Myth Busters 3 – Myth: “You must have a gate resistor”

Hi Nick,

I'm not saying that limiting IGS is not sometimes a good thing with some MOSFETS. After all, there does have to be a limit. It only makes sense.

I was more trying to get a spec jockey to come up with a reason for their statement given that I have never seen a spec for absolute maximum gate current in a MOSFET datasheet. The person in question has, at least indirectly, admitted that they will only believe manufacturers datasheets. I just find it odd that that same person is continually commenting on others findings for parameters that are not in a datasheet. I wonder where they get their information.

BillO:
whatever information you do have is based on what? Speculation? Hearsay?

You're the one basing his argument on some particular chip samples and a whole lot of speculation.

I prefer to rely on the written information provided by the people who designed/made the chip. We've already seen that they do change their production processes every now and again. Maybe the "250mA" figure that you so eagerly dismiss was for a chip from an older production process.

fungus:
... some particular chip samples and a whole lot of speculation.

Actually doing tests is speculation? Hmmm...

Maybe the "250mA" figure that you so eagerly dismiss was for a chip from an older production process.

  1. Where did that value come from and does it come with other parameters, like time?

  2. How is that pertinent today?

  3. I eagerly dismiss it because I did not find it in actual tests

  4. How do you deal with parameters that are not in datasheets?

BillO:

fungus:
... some particular chip samples and a whole lot of speculation.

Actually doing tests is speculation? Hmmm...

[/quote]

The datasheet clearly says that pin/chip damage may occur above 40mA (and that 40mA is a stress rating only and not recommended for long term use).

The famous "pin output voltage drop vs. current" graphs only go up to 20mA. Beyond that, it's 100% pure speculation, yes.

BillO:
4) How do you deal with parameters that are not in datasheets?

Maybe they're left out for a reason, ie. to allow them to change their production processes without breaking the designs of people who follow the datasheet.

What you're doing is equivalent to using undocumented APIs in an operating system. Your experimental data may show that the functions work perfectly, but there's a reason they're left undocumented.

PS: Why aren't you experimenting with running it at 9V? That's just as valid, right...?

fungus:
The famous "pin output voltage drop vs. current" graphs only go up to 20mA. Beyond that, it's 100% pure speculation, yes.

They used to go to 40mA in the early (pre-328) days. BTW, do you understand what these graphs tell you? Have ever taken the time to do some analysis? The physics behind these curves is real, well understood and predictable, not speculatory. I will be presenting more on this in a few days but I do not expect you will understand it more then than you do now.

Maybe they're left out for a reason, ie. to allow them to change their production processes without breaking the designs of people who follow the datasheet.

What you're doing is equivalent to using undocumented APIs in an operating system. Your experimental data may show that the functions work perfectly, but there's a reason they're left undocumented.

While this does seem to be a well thought out answer to a question, it's not the answer to the question I asked. But thanks anyway.

PS: Why aren't you experimenting with running it at 9V? That's just as valid, right...?

Now who's speculating? Who said I haven't? In any case, over voltage and over current are two entirely different beasts. I'm not sure every one knows what happens when you use too high a voltage on semi-conductor devises, but I do and it's an experiment I actually had to do many years ago when I worked for the University of Toronto. I have no need to repeat it at this time.

fungus:
The datasheet clearly says that pin/chip damage may occur above 40mA (and that 40mA is a stress rating only and not recommended for long term use).

What I want to know and want what the data sheet does not specifically tell us includes, but is not limited to, the following:

  1. How long 40mA is safe for? (They tell use it's not safe for a long time, but how long is that, specifically. 1s, 10s, 1ms, 1us?. Most device datasheets give this kind of information. It is useful information.)

  2. They do not tell us what the recommended long term maximum output current is. (Is it 15mA, 20mA, 25mA, 30mA, 35mA)?

  3. Rise and fall times of all outputs and under what test conditions.

  4. Inherent internal capacitance of each pin under the various possible configurations and under what test conditions.

...

In fact, the datasheet is splendidly uncluttered with many specifications of interest to the applications these device are put. I note that the specifications in these areas supplied by Microchip are better. Not great, but definitely better.

BillO:
What I want to know and want what the data sheet does not specifically tell us includes, but is not limited to, the following:

  1. How long 40mA is safe for? (They tell use it's not safe for a long time, but how long is that, specifically. 1s, 10s, 1ms, 1us?. Most device datasheets give this kind of information. It is useful information.)

  2. They do not tell us what the recommended long term maximum output current is. (Is it 15mA, 20mA, 25mA, 30mA, 35mA)?

Sure we'd like to know but..... it is impossible for them to give these informations. They would do the tests at 25°C, which is not really the every day T° .... and users wouldn't notice that, they would only see the values they are interested in...you can easily imagine the consequences :wink:
When they give safe values, even if the test conditions were 25°C ... , they know that +/- 15° won't destroy the chip, which is not true when approaching the limits :wink:

@BillO: you have to understand the atmega328p is a cheap mcu for general use. You will not get any more information than you may see in the datasheet. And you have to accept the messages there.

If you need devices for heavy duty applications, where the chip maker will share with you the parameters you want to know (and will provide all possible measurements and tests for you), just ask for military or space grade mcus (IBM, Sandia Labs, etc.). FYI - such a device costs $50k - $200k a piece.

@pito and alnath

Well, that's why I do these tests.

  1. How long 40mA is safe for?

It is not, the term absolute means absolute. You can read, it says 40mA is a stress rating only.

Most device datasheets give this kind of information.

No they don't. The ones that do are for components designed to take high pulsed currents.

They do not tell us what the recommended long term maximum output current is.

Yes they do, all the ratings are for 20mA.

Rise and fall times of all outputs and under what test conditions

Your right they don't say that. Do you understand why?
You are very keen on telling other people they will not understand stuff, have a word with yourself.

Inherent internal capacitance of each pin under the various possible configurations and under what test conditions.

Please try and live in the real world.

I note that the specifications in these areas supplied by Microchip are better.

So use their chips and stop beefing.

I think nobody in this discussion has something against your tests. I am also happy to know the 328p survived shorts and that it draws 88mA when shorted. I would guess the chip could work at 8Volts Vcc and 130degC :slight_smile:

You asked primarily whether the gate resistor is necessary when driving a power mosfets from the atmega328p chip. The answer from people with experience is Yes, because:

  1. it limits the inrush current into the mosfet's gate to the 20mA output limit according to the atmega328p datasheet
  2. it blocks gate ringing with poor wiring
  3. it protects atmega328p when the mosfet's gate shorts.

PS: In addition, I always use a pull-down resistor from the gate to ground. When the atmegas output pin floats (ie. after reset or because of a sw bug) the mosfet might be switching the load erratically, or the mosfet could overheat itself when gate floats in linear region).

power mosfet from arduino.jpg

It has.. @BillO, been my Experience that the data sheets are right more often that the interpretations of them are.
Your quest for absolute numbers smacks of OCD. Further there is a large body of evidence that exceeding the extremely loose specifications will lead to grief, Are you attempting to delineate grief?
I for one choose not to heat up a processor any more than required so I keep the port loading down to the median of what the port is specified to supply. This is because good design is NOT a top down process only. The tasks assigned to the processor must be evaluated from both ends. Can the port and code provide the correct waveforms and what is the total impact of the actions controlled by the port on both the port and the controlled circuit?.
This consideration led me to the gate resistor to fix an unreasonable load on the power supply. analysis found the clock on the processor stopped and it required a reset to restart the clock. The circuitry involved including the processor had a quiescent current of 1.95 mA and for a few microseconds there was a short circuit on that port. Limiting the current to ~50 times that current or (5V/47R = 106 mA vs 1.95 mA..), allowed the 1uF cap that was the Vcc pin bypass to supply the current and not through the loss of the series inductance of the supply trace. This was proven when I increased the cap to 10 uF and had the failure rate go from it's initial 1 in 1000 to 1 in 5000. An increase in the resistor size caused the Fet to stay in the ohmic region for too long thus limiting the available current from a 4700uF capacitor used to control the solenoid and charged to 12V DC. 47 Ohms was my initial choice because that was the first resistor I found 'in the ballpark' and close to hand. Values to 470 ohms were evaluated and the 47 ohm resistor was found to be appropriate. Basically it worked and I could no longer see the spike on the Vcc supply at the controller Vcc pin. Testing proved the evaluation in that the device worked far beyond it's design specification. Similar testing was applied to 10 units total and no further failures were noted. End of story... This was the first time I had used a mosfet to control a heavy inductive load especially one with a widely varying inductance (bobbin out vs bobbin in) and with 300 meters of wire in series with it. The design worked well and I kept the solenoid activated for 100 mS after the initial pulse to control the Back Emf pulse by dumping it (what didn't get sloughed off by the snubber) back into the capacitor used as the current source for the load.
In closing I have to quote Nick Gammon:

I agree it's important to test things, but even then to be aware of possible errors in your test procedures. That's why I am happy if my tests can be reproduced. And even happier if they are backed up by the theory behind the test.

.
All of my initial testing was very carefully evaluated in the testing of the other 9 devices, all were within acceptable norms.

Doc

I will also need to re-do my "short circuit" test.

I think you would need to repeat the above general idea but have a short instead of the MOSFET. Then the scope trace (on the triggered point) should show you the instantaneous current during the "short" cycle.

I don't really want to make predictions, but my guess is that you might find it exceeds 88 mA for some nanoseconds and then drops back to 88 mA.

My guess is based on the fact that we have both measured more like 440 mA into the MOSFET for a short time. So it is reasonable to suppose we would see a similar if not greater amount into a short. How long before that drops back to (around) 88 mA would need to be experimentally verified.

Grumpy_Mike:
It is not, the term absolute means absolute. You can read, it says 40mA is a stress rating only.

Yeah, I think I can read. So, then 38mA should be fine, right?

No they don't. The ones that do are for components designed to take high pulsed currents.

OK. If you say so.

Yes they do, all the ratings are for 20mA.

Well, I'm not going to accuse the famous Grumpy_Mike of not being able to read a datasheet. Far be it from me. However, I'd like to point out that they are operating conditions for the specific specification quote, plus they go on to say:

Although each I/O port can sink more than the test conditions (20mA at VCC = 5V, 10mA at VCC = 3V) under steady state conditions (non-transient), the following must be observed:
ATmega48A/PA/88A/PA/168A/PA/328/P:
1] The sum of all IOL, for ports C0 - C5, ADC7, ADC6 should not exceed 100mA.
2] The sum of all IOL, for ports B0 - B5, D5 - D7, XTAL1, XTAL2 should not exceed 100mA.
3] The sum of all IOL, for ports D0 - D4, RESET should not exceed 100mA.

Your right they don't say that. Do you understand why?

No, illuminate please.

Please try and live in the real world.

There is no need to be a horses behind. I fail to see where this might a difficult thing for them to test and include in section 31.

So use their chips and stop beefing.

I do and I'm not. Just stating some observations. Oh, but that's right, I remember now. I'm not allowed my own perspective on things.

[quote author=Nick Gammon link=topic=176968.msg1316025#msg1316025 date=1373836831]
I think you would need to repeat the above general idea but have a short instead of the MOSFET. Then the scope trace (on the triggered point) should show you the instantaneous current during the "short" cycle.[/quote]
That's the plan.

I don't really want to make predictions, but my guess is that you might find it exceeds 88 mA for some nanoseconds and then drops back to 88 mA.

I agree. That's what I'm expecting too.

My guess is based on the fact that we have both measured more like 440 mA into the MOSFET for a short time. So it is reasonable to suppose we would see a similar if not greater amount into a short. How long before that drops back to (around) 88 mA would need to be experimentally verified.

My initial glance at the "cleaned up" test rig (no breadboard, shorter wiring, proper grounding, ...) has promised much lower current than the 440, but to be honest, I did not get more than a few minutes at it before I had to bail.

(See, time matters!)

Just tested a short of the pin to ground (no MOSFET) through a 0.47Ohm resistor using the PWM sketch from the early posts.

Results:

Steady State voltage: 40mV
Peak voltage: 150mV

So Steady State current sourced through resistor is:
I = 0.04/0.47 = 85mA, this is pretty much dead on the 88mA short circuit current I have measured in the past. A second test I did yielded 42mV at steady state which equates to 89mA, so that's a great sign.

And the peak current:
I = 0.15/0.47 = 319mA which is quite a lot! >_<

Docedison:
@BillO, been my Experience that the data sheets are right more often that the interpretations of them are.

Agreed. I've got no real problem with what's there (with one exception. There is a big error in the datasheet, but it's easy to see, so I ignore it). It's what is not there.

Your quest for absolute numbers smacks of OCD.

Well, that's a little unfair.

Really, I just want to know. It's one of the ways I get my enjoyment from this hobby, I do design and build projects, and I actually do keep my designs to "within spec". However, sometimes I just want to investigate something. I find it mildly entertaining though, how others can get so upset by it. Blasphemy, I tell you! Burn him at the stake!

Docedison:
Your quest for absolute numbers smacks of OCD.

I would like to keep this thread on the topic of whether or not the gate resistor is required, and if so, what value. Please avoid personal insults.

And the peak current:
I = 0.15/0.47 = 319mA which is quite a lot! >_<

These measurements appear to be in the ball-park of earlier observations. What was the length of that peak pulse?

Forgot to measure the period, so repeated the test and zoomed right in.

This time the spike was around 135mV, though over several periods it varied, with several being around 145mV, another being 150mV as before. The transient period lasted around 40nS, with the first and largest spike lasting around 8nS

And the peak current: I = 0.15/0.47 = 319mA which is quite a lot! >_<

Why do those transients look so much like the stuff that the guys in the 2 youtube videos mentioned earlier said one needs to be extremely careful to determine whether they're:

(a) from your setup, or
(b) from your scope and probe, or
(c) extraneous pickup, or
(d) are actual valid measurements on the system being measured?

Why is there so much multi-cycle ringing in the last picture especially, right at the start of the pulse response? That's the sort of thing the guy in the first video immediately pointed a finger at, and said "hmmmm...".

(a), (b), (c), or (d)?

And why is there so much noise [ringing] at exactly the same frequency as the major ringing present even before the response begins?