Better power MOSFET models in LTSpice

The tracer came today. Pretty clever device, and not badly made considering its a HK. I'm getting to understand how it works. Offset was way off, replacing an ancient 741 opamp with a TL071 fixed that, so it appears to be working as designed again now.

Can you define. "constant power curve"? It does have switchable current limit resistors and a circuit to interrupt the voltage sweep when the limit is reached. Which seems to work ok for bjts. But--

I didn't quite think through the ramifications of the way it handles FETs. As built, it only does depletion mode fets (I also did a quick mod with a dpdt switch to invert the gate bias circuit for enhancement mode MOSFETS, btw). But all trace runs start from zero bias. I can limit the number of steps, but that doesnt help with a depletion device when step 1 is the max current one! Probably fine for little jfets or LMD150, but my first DN2450 burned up in my attempts to give it enough Vcc while trying to comprehend what was going on. I'd like to come up with a clever way to make it start from negative and go toward zero with a prescribed maximum so I can avoid the high dissipation region until needed.

Another problem is that its voltage drive is though a voltage divider (within a feedback loop), so large gate capacitance makes it go ugly. Next up is to toss a little buffer in there for that, with a switch to take it back out when doing current drive for bjts. That shouldn't be too hard.

So this is turning into a redesign, might not be worth it. I'll see what I can do picking off the low hanging fruit first, with luck maybe enough to get the DN2540 curves that started this..
 
Member
Joined 2011
Paid Member
Datasheet curves are taken with 300uS pulses or so, so that the device doesn't have time to warm up significantly.

It's interesting to calculate the time needed to sweep out the full I-vs-V curve of a semiconductor device, using this pulse method.

If the device is operated at 1% duty cycle to keep dissipation small, then the on-time is 300usec and the off-time is 29,700usec. One measured IV point every 30,000usec. 30 milliseconds per datapoint. If you're sweeping ten base-current curves, at 200 datapoints per curve (measuring Icoll at 200 different collector voltages), that's 2000 measurements. At 30 milliseconds apiece, that's 60 seconds for the full IV sweep.

After about the fourth or fifth transistor, it starts to get real old, real quick, waiting 60 seconds for the damn I/V sweep to finish. Especially if you're used to the old analog-style curve tracers that gave the curve instantly.
 
That does throw another curve into the project. I can pretty easily pulse the voltage bias (a 555 timer and a CD4053 analog switch, just checked i do have some here). They could be worked into the buffer/bias offset circuit. I think I'll go for about 5% duty cycle and make it variable so I can see when heating starts to change things. Not sure if the output current sense circuit will be fast enough though... TBD
 
Last edited:
Member
Joined 2011
Paid Member
2% duty cycle and 0.3ms pulse will for the same number of data points ("If you're sweeping ten base-current curves, at 200 datapoints per curve, measuring Icoll at 200 different collector voltages"), plot a single curve in ~3seconds.

Yes, the first curve at the first value of base current takes 3 seconds; and all 10 curves for all 10 values of base current takes 30 seconds. Since what you actually want is the complete plot with all 10 base current curves, you must wait the full 30 seconds.

However, I personally don't like the idea of using a duty cycle as large as 2%, and here's why. To properly characterize (this JFET in a TO92 package) we need to take data at Vds=10V, Ids=500mA ... which is 5 watts of dissipation during the measurement. A LOT more than the 0.625 watt rating of the plastic package. Measuring at 2% duty cycle drops the long term average power to 100 milliwatts which is still too high for my tastes; I am concerned that 100mW will heat up the device and distort the data. Fairchild gives a small hint that they agree; they say "duty cycle less than 2%" in that datasheet, footnote 5 on page 2.
 
Last edited:
A number of manufacturers state 2% duty factor, that's why I used that as an example.

2% duty factor for a 625mW device is an average of 12.5mW
Look at a few curves, they stop at lower voltage when the current is higher.
This is presumably done to minimise the thermals.
 
Member
Joined 2011
Paid Member
I wonder how they measured the data shown in Fig.2 and Fig.3 of the J105 datasheet; it is a TO-92 device with 625mW max dissipation. Grab a couple I-V points from the pentode (near-horizontal IDS) region of Fig.2, and multiply IDS times VDS. Cowabunga!

Since IDSS must be measured in the pentode region of the I-V curve (where VDS > |VGSoff|), the J105 requires VDS >= 4.5V to measure its IDSS, according to the data in Fig.3. So, when you're measuring IDSS, the part burns 4.5V*0.5A = 2.25 watts. Biggger than 625mW. I have placed a red dot at the J105 point on Fig.3.

Fig.2 looks to me like the J106 (the middle device in the J105 J106 J107 family that this datasheet represents). The Fig.2 device's IDSS appears to be about 400mA, which is within spec for a J106 but not quite within spec for a J105. Here's the datasheet itself.

_
 

Attachments

  • j105_fig2.png
    j105_fig2.png
    59.2 KB · Views: 526
  • j105_fig3.png
    j105_fig3.png
    270.9 KB · Views: 519
Last edited:
I wonder how they measured the data shown in Fig.2 and Fig.3 of the J105 datasheet; it is a TO-92 device with 625mW max dissipation. Grab a couple I-V points from the pentode (near-horizontal IDS) region of Fig.2, and multiply IDS times VDS. Cowabunga!

Sometimes TO126 or even TO220 dies are put in TO92 packages. The die is the same and they just use the same data for each datasheet.

Even if not, constant power PWM testing automatically allows you to make pulsed measurements at high power because the average power remains low. The thermal mass of the die and package absorbs the transient power.
 
Member
Joined 2011
Paid Member
... constant power PWM testing automatically allows you to make pulsed measurements at high power because the average power remains low. The thermal mass of the die and package absorbs the transient power.
It also gives you the pleasure of waiting, minutes per device.

At first blush this seems inconsequential; the alternative is no measured data at all. Get exactly what you want with a little bit of inconvenience, or don't get anything at all. That sounds like an easy choice!

Until it's you, sitting in the lab, measuring a box full of parts. Yourself. Then it's excruciating.
 
OK, got some data, at least a first shot. I did it on a TO-220 version of DN2540, as that was in the socket when I gave the revised curve tracer its test run. I'm actually interested in the TO-92 version, I'll get data for that in case it's different, after Keantoken tells me whether this format data is good enough for him to work with.

Here's some Xcel graphs of what it looks like:
DN2540%20TO-220.png

I had a bum point (at 4Vds, -1.31Vgs) that I fixed by eyeball, the rest are as measured. The actual data can be downloaded here.

Can you work with something like this, KT? I only measured up to 40Vds, the tracer gets a little wonky above that. Also, I'm not using the part that high in voltage (or current, for that matter). I did current up to about 700mA, the FET wanted to oscillate right around there even with all the caps and beads.

___

The modification of the old Heathkit curve tracer turned into a lot more of a project that expected.... but I guess should have expected that :rolleyes:.

I have it able to measure in 300msec pulses of varying duty cycle or continuous (see pix). The results are pretty much the same either way until you get up near the 100s of mA. I also modded it to put a "Marker" dot on the scope (adjustable with two pots and pulsed into the X/Y signals via analog switches) so that I could get more accurate readouts of data points - just move the dot onto the part of the curve to measure, then read the voltage on the pots with a couple DVMs. I also provided a DC offset adjustment for Vgs, a reverse switch for the stepping waveform (so it could start at low current rather than highest current). Blew the power supply regulators.... twice.... and fixed those. Updated most of the opamps for better bandwidth to handle the pulsing.

Pulsed.jpg

Note the marker dot (positioned at center screen for the photo).
Continuous.jpg

It was a pure design-as-you-go-along effort. The box looks pretty nice on the outside (notice that it has grown 4 new knobs, 2 new test terminals, and 3 new switches!), but is a frightening mess inside now, wires everywhere, black tape, patched-in circuit board, etc. Also a few caps and bigger ferrite beads thrown in to try to settle down the FETS when they get to higher currents. It also takes a while and some effort to record a set of curves, followed by time with a spreadsheet to correct the voltages for the settings of various switches. Which is why I haven't gone ahead on the TO-92 version of DN2540 (or LND150 or IPP80N06S2L) yet.

In%20Use.jpg

I'm not writing the box up for the test gear forum, as I don't think there are many of these old tracers around anymore for someone to start with, and I didn't keep up very well on notes of all the stuff done or what exact circuits finally worked. I probably have more time spent on the tracer than I'll spend on the amplifier I want the SPICE model for! (The model is almost academic at this point, as the amp is working already on breadboard. But a SPICE model might help tweak it in a bit better). But, hey, now I have a curve tracer....

Let me know, KT.
 
I don't think I can do a Vgs/Id sweep (could reconstruct a version of it from the other curves, though). The Heathkit uses a full-wave rectified (but not filtered) output from it's power transformer as the source for horizontal sweep! Clever, given that making sweep ramps would have cost actual money back in those days. I briefly considered redoing the horizontal sweep with an actual ramp and power amp, but.....

The Vgs (or Ib) steps are done with a TTL counter chip and weighted resistors.
 
Thermally stable, yeah a problem. Here is data as far as I could keep it sort of happy at 40V and 2V. I have to vary the offset voltage to get the values (changed to a 10-turn pot), and use steps (and pulsing) to sneak it to the higher values. You can see there are some not-so-smooth parts on the graphs.

I'm actually using the part at the moment at 2V (in a cascode and around 10mA), so that's really the most important to me personally.

http://libinst.com/random/DN2540%20TO220%20done.xlsx

IdVsVgs.png
IdVsVgs2Vds.png

(this is still the same TO-220 part, BTW)
 
Last edited: