John,

As mentioned, I didn't design this, so I do not know the theory behind the decisions made to have only one resistor, 20 FETs or anything else. As for a schematic, all I can do is to post a picture of the board. Let me know if you would like me to.

OK, you bought a kit.

However, all I am really interested in is what goes into choosing a gate resistor. Do I simply choose the value the manufacturer lists under their test conditions, or is it even that important?

Contact the kit manufacturer. What resistor does it recommend? Maybe, ask for a refund under warranty.

Most of the research I have done (including reading several documents from IR) suggests that the resistor is more or less guesswork.

While it is true the final value is determined empirically, it is not entirely guess work. There are good starting points, and with a scope, you can easily see when oscillations are damped. Circuit inductance and capacitance for each gate affect the final value. One thing to remember is that a single, 20-ohm resistor feeding the gates of 20 power resistors is equivalent to a 400 ohm resistor feeding one, at least in terms of turn on time. It will be relatively slow.

Honestly, I'm leaning towards leaving the 20k gate resistor in for the new mosfets. My pcb is already designed and built, so I am now aiming to make the best of it.

A single 20K gate resistor feeding 20 power mosfets is like feeding one with 400K. They will turn on, eventually, but may burn up first.

As for discharge time, there is a diode and resistor parallel to the gate resistor that is there to control turn off times. After reading the second article you posted, I see how important turn off can be when dealing with parallel mosfets. It seems to me that you would almost not want any resistance at all on the diode, to allow them to turn off even faster since miller charge can tend to keep the quicker to turn off FETs to stay in linear mode for longer, creating heat.

Yes, some people use just a diode, no resistor, in the turn-off circuit. That design is shown in the IR application note.

In sum, I think it is a mistake to use a single gate resistor. If you consider the discharge time and hence maximum current, my gut tells me that 20 mosfets is overkill in this application. Since they are in parallel, you could try using just 5 to 10 for starters. If the PCB isn't drilled for the resistors, you can consider an off-board construction technique like this:

We have not discussed what you paid for the kit/design. I hope it was not much, as it does not sound like a well engineered design considering its use of a single, high-value gate resistor. At some point, you may decide to write it off as education.

The idea of using mosfets for a CD welder occurred to me too, as one might be able to control the welding cycle(s) better. For example, some CD welder vendors recommend using two pulses, one to clean and one to weld. That could be done easily in theory with mosfets; however, I would want to see actual results obtained under objective, controlled conditions before making a design to do that. I have not seen any such data. If you use a single scr (or mosfet), you could always just fire it twice. That's what I do. The additional amount of time to assemble a pack for a hobbyist is insignificant.

Finally, 10 mil Ni seems a bit heavy gauge for battery tabs, but that is just a first impression. I suspect 15V from audio (?) capacitors may not be sufficient for that thickness. Remember, a lot of factors will limit the maximum current you can achieve, not just the mosfets or size of cable.

John

 

The 20k value, as well as the IRFP2907s, are what was recommended, and the guy seems to get a bit annoyed when asked about resistor values if I were to use different mosfets, hence why I am here.

The turn on time definitely seems slow to even me, considering that the manufacturer's specs were quoted using 1.2-ohm Rg. I will definitely lower that value, as well as possibly remove the resistor in the turn off circuit.

In regards to the welder, it is two separate boards. One is the control board and the other is the mosfet board. The control board is processor controlled with dual pulse adjustments and voltage adjustments, controlled via 3 pots and an LCD. The charge/discharge circuit for the main caps is also mosfet controlled. The mosfet board is as has been described. Both are two sided heavy gauge PCBs. I paid $180 for the PCBs and processor.

I chocked the whole project up to education before I even started, as I had an opportunity to get a Sunstone welder for only around $1000 second hand. I enjoy this type of thing and being new to circuits, I was intrigued. I've leaned more in the last 6 weeks with this project than I feel I would in several classes in school, and I think I have finally settled on a major! Good real world experience is worth a lot in my opinion, so there is no way this is a waste to me.

The reason for this project was to build battery packs from A123 26650 cells for starting motorcycles, hence the heavy tabs. I've been building them for people in the CMRA (local motorcycle racing club) for a few months and got intrigued by CD welders.

While the welder was working, I actually got it set quite well and successfully built two batteries. At 15.5 Volts with the first pulse 3.5 milliseconds and the second pulse 14 milliseconds, I was getting great welds on both ends of the A123s.

After what I have learned the past few days, I have to agree. The welder definitely should have been constructed using individual resistors. However, in an effort to get by with what I have, could I lower the charge resistance and remove the turn off resistor in an effort to speed up on/off times and maybe make it more reliable? Do you think he was maybe charging the FETs slowly in an effort to keep them balanced?

Worst case scenario would be to use the board for the package, but externally wire up the gates with individual resistors. Wouldn't take long.

For now, with the welder constructed the way it is, what value gate resistor would you recommend to have the best chance of (temporary) reliability? Should I remove the turn off resistor?

I believe I am definitely going to redesign the mosfet board down the road. I now have 17 good IRFP2907s I can use when I do. 3 Computer grade caps are also in the future. The control board is designed for up to 20 volts, but these caps only handle 17.5-18.5 max. The new caps should significantly lower the ESR as well.

 

Power board: Can you attach a photo of both sides? In fact, it would be nice to see pictures of the whole device.

Gate resistor: You are charging N x capacitance of each mosfet. The current capability of your gate driver may be an issue. I believe the first link (the one on theory) goes through a calculation. Was 9A mentioned somewhere? If it is 9A, that should be plenty. As for the value, I would consider just trying something between the manufacturer's recommendation (1.2 ohm?) and 10 ohm. The board example I showed was for a motor control and used slightly higher individual resistors.

Turn-off resistor: You may not need a turn off resistor, depending on the design of your gate driver. If it sinks current, then a turn-off resistor may not be necessary. However, including one is like insurance. In the example I posted above, the driver sinks current, but that high value turn-off resistor (47K) was added after the fact to insure the top mosfets were off when power is applied. An anti-parallel diode will speed up turn off, if the driver sinks current. Note, if your driver will not sink current, then you need a turn-off resistor. A diode in that case is unnecessary. I would use a turn off resistor at least 10X and up to 100X the gate resistor.

Experience to date: That is good to hear that it actually works. Do you see a difference between two pulses and just one? What are you using for your welding electrodes? I use copper electrodes and have considered trying tungsten. My tabs are 3 mil or 5 mil Ni x 1/8" at 17.5 to 18.5 volts with a minimum of 4 welds per joint/contact. My capacitor bank is considerably smaller than yours, which probably explains the higher voltage I need. The packs are used only for relatively low current of a few amps. The weak point in the weld is the strip. The thin strips will just tear, but that is plenty strong for my application.

My project, which is posted in the project collection here, also uses a microcontroller for timing. If two pulses are truly an advantage, the programming would be easy and I would probably switch to mosfets.

John

 

I have many points...

First, I stick to my guns that a gate resistor is often not *necessary*. Especially when driving a MOSFET from a low current source like a microcontroller, it's not needed. But like John said, it can be "sort of" simulated but it's more of both a gut and empirical thing. You get a feeling for it, then you use the scope to check it.

In this case, if the goal is fast switching as I think it is, then I really think a single 20 Ohm resistor for all 18 FETs is too limiting.

I am going to attempt some math to estimate the timing of this thing. Let's assume the old component (IRFP2907). I'll use a total gate charge for turn-on of 500 nC, a reasonable value by the data sheet (this is ballpark estimation). For 18 of these FETs, total gate charge is 9 uC (microCoulombs).

Let's say we're starting from 0V on the gate, with a 15V supply to the gate driver, which can handle 9A. The 20 Ohm resistor will limit the initial current to 0.75A (= 15 V / 20 Ohms). Right there, you see how this resistor is limiting us and we're not even seeing much of the gate driver's 9A capability. This 0.75A is shared among all 18 MOSFET gates.

Then, let's say we start to reach Miller plateaus around 5V, as shown in the data sheet graphs. At that point, the gate driver is pumping out 15V. The 20 Ohm resistor is dropping 10V (from 15V to 5V). Therefore, the current is limited to 0.5A among all 18 MOSFET gates.

I'm going to use a figure of 0.6A (between 0.5A and 0.75A) to ballpark the turn-on times, i.e. time for the gates to punch through the Miller plateau and be on their way to fully enhanced conduction.

A Coulomb is 1 Amp-second. So 1 Amp would supply 9 microCoulombs in 9 microseconds. We're using 0.6A, so we're looking at 15 microseconds to punch through the Miller state.

In a situation like this, I would want that to be as short as possible, more like 100 nanoseconds, a factor of 100X shorter duration.

I'm in favor of changing that single gate resistor to 1.5 Ohms or so. That will limit the initial total gate drive current to under 10A and it will be around 6A when the things hit the Miller plateau. I think that would be a good compromise between the gate driver capability and the requirements of the switching times. At least that will reduce the turn-on time down to around 1 microsecond.

There *might* be something about welding that I don't know... *maybe* there is a good reason to limit turn-on time. Maybe that makes the weld better or ramps up the heat instead of "blasting" the metal under weld... I've spot welded some metal in my day but never built a welder or even saw their schematics.

(side note -- I think that when Matt is saying "20k gate resistor" he's really saying "20 Ohms resistor".)

 

I'm in favor of changing that single gate resistor to 1.5 Ohms or so. That will limit the initial total gate drive current to under 10A and it will be around 6A when the things hit the Miller plateau. I think that would be a good compromise between the gate driver capability and the requirements of the switching times. At least that will reduce the turn-on time down to around 1 microsecond.

There *might* be something about welding that I don't know... *maybe* there is a good reason to limit turn-on time. Maybe that makes the weld better or ramps up the heat instead of "blasting" the metal under weld... I've spot welded some metal in my day but never built a welder or even saw their schematics.

(side note -- I think that when Matt is saying "20k gate resistor" he's really saying "20 Ohms resistor".)

I agree with the two bolded statements; although, I have never seen the original reference by "Matt." A higher value resistor might be needed, if you use one for each gate.

As for the ramping effect, that is a good point. One thing that is easily, even spectacularly noticed is when the electrodes make poor contact to the substrate. You do run the risk of blasting a hole. Lots of sparks. So, be sure to wear safety glasses.

John

 

Yes, I mean 20-ohm, as was originally stated back on page 1. As for ramping, seeing as how the designer already messed up once by using a single gate resistor, I hardly feel he intended to slow down turn-on that much. As to why he would slow down turn off, I don't know that either.

I'm definitely going to get a different value resistor though for the new FETs (which should be here today since mouser does such a spectacular job shipping quickly). I have classes today, so I won't be near the welder for pictures, but I'll get some soon. Fry's is right by my campus, so I'm just going to get a resistor there. Probably going to pick up a few of different values.

Jpanhalt, my electrodes are 3/16" copper encased in 1" delrin rod. I did weld at the same voltage with a single pulse for 10ms, but it wasn't long, and the welds look a little better with two pulses. I'm sure if I raised the voltage on single pulse, it would do just fine. I can't tell you how much better, but if you have one that works now, I wouldn't bother unless you just wanted to. If you did though, you could share the design with us!

Thanks for all of the help here guys! I think I have a pretty good grasp! I'll be sure and report back every step of the way.

Side note: Would you guess that the dual pulse models from sunstone utilize mosfets for firing?

 

picked up two each: .47-ohm, 1-ohm and 1.5-ohm. Those will allow me to make many different Rg values.

 

FETs are here and I am installing right now. What is the verdict on turn off resistor?

Driver datasheet:
http://focus.ti.com/lit/ds/symlink/ucc27322.pdf

On page 7/30, it states that is sinks approximately 9A. Later it also states that it can sink slightly more than it can supply, to ensure the mosfet is fully off and stays off. In theory, I don't then need the resistor, correct? Should I just leave it anyway? I'm thinking that even though the FETs will stay in linear mode longer with it, it could help in turning them off at the same time to avoid over-driving some that turn off slightly later. Is that a sound theory?

For reference, I'm starting with a 1.5-ohm gate resistor as those much wiser than I have suggested/recommended as a good starting value.

EDIT: Couldn't the resistor also in theory take away some of the power the driver has to dissipate, to prolong it's life? The turn off resistor is a a 10k-ohm (no mistake here) resistor.

 

From reviewing the datasheet, the turn-off resistor is not needed. As I suspected, the driver output section will sink and source current.

As mentioned with the motor drive unit example I posted, I included a turn-off resistor to ensure the mosfets were off when power was first applied. If even one mosfet were on, say from static, then there would be a dead short across it from a high-power storage battery.

As I understand your circuit, in the event the mosfets are on at start up, the electrodes would also need to be shorted to have any current. Even then, that current would be limited by your supply, which is probably less potent than a storage battery. So, I think the risk is negligible and the resistor can be omitted. However, if your gate resistor is less than 10 ohm, and the turn-off resistor is 10K, that is a 1000:1 split. I don't believe you will notice any change whether it is there or not. Flip a coin. If it is already soldered in place, it is not worth the effort to remove.

John

 

Thanks John! Leaving it I am. I'll let everyone know how it works out when I'm done.

 

So, the designer got back to me regarding the gate resistor. He says the gate resistor was specifically calculated based on di/dt specifications of 100A/μs, as listed in the datasheet. The IXYS parts have the same "condition" called out in their datasheet as well, so should this be a concern? Leave the resistor at the designers specs?

EDIT: He also has a source that tested the ESR of my main caps, and came up with a value of < .0005 ohms when paralleled. If correct, that could produce some incredible amounts of current, with the limiting factor being the FETs. Rds(on) would then have a huge impact. I can definitely understand why the aforementioned ramping would need to be considered. Does anyone have any idea where to start with calculations? I was hoping some of the graphs in the sheets would be useful for estimations, but not so much.

So much to consider!

 

Re. ESR: Do you really mean that many zeros?

Re. di/dt: Yes, that can be an issue. How did the designer estimate it? A di/dt calculated only from the esr of your paralleled capacitors and Rds(on) of the mosfets is optimistic. You must consider inductance in the system (e.g., leads, gate resistor, PCB traces, capacitors) as well as other sources of stray capacitance. Moreover, with a single resistor, the mosfets are turning on at different rates/times.

John

 

On some further thought, I think it's best how the designer has created the circuit, with a single gate resistor. This will limit the rate of increase in voltage of all 18 gates. There will still be resistance and inductance due to the circuit board traces, but the gates will tend to be at the same voltage and act like one big capacitor.

I am *not* very knowledgable about the dV/dt or dI/dt specs, but I was reading some about it, and from my quick ballpark calculation in a previous post that the transition from off to fully on would be on the order of 1 microsecond, I don't think that it is approaching the maximum rates in the data sheets of 5 V/ns or 3 V/ns. I think it's a couple of orders of magnitude in the safe direction.

I find figures 15 and 18 in the IXYS part data sheet to be instructive. They give the turn-on and turn-off times based on gate resistance from 5 to 20 Ohms. They are on the order of 40 to 200 ns. They get higher, of course, with greater resistances.

If you use a 20 Ohm resistor for all 18 MOSFET gates, then each one effectively sees 360 Ohms, in respect to how fast the common gates change in voltage. If you use a 1.5 Ohm resistor, then each one effectively sees 27 Ohms, which is still past the high end of the graphs.

My recommendation is to keep the single gate resistor topology, but lower that frmo 20 Ohms to something like 1.5 Ohms.

That's my analysis. As I've said, I have not designed circuits for such high power levels before and I'm interested in any and all criticism.


Here are some useful resources:

dV/dt ratings for low voltage and high power MOSFET
(This one explains the meaning and reason for the dV/dt spec on data sheets.)

Paralleling of power MOSFETs for higher power output
(This one is especially reassuring. It says in the bullet points around Figure 2, that MOSFETs in parallel tend to balance with one another, despite differences in specs like gate-source capacitance, Miller capacitance, and threshold voltage. It also says that it's best if the multiple gates are *not* decoupled by gate resistors.)

Analysis of avalanche behavior of paralleled MOSFETs

 

Re. ESR: Yes, 3 zeros. .0005. I did not measure this, I do not know common values of similar units and I have no idea if it is accurate. The original .007 number I gave earlier was for each cap, supposedly.

Re. Gate Resistance: Quote from designer in regards to individual vs shared gate resistor: "The gate pulse is supplied with a constant voltage and your resistance and voltage does not change when turning 1 or 18 mosfet on, so your current and the speed that the fets turn on will stay the same."

I do not have enough experience for things like this to be instinctual, but is he correct that in reference to turn on, shared vs. individual is the same?

If so, 20-ohms would be a good value.

 

The time constant for charging a capacitor from a constant voltage is RC. Despite what the developer of the circuit says, the resistance (R) and capacitance (C) affect the time to reach a certain voltage. More mosfets in parallel gives more capacitance.

See: http://hyperphysics.phy-astr.gsu.edu/hbase/electric/capchg.html

John

 

If there were one resistor per MOSFET then he'd be right, as long as the gate driver had enough current capability to maintain the same voltage against the resistors.

However, if there is one gate resistor that leads to all 18 MOSFETs in parallel, then like John said, their capacitance all adds up and it takes that many times the current to change their voltages and turn the MOSFET on or off.

In short, I second John's answer and I am pretty sure that the designer is mistaken on this one.

Are you totally sure that the circuit is as I understand it to be, as follows?

gate driver output ----> 20 Ohm resistor ------> gates of all 18 MOSFETs


- Sage

 

By the way, a different point:

Whatever resistor you use, it should be beefy. I know the current is only flowing for the briefest period of time. A 1/2 Watt resistor would *probably* work, because of that.

But still.

If it's a 20 Ohm resistor, it could theoretically see a peak of 11.25 Watts (15 Volts * 15 Volts / 20 Ohms).

If it's a 1.5 Ohm resistor, it could theoretically see 150 Watts (15 Volts * 15 Volts / 1.5 Ohms) -- that is assuming that the gate driver can really deliver 9.3 Amps and is properly bypassed with a capacitor, and the circuit board traces are beefy enough. But in any case it will probably see 50 Watts for ~ 100 nanoseconds.

This is only for the *briefest* period that it sees the peak power, and then that tapers quickly within a microsecond to nothing, then 15 milliseconds later on the turn-off it sees the same thing.

Maybe other forum readers will know more about the effects of large pulse currents on resistors. Obviously, the 20 Ohm resistor worked even though it was seeing 11 Watts.

Anyway, if I were you, I'd feel more secure if I used those large beefy resistors that are rated for 5W. You can order them from suppliers, and most Radio Shacks have them in 1, 5 and 10 Ohms. You could parallel four 10 Ohms and get 2.5 Ohms with 20W rating. That would be great.

 

I would still like to see a picture of the mosfet board and driver board.

Nevertheless, here's what I would do. Upon reviewing the original posts, it seems the mosfet(s) blew because of an accident. We don't know if the design had anything to do with that. I would keep with the original design, including a single, 20-ohm gate resistor. I might cut the number of mosfets in half, but that is just based on my gut feeling that there are too many. The single-pulse current capacity of the mosfets it quite high.

See what happens. If you still blow a mosfet, then try modifying the gate connections to use individual resistors (like in the example I posted). BTW, I use only 4 awg welding cable in my welder. They are a bit less than a 1 meter long (ca. 75 cm). Depending on how they are bent/looped, they really jump when I hit the switch. Do your larger cables jump?

We gotta get this thing working, so you can start making battery packs.

John

 

I'm a step ahead on using the called-for value first before making modifications. Just a gut feeling I guess. After he replied explaining that an oscilloscope was used to narrow down a correct resistor value based on circuit inductance (this part sailed smooth over my head), and that he believed my mosfets called for a similar di/dt condition, I started to have some more faith in him that his design was sound. It also seems as though I am coming across just as much information stating that a single gate resistor is better as I was before regarding individual resistors. If you think about it, the individual models would have to be pretty precise values at this high current, to prevent some fets from getting huge loads. That's my theory at least.

One thing he did state was that my fets had a significantly lower Rds(on) rating, meaning that they would be more likely to blow from high currents. He said to be careful welding too long at high voltages with my selected FETs. Hopefully that interprets into being able to weld my tabs at lower voltages, otherwise that could be an issue given the previous setting requirements (15.5v).

As for wattage, the original design called for a 1w resistor, however, I did the math too and decided that since the potential was there and the 2w resistor was the same price, I'd go with it. If you find it a possible issue, I can monitor temps after some welds and report back.

Oh, and John, my cables tend to jump a bit, but nothing dramatic. Then again, they're pretty heavy!

One other thing: My board has vias placed about it that connect the top and bottom layers together, as well as some places that need to be jumped over the gate drive traces. I beefed up the vias with wire pushed through each one and soldered and changed the jumpers from 14 awg wire to .01"x1/2" nickel plating. That should beef it up a bit and help to displace the load equally to the fets farther from the electrode connection.

I'm going to post up some descriptive content pretty soon (pictures, diagrams, etc..), so everyone can see what exactly they've been fueling!

PS: I love this forum!

 

Everything tests good, but I remembered upon starting up that my control pots were having issues. The Pulse 1 settings only go down to ~ 9ms, and will not stay at a particular value. There is a 0.1μF Capacitor on each pot, and the pot that controls my pulse 1 width is bouncing around like crazy. I switched the cap with the pulse 2 pot's cap and the problem followed. The two "good" ones read 0.11μF while the problematic cap reads 0.15μF. Well within tolerance specified, but obviously outside of the acceptable tolerances of the circuit. I'll pick up some with tighter tolerances tomorrow to replace all 3, since it'd suck to have the one on my voltage pot go nuts, cause my charge mosfet to blow and blow my controller, processor, LCD and everything in between! The charge mosfet blew once before (rather violently) from a faulty pot, but luckily only a capacitor was damaged in the extravaganza.

Problem part: http://www.mouser.com/ProductDetail...=sGAEpiMZZMt1mVBmZSXTPCB4s6x5SlOX3BJ4FBUhgY4=

Another update tomorrow!