FET gate inductance / Crss and oscillation

[BrianHG]

At 200ns/div, we are getting into the noise introduced through the limitations of your layout, the bread-board & wiring.  Something which would barely be there with a well designed PCB.

So to design this for a PCB (ie choose the correct value for gate resistor), do I basically have to know the inductance of the gate trace?  Which I can know from one of those PCB calculator tools?  And then that's a ballpark and I would have to try different resistor values to really dial it in?

Or is it the case for a circuit like this, where rise time isn't critical, that a ballpark figure is good enough and I just add (say) 5% to account for uncertainty in the inductance value and resistor tolerance?

For a PCB, I would put a series resistor for the gate in case you do want to soften the drive, but, already looking at the vast improvement in your new scope output, you may just put in an anywhere from 0 ohm to 2k as you see fit.  As for the LED, what length of wiring will you be using?  You might ass a cap at the led if you don't need switch speed in the Mhz, or even Khz.  This would pretty much nullify any remaining signal bounce by another 3 or more fold, so long as you PCB has proper power supply decoupling caps.  At this point, such small bounce would only be of interest in analog applications, or, extra low EMI design considerations.

[MagicSmoker]

10uf film bypass cap on drain
no gate resistor

the crazy ringing is gone!  Still some oscillation but this looks much better.

With the bypass cap the drain voltage never (1s) quite reaches the steady state of 2.0V.  Compared to without the bypass cap it got there in about 2ms.  I suppose it would get there faster with smaller bypass cap but at some point I would start to introduce the wild ringing again.

Wait, did you wire the capacitor across the power supply near the MOSFET, or to the drain terminal (and, presumably, ground/source) of the MOSFET? I wanted you to do the former, and while the capacitor in the latter position will reduce turn-off ringing, it is also very hard on the MOSFET since every time the MOSFET turns on it effectively shorts out the capacitor.

3.3V is indeed the "everything" voltage.  Why should I prefer a BJT in that case?

AIUI BJT switch will consume more power (current) as I will need some Ib to keep the transistor on.
...
As long as I am operating in the linear/ohmic region of the MOSFET (which I am; Vgs-Vth > Vds), I would have thought that I am good.  What am I missing here?

Yes, a BJT requires continuous base current to remain on, but since you are switching ~20mA max (actually, not even that much - see below) and the typical current gain of (Beta) of a small signal BJT is around 100, this means you'll need about 200uA of base current or so.

The reason why a BJT is preferred here is because it will turn fully on with just 0.6V applied to its base, while the MOSFET's gate threshold voltage, Vg[th], is usually in the range of 1.5-3.0V, and it will behave more as a resistor than a switch until at least 5V is applied to its gate. You might think that is a bonus in your application as it theoretically saves the need for a current limiting resistor for the LED, but another problem is that Vg[th] varies quite a bit from device to device and, worse, with temperature. Drain-source channel resistance is also a square law function of gate voltage, so when operating near the threshold region there will be huge changes in LED brightness for very small changes in either Vg[th] or applied gate voltage.

[T3sl4co1l]

3.3V is indeed the "everything" voltage.  Why should I prefer a BJT in that case?

It's not enough to saturate "logic level" MOSFETs that are rated Vds > 20V or so.

Lower voltage MOSFETs have higher gain, with 6-10V parts being 1.8V logic level compatible; some are available in battery-management-scale parts with Rds(on) < 1mohm!

Gain quickly drops to ordinary levels as you go up in voltage.  Hence there's no need to go over about 10V drive, for MOSFETs rated 30V to 2500V+.

"Logic level" MOSFETs, at regular voltages, are just regular devices modified for lower Vgs(th).  Their performance is marginal at 5V, because Rds(on) isn't quite saturated at that level, and the lower gate voltage takes longer to cross the switching threshold.  It's good enough for simple things like relays, but I don't recommend it for power switching applications.

AIUI BJT switch will consume more power (current) as I will need some Ib to keep the transistor on.  Whereas with the MOSFET the transistor doesn't consume/dissipate power during the on-time, only during the transition off-on and on-off.  But in that understanding I'm not considering the voltages of the switch device vs the load.

As long as I am operating in the linear/ohmic region of the MOSFET (which I am; Vgs-Vth > Vds), I would have thought that I am good.  What am I missing here?

Look at it by overall efficiency: you're spending < 1/10th the current, and 0.7V, whereas the load might be dropping well over 7V.  That's over 40dB power gain already!

BJTs get less attractive if the load is externally powered, and internal power is limited (e.g., a battery powered device with contact closure output), or if the load current varies and isn't always at high current (in which case you're wasting all that base current for no reason, most of the time).

There are some ways to arrange a circuit to deal with this, for example drawing parasite power from the load during the off state (a few uA will be nothing but leakage for most applications, but could be sufficient to keep the battery topped up), or controlling base current to keep collector voltage just barely saturated (the Darlington is a simple arrangement of this, but Vce(sat) is rather high; more creative arrangements can achieve better saturation, and with less current consumption and more speed).

BJTs start looking particularly attractive at low voltages, particularly in applications that can spare the base current.  Battery powered switching regulators are a good example: if it's always operating at design current output, then efficiency will be high -- you don't much care about base driving power exactly, as long as overall efficiency is good.  Switch outputs from non-micropower, low voltage ( <= 3.3V) logic is also a good candidate: the BJT will save cost, switch better from the logic source, and you don't mind the base current.

Low voltage and low-Vce(sat)  BJTs also have improved hFE, so that you can switch them at hFE(sat) = 50 or 100 even.  When you're doing this at power switching frequencies, you're mostly delivering turn-on and turn-off current -- because the base-emitter junction is charge-controlled, just like the MOSFET is.  The BJT starts to look more like a leaky MOSFET with super-high gain, than a BJT!

Tim

[electrolust]

Wait, did you wire the capacitor across the power supply near the MOSFET, or to the drain terminal (and, presumably, ground/source) of the MOSFET? I wanted you to do the former, and while the capacitor in the latter position will reduce turn-off ringing, it is also very hard on the MOSFET since every time the MOSFET turns on it effectively shorts out the capacitor.

I did the former, or tried to.  The cap is across the power rails of the breadboard.  Not right at the FET terminals.  But how close is close?  There is a short jumper wire from the + power rail to the drain (and the probe attached right at the drain), and a longer but still quite short jumper wire from source to - power rail.

Isn't the only difference in the 2 configurations, an inch or 2 of wire?  I understand that for small bypass caps, you need them to be close to the power supply pins of the device you are bypassing, and the closer the better, but for 10uF is 1-2" of wire significant?

[electrolust]

As for the LED, what length of wiring will you be using?  You might ass a cap at the led if you don't need switch speed in the Mhz, or even Khz.  This would pretty much nullify any remaining signal bounce by another 3 or more fold, so long as you PCB has proper power supply decoupling caps.  At this point, such small bounce would only be of interest in analog applications, or, extra low EMI design considerations.

I need 30-40mm of wire to the LED.  It's kind of a heartbeat LED and the on time is only tens of ms, and a period >=250ms.  I guess that implies hundred Khz to single Mhz switching speed.  Any cap would have to be at the distal end of the LED wire (far from LED, close to PCB).  I suppose, like the gate resistor, I can have a spot for it on the PCB regardless, and experiment with different values maybe ending up with 0uF.

VCC -- R_L -- LED -- FET -- GND

Would I put the CAP at the R_L side or the FET side of the LED?  (bypass to GND, not in series)

[electrolust]

BJTs start looking particularly attractive at low voltages, particularly in applications that can spare the base current.  Battery powered switching regulators are a good example: if it's always operating at design current output, then efficiency will be high -- you don't much care about base driving power exactly, as long as overall efficiency is good.

Ah ha ... got it now

(plus all the other considerations)

[electrolust]

Hey, for a single LED (20mA) why don't I do away with the BJT or FET altogether?  My MCU can source 40mA so if I just put an appropriate resistor in line with the LED ...

Sure, now I have a power dissipation concern at the MCU but at half the rated output and at 2% to 20% duty cycle I'm thinking it's probably manageable.  There might be a lot of voltage drop at 20mA but as long as there is enough margin above the LED's Vf it's ok isn't it?

[MagicSmoker]

Okay, sounds like you put the cap in the right spot. There's no problem with the LED being a few cm away - the main issue is with the >1m long wiring between MCU and MOSFET.

Don't drive LEDs directly from an MCU unless you really limit the current. I know the datasheet for the MCU probably says each pin can handle 20mA or whatever, but take a closer look and you might find a rather low total current limit for all I/O, perhaps as low as 100mA. An even worse sin, though, is to connect an MCU pin directly to the outside world without any sort of protection against static discharge, picking up/radiating EMI, etc. So, just don't do it.

[T3sl4co1l]

MCUs driving LEDs is fine if:
- Total supply current is observed.  Usually something like 10-20mA per pin, and a total of 100mA per IO bank, so 5 or 10 painfully bright LEDs.  Or double that, if you use half the LEDs to GND (uses the pin's high side PMOS) and half from VCC (low side NMOS).
- LED voltage is available (i.e., 3.3V logic can't consistently drive green superbright, blue and white LEDs)
- The LEDs aren't distant, say on cables (EMI hazard)

Remember EMI works both ways.  Maybe you don't personally care about EMI emission, but ambient EMI, and transients like ESD, can blow up your MCU pins, or upset operation (RFI causing protection-diode conduction is well known to upset the analog section of most MCUs).

If you don't meet one or more of these conditions, then a lever shifter, logic buffer, external switch, IO expander, etc. is a good idea.

A 'gold standard' example of all three options:
- RS-422 transmitter(s) at the MCU.  Driven by UART or SPI.
- Differential transmission line, with protection and filtering (this, and the robust RS-422 channel, wins you a ton of EMI capability).
- RS-422 receiver (and line terminator) at the LED board
- Serdes or shift register or IO expander, with enough voltage and current capacity for all the LEDs attached.

Tim

[BrianHG]

As for the LED, what length of wiring will you be using?  You might ass a cap at the led if you don't need switch speed in the Mhz, or even Khz.  This would pretty much nullify any remaining signal bounce by another 3 or more fold, so long as you PCB has proper power supply decoupling caps.  At this point, such small bounce would only be of interest in analog applications, or, extra low EMI design considerations.

I need 30-40mm of wire to the LED.  It's kind of a heartbeat LED and the on time is only tens of ms, and a period >=250ms.  I guess that implies hundred Khz to single Mhz switching speed.  Any cap would have to be at the distal end of the LED wire (far from LED, close to PCB).  I suppose, like the gate resistor, I can have a spot for it on the PCB regardless, and experiment with different values maybe ending up with 0uF.

VCC -- R_L -- LED -- FET -- GND

Would I put the CAP at the R_L side or the FET side of the LED?  (bypass to GND, not in series)

This means less than 1Khz, unless you are using a 1 watt led and attempting a strobe quicker than the sensitivity of the human eye to artificially dim it.  In fact, you should be operating below 100hz.

I think you circuit is fine and functional.  That tiny remaining little ring you see on you scope will shrink even further off the breadboard on a normal PCB and you are measuring it is in the ns.

[electrolust]

This means less than 1Khz, unless you are using a 1 watt led and attempting a strobe quicker than the sensitivity of the human eye to artificially dim it.  In fact, you should be operating below 100hz.

oh yeah.  I should have said 100hz to 1Khz.  I was off by a prefix (10^3).  But yeah if you consider the signal itself is 4Hz, I can probably even tolerate switching speed as slow as 10s of hz.  (assuming large latency is acceptable, which it is)

I'll try to revisit this in a few weeks after I've got the PCB and will post final results.