Re: [CAD_CAM_EDM_DRO] Magic Smoke escapes

Chris Stratton wrote:

> Milled some nice little circuit boards on the light machines sherline
> using EMC and a random dremel bit (in the sherline headstock). First
> real part I've made with any of my CNC projects. I don't have a
> powered Z axis, so I milled the traces as one unbroken path and put in
> M00's at each through hole location so I could lower the Z by hand to
> create a center hole that could later be drilled through. It actually
> all worked pretty well, although I don't think I will try doing PCB
> layout on graph paper and manually generating a toolpath again!
>
> Unfortunately, the circuit doesn't work as well as the boards look, at
> least not yet. Each one is a half bridge with an IR2104, bootstrap
> capacitor and diode, two IRF530's, sense resistor, recirculation
> diodes, and other stuff to make up essentially the circuit from the
> first page of the 2104 data sheet. So far I've let the magic smoke
> out of 2 or 3 FET's, exploded a sense resistor, exploded the 7812
> regulator that was powering the 2104, and blown a power supply fuse!
> Between all of that I did learn quite a bit though.
>
> I suspect that despite the 2104's best efforts, there is some shoot
> through occuring. The datasheet shows series resistors on the MOSFET
> gates, I used 1k ohm as in the DPRG ("untested") discrete bridge
> circuit.

AACK! That's way too much! That causes the transistors to turn on
and off INCREDIBLY slowly. I am using a 13 Ohm resistor in my
PWM servo amp. Anything much over 25 Ohms is putting the transistors
in the linear region for a long time.

> In looking on the scope I'm seeing a noticeable ramp down of
> the high side gate, and I suspect it is still on when the low side
> switches on. So one action item for tomorrow is to try much smaller
> resistors. (Let's see, somewhere I heard I figure of about 1000 pf
> for gate capacitance. RC for that and 1k is 1 us, wheras the
> off-before-on anti-shoot through sequencing is on the order of a
> hundred nanoseconds).

Gate capacitance can easily be 5000 + pF in the larger fets. There is also the
Miller effect, where capacitance between gate and drain causes more charge
to be needed to turn the transistor on than the simple capacitance would indicate.
Your gate ramp time is likely to be 5 - 15 uS, when including the Miller effect.
Unless you have allowed at least 15 uS dead time between turning off one
transistor and turning on the other, shoot-through is possible, as the gate capacitances,
Miller effect and gate threshold vary significantly from one transistor to the next.
If you don't have a dead-time circuit, then what you have is two linear transistors
in series, both partially on at the same time. This can be worse than having them full
on at the same time, due to dissipation.

> Another thing I'm trying to figure out is if I should put the 2104's
> ground on the FET (high) side of the sense resistor, in which case it
> may misinterpret input signals, or if I should connect it to the
> actual ground and let the resistor voltage potentialy add to or
> subtract from the gate drive.

It depends on the expected voltage across the sense resistor. I believe most of
the IR drivers can accept up to minus 5 Volts from the low-side ground to the
signal ground. See the data sheets.

> I'm probably going to take it in to work and hook it up to a power
> supply with current limit, and try to find something with substantial
> inductance that also has a high enough DC impedance to not fry things
> if it gets left on 100% duty cycle.

If you are allowing the high side drivers to be on 100% of the time, the bootstrap
capacitor can not be trusted to hold a voltage sufficient to turn the high side
transistor on indefinitely. It will drain, and either the high side will shut down
on undervoltage, or the high side transistor will become resistive, which will
get it very hot. If you allow the high side to be on for more than a millisecond or so,
then you need a charge pump or some other circuit to keep the capacitor
for the high side gate bias charged. I did this on my PWM servo amp.

Jon