Bipolar output isolation and grounding switch

[ezalys]

I have a 20-bit bipolar output DAC that's driving a very sensitive load, and I want to isolate and ground the load during power-on and power-off. The obvious solution is an SPDT relay that clicks between the DAC and ground, and requires power to connect the DAC. I understand however that there may be a clever way to do this with active devices.

The tricky bit is that with a FET, you need to supply the voltage needed to get the device to go into saturation. I don't know how to do this when you're isolating a bipolar output DAC. Has anyone seen any circuits that have figured out how to do this? My first whack at the problem is using a photocell from gate to source to bias the FET.

[ejeffrey]

Laser diode drivers have used jfets for this purpose. Do you have a maximum transient voltage spec? 

[ezalys]

The transient voltage can vary from device to device. The DAC is a general purpose thing -- some devices that will see it want maybe a volt per minute max, but the DAC may also used to generate control signals which have no meaningful transient spec. That's why a switch sounded like an appealing idea. Check power good, connect load, and wait to see if the 60 Hz goes away to disconnect the load after ramping the voltages down.

[moffy]

Do you have a good reason not to go with your obvious solution, a SPDT relay? It might be the best.

[ezalys]

A fellow voltnut had the insight that there could be non-negligible thermal EMF contribution if I'm not using a latching relay. The DAC I'm using has superb stability, and I'm using an LTZ1000 as a reference, and I'm just curious if there's a nice way to avoid spoiling this spec. He had also suggested a nice design using a latching relay. I'm also curious if there's some clever circuit using FETs that would do the job, just for my own edification and potentially to avoid corner cases of the relays being mechanically jostled into the wrong state before power-on. There are of course ultra-low thermal EMF relays, but they can get pretty expensive.

[DavidAlfa]

JFETs, contrary to regular MOSFETs, conduct when Vg=0.
N-JFET turn off when Vg<0V (negative)
P-JFET when Vg>0V

You need nothing fancy here, get a p-jfet, turn it off by applying 5-8V to the gate and a pull-down resistor.
When the power goes off, the jfet will discharge and turn on.

[ezalys]

I haven't specified JFET or MOSFET. I'm just asking if there are any circuits which would meet my specifications.

[ezalys]

I suppose this could work for crowbarring the output, but only for positive voltages. A JFET attached straight from ground to the output will see Vds positive and negative.

[T3sl4co1l]

There are also depletion mode MOSFETs, functionally same as JFETs.  I forget if P-ch are widely available, but likely moreso than P-JFETs which are almost entirely extinct.  Anyway, complements shouldn't be important as long as isolated gate drive can be provided.  Likewise, normally-closed SSRs are available.  (SSRs will have a similar thermal EMF concern as with regular relays of course.)

Is ESD protection a concern as well?  Shunting an unpowered output seems like unexpected behavior to me, also seems like taking on additional responsibility that might not be there otherwise (i.e. in event of transients like ESD, or cross-wiring, what is supposed to happen?).

Tim

[ezalys]

Maybe it's just cargo culting but typically we always like to have outputs either driven or grounded. The old keithley 230 seems to use some relays in the configuration I had described to shunt the output to ground, so it didn't seem like such a crazy extra thing to do.

And yes, we're pretty concerned about ESD. We haven't properly applied the human body model or any of these other standard engineering methods as we should, but grounding the output seemed like a not unreasonable step to take.

[ezalys]

Sorry -- there actually is a concern. Suppose we use an active device as a switch. It'll leak to some degree -- much worse than a Relay, for instance. We can take the PVA33NPbF as an example, which gives an open resistance of 10^10 ohms. If the load presents a comparable impedance, the switch won't really provide much isolation. This is a bit of a contrived situation, since with a 100 pF load that's a 1 second time constant. I guess a meter of coax would eat that up just fine. But still... it feels close enough to be a concern. Even though the time constant is long, it still just means there's a pole at 1/(2*pi*R*C), some rather attenuated transient might still get by. Just feels better to clamp it.

[moffy]

Relays have thermal emfs and semiconductors have leakage currents and larger resistance etc. From what I can see low emf reed relays seem to be the preferred solution for high accuracy signals, but they are relatively expensive. You mentioned latching relays as a possibility, if you had one with an extra set of outputs you could use those to monitor its state. e.g. https://au.mouser.com/datasheet/2/307/en_g6k-348800.pdf
Certainly a lot less expensive than the low emf reeds, might be suitable.

[T3sl4co1l]

What does switch leakage matter if this is shunting a DAC output?

Or by "isolate" do you mean cutting ground as well?  That's almost definitely not a good job for a SS switch.

Tim

[moffy]

Leakage is a problem when not shunting the DAC output, because of finite output impedances it will induce errors, unless I misunderstand your point? The leakage current will be there when the semiconductor switch is supposed to be off.

[T3sl4co1l]

It's a DAC, it's got all the sense and feedback right there, or as low a source impedance as you like; even a basic DAC like MCP4922 has a Ro of some ohms, that's 32 bits against a 10^10 ohm shunt, vastly in excess of its 12b capability.  I assume any DAC for higher bits also uses buffering (whether internal or external); wiring an R2R or whatever straight to an output would be silly.  And, with other connections, leakage can be bootstrapped anyway.  (Which also goes for ESD protection devices.)

Unless this is a far more limited instance than I imagine, but I've seen nothing to suggest this would be a problem.

Maybe there's something about the "isolate" vs "ground" functions that I'm not imagining the topology of.  Isolation seems redundant as an unpowered DAC isn't going to produce much output anyway, so the shunt dominates; that seems pretty trivial to me.

Tim

[ezalys]

Ah I think I'm being confusing by conflating the isolating and grounding switches I'm proposing. The isolating switch disconnects the DAC from the load during power-up. My experience has been that when opamps don't have their rails in spec, their output can basically do anything. I figure I'll just disconnect the DAC from the load during power-up.

If I just ground/crowbar the output stage, that's somewhat unkind to the output opamp. If I just isolate and don't ground, then the load sees the DAC through the open state resistance of the switch. Which, eh, is maybe fine. I gave my rationale for being paranoid about this above. I’m considering the situation that the 10^10 ohms is the thevenin impedance of the misbehaving output stage, forming a voltage divider with the load.

[moffy]

Sorry to bother, ezalys, but could you draw what you want, just for my own curiosity.

[ezalys]

Just this, With S1 open and S2 closed (as shown) during power up and power off, and S1 closed and S2 open during the normal operating mode.

[moffy]

thank you for that, it is what I thought, hence your need for the SPDT switch/relay.

[ezalys]

Maybe I’m asking the wrong question. Is there some condition under which I can short the opamp output to ground? With this configuration, when unclamped, the signal never passes through a relay, and you never take a hit due to the seebeck effect. Is it still mean to crowbar the opamp output if + and - are shorted together? I guess it probably still is — is there some way to make this a valid approach?

[moffy]

Unfortunately for the output of the opamp it is probably undefined during the PSU switch on phase. You might suffer an undefined surge on the output. This is similar to avoiding the switch on thump of audio amplifiers, and they follow a similar approach of disconnecting the speakers, with a relay, for a given time after switch on until the amplifier settles.

[T3sl4co1l]

Oh -- I remember now!  FYI for context, OP is working with very sensitive loads (semi/superconducting junctions in a cryostat or something like that?), so an op-amp generating a few erroneous volts during startup could be a game-ender.

Context is indeed important.

Tim

[Marco]

JFET to ground the output, SSR to disconnect it from the source (JFETs aren't exactly low ohmic, but low enough in combination with the SSR).