From: Winfield Hill
Subject: Strange going-on with a floating MOSFET gate
Date: 4 Jan 2003 15:43:49 -0800
Organization: Rowland Institute
X-Newsreader: Direct Read News 4.20
Reporting on some strange business with a floating MOSFET gate...
John LArkin wrote...
> I did a post a while back: a 2N7000 with a 9-volt battery and an LED
> in its drain, gate hanging open. This can be set on or off and stay in
> that state for many days. You can also perch the gate voltage just at
> the turnon point and get the LED to stay on dimly, with no perceptable
> brightness change over many hours.
This sounded interesting, so I figured it'd be useful to take some
detailed measurements using the data logging capability of my HP
34401 multimeter. My intention was to track a floating gate voltage
by measuring the drain current. First I characterized a subject
2n7000 mosfet over the vicinity of 0.3 to 2mA, making plots of Id
vs. Vgs for fixed Vd, transfer plots of Id vs. Vgs with fixed Vd,
Id vs. Vds with fixed Vgs, and Id vs. Vd with the gate floating, etc.
After wiring a setup with clip leads, I did a bit of experimenting,
setting the gate voltage to 2.00 volts for about 0.5mA drain current.
Right away one sees a steady Id with the gate floating, as expected.
With a sub-uA measurement-noise level in the case of my bench hookup.
The steady drain current reading resulted whether the drain voltage
was higher (e.g. 4 to 6V) than the gate, or lower (e.g. 600mV).
But what I didn't expect was what happened immediately when removing
the clip lead to the gate. Without fail the current would instantly
drop by about 2x, indicating a roughly -75mV change in gate voltage.
Sometimes it would shut off, indicating a larger gate-voltage drop.
This effect was independent of the drain voltage (the same whether
it was positive or negative with respect to the gate) or the drain
current (over the 100x range I tried). This -75mV change indicated
that about 4.5pC of charge was being removed from the gate by my
action each time (say 3 to 10pC), which was a surprise to me.
To continue the experiment I had to set the gate voltage higher than
intended before removing the lead. With the drain voltage at 695mV,
an open gate lead and a steady 540uA drain current, implying 2.04V
on the gate, the HP34401A was left to gather data for two hours. It
recorded one data point every 15 seconds, 480 measurements in total.
After two hours, I extracted the data and plotted the readings.
For the first 156 measurements (about 40 minutes) the current didn't
change at all (except for < 1uA measurement-to-measurement noise).
The drain current change, if any, was under 0.1%, corresponding to a
gate-voltage change of under 100uV. This means any charge injection
would have to be under about 3fA on average. Given Vgs at +2.0V and
Vgd at +1.4V, the gate insulation resistance would have to be over
600000G. or 600T ohms. All very well and good.
But then something very strange happened. At reading #157 the drain
current jumped a few uA. Thereafter it increased by a few uA at each
15-second recording for a steady rate of 5uA/minute. This means that
the gate voltage was increasing by about 1mV per minute. I thought,
aha! given the FET's -2mV/C tempco that's like a temperature decrease
of about 0.5C per minute. But oops! this funny business went on for
the rest of the experiment, for a total change of +45mV in 80 minutes.
I know the lab temperature didn't go down by 22C, to below freezing!
OK, fine, you say. The steady change of +45mV total resulted from a
leakage current _into_ the FET's gate of about I = C dV/dt = + 0.5fA.
While there'd be no problem with observing this low leakage current,
we know this explanation cannot be correct, because all the other
FET voltages were _lower_ rather than higher than the gate voltage.
. ,-----+---- current -----,
. | | recorder |
. D volt +
. float --- G meter supply
. 2.04V S 695mV -
. | | |
The gate voltage should have dropped, and the drain current should
also have dropped. Instead the current rose from 540uA to almost
900uA during the two hours. !!!
An additional very strange thing happened, twice. More later.