8
\$\begingroup\$

Application: I have a copper mesh (10cm x 10cm square) in a vacuum chamber connected to a BNC connector by a 24-cm-long copper wire. The goal is to switch the mesh voltage (referenced to ground) from 8 V to ~0 V quickly. (This will switch the electric field in the chamber, which is a control mechanism for our atomic physics experiments.) It is essential that roughly 500 ns after the switching starts, the signal settles to <10 mV (~<0.1%). The mesh is floating; it is not terminated in the chamber.

Problem: There is a "hump" at the bottom of my inverted square pulse. I need to flatten it.

Circuit: I have settled on a simple MOSFET switching circuit:

circuit schematic

Description: The MOSFET (ZVN2110A-ND, N-Channel Enhancement Mode) is driven by a IRS2117PBF-ND driver, which outputs a 15 V positive pulse. The baseline of this trigger pulse floats on V_S, which is tied to V_LO by a small resistor. The mesh is connected to Point B. The output low-pass filter was an attempt at fixing the problem. All resistor values were determined experimentally (i.e. by initially using potentiometers). The result was hard-wired using a "dead-bug" style on a copper-clad board.

hardwired dead bug enter image description here

Probe Details: To simulate the mesh, I soldered a 24 cm wire to a piece of copper-clad perf board and connected it to the circuit output (Point B). I probed the signal on the perf board with a Tektronix probe (500 MHz, 8.0 pF, 10MOhm, 10x) into a Tektronix scope (TDS3012 100 MHz digital scope).

Observations: It switches quickly enough (although I could speed it up by removing the filter), the ringing amplitude and duration is tolerable, but on the (essential) microsecond time-scale, there is a large "hump" and droop/sag of 20 mV (labeled in image by red line). This is unacceptably large and makes it impossible to do our experiments, which take place from the moment of switching until about 10 microseconds after switching.

scope trace 1scope trace 2

Details of Application: We use electric fields to tune atomic resonances in our experiments. Scanning the electric field applied to the atoms lets us record a "spectrum" of these resonances showing their location and shape. The widths and separations of these resonances are on order of 1-10 mV/cm (very small!). To apply the electric field, we place the atoms between two flat copper pieces of mesh, separated by 1 cm. The E-field between the copper mesh pieces is just the potential difference between the mesh pieces (1 V difference equals 1 V/cm E-field, a 1-to-1 conversion). In collecting a spectrum, we sample an E-field value by switching to the corresponding voltage and waiting a few microseconds before detection. If the voltage (and thus the E-field) drifts during the sampling period more than the size of the resonances (<10 mV) the resolution is degraded to the point where our spectrum picture becomes blurred beyond recognition.

Additional Thoughts: I have considered the possibility that the MOSFET is heating up, thereby changing its on-resistance (normally ~4 Ohms). To test for this, I tried two things: (1) placing two MOSFETs in parallel, and (2) replacing the ZVN2110A with a IRF1010EZ MOSFET that has a much lower on-resistance (100 mOhm). Neither things helped, the "hump" is still 20 mV and still lasts a few microseconds. It seems to me that increasing the pull-up resistor (as suggested in the comments) could also help, so I will try this.

Update 1: I have tried increasing the pull-up resistor from 470 Ohms to 10 kOhms. There was no effect on the output; it still has the 20 mV "hump" after the initial ringing.

Update 2: Disconnecting the "mock-up" wire + mesh from the circuit and probing Point B directly has no effect on the measured signal.

Update 3: Below are traces for the corresponding points in the schematic above:

enter image description hereenter image description here

It appears as though the "hump" appears on the gate pulse, too. Point "D" right near the FET does not look any different than probing the mesh.

Update 4: I have (1) increased the pull-up resistor to 1kOhm, (2) removed the filtering 1000pF resistor, (3) disconnected the mesh, (4) added two "jam can" 470uF electrolytic capacitors to the rails, and (5) replaced the pulse generator with a faster one (Agilent 33250A). New schematic and traces:

enter image description here enter image description hereenter image description here

Even with a faster trigger pulse for the FET driver, the problem remains. The "jam can" caps do seem to filter out some high frequency oscillations, but the "hump" remains.

\$\endgroup\$
  • 2
    \$\begingroup\$ It's not certain what you mean by droop in this context. I suggest posting a picture with the "droop" portion clearly labelled. Also, please explain why this droop makes the circuit unsuitable. Active hi/lo drive may help. \$\endgroup\$ – Russell McMahon Aug 28 '12 at 7:56
  • \$\begingroup\$ All I can suggest is that your pullup resistor is very low. I thought the impedance of this mesh was high (not deliberately connected to anything), so a much higher pullup, like 10 kOhms, should keep it at the high voltage well enough. Like Russell, I don't know what you mean by "droop" since the usual definition doesn't seem to apply. Show a scope trace and explain exactly what you don't like about it. \$\endgroup\$ – Olin Lathrop Aug 28 '12 at 12:33
  • \$\begingroup\$ I've reworded the post to focus on "settle time", rather than "droop". The link under "Images" provides scope traces that highlight the problem. (Note: Until I have a reputation of 10, I cannot post pictures or more than 2 links.) I will try a 10kOhm pull-up resistor and post the results. \$\endgroup\$ – higgy Aug 28 '12 at 15:30
  • \$\begingroup\$ An upvote to your question should fix the reputation problem :) How big is your mesh? I assume the 8v is measured from one mesh to the other. Voltage has to be referenced to something, after all. I'm looking for capacitance that could hold up your voltage. There's 1000pf on the output. There's the capacitance of the plates. There's the amplified (Miller) capacitance of the FET. But none of this is very convincing. Also, combined with the inductance of the leads, there could be RF resonance, (the ringing) but that seems to settle fast enough. \$\endgroup\$ – gbarry Aug 28 '12 at 16:34
  • \$\begingroup\$ @gbarry: For this discussion, reference 8V on the mesh to ground. The mesh is approx. 10cm x 10cm square, composed of very thin copper wire (~95% optical transparency), supported on the sides by a steel frame, and electrically isolated from the chamber. An identical mesh is parallel to it and separated by 1 cm; it is grounded. The 24-cm-long wire connecting the mesh to the "outside world" (i.e. the BNC feedthrough) is unshielded and may have a few loops in it. \$\endgroup\$ – higgy Aug 28 '12 at 16:41
2
\$\begingroup\$

If you look at the characteristic frequency of the hump it's on the order of 100's of KHz. the only thing n that circuit that has a dominant pole in that range will be the power supplies. Look at the lower rail and see if it correlates to the hump.

\$\endgroup\$
  • \$\begingroup\$ The lower rail is powered by a HP E3620A (25V, 1A, Dual Output). The upper +15V rail of the MOSFET driver is powered by the other output of this supply. The drain side of the MOSFET is powered by a MPJA 9312-PS (120V, 1A). The only thing I could find corresponding to a time/frequency at first glance for the HP supply was "load transient response", which is supposed to be <50 microseconds. That sounds slow. Could that be the issue? \$\endgroup\$ – higgy Aug 28 '12 at 20:42
  • 1
    \$\begingroup\$ Yep, this essentially what I pointed out. Your rail is bouncing. Control loops in the Power supplies tend to be lower frequencies. -> Solution? put some jam can caps on there to provide charge storage. \$\endgroup\$ – placeholder Aug 28 '12 at 20:50
  • \$\begingroup\$ Gbarry's comment about driving directly is a good one if you can run it off the lower rail. \$\endgroup\$ – placeholder Aug 28 '12 at 20:53
  • \$\begingroup\$ If it turns out that the Agilent pulser isn't the problem (see @gbarry's comment) and I try this, how should I wire the caps into the circuit? A cap from V_LO supply to ground? (If it's complicated, could you link to a drawing?) Also, could you suggest a cap value (would 0.1uF do?), and what do you mean be "jam can"? \$\endgroup\$ – higgy Aug 28 '12 at 21:31
  • 1
    \$\begingroup\$ Jam can" -> i.e. Very large, the size of well ... a jam can (don't ask me why that term is used, jam isn't sold in cans any more). You are actually in luck here, the response is slow so even an electrolytic (Al) will work. Just wire it across the + and - terminals of the Power supply to test and dead bug later. Pay attention to polarity. Current limit supply whilst turning on or in rush current will be too high. get 100's of uF size. If you put your probe on the lower rail you can verify this hypothesis right now however even before wiring in a cap. \$\endgroup\$ – placeholder Aug 28 '12 at 21:47
1
\$\begingroup\$

I would bet that the hump, as you call it, is caused by the capacitance of the mesh and the inductance/impedance of the 24 cm cable. Here are some things to try:

  1. Reduce the length of the 24 cm cable. This will reduce the inductance/impedance of the cable and allow for faster discharging of the mesh.

  2. Make the 24 cm cable thicker. Same concept as #1.

  3. Move the MOSFET right next to the grid, inside the chamber. Same concept as #1, but taken to an extreme.

  4. Any wire that is carrying the mesh discharge current must be as short and thick as possible. This includes any ground wires.

Some of these, maybe most, are going to be impractical to do during "scientific operations", but they are worth doing anyway just to help narrow down where that hump is coming from.

\$\endgroup\$
  • \$\begingroup\$ As you anticipated, all of these suggestions would be difficult. They would require opening the chamber, which would delay our experiments by about a month (getting to ultra-high vacuum requires several pump stages and baking). I can still test your theory by using my "mock-up" wire + copper-perf-board to see if the hump can be reduced with shorter/thicker wire, etc. \$\endgroup\$ – higgy Aug 28 '12 at 18:57
  • \$\begingroup\$ Update: The traces in the OP were from probing a copper-clad perf board connected to the "Point B" by a 24-cm-long wire, to imitate the wire+mesh. I simply disconnected it and probed "Point B" directly. The signal was unchanged, unfortunately; the "hump" is still there and the same size. \$\endgroup\$ – higgy Aug 28 '12 at 19:22
1
\$\begingroup\$

It might be instructive to know what the voltage was doing (a) out on the grid, (b) at the resistor connected to point "b", (c) right at the drain of the FET, and finally, (d) at the gate of the FET. It may be inductance/capacitance in the wiring, but it might be the FET doing something other than what we expect.

I wonder if you could drive the grid directly from the IRS2117, since neither your voltage nor your current are extreme. A gate driver is designed to drive the capacitive load of the FET's gate, and this seems to be the nature of the original problem.

Finally, if you have to go extreme, some kind of control loop scheme may be necessary, where you have a negative supply and actually drive the output negative until it reaches zero (this pulls current from the grids)...then you bring a feedback line in from the output to control this driving circuit so that it applies just the right drive to get this behavior.

Edit: I just noticed V LO. What voltage is that? I think most of my answer just went away...

\$\endgroup\$
  • \$\begingroup\$ I've taken traces for most points on the circuit, but I'll have to wait 'til later to post them. Interestingly, the "hump" appears on the gate pulse and also on the input trigger pulse to the driver. So perhaps my Agilent 33220A is the problem and it's just propagating through the circuit. V_LO, for this discussion, can be taken to be 0V. See comment to @rawbrawb about power supplies. Ultimately, we would like to scan V_LO from 0-7V to record atomic spectra. \$\endgroup\$ – higgy Aug 28 '12 at 21:25
1
\$\begingroup\$

Fiirst, I assume that you are measuring the signal of interest at point B in your circuit.

Second, I presume that you have calculated the RC time constant that your circuit has to deal with - my estimates are (for short direct leads outside the vacuum system): C~100pF, R~600 Ohms, therefore t~0.1usec. To reach 0.1% of the signal requires ~7 time constants or ~0.7usec.

A problem with the circuit, as given, is that the output capacity of the MOSFET is 25pF, the input capacity is 75pF and the transfer capacity is 8pF. Also, the gate charge that has to be removed is 1n Coloumb.

As you have noted, the signal generator output is being transfered through the driver to the input and then to the output of the MOSFET. Also most pulse generators do not reach a true zero volts in their rated fall times - the time is usually specified as the 90% to the 10% time.

A better solution is to use a CD4010UB gate to replace both the driver and the MOSFET - connect the signal generator to the gate input and the gate output to the 600 Ohm resistor attached to point B. Unfortunately the '10 it is probably no longer availabe - I couldn't find one with a search.

The 'second best' part would be the CD4009UB hex inverter (available from Digikey p/n 292-2030-J-ND $0.55).

The 'trick' is that the part has seperate power supply connections for the input and the output sections of the gates. The input connection (Vdd) should be set to the highest voltage that you will need on the output and the output connection (Vcc) set from 0 to Vdd.

In spite of the data sheet, I have used this configuration with Vcc from -0.3V to Vdd with no problem.

You will have to adjust the 600 Ohm resistor to compensate for the gate's internal resistance - ~200 Ohms - or you could parallel all six gate inputs and their outputs. If you do not parallel the other five gates, you should connect their inputs to Vdd - do not let them float.

\$\endgroup\$
0
\$\begingroup\$

There is a high probability you are measuring your scope's overload recovery. Consider the scope screenshot below:

scope overload recovert The voltage measured by the blue trace does not exist. As you can see, on the left side of the display, the trace went offscreen and clipped the highspeed opamp inside the scope analog frontend. This causes all sorts of nastiness, like differential heating in the input stage, upsetting of bias points, etc. As a result, the opamp needs several tens of milliseconds to settle... amazing for a chip which has hundreds of MHz bandwidth, isn't it?

Read the bonus stuff (pink background) in this Jim Williams document:

http://cds.linear.com/docs/en/application-note/an10f.pdf

I'm not saying this is the culprit, but it is likely. When the trace clips, even for a µs, the scope should not be trusted. Any linear circuit which clips, or nears clipping, even for an extremely short time (like 1ns), cannot be trusted for precision or settling until we're absolutely sure everything has cooled down, every stored charge in every integrating capacitor has returned to nominal value, etc...

This includes an opamp which goes into slew rate limiting, by the way. The recovery time is the settling time mentioned in the datasheet, and it is much longer after slewing than after processing a slew-limited pulse of same amplitude. Please note datasheet specified settling time usually implies the opamp DID NOT clip!

In order to measure your settling time, you will need special measures, most likely an analog switch to only let the voltage to be measured through a few tens of nanoseconds AFTER it is inside the scope's range...

You could also use a good precision opamp (specified for quick and accurate settling time, much faster than what you are trying to measure) and limiting diodes in the feedback network. Slow down the MOSFET switching until no spikes upset the opamp.

For the same reason, the flatness of your pulse generator output pulse cannot be measured with the scope.

http://cds.linear.com/docs/en/application-note/an79.pdf

Have fun! When Jim Williams app notes need to be brought in, you know you're in trouble! These are very delicate issues...

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.