Clifton Laboratories 7236 Clifton Road  Clifton VA 20124 tel: (703) 830 0368 fax: (703) 830 0711



To search within the Clifton Laboratories site, enter your search term below.


Software Updates
Softrock Lite 6.2
Adventures in Electronics and Radio
Elecraft K2 and K3 Transceivers



Diode Turn-on/off Time and Relay Snubbing

Revision History
29-30 December 2007; First released
08 April 2009; Revised to add relay snubbing measurements
09 April 2009; Revised to add section "Why Use a Snubber"?
11 April 2009; Added RC snubber section
12 April 2009; Fixed bad graphics link

Table of Contents and page links
Recovery Time Conclusion


This page presents turn-on and turn-off time measurements for five power diodes:

  • 1N4005
  • 1N4007
  • 1N4001
  • HEP108
  • 11SQ05

I took these measurements to answer a debate on the PIC mailing list concerning the turn-on time of typical silicon power diodes. In particular, whether a silicon power diode, such as the common 1N400x series, has an appreciable turn-on delay such that it renders it unsatisfactory to clamp the inductive kickback of an inductor, such as a relay coil, when switched with a MOSFET.

On 08 April 2009, I've extended the measurements to include release voltages and operating time for a typical relay without a snubbing diode and with 1N4148 and 1N4007 snubbing diodes. As my December 2007 measurements demonstrated, a 1N4007 power diode is quite effective as a relay snubbing device, since the diode's turn-on time is the critical element.

As a preliminary matter, it's well known that standard silicon power diodes, such as the 1N400x series have a significant turn-off time, commonly called the "reverse recovery time."

The reverse recovery time involves a diode switching from forward biased (conducting) to reverse biased (non-conducting). The diode's PN junction, when conducting, has excess minority carriers and it requires a finite time for these excess minority carriers to be neutralized when the PN junction switches from forward to reverse bias. For more detail, see Standard silicon power diode, intended for use in 50/60 Hz power systems, exhibit a reverse recovery time of several microseconds, although higher performance devices are available.

Less common considered is the forward recovery time, i.e., the time it takes the diode to conduct when switched from reverse to forward bias. This mode is involved in the typical inductive clamp circuit, as can be seen from the circuit fragment below. When the relay coil is energized (the 2N7000 MOSFET is biased into saturation), diode D8 is reversed biased. When the 2N7000 MOSFET is turned off, the relay's magnetic field collapses and induces a reverse voltage in the windings, which forward biases D8. (Lenz's law says the induced voltage is reversed.)

If D8 has a slow forward recovery time, it is possible for the voltage induced by the collapsing field to rise to levels that might damage the switching MOSFET before D8 begins conducting. (This is a simplification, as the MOSFET will breakdown at some voltage level and will thus limit the voltage. The question is whether the energy deposited into the MOSFET in this event is sufficient to damage the device. We'll ignore this aspect of the analysis.)

The circut fragment below shows typical use of a diode to clamp an inductive switching transient. The time constant is approximately L/R where R is the relay coil resistance and L is the coil inductance. For an Omron G2RL-24-12DC relay, (12 volt DC coil) the coil resistance is 360 ohms. Omron does not specify the armature inductance, but based on measurements I've made for similar relays, it is in the range of 3 H when the armature is closed. L./R is thus 8 milliseconds. The relay's opening will be delayed by a time constant or two, depending on its release current performance. In addition, of course, there may be extra delay due to the mechanical structure of the relay armature.


Forward Recovery Time Test Setup

To assess the forward recovery time of typical diodes, I used the test setup illustrated below.

To see both the forward and reverse recovery time, I set the 8012B to provide a bi-polar pulse, i.e. from a negative level to positive level. The image below shows the test signal., with a peak-to-peak voltage of approximately 7.8 volts, swinging from +3.9V to -3.9V. The rise and fall times are less than 5 ns.

I installed the diodes in a home made test fixture that I use for RF network analysis. The diode is installed shunting the pulse to ground, such that a positive pulse puts the diode into forward conduction.
Test fixture with diode shunting the coaxial center conductor to ground.

The 1N4005 is a member of the venerable 1N400x family of silicon rectifiers. I don't know when they were first introduced, but I recall purchasing some in early 1970's so they have been around for 35 years plus. Of course, today's 1N400x diode is not necessarily the same as one fabricated 35 years ago. I have no idea when the 1N4005 I tested was manufactured, as it's been in my junkbox for an unknown time. (My 1971 GE Transistor/Diode handbook lists the 1N400x series, so the original introduction date must have been in the mid to late 1960's.)

Fairchild's data sheet for the 1N4005 (the datasheet is for all 1N400x series) can be viewed at It notes the 1N4005 has a reverse voltage breakdown of 500 volts, providing no data on either forward or reverse recovery time. Like all other members of the 1N400x family, the 1N4005 is rated at 1 A forward current.

The horizontal sweep time is 1 μs/div. The vertical spikes show the applied pulse transition times. Voltage above the center graticule line are positive, indicating forward bias and those below are negative, where the diode is reverse biased.

Note how fast the transition from reverse bias to essentially steady state forward conduction is. The forward bias transition is way too short to see with the 1 μs/div sweep speed. There's a slight rise in forward voltage for 200 ns or so after the diode transitions from reverse to forward bias, but it's only a tenth of a volt or so, negligible in the application being considered.

The reverse recover time, in contrast, is remarkable. Note how the 1N4005 continues in forward conduction state essentially unchanged for 400 to 500 ns after the driving waveform switches from positive to negative, with only a slight drop in voltage. After 500 ns or so, the diode begins to cease conduction, although the total time to stabilize in reverse mode approaches 3 μs.  Of course, the reverse recovery time is not of significance in an inductive clamp application.


The 1N4007 is the 1000 V reverse voltage member of the 1N400x family, with the same 1 A forward current specification.

As the illustration below shows, it's not much different than the 1N4005. The forward recovery time is very short, on the order of a few nanoseconds, with a similar small transition after switching to forward conduction.

The reverse recovery time is similar to the 1N4005.

It's commonly understood that the 1N4007 has an special junction structure resembling a PIN diode and it has, of course, been widely used as a "poor man's PIN diode" in RF power switching. Elecraft, for example, uses 1N4007 diodes for transmit/receive switching in both its 10 watt and 100 watt option K2 transceiver.

Used as a quasi-PIN diode for RF power switching, the 1N4007 takes advantage of the lengthy reverse recovery period, which prevents it from rectifying the high frequency signal it is switching. (This is because the reverse voltage half-cycle period of, say, a 3.5 MHz signal is much shorter (140 ns) than the reverse recovery time. Hence, if forward biased with DC, any reverse bias from RF negative half-cycles don't last long enough to neutralize the 1N4007's excess minority carriers. Hence, the 1N4007 stays forward biased and presents a low impedance to the full RF cycle. There's a minimum frequency for this effect which we see from the trace below. With a bias current in the 60 mA range from the 8012B pulse generator, we see the following voltage waveform.

I measured the forward and reverse current with a Tektronix P6022 clip-on current probe, set for 1 mA = 1 mV. The figure below shows the diode current.

The figure shows during the forward bias period, the current is 67.5 mA. Then, when the 8022B pulse generator reverses polarity and reverse biases the diode, we see the current approximately doubles during the recombination time. This can be considered as the sum of the recombination current in the PN junction (68 mA) plus the negative current from the pulse generator (68 mA) or a total of 136 mA. As the excess minority carriers are neutralized, the diode assumes its normal reverse bias state of being essentially non-conducting, with negligible current flow.

Note that since the probe measures current, the forward turn on voltage elevation appears as a reduced current.



Spehro Pefhany has provided a link to the forward recovery time graph presented below for the 1N400x series diodes. Unfortunately, my 68 mA forward test current is just off the horizontal scale, so we have to extrapolate a bit. Also, the magnitude of the Vf elevation during the forward recovery time is not provided.

Nonetheless, the forward recovery time graph is in general agreement with the data I measured.

Let's look at the 1N4007 in more detail with an annotated, expanded view of the transitions. The illustration is from an 1N4007, but the concept applies to all diodes.

The transition A' to A takes the pulse generator goes from negative to positive, and thus the diode from reverse bias to forward bias. The 'scope sweep shows the transition to be quite fast, as A'-A appears as an essentially vertical line. With 500 ns/div, we expect to see a transition time that is over 50 ns to show as a sloping line, so we can say that the 1N4007's turn-on time, or forward recovery time is < 50 ns. At most, we see a small elevated forward voltage effect for a couple hundred ns after the diode becomes forward conducting. The diode's forward voltage starts around 1.2 volts, dropping to the steady state level of 0.8 volts within 250 ns or so.

At time B, the pulse generator output switches from positive to negative. When negative, the 1N4007 is reverse biased and should carry no current. Accordingly, we should see on the oscilloscope the full negative pulse voltage, around -4 volts at time B. However, the 1N4007 continues in forward conduction at time B, and indeed even by time C, the diode voltage is still > 0. In other words, during the time from B to C (approximately 500 ns) the 1N4007 is still in forward conduction mode, despite the fact that the diode is reverse biased to -4 volts by the pulse generator. Rather, the diode stays in forward conduction during the time B-C as it takes time for the excess minority carriers in the 1N4007's PN junction to be neutralized by the reverse bias.

In fact, it takes until time D for the last excess minority carriers to be neutralized and the 1N4007 to become fully reverse biased with essentially no current flow. This is a full 3 μs between the time the diode is reverse biased and the time that current flow actually ceases.



The 1N4001 is the lowest voltage rated of the 1N400x family at 50 volts. Its reverse recovery switching time behavior is similar to other members of the family, as evidenced in the curve below.

However, the 1N4001's forward switching shows considerably less of the increased forward voltage effect seen in the 1N4005 and 1N4007 devices.

In fact, if we expand the sweep to 20 ns/div and look at the 1N4001's forward turn-on time, we see two points of interest:
  • The turn-on time is about 4 ns, which represents the 2246's bandwidth (100 MHz) and the rise time of the 8012B pulse generator (less than 5 ns). The 1N4001's turn-on time is thus less than the test equipment used in my test setup.
  • The excess forward voltage, Vf, is essentially zero. The overshoot and ringing shown in the 20 ns/div sweep is likely an artifact of my test setup and cable ringing.
HER 108

The HER 108 is a member of the HER 10x family, similar to the 1N400x group, with the part number's last digit increasing with increasing voltage. The HER 10x diodes are all rated at 1.0 A forward current, and the HEr 108 is rated at 1000 V reverse voltage. For more detail, see Rectron's data sheet for the family at  .

However, the HER 10x family is designed to address the reverse recovery problem seen in the 1N400x and other similar run-of-the-mill silicon power rectifiers, with a maximum reverse recovery time of 50 ns (for the HER 101-105 devices, through 400 volts) and 75 ns for the HER 106 - 108 parts (600 - 1000 volts).  No forward recovery time is defined in Rectron's data sheet.

The illustration below confirms the major improvement in reverse recovery time compared with the 1N400x diodes.

The forward recovery time is still short, on the order of a few nanoseconds, but clear signs of different behavior are seen. The initial conduction voltage is around 2 volts and it does not drop to the steady state value of 1 volt for about 500 ns. Both the voltage elevation and response time in this regard are worse than the 1N400x devices. However, for diode clamping, the HER 108 seems adequate. Of course, there's no real reason to pay for the HER 108's improved reverse recovery time in this application.



The 11DQ05 is a 1.1 A, 50 volt Schottky diode and more details may be found at

We expect a Schottky diode to have fast forward and reverse recovery times, and the 11DQ05 does not disappoint.

To better see the 11DQ05's performance, I've increased the sweep speed to 0.5 μs/div in the image below. As the illustration shows, both the forward and reverse recovery times are not easily measurable with this sweep speed, and are on the order of a few nanoseconds.

Also, the Schottky's lower forward bias voltage drop shows clearly, with about 0.4 volts across the diode when forward biased.


Recovery Time Conclusion:

There's a clear difference in reverse recovery time amongst the diodes examined, with the old 1N400x series having by far the worst reverse recovery time. The HER 108 improved silicon diode provides a much better reverse recovery time, and the 11DQ05 Schottky is even better.

However, for a diode to be used as an inductive snubber, the evidence is not as clear. All of the diodes transition from reverse to forward bias with great rapidity. At worse, the HER 108 has a period of higher than normal forward voltage, but the elevated voltage is not that great when considered in the context of an inductive snubber.

Snubber Measurements Introduction

To one and for all put to rest the myth that a silicon power diode such as the 1N4007 cannot be used as a relay snubber due to its recovery time, I set up a simple test fixture, illustrated below.

The HP 8904A multi-synthesizer is configured to provide a 10 Hz square wave (0→5V) driving a 2N7000 MOSFET  transistor to turn a  typical relay on and off. DC power is provided by an Agilent E3631A power supply, set for 13V output. A Tektronix TDS430A digital oscilloscope (400 MHz) is used to view the circuit's waveforms. The 2N7000 thus turns the relay on for 50 ms and off for 50 ms.

Two measurement setups are used:

  • Voltage spike�Channel 1 to test point 1; channel 2 to test point 2A.
  • Relay operate time�Channel 1 to test point 1; channel 2 to test point 2B.
Snubbing Diode Test Setup
The relay used is typical of small footprint relays found in amateur radio gear. It is a Fujitsu FTR-F1CD-012V part, with a data sheet available (as of April 2009) at

The key specifications of the relay are 12V nominal operation (minimum pull-in voltage 9 volts; release voltage 1.2 volts), DPDT ("C-form") contacts rated at 5A, 24 VDC. Operate time is 15 ms and release time (no diode) is 5 ms. DC resistance 360 ohms.

The data sheet does not specify the relay coil inductance, so I measured it with a General Radio GR-1650A RLC bridge, both in the un-energized and energized mode. (The GR-1650A and B bridges are quite valuable for measuring relays and capacitors with DC bias, a feature not always available with lower end digital instruments.)

I measured:

Parameter Value Computed Stored Energy @ 13V, 38 mA
DC Resistance 341 ohms  
Inductance @ 1000 Hz, un-energized 570 mH 828 x 10-6 joules
Inductance @1000 Hz, with DC bias applied to pull in relay 680 mH 982 x 10-6 joules


A relay, of course, has different inductance when energized and non-energized for two reasons. One is mechanical; the magnetic circuit no longer has an air gap, or if an air gap is designed into the relay, the gap distance is reduced. This will increase the inductance. Second, the DC bias will shift the B-H point around which the incremental inductance is measured. In general the B-H curve shift will usually decrease the inductance.

It would be possible to separate these two effects by disassembling the relay and mechanically positioning the armature to simulate energized and non-energized positions whilst measuring the inductance. I'll leave that as an exercise to the interested reader.

An inductor, including the relay coil, stores energy in its magnetic field, with the stored energy (in joules or watt-seconds) equal to:

 E_\mathrm{stored} = {1 \over 2} L I^2

I've computed the stored energy in the above table for the current the relay consumes at 13V. It's roughly 1 milli-joule when energized.

This stored energy must "go someplace" when  the relay is de-energized. The collapsing magnetic field resulting from opening the circuit when the 2N7000 goes into a high impedance state, causes a voltage to be induced in the windings opposite in polarity to the energizing field. (Lenz's law.)  From elementary circuit theory, we know:

E = Ldi/dt

where E is the induced voltage, L is the inductance and di/dt is the time rate of change of the current. When a fast device, such as a switching transistor, opens the relay coil circuit in a few hundred nanoseconds, di/dt becomes quite large.

In the absence of a snubbing diode as shown in the illustration, the collapsing field can generate a significant voltage, sufficient to easily cause the 2N7000 to break down. (The 2N7000's rated maximum drain-to-source voltage is 60 volts.) This completes the circuit and allows the reverse current flow. Or, if a snubbing diode is present, the diode goes into forward conduction and allows reverse current to flow through the relay coil. In either case, the energy stored in the coil's magnetic field is dissipated as heat in the relay's windings.

This raises an interesting question�what would happen if we had a perfect switch. Would the collapsing field reach an infinite voltage? The short answer is no. Our look at the circuit is simplistic, and in fact the relay windings have distributed capacitance so that an AC current flow can occur even with one end of the inductor dangling in space, disconnected from the rest of the circuit. What happens is a dampened oscillation occurs (RLC circuit) with a resonant frequency determined by the inductance, distributed capacitance and resistance of the relay coil, with the stored energy dissipated in the coil  resistance over some number of cycles. Although not infinite, it's easy to generate several hundred volts in this fashion.

Voltage Measurements with and without Snubber

First, what happens without a snubbing diode? Channel 2 (blue) shows the voltage at the 2N7000's drain; it reaches 68.8 volts and starts conducting. Since the 2N7000's maximum rated drain-source voltage is 60 volts, seeing a breakdown at 68.8 volts isn't surprising. It is worth noting that the 2N7000 isn't damaged by this breakdown, or at least the damage is not apparent in this rather non-critical application. The peak current through the 2N7000 is limited by the relay's 341 ohm resistance, which for 68.8 volts corresponds to 200 mA.

Next, we add a 1N4148 snubbing diode. The diode limits the peak voltage across the 2N7000 to 14.4 volts.

Substituting a 1N4007 silicon power diode, we see the same peak voltage, 14.4 volts, across the 2N7000.

Let's look at these waveforms with a faster sweep to see if perhaps the 1N4007 is allowing a glitch or spike  too narrow to see with the  500 μs/div of these oscilloscope captures. We should see a more accurate voltage reading as well, as whatever happens is rather rapid and 500 μs/div is too slow to accurately capture the transition.

First, without a snubbing diode, at 50 ns/division. We see the voltage rises from 0 to 74 volts in a bit less than 300 ns.

With a 1N4148 snubbing diode, the peak voltage is 13.8 volts and we also see a damped oscillation with a period of 100 ns, or 10 MHz. If the relay coil inductance is 570 mH at 10 MHz (which is doubtful at best), for 10 MHz resonance, the distributed capacitance of the winding is 4.5 x10-16 F. This is clearly an absurd value for a relay coil's distributed capacitance, so it's likely that the ringing is due to the oscilloscope probe's long ground lead. (I've written about this before.)

With the 1N4007 diode, the waveform is quite similar to that seen with the 1N4148.

From these measurements, which support the earlier turn-on time data, it is time to put to rest the mistaken view that a 1N4007 diode cannot be used to snub a relay.

And,  there may well be times when a 1N4007's higher current rating is necessary to handle a particular relay. The small signal relays used inside receivers and transceivers generally have sufficient coil resistance to limit the voltage spike current to a few hundred milliamperes, but one should consult the relay data sheet before concluding that a 1N4148 will work within its ratings.

Why Use a Snubber?

It's not a good idea to allow the relay's Ldi/dt voltage to cause the switching transistor to break down. Dave Gilbert, AB7E, excellently summarized the reasons why you should rely upon the switching transistor to handle the relay inductive spike switching transient in a post on the Elecraft reflector. Dave has allowed me to quote his posting:

I worked in the discrete semiconductor industry for 30 years (engineering, manufacturing, management) and that statement [that the switching device breakdown is an acceptable spike limiting methodology] is simply not true at all.

Zener diodes are designed to break down uniformly across the entire junction, but most other semiconductors are not.  A reverse voltage breakdown of the collector-base junction will in most cases force current through an extremely small region of the junction (often referred to as a "puncture") and fuse the silicon at that point, rendering the device useless.  The current required to do so is much less than would be required to heat the device even a few degrees, and the time required to destroy the device is short.  Junctions with steeper diffusion gradients (RF devices, switching transistors, etc) will fail more easily (sometimes in microseconds), while sloppier junctions (power devices, etc) will take considerably more abuse.  It is not possible to predict the current at which catastrophic failure will occur.

Emitter-base junctions are typically more graded and won't fail catastrophically as quickly, but repeated reverse bias conditions will degrade the transfer gain of the device by creating defect centers that kill the carrier lifetimes in the base region.

The inductive kick from even a small relay is sufficient to puncture the junction of many commonly used transistors or driver ICs, and there are reams of failure analysis reports documenting that fact.  There isn't a manufacturer on this earth that will honor the warranty for a device used as you describe unless the device was specifically designed to survive it.

 Dave   AB7E


SPICE Simulation

Simulating the circuit with LTspice IV, without a snubbing diode shows the maximum voltage across the 2N7000 is over 500 volts. This simulation indicates that the 2N7000 is not well modeled when it comes to breakdown.

Adding a 1N4148 diode to the simulation shows a response time quite a bit faster than the measured data. In part this is likely due to the rather slow transition of the HP 8904A synthesizer which causes the 2N7000 to operate in the linear  region for a few tens of nanoseconds.

Changing the simulation to use a 1N4007 snubbing diode shows virtually identical response.

Release Time Change

Adding a diode will slow the relay's release/operate  time as the current decay period is longer. It's not possible to show the relay switching time without a diode because the 2N7000 breaks down and acts as a 70 some volt Zener diode. (Not a good thing for the 2N7000, by the way.) We can measure the times with the three combinations, however, and compare them with the specification, which is 5 ms in the absence of a diode. (Yes, I could use a high voltage MOSFET, or, for that matter, a mercury wetted relay to switch the relay coil. I'll also leave that as an exercise for the interested reader.)

In the plots below, channel 1 is the 2N7000 gate voltage and channel 2 is the normally closed contact voltage. (+13 volts is applied to the common contact.) The time differential between the gate voltage going to zero and  the relay voltage going to +13V represents the relay's release time.

With no explicit snubber diode (leaving just the 2N7000's breakdown Zener effect) the release time is 2.68 milliseconds, but note the contact bounce does not stabilize completely until 8 ms. (The spike at switch  time is due to the inductive voltage from the coil being coupled back into the oscilloscope probe. The rest of the noise is contact bounce.)

With a 1N4148 snubbing diode, the release time increases to 11.2 ms measured to the first contact point, or, if allowance for contact  bounce is made, around 18 ms. (The time axis is slowed on the next two plots to 5 ms/div.)

With a 1N4007 snubbing diode,  the release time is identical. The difference in contact bounce is not significant in that bounce varies a bit with each switching cycle.
Release Time Conclusion

Adding a snubbing diode slows the relay's release time. There are techniques that reduce the voltage spike with less effect on switching time but I'll leave that discussion for another time.

RC Snubbing

Before solid state diodes, inductive loads were snubbed with RC networks. It's still a useful technique and worth exploring it allows some control over release time, albeit at the price of a higher voltage spike. (A Zener diode can be used in a similar fashion, of course.)

The schematic fragment below shows the RC network I designed to snub the Fujitsu FTR-F1CD-012V relay.

RC Snubber Network for Fujitsu FTR-F1CD-012V Relay

How did I select these particular values? There are several approaches to designing a snubbing network, and I treated it as a Zobel network. Wikipedia has an excellent discussion on Zobel networks, which I will not repeat here.

If you are not familiar with a Zobel network, it is one method of canceling reactance and converting a reactive network to a purely resistive network. The voltage at the junction of L1-R1 equals the voltage at the junction of R3-C1 for any frequency. It can be shown that the result is a purely resistive network when viewed at the network terminals.

The relevant equations for the above L-R network are:

R3 = R1

C1 = L1/R12

R3 is simple enough; it equals the relay coil resistance. The nearest 5% standard value, 330 ohms, can be used for our purposes.

C1 = 0.68 / (341 x 341) = 5.8 x 10-6F or 5.8 μF. The nearest standard value is 5.6 μF.

Voltage Rise with RC Snubber

Let's see how the RC snubber does in the test network. My junkbox did not yield a 5.6 μF capacitor, so I tried 1, 4.7, 6.8 and 10 μF parts. These should  be non-polarized capacitors, but only the 1 μF I used was non-polarized. I'll discuss capacitor selection later on this page.

1 μF & 330 ohm Network

4.7 μF & 330 ohm Network
6.8 μF & 330 ohm Network
10 μF & 330 ohm Network
Although based on a small sample of values, as we move away from the computed value,  the peak voltage across the 2N7000's drain increases. This is expected as the network becomes reactive (either inductive or capacitive) as C1 departs from the correct value 5.8 μF. (Complicating things, the relay's inductance changes as the armature moves, so C1 has no single "correct" value. Fortunately, this relay presents only a small inductance change between engaged and released armatures.)

The 4.7 and 6.8 μF parts bracket the computed 5.8 μF value and show the lowest voltage rise, 13.2 volts. Based on voltage rise considerations, any value between 4.7 and 10 μF are acceptable. Even a 1μF capacitor keeps the 2N7000 drain voltage maximum to around 50% of its rated value.

How does  the RC network alter the relay's release time?

As seen with the diode snubber, there's a price to be paid for snubbing the inductive spike voltage; slower release time. The same thing is seen with an RC snubbing network. The test setup is the same as used in the diode measurements.

1 μF & 330 ohm Network

4.7 μF & 330 ohm Network
6.8 μF & 330 ohm Network
10 μF & 330 ohm Network
The 1N4148 diode snubber release time is 11.2 ms. With the close-to-optimum 4.7 and 6.8 μF capacitor, 330 ohm  resistive snubber network, the release time is 9.6 and 10.3 ms, respectively, a modest improvement at best. If we allow the 2N7000 switching MOSFET's drain voltage to rise to 30 volts (still only 50% of its maximum 60 volt rating) by using a 1μF capacitor, the release time improves to 4 ms.

Of course, relay contact bounce adds additional waiting time in some instances, depending on the load the relay is switching, but 4 ms is a marked improvement over 11 ms with a diode snubber.


Voltage Across the Snubbing Capacitor

Let's look at the voltage across the snubbing capacitor during  the switching interval. I've attached the oscilloscope probes to points [1] and [2] of the snubbing capacitor. Channel 1 is connected to [1] and is the black trace in the images. Channel 2 is connected to [2] and is the blue trace. I've set both channels to the same gain and the same zero reference point.

Oscilloscope connections are to [1] and [2]. Channel 1 to [1] and Channel 2 to [2]

4.7 μF & 330 ohm Network

 I've annotated the oscilloscope capture to identify the polarity reversal area. When the 2N7000 is pulled low and the relay is enabled and for about 5 ms after the 2N7000 goes high, C1's terminal [1] is positive with respect to terminal [2] by 13 volts (during the relay operate period) and a lesser amount after the 2N7000 goes high. The polarity then reverses, with terminal [1] becoming, at the peak point, around 4 volts negative with respect to terminal [2]. The polarity reversal continues for 15 milliseconds or so. The reversal occurs because L1 is sourcing current into terminal [2]. (I use the positive current convention in this discussion.)

1 μF & 330 ohm Network

With the sub-optimum 1 μF capacitor, we see a much greater reverse voltage, 15 volts or so, compared with the 5 volts reverse polarity seen in the close-to-design-value 4.7 μF capacitor. 

6.8 μF & 330 ohm Network

With the second close-to-optimum 6.8μF capacitor, the reverse polarity area is roughly similar in duration and magnitude. (Note the change in horizontal axis sweep time in the next two plots.) The worst case reverse voltage is around 4 volts.

10 μF & 330 ohm Network

With 10 μF, the reverse voltage magnitude is even smaller, around 2 volts. As the capacitance increases, the  reverse voltage magnitude will decrease, of course. One can think of this as the capacitor being able to absorb more current from the inductor without a corresponding voltage rise. (remember V=q/C where q is charge, C is the capacitance and V is the voltage. When combined with i = dq/dt where i is the current and dq/dt is the time rate of change of the charge, it should be apparent that as C increases, more charge (and hence more input current) can be held with the same voltage across the capacitor.)

To avoid capacitor failure, therefore, it's necessary to employ a non-polarized capacitor in the snubbing network. This may be accomplished by using a non-polarized capacitor or by other methods such as back-to-back polarized capacitors with parallel diodes to prevent reverse voltage. However, that's a lot of extra effort to go to in order to avoid buying a non-polarized part.


RC Snubber SPICE Simulation

Properly done, computer simulation using SPICE can be amazingly accurate. As an example, the plot below shows the voltage at points [1] =V(junction) and [2]=V(drain) with a 330 ohm / 1 uF RC snubbing network. It compares quite closely with the measured data presented earlier. The drain peak voltage, for example measured at 29.6 volts [channel 2 in the oscilloscope capture] compared with 28.1 SPICE predicted. Likewise the peak voltage at point [1] [channel 1 in the oscilloscope capture] measured 23.8 volts peak, compared with a SPICE predicted peak of 24.9 volts. 

The main difference is  the time scale; the SPICE simulated drain voltage oscillation has a period of 5.5 ms, whilst the measured period is closer to 10 ms, nearly twice as long. I'm not sure what's behind  this divergence.

The SPICE simulation holds the inductance and capacitance constant. However, the relay coil inductance is not a constant; we know it  varies from 680 mH to 560 mH as the armature releases. In addition, the 1.0 uF snubbing capacitor is not constant either. The part I used is a high dielectric constant capacitor normally used for bypassing. As can be seen at these parts possess a significant C versus V variation. SPICE permits modeling both non-constant inductance and  capacitance, but I have not  taken the simulation to that step.

SPICE Simulation 330 Ohm R and 1 uF C.