Fast optocoupler shutoff

[matthuszagh]

I'm designing an over/under-voltage protection circuit that needs to have really fast shutoff (ideally in the low 10s of ns). The downstream circuit is connected/disconnected with a bidirectional mosfet switch whose gate is driven by an optocoupler (probably the VO1263). The optocoupler is needed to isolate the gate from the comparator circuitry. Otherwise the gate-source and gate-drain ratings would be greatly exceeded (input can go to +/-100V). I've attached an image showing this part of the circuit.

My question is: how can I make this shutoff as fast as possible? The turn-on time isn't critical (within reason, anything in the ms range, or even up to a second is fine). Here's what I've come up with so far. I'm using a P-channel JFET to switch in and discharge the charge stored on the MOSFET gates (the capacitance is significant, as these MOSFETs are designed for high VDS and low Rds(on)). This is discussed in the VO1263 and VOM1271 datasheets. Additionally, I'm planning to add a bleed resistor (unlabeled resistor in the schematic) to hasten the turn on time of Q4. And, I can use JFET with a low on-state channel resistance, or parallel several. There's a tradeoff here, since I don't want the JFET gate capacitance too high. I'll use the smallest possible bleed resistor such that the optocoupler can still bring the MOSFET gate voltage high enough (optocoupler short-circuit current is in the 10s of uA). To be able to make this bleed resistor smaller, I'm planning to parallel a number of these optocouplers (I think 10, but that's TBD). Another thing I'll do is to make the JFET gate threshold voltage as high as possible so it kicks in as quickly as possible when the gate starts to discharge.

Are there any other options? Have I overlooked anything in my various optimizations discussed above?

[moffy]

If you need 10s of ns shutdown then you will need a dedicated separate high speed shut down circuit. The photovoltaic coupler is orders of magnitude too slow, you would need something like a fast digital isolator to get the response you need.

[matthuszagh]

I don't understand how that would work. As far as I know, all digital isolators require power rails on the output side. That will exceed the MOSFET Vgd/Vgs rating.

[matthuszagh]

What is it precisely that makes the optocoupler shutdown slow? If the photodiode stops generating current quickly after the LED is shutoff, then it's just that the load capacitance has to discharge through the parallel resistance. The VOM1271 uses a parallel 200pF, 10Mohm load to measure shutoff time. I can use a much lower parallel resistance than 10M (probably 30k) and the JFET gate only has 5.5p of capacitance.

[matthuszagh]

I'll try to setup some measurements to actually verify this, but I looked up latencies of LEDs and photodiodes. LEDs apparently have latencies of single-digit to 10s of ns and photodiode latencies are in the picosecond range. So, I'm maybe a bit out of my budget when I consider other contributions to shutoff, but not by much. I don't see why I'd be orders of magnitude off.

[moffy]

What is it precisely that makes the optocoupler shutdown slow? If the photodiode stops generating current quickly after the LED is shutoff, then it's just that the load capacitance has to discharge through the parallel resistance. The VOM1271 uses a parallel 200pF, 10Mohm load to measure shutoff time. I can use a much lower parallel resistance than 10M (probably 30k) and the JFET gate only has 5.5p of capacitance.

The datasheet: https://www.vishay.com/docs/84639/vo1263aa.pdf on Page 2 "Switching Characteristics" quotes a "Turn off time" of 472us. When the LED stops shining there is still a lot of conducting charges in the photovoltaic diode that need to be swept out. Because you want these charges to last a long time for efficiency sake the device is very slooow. And yes for a digital isolator you would need a supply on both sides, but if you want ns response that is part of the cost.

[matthuszagh]

The datasheet: https://www.vishay.com/docs/84639/vo1263aa.pdf on Page 2 "Switching Characteristics" quotes a "Turn off time" of 472us.

Ok, but that's with a 1M, 15p load. Isn't that going to slow the turn-off time? The VOM1271, which has an integrated fast turnoff (I think this is just a p-channel jfet and resistor) is more than an order of magnitude faster, indicating at least some of the turn-off time isn't just optocoupler delay.

When the LED stops shining there is still a lot of conducting charges in the photovoltaic diode that need to be swept out. Because you want these charges to last a long time for efficiency sake the device is very slooow.

Hm, ok this I don't know much about. I'll try to find out more. If you have any resources I can read that would be much appreciated!

I think I'm going to setup a measurement circuit like the one attached. The load resistor will be small enough that the optocoupler still sources its full short-circuit current, but large enough to be clearly visible on the scope. The opamp is just there to isolate the test circuit from the scope input and attached cable capacitance. You might well be right, but I want to be sure of how much of this delay comes from the actual optocoupler and not the specific load conditions used for the datasheet values. Oh and I neglected to show the scope measuring the input signal too, but of course it will to measure the delay between input and output turn-off.

[moffy]

The turn off time measurement has an If of 20ma, or it is discharging at a constant 20ma, not the 1M/15p.

[matthuszagh]

The turn off time measurement has an If of 20ma, or it is discharging at a constant 20ma, not the 1M/15p.

IF is the LED forward current. The short-circuit output current is in the range of 10s of uA. 20mA doesn't mean the output is discharging at that current.

[matthuszagh]

Here's a scope capture of the VOM1271 turn-off into a scope input (1M||10p, plus cable capacitance). This is not the circuit I proposed above, which I have not yet had the time to setup. This isn't an ideal measurement, and the time scale is too coarse, but clearly the photodiode stops generating current pretty quickly after the input current shuts off. This looks to me like the photodiode shuts off and the 10p cap (plus cable) start discharging through the 1M resistor. That drop off is, I think, the internal JFET kicking in.

[moffy]

The turn off time measurement has an If of 20ma, or it is discharging at a constant 20ma, not the 1M/15p.

IF is the LED forward current. The short-circuit output current is in the range of 10s of uA. 20mA doesn't mean the output is discharging at that current.

Good point.

P.S. The gate capacitance of the MOSFETS could be many nF.

[Marco]

Your bleed resistor will need to be upwards of 220k, even 10p is a lot then.

I'd look at using a 6n135 just for turn off (the Avago datasheets have some fast linear circuits you could adapt).

[matthuszagh]

P.S. The gate capacitance of the MOSFETS could be many nF.

The MOSFET gates will discharge through the JFET channel. When the JFET turns on, it has a channel resistance of 85ohms (at least the one I'm currently looking at). There are JFETs with even lower channel resistance, but generally you trade off gate capacitance. The bleed resistor does not need to discharge the MOSFET gates because the 2M resistor isolates them. It only needs to discharge the JFET gate, so the total relevant capacitance across the photodiode is still just 5.5p.

[matthuszagh]

Your bleed resistor will need to be upwards of 220k, even 10p is a lot then.

No, I don't think so. This is the point of using multiple optocouplers in parallel. I should be able to divide the bleed resistance by the number of parallel optocouplers. I was planning to use something like 30k bleed resistance and 10 optos (5 of the VO1263s). That should turn on the JFET in under 50ns (ignoring any optocoupler delay). That's still a bit slower than I'd like, so I may add some more in parallel.

I'd look at using a 6n135 just for turn off (the Avago datasheets have some fast linear circuits you could adapt).

Unfortunately the 6n135 requires a power supply at the output (all fast shutoff optos do, I think). That will exceed the MOSFET Vgd/Vgs ratings. I'll take a look at the Avago datasheets.

[Zero999]

How much current?

You won't be able to turn it off in 10ns. Use a gas discharge tube or transient voltage suppressor, as well as switching off the MOSFET.

[Kleinstein]

Photodiode / PV cells are slow to get good efficiency.  In PV mode I would expect a silicon photodiode to take some 5-50 µs decay time. That is just the minority carrier lifetime in relatively clean silicon.  The PV couplers may already include an extra circuit to turn of a bit faster on the scope this is seen as the fast drop after some 40 µs.

Normal optocouplers with a phototransistor output can be resonable fast on turn on, especially when driven hard. It can still need quite some time to discharge the capacitance and maybe drive another transistor as emitter follower. Discharging the PV coupler may take some time, they may well act like a capacitance (e.g. 50 pF range) a bit similar to the storage time or reverse recovery of a very slow didoe.
The 6N135 chould get away from the supply of the PV coupler still the turn off time is more in the 500 ns range, maybe 100 ns with high drive current.

[matthuszagh]

How much current?

You won't be able to turn it off in 10ns. Use a gas discharge tube or transient voltage suppressor, as well as switching off the MOSFET.

Yeah I was planning to put a TVS diode downstream. I may also put a small resistor in series with the downstream circuitry (0V-biased parallel n-channel JFETs) to provide some additional current limiting, though I can't get away with that much series resistance. A suitable TVS diode will be able to handle the current, assuming I get the MOSFET to switch off in time. I do want to limit the amount of time I short out the upstream device though.

[matthuszagh]

Photodiode / PV cells are slow to get good efficiency.  In PV mode I would expect a silicon photodiode to take some 5-50 µs decay time. That is just the minority carrier lifetime in relatively clean silicon.  The PV couplers may already include an extra circuit to turn of a bit faster on the scope this is seen as the fast drop after some 40 µs.

Normal optocouplers with a phototransistor output can be resonable fast on turn on, especially when driven hard. It can still need quite some time to discharge the capacitance and maybe drive another transistor as emitter follower. Discharging the PV coupler may take some time, they may well act like a capacitance (e.g. 50 pF range) a bit similar to the storage time or reverse recovery of a very slow didoe.
The 6N135 chould get away from the supply of the PV coupler still the turn off time is more in the 500 ns range, maybe 100 ns with high drive current.

The 6N135 looks really helpful here. Thanks for the suggestion!

[Doctorandus_P]

6n135 is rated for 1Mb/s, while the 6n137 is rated for 10Mb/s, also if you go to fiber optic modules such SFP (Small Formfactor Pluggable), they readily go to 10Gb/s but those are already getting expensive. You can go quicker, but that gets more expensive exponentially still.

https://en.wikipedia.org/wiki/Small_Form-factor_Pluggable

And for speed, I would probably go for a dedicated gate driver IC. That would need a separate (isolated) power supply, but that can be a simple EUR4 sil module.

And are you sure this is not an https://en.wikipedia.org/wiki/XY_problem
Why do you need that speed in the first place? Maybe it's a better option to add some kind of inductor or filtering on the "other side" to lower di/dt.

[matthuszagh]

6n135 is rated for 1Mb/s, while the 6n137 is rated for 10Mb/s, also if you go to fiber optic modules such SFP (Small Formfactor Pluggable), they readily go to 10Gb/s but those are already getting expensive. You can go quicker, but that gets more expensive exponentially still.

https://en.wikipedia.org/wiki/Small_Form-factor_Pluggable

And for speed, I would probably go for a dedicated gate driver IC. That would need a separate (isolated) power supply, but that can be a simple EUR4 sil module.

And are you sure this is not an https://en.wikipedia.org/wiki/XY_problem
Why do you need that speed in the first place? Maybe it's a better option to add some kind of inductor or filtering on the "other side" to lower di/dt.

Ah yeah of course, I can just create an isolated supply. That would simplify this tremendously and then I can use proper, high-speed optocouplers.

The question of how much speed I actually need is a bit tough to answer. I'm trying to protect a sensitive, downstream preamplifier. The first stage is some n-channel JFETs in parallel biased to 0V. I need to be able to handle hotplugging power supplies (DUT) up to +/- 100V to the input. There are at least two parts to this challenge: protecting the JFETs and not shorting out the DUT for too long. Eventually I'll do some surge testing of these JFETs, but otherwise its difficult to get data on how much power they can actually dissipate for very short periods (ns to us durations). Anyway, generating an isolated supply should make this whole thing a lot easier. Thanks for the suggestion!

I will add some additional downstream protection, including some small series resistance and a TVS diode. If the TVS can't handle the current duration, I may also add a limiter like the BAV99, we'll see. As for a series inductor, I'm not really a fan of that solution. That limits the max voltage, but prolongs the duration, which could still blow up these JFETs. I looked at ferrite beads too, but they were also too inductive at applicable frequencies.

[Kleinstein]

The inductor will prolong the pulse, but after some time the JFETs will turn off and than the current will go down quite fast.  The tricky point could be ESD, as this gives a quite high voltage (e.g. kV range) and thus still a quite fast rise in current. With a more moderate voltage the inductor can be a good solution to limit the very fast part. 1 kV and 1 mH would give 1 A/µs for the speed that the current can go up - though only until the inductor saturates, which may not be very long. It also helps with EMI.

There is also the option to add a local current limit to the MOSFET part. So a series resistor between the MOSFETs and BJTs to short out the gate to source of the FETs.  With some 60 ohm and some 600 mV for the transistors to turn on one would be in the 10 mA range. This may be acceptable for the amplifier and DUT side for a short time until the optocoupler part turns on (some 1-10 µs range) to reduce the current even more.

Some Keithley meters like K2001, K2000 use such a protection.  As a simple part they use an optocoupler with AC drive (2 LEDs) also for the clamping part. So a rather direct feedback loop with no many extra parts. The steady state courent would than be at the PV drive current divited by the CTR (around 10-100%), so in the 10-100 µA range, which is normally OK.

[matthuszagh]

There is also the option to add a local current limit to the MOSFET part. So a series resistor between the MOSFETs and BJTs to short out the gate to source of the FETs.  With some 60 ohm and some 600 mV for the transistors to turn on one would be in the 10 mA range. This may be acceptable for the amplifier and DUT side for a short time until the optocoupler part turns on (some 1-10 µs range) to reduce the current even more.

I looked into a BJT current limit, but unfortunately the thermal noise from the series resistor is too much. Even an order of magnitude less than this would dominate the preamp input noise. I may accept some degradation in input noise to add a protective series resistor, but 60ohm is too much.

[Marco]

Depending on the duty cycle you don't really need an additional power supply, the VO1263 is a power supply.

[Kleinstein]

If 60 Ohm is way to high this means a really low noise amplifier.
One may consider having something like MOVs at the input to catch ESD or other high voltage peaks well higher than the normal DUT and than rely on an inductor (possibly physically relatively large, like a flyback transformer) to slow down the rise in current, before the protection can kick in.  For a short time the LED in the OK should be OK for maybe 100 mA. So with lets say 100 V and 1 mH one would have 100 A/ms and thus 1 µs to reach 100 mA. With higher current to the LED the turn on time is expected to go down. So even "normal speed" optocouplers like TLP290 can turn on in less than 500 ns.
A fast rising voltage would also need extra current to discharge the gate. So one may want fast switching fets for not too high a voltage.
A problem could be that the clamp voltage from the LEDs may reach 1.5 or maybe even 2 V at higher current.

[Zero999]

How much current?

You won't be able to turn it off in 10ns. Use a gas discharge tube or transient voltage suppressor, as well as switching off the MOSFET.

Yeah I was planning to put a TVS diode downstream. I may also put a small resistor in series with the downstream circuitry (0V-biased parallel n-channel JFETs) to provide some additional current limiting, though I can't get away with that much series resistance. A suitable TVS diode will be able to handle the current, assuming I get the MOSFET to switch off in time. I do want to limit the amount of time I short out the upstream device though.

You still haven't said how many amps it needs to pass.

How about an SCR in series with a depletion mode MOSFET? The SCR and MOSFET can be coupled to the drivers via pulse transformers. The SCR will need a short pulse to turn on and the MOSFET a pulse to turn it off for long enough for the SCR to latch off.