Heat Treatment Oven

Sorry, I was speaking in shorthand late at night.  If a Zener or MOV snubber works for you go with it.  If it doesn't we have to go to something like a QuenchArc, which is a metalized capacitor plus a resistor.  I was just thinking about the capacitor part.  You want to use the smallest quenchark (big R, small C) that reliably works.  If your load is usually on, put it across the switch.  If the load is usually off, put it across the load.

As far as the coil that drives a relay or valve goes, I wasn't talking about a capacitor large enough to change the net inductive character of the coil under our normal frequency of operation.  I was thinking of an RF short to try and keep things quiet.  A bead in series never hurts here, neither.

The thing to do is model your scheme in Spice using a good model and good data.

And then see if your scope agrees!

Even that said I never use electromechanical relays of the type I think we are describing, so I could be beyond rusty.  The only electromechanical relay I use is a big bistable from Panasonic, and this always with a microprocessor.  I can suss its state from a second set of contacts every service routine.  This is only for a high-current load that is on for long periods.  Otherwise, it is Si all the way.  One output structure that works switching AC into inductive loads is dual SCR rather than triac, as now we have less ways to turn back on.   Yeah, if you want to snub with a cap, ya need a resistor in series, I just thought that was beyond a given.

Sure, you want your DS channel current to be significantly greater than your actual current.  This is part of the DC analysis.  MOSFETs at these bias levels have positive tempcos and we don't want to warm them up into a regime where they perform poorly and burn fingers.

The issue with quenching spikes is not about DC current levels in the conduction channel whatsoever, it is about the competing parameter, channel breakdown voltage (and AC analysis) when the channel isn't enhanced.  The inductive load will spend its magnetic field attempting to maintain the current through it (and anything in series with it)  when anything acts to change that current.  The spike tapers off logarithmically because its energy source is in the past.  But at the moment of disruption, the current is maintained.  If our Q suddenly goes High-Z, the drain potential will spike to maintain the current level through the conduction channel.  Quenching is about finding an alternate route for this current, so that our voltage spike at turnoff doesn't rise to Q BV.

Varicaps are devices used to tune stuff in RF, like VCOs for PLLs and resonant traps.

Given like technology, diode capacitance is a convenient way to express speed.

Working this feat in an HV regime brings us into the fascinating world of spark-gaps.  The simplest variant looks like two carpet tacks of noble metal in a gas-filled bulb.  Back in the fifties, they developed triggered spark gaps, kinda like little lightning ionizing a path for big lightning.  Then even more sophisticated devices.

One doesn't want to use a 1n400X-type diode on spikes.  This is a power supply diode.  Its fastest intended application would be to rectify French line power.  It is not intended to be used upwards of a KHz.  Another way to say it is its capacitance is way to high.  The diode, provided our currents our sub-Ampere, to use is a 1n914 or 1n4818 because they are gold-doped to minimize minority carrier lifetime.  Shottkys are fast, low-drop, but their BV is limited.  So if we are wrestling in the open class, we need UF diodes.  The degradation on power Qs  by hot electron comes on slowly.  So people who clamp with inappropriate diodes on their bench and proclaim 'it works' don't seem to 'get it.'  Also, no benefit from a fast diode will ensue unless the far end is connected to a fast capacitor through a low-Z route.  Margin for capacitor BV, probably a resistor across it.

Here is a useful circuit idea:  Get  a small line transformer, output around 6-0-6.  Connect this to a tiny bridge rectifier.  Connect this to an opto-isolator, insert R for current-limit.  Connect opto output to a resistor.   Connect this to a schmidt-trigger.  [Connect this to a %2 circuit?] Connect this to an interrupt line on your processor.  Now your uP knows when the zero-crossings are.

The network that will take a reactor and, using a reactor of the other type and a resistor to make the whole thing look purely resistive is called a Zobel network, there is a good rundown on wiki.  The reason you do not want to do this, typically, for a switching circumstanse is that the load will ring for a long time if you do this.  The reason why pure-diode clamping is unideal is that the diode must go through a thermodynamic phase change before it begins to help, this ensures an RF event, at the very least.  Truly excellent clamps use diodes and R's and C's.  Pure R & C is OTT.  Use SPICE and good data to design a cost-effective clamp.  Put a bead on the supply lead for the coil.  Ceramic capacitors are a good choice here, I'd go with metal film or counter-inductively wound wire-wound resistors in the high-power regime.

Electronics, by now, shouldn't just 'work.'  Electronic products should not unintendedly leak RF energy nor should they complicate the power supply.

Don, experience and simulation show that the 1N4001 is more robust and works BETTER than a 1N4148. Other diodes may work better than either but fast or fancy ones will cost more. Using a passive RC network works fine and does not result in ringing of the load current.

I've attached a rather huge screenshot of the three options simulated in LTSpice.

The netlist for the RC simulation follows:

* C:\Docs\Rabbits\relay_zobel.asc

L1 N001 N002 .47

R1 N002 N004 288

V1 N001 0 12

M1 N004 N006 0 0 NDC7002N

V2 N005 0 PULSE(0 5 1m 50n 50n 20m 40m 1)

R3 N006 N005 1k

R2 N003 N004 288

C1 N001 N003 4.7µf

.model NMOS NMOS

.model PMOS PMOS

.lib C:\Program Files (x86)\LTC\LTspiceIV\lib\cmp\standard.mos

.tran 50m

.backanno

.end

The relay is an OMRON G5V-2 type, 12V coil, 288R, 0.47H off, 0.74H on. The simulation makes no attempt to model the effect of the armature moving whe the relay releases.

In case it's not clear from the picture the 1N4001 ciruit rings for about 0.2mS at about 29kHz, the IN4148 for about 0.5mS at 42kHz and the initial amplitude of the ringing is about twice as large for the 1N4148. In either case the frequency and amplitudes are so low that there would not normally be any EMC issues.

The passive network works better with no HF ringing at all.

If you have any other data, measured or simulated, to suggest otherwise, I would be interested to see it.

MK