks for sharing!
I have switched from @ "thermal equilibrium" to @ 27degC.
For 500mA, the Vgs drop by about 0.4mV per second after power on due to heat, very clearly visible on my Agilent 34401A meter.
It is easy enough to get the reading within 2 seconds and the reading have shifted less than 1mV, not a big deal.
With a bit of air-con it is easy to keep 27deg in the room in Singapore.
Between days I typically get 1-3mV off from the same DUT.
But the differences is uniform across days, that is if I get 2mV higher today, most of the device will be higher by 2mV +/-1mV at most.
The difference between day is due to ambient temperature, the longer I turn on my air-con, the closer it cools down towards 27degC (yes it's hot in Singapore), and the higher the number.
With this 27degC method, the repeatably improved substantiated compared to @ equilibrium. Previously I was struggling to get 10mV repeatability.
Another point to you: at 2A I believe you will get quite a few mV of voltage drop due to connection (socket) and wiring. When the socket is new the V drop is consistent so repeatability might not show, but soon the socket will get loose.
4Wire kelvin connection will really be needed at 2A, if you really care about the single mV digit.
I'm using some DIY kelvin clip for the mosfets, crude but it works.
Regarding tracer, I'm having some Agilent toy on my way!
The difficulty will not be the measurement, but the data management.
Sorted at 1 single current I just need to do a simple sort and sell from bottom up or top down.
Now imaging I have 1000 MOSFET, all tested at 100 data point.
Different customer require matching under different current, some wants curve matched, and all want different numbers in a set (quad, hex etc).
And I want to offer the best matching while not wasting any device.
Without some serious simplification, it's harder than most PhD topics.
PS: I promised 10mV, not 10mA. Measurement error was not taken into that 10mV.
Toshiba 2SK170-GR/BL/V, 2SK246/2SJ103, 2SK369-BL, Semisouth JFet R100, R550 & R085, Keratherm Red Isolators, AD1865N-K,
2SK1530/2SJ201, 2SK2013/2SJ313 ALPS RK27 20KA*2/50KA*4
Last edited by zhoufang; 26th May 2012 at 10:38 PM.
|27th May 2012, 07:55 PM||#6|
Thank you for the comments.
The values are of course in mV and not mA.
I agree that the temperature of the device is the most critical parameter to keep under control and presumably what will eventually limit the precision to which the measurements can be conducted reproducibly. If one want to have <1 mV reproducibility the temperature would need to be controlled to a fraction of a degree.
Although it goes a bit beyond my capacity, I'm sure that the data management issue with traces could be fairly easily solved by somebody trained in dealing with this kind of data sets.
I certainly hope you have more interesting and difficult projects to offer your PhD students!
I'm sure my friends and colleagues at A*STAR in Singapore could solve the problem in a day or two.
|28th May 2012, 10:43 AM||#7|
Join Date: Nov 2008
Ha, so you know the Singapore scene!
Sure it is easy to write a code the get you a best matching quad, then get you another best matching hex. then another quad.
The difficulty is to maximising the number of matching sets that I can get from the batch, the above method will leave you a sparse map of unmatchable device in the end.
For power device used as parralleled output pairs, I think there 1mV precision is not really needed or even desired. Your typical heatsink will not have a uniform temperature, usually hotter at the centre. And you isolator thermal conductance variation (due to non uniform force applied) will add another few degC of variation to Tj. So the Tj is not going to be the same amount parallel device.
It might well be better to have device in a set varies by a few mV, then use the higher Vgs ones for the hotter region, and use the lower Vgs one for the cooler region.
|Thread Tools||Search this Thread|
|Thread||Thread Starter||Forum||Replies||Last Post|
|Cable/Wire Measuring Instrument(s)||dnlyko||Analogue Source||1||5th November 2010 04:03 AM|
|my pre instrument||digi01||Chip Amps||7||30th January 2008 01:14 AM|
|41hz Amp4 - Telling MOSFETS apart, and measuring Inductance without a metre||Peteruk||Parts||2||27th July 2007 12:58 PM|
|n/p Mosfets vs depletion mod mosfets||marcello7x||Car Audio||6||4th October 2006 05:01 AM|
|New To Site?||Need Help?|