By having parallel devices that are capable of carrying the full load, you're increasing your MTBF of that particular section of the circuit. Then you'll need to look at the MTBF's of the remainder of your circuit to see what it's weak points are - like perhaps the Zener diode, or the gate drivers, or perhaps the width of the board traces. You can spend a lot of time re-engineering things to be "bulletproof".

Having two power MOSFETS in parallel on one hand is convenient, because if one starts getting warm, it's internal resistance increases, and the other MOSFET will carry the load until the hot one cools down.

OTOH, you double up on your gate capacitance and output capacitance. The former requires more robust drivers to keep the gates supplied, minimizing transition time and resultant heating. Increased output capacitance will lead to larger snubber capacitors being required. It will also increase your costs of both the BOM and assembly time.

Oh, for your negative supply - have a look at Linear Technology's LT1054. It's a switched-capacitor voltage converter that can invert your positive supply (up to 15v) at up to 100mA.

 

What's the most common failure mode? Can they fail in another mode?

I vote for single. I am not sure redundancy helps, if you blow a gate to be short-circuit to D or S.

John

John brings up a great point.
If the MOSFET shorted drain to source, the injector would be constantly energized, thus open. If the injector was for a piston or rotary engine, and the engine was stopped, then ignition turned on for a few moments (perhaps to warm up diesel glow-plugs or the like) the combustion chamber would be partially or fully filled with fuel, causing "hydraulic lock" and possibly severe damage.

 

Two engines are put on an airplane, so if one fails, the airplane can still make it to an airport for landing. The MTBF for any "engine" is halved, but hopefully, the MTBF for a crash is more than doubled.

There are two issues raised by the injector problem:

1) If one mosfet fails, will the other continue to work normally (as already mentioned)
2) If one fails, and you don't know it has failed, then you are back to a one-mosfet injector in which the remaining mosfet is presumably stressed more than the others and may be more likely to fail.

So, to get the full advantage of redundancy, you need to ensure #1 is true, and you need a way to detect if only one mosfet has failed.

John

Edit: In my examples, I mean of course, MTBF from random causes. MTBF from a common cause, such as running out of fuel in the airplane, is not changed or may be made worse from added stress (e.g., faster fuel useage and less reserve fuel in measures of time). For the injectors, if the ringing from two is worse than from one, and if the ringing adds to mosfet stress, then two might actually be worse.

 

ah, let me explain further.

my initial design parameters was to choose a power fet that was extremely robust in its current handling capabilities which could be used for multiple heavy load devices i am working on. the IRF1404's handle a crap load of current with very little 'on' resistence (low heat, no heatsink needed, etc). it was my understanding (probability) that a fet would die (stop conducting) before a D-S short would occur, so parallel fet's took care of that.

the 2nd part of my module, which includes the IA and some 4bit counters is to detect a missed current pulse. the IA output is suppose to trigger the MR of the counter, and the counter will advance when a pulse signal is seen from the external signal generator. i watch bit 2 of the counter. if bit 2 goes high i know someting went wrong with the specific injector pulse (bad inj, bad fets, shorted fet, etc, etc). in norm operation the IA will keep resetting the counter so the counter should not advance.

in the case of two parallel fet's and one just dies, the other will work just fine. even if one dies i'm not so worried. if for some reason both die then all 6 channels on the counters will quickly count to 2 and every alarm will signal.

a D-S short is possible. it suspect its not likely. injectors are typically controlled by single npn darlingtons and i dont here any cases of the npn staying on and inj flooding the intake... this stuff all built on probabilty, so i thinks it's not likely to happen (with a solid design, etc).

 

looking for some feedback to what i am seeing on my scope. let me set the scene.

It looks like the FET capacitance and the coil are creating an oscilator.

Clearly a snubber would fix the ringing.

Is the MOSFET being over voltaged ?
When the coil switches off there will be massive back EMF.
This can completely screw up the FET if your not careful.

I did a similar job on a motorbike coil system and we had to use FET's with a voltage of around 500 to stop them being destroyed. Clearly with a coil I couldnt use a snubber as that would quench the spark !

 

It looks like the FET capacitance and the coil are creating an oscilator.

Clearly a snubber would fix the ringing.

Is the MOSFET being over voltaged ?
When the coil switches off there will be massive back EMF.
This can completely screw up the FET if your not careful.

I did a similar job on a motorbike coil system and we had to use FET's with a voltage of around 500 to stop them being destroyed. Clearly with a coil I couldnt use a snubber as that would quench the spark !

my design has 3 34v zeners in parallel protecting the fet. my design is slightly flawed, but i'll try to save these 1st 30 boards. then i'll move some things around.

 

just wanted to add some follow-up on this.

i added a snubber.... actually does great job of taking out the ringing. i added 5nF with a 10k pot (pot was set around 1k in this pic). however, this is subject to change because i am swapping out my fet's with different ones.

 

SgtWookie,
in post #8 you mention less gate charge with the IRZ24 fets. but if i redesign to have a fet for each channel (with Rsense after the fet) then the gate charge is 6 times greater than the single fet since i'll be tying all the gates together.... so gate charge is probably back where it was with dual 1404's. i still have to test my current op-amp setup and if it doesnt work then i'll have to definitely do a redesign with Rsense after fet, etc.

 

Hmm - so you're using dual 1404's to energize ALL of the inductive loads at once, and taking the reading from Rsense on the supply side?

Having visited this topic so seldom, and not having a full picture of what the circuit is doing certainly can limit the usefullness of my replies.

 

Hmm - so you're using dual 1404's to energize ALL of the inductive loads at once, and taking the reading from Rsense on the supply side?

Having visited this topic so seldom, and not having a full picture of what the circuit is doing certainly can limit the usefullness of my replies.

i think in your post #8 the reference to the IRFZ24's was before i mentioned my issue with the op-amp. op-amp being on the supply side will have its inputs reach the flyback voltage when inductor turns off. doing so may have unexpected results on the op-amp output.... the flyback is defintely higher than the rails of the op-amp, but if there is no difference between the inputs then op-amp should not try and amplify...... but we'll soon see.

 

so i swapped out the 1404's with a pair of IRFZ24's. snubbed out the ring with a 10nF and resistor, and the switching looks real good and no ringing (only a single dip in the ring and then flat).

i noticed something odd. independent of duty cycle, as the frequency went up the fet's got hotter. odd was as i approached 1kHz (was at 687Hz @ approx 1.87amps) one fet got hotter than the other. one fet was at 122F while the other was at 100F.

so two questions this post:

1. why do they get hotter as frequency goes up?
2. why does one get hotter than the other (manufacturing differences)?

 

1. why do they get hotter as frequency goes up?

They are spending more time in the linear ie transition region. If you want them to stay cooler, increase the gate drive current and decrease transition times for both turn-on and turn-off.

2. why does one get hotter than the other (manufacturing differences)?

That could be. It could also be that the signal path for the hot one's gate has more impedance than the other, or the hot one's source or drain paths have less impedance than the cooler MOSFET. Long, skinny traces will have a fair bit of inductance. Long, wide traces will have parasitic capacitance from the ground plane on the other side of the board.

 

on/off time is on the order of 2us. its not really a issue for me since my max frequency will be around 60Hz. was just wondering why they get hotter as freq goes up.

the fet's are soldered to very large trace paths. minimum current path trace width is .25" and as the channels join the common area the trace path gets larger, then end in a large fill area of about 3/4" x 7/8". this fill area is where the fet drains are soldered. the source of the fet's are also soldered to a large fill area. the source fill area is terminated directly to ground with a piece of 12gauge copper. no ground plane on bottom of board. this board is dense with parts and as such i used both sides for traces, etc.

the traces for the gates are also fairly beefy. since the two fet's are directly adjacent to each other the gate trace from one goes to the other then over to the pfet gate driver...... i would think impedance on the gates are exactly the same....

 

fixed my op-amp issue (proper power, etc). the design works as expected and it seems ok even though the amp inputs "see" the flyback voltage (34v), but since it "sees" it at the same time on both inputs the net diff is zero and the amp does not amplify, and the output drops to zero. i used the ICL7660 to build a quik dc-dc converter. with just two small 4148 diodes and four smt 10uF ceramics i got +9v and -5v to the amp..... since the amp was working as expected i moved on to my 4bit counter. here's a mpeg of it in action.

what you see is realtime. the green light is the signal generator into my board and is pulsing at a slow rate of about 1Hz @ 50% duty cycle. in the movie 0-7 sec is when the injector being monitored is working normally. right around 7sec you see a shake in the video, thats me disconnecting the injector. as you can see on the 2nd green flash the red alarm led's light and stay lit up via scr's. this board has 6 channels on it, but just 1 was being monitored during this test.

mpeg video approx 6.75MB
click here for video

 

Out of curiosity, what will you be using this for? Is this just an experiment or will this be used for a certain application at some point in time?

 

Out of curiosity, what will you be using this for? Is this just an experiment or will this be used for a certain application at some point in time?

its a unit that drives additional (aka "aux") fuel injectors. just a test project.

 

Very cool, I'm pretty familiar with the applications being a bit of a "car guy". A couple years ago I turbocharged a little sport compact myself and was researching this option to meet fuel demands. For cost and simplicity I went with a rising rate fuel pressure regulator instead but I would have loved to have the specific control a standalone system would have given me.

At the time my options for a standalone system were in the neighborhood of $1500+. After I went the other route I found a system that catered more to the low buck crowd. If you're interested in researching your future competition you might want to check out a product called Megasquirt.

Best of luck to you on your project. I'll definitely be following this thread to see how it goes. Being a guy that has a heavy automotive background and I'm just now getting into the electronics side your project really piques my interest.

 

TanTJ - yep, the megasquirt standalone is the grandfather of standalones, but today there exist many piggyback units that can manipulate the oem ecu to get about same functionality as a megasquirt.

as for my project, i may have mis-stated the intent. its really a project for my own turbo setup and i have a piggyback MAP-ECU2.