And by the way the "simple series resistor" would be way greater nightmare to design, even ignoring the corrosion problems.
\First 2x 100W dissipation is ridiculous power to handle in the crammed car environment without forced cooling. Then deal with the failure of such cooling.
Unlike 1mOhm MOSFETS, a lamp short circuit won't just blow the upstream fuse, because the resitors would limit the current (to about 45A, so not that much above the 20A normal working current of each lamp). But it would cause them to generate about 10x as much heat than in normal operation, so a short circuit would be an event really asking for severe car fire.
So no, something that may seem simple can really very quickly become unsolvable nightmare.

Nothing unsolvable. Ballast resistors can be mounted on the lamp rack externally. Some kind of thermal cutoff like a thermal fuse can be added. Still much easier than using a bunch of complicated power electronics and worrying of interference generated. 100W dissipation may look huge in a scale of a small home appliance, but is really minute in a scale of a large car...

But I still think 40+A consumption can be problematic, electronic dimmer or not, by taxing car's alternator and taking almost all charging current available, especially at low RPMs. 

We are talking about cars. There are quite strong restrictions on what you can and can not do. One example it can not weigh half a ton. Other there events like car crashes you have to address. There is rain outside. There is potentially combustible dry organic mess falling onto the car you should not set on fire.
You can not have any open live voltage. Small collision and you have a potential short circuit and so fire.

Today (since 2000's) the car alternator is designed to deliver more than 100A even on the very small basic models, more often it is in the 200..400A range, half of that should be available at idle. This is one of the reason for the dimming: To also reduce the current consumption when the full power is really not necessary. And that means just alone a good reason to not use any energy wasting dimming - with resistors the current consumption would be about 30A, but with this PWM just 20A. I would assume the full power will be used only at higher speed when the engine has sufficient rpm anyway, or a special occasion for a short time.

But otherwise yes, 2x250W is ridiculous power for a car and it may need an alternator upgrade, mainly if it is going to be used for longer periods of time. Running alternators constantly at more than about 30..50% of their sticker current rating is running them very hot and so will have impact on their lifetime.

I would take a good old IR2153 to do this kind of project. Heck, even Helvar used these in their dimming fluorescent ballasts)

The problem with these ready made drivers is, they do support alternate switching, only few support both channels off at the same time, some may support both on (those for current fed boost push pull converters),
but here we need all three modes within one system, both transistors on for the full power, alternating for the "50%" and both off for (fault) shut down. And it is the combination of all three modes within one driver system what is unobtainable with ready made IC's.
So you either make something really from scratch (my approach), or select some IC that could be "bent" to do some of the features and then attempt to attach the missing functionality externally by some glue-on circuit (the way Ash approched it) or implement the functionality into a microcontroller (the way these things are normally done in the industry, as it offers to make the function really matching what the application needs yet it yields the simplest HW so cheapest high volume production).
Each of these ways bring their complexities, neither is wrong, just all of them are somewhat complex and the point is to find the simpler one, also taking into account the capabilities of the one who will be building it.

Well OK, with IR2153 and likes, and by slightly optimizing the schematics, you can do exactly what Ash variant does, at about 1/5 of the parts count

You need shutdown? OK short the timing capacitor to the ground. You need both mosfets on? OK, feed the gates +12V directly with a pair of diodes, or just cut off IR2153 ground as a rough hack, that will do, too)

UVLO? No need to mess with LM2903 and a ton of discretes, the circuit is already on the chip. I'd say Ash circuit is nominating for a Guinness for the most 2N700X transistors used in a small project, though I sure have seen the other contenders

 



   

By the way I was just quickly scanning the MOSFET offerings and it seems the mOhm FETs at 60V are still available with low Vth (logic drive) variant, e.g. two parallel IRLB3036PbF and there are many more.

To me this makes the microcontroller way simpler than I originally thought, the 4.5V minimum gate drive means no need for extra gate driver complexity, at 100Hz the microcontroller has plenty of drive power to drive them virtually directly, just via some protection RDR network:
To prevent latchup excitations from some Cdg coupling - one 100Ohm resistor from the MCU to a "point", from there another 220..470Ohm resistor to the MOSFET gate and a Schottky diode with anode on GND and cathode on that "point". When a negative edge apepars on the drain and couples via Cdg as a negative spike on the Gate, the Schottky clamps the "point" to about -0.3V, which first is below the 0.7V needed to activate the parasitic bipolars (and so eventually the latchup) within the MCU IC and seconds even if it exceeds it, it limits the current to the output so it won't even approach the threshold triggering the latchup. Bteween Gate and SZource then nees to be about 47kOhm bleeder, ensuring the MOSFET stays OFF should the  MCU not start for some reason.

You can start with Arduino, you just need to replace the 5V regulator with something compatible with automotive (at least 40V input electrical rating, add the filters and reverse polarity diode,...) and reconfigure the fuses to use the internal RC oscillator instead of the default quartz (for a reliability reason - quartz tends to fail with even very small leakages on the board, so a bit of humidity and it would fail). And you should activate the toughest brown out detection threshold (the 4.75V) as the undrvoltage protection, in order to ensure sufficient MOSFET drive.
The temperature can be measured via the MCU itself, or using NTS+resistor divider on an ADC input. Similar for overvoltage protection (use 100kOhm/10kOhm resistive divider to convert the 0..50V on the battery input into 0..5V for the MCU to process, I think the VDD would be a good reference for both).
And obviously set and use the WatchDog properly...

In the code

@Ash @Medved So would Ash's circuit be functional and reliable? I quite like the added protections and considerations in his circuit. but when it comes to PCB level design, I don't quite have the training/skill yet to determine whether it is a functioning unit. I can 100% follow the schematic and build the circuit to see if it works (once I get the components mailed in) but determining it is a circuit that will function in the environment/system it will be subjected just reading the schematics alone... not so much.

I like your circuit aswell @Medved due to it's simplicity, however I am having trouble following the schematic. it seems V1, V2, and V3 are voltage sources? or are those momentary switches? I also have never seen an inductor symbol used for a light bulb before. unless those are actually inductors?

It is a simulation schematic, it was intended to show the generation of the gate control signals in all 3 modes: Shut down (for faults), alternate (for "Half" setting) and full ON (for "Full" setting). It is missing the fault detector comparators (under and overvoltage, thermal protection), these are essentially the same as in Ash's circuit (just the polarity has to be the same)
Vxx are indeeed voltage sources. V2 (a constant 14V) represents the main power from the battery, V3 represents a control signal switching between "Half" and "full" power, V1 simulates generation of the fault shut down signal (coming from the protection comparators).

However both of the "pure HW" circuits are rather complex and are lacking a short circuit protection (they rely on the upstream fuse and sufficient wiring resistance), because that is not that trivial thing once you don't want a lot of nuisance trips. I do not see a simple way to do it without a microcontroller. Beside some way to observe the current (that is the easy part - just monitoring the Vds is enough) it needs to be rather fast responding, but requires time masking (so it won't trip on turn On or turn OFF transients; still doable in few HW components), automatic restart logic (the fun starts), but with some time out (so it won't be attempting to endlessly "restart" into a real short circuit; now we are cooking) and on top of that make it so it won't be fooled by the 100Hz PWM as "manual resets"; now we are cooking on gas), this protection will most likely prevent the fuse from blowing, but each restart attempt into a real short circuit is a stress so it should manage everything right.
In a microcontroller the only thing you need is some Drain voltage monitoring (a Vbe as a threshold is good enough into a digital input) and do all the logic in the code (interrupt to immediately shut down, masking it when OFF and for some time after turn ON, then few timer counters to count the restart attempts; about 50 lines of code in C).
And most important the HW is so simple and can be easily made very dumb so with very little chance of something being wrong, so when something isn't working right, you can update the code easily without the need to modify the HW...

Here is a proposed parts-optimized version of analog solution. D1-D4 hack is meant to keep IR2153 UVLO to work also in 100% mode. I assume a 100% switch is located on the control box or nearby, otherwise an extra small capacitor may be necessary.

EOL circuitry is omitted, as it seems not much needed, but if wanted, can be added to Q3 gate.

I think you hit the nail for a non-mcu implementation, the diode ORing for the 100% mode is indeed a very nice trick in my books...
Yes, it needs all the short circuit detection and overvoltage logic, the open load logic could easily stuck the CT to one or the other side (depends on which side is faulty so which needs to remain ON) if lighting the other lamp 100% is preferred, or via the Q3 to shut everything down.
But all the fault logic is a matter of quad comparator (LM339,...), a reference (TL431 or so) and a few resistors and maybe a few bipolars (the short circuit). The undervoltage is already within the IRS2153, so you need just the overvoltage, thermal shut down, open and optionally shorted load.

The open load detection is there to prevent the thing from constantly pulsing 20A into the electrical system, so it should either shut down both lamps (via grounding CT) or stop the oscillator so the other lamp stay lit (pulling the CT up or down so it stays in that state, but won't go below the ShutDown threshold; two transistors pulling it via a resistor to ground or VCC will do the job...). Which way depends what is less of a problem. If both shut down, indicator LEDs need to be added, so the thing will signal which of the lamps is faulty, so it becomes easier to diagnose and replace the faulty lamp in the wild...

Here is a proposed parts-optimized version of analog solution. D1-D4 hack is meant to keep IR2153 UVLO to work also in 100% mode. I assume a 100% switch is located on the control box or nearby, otherwise an extra small capacitor may be necessary.

EOL circuitry is omitted, as it seems not much needed, but if wanted, can be added to Q3 gate.

It appears the IR2153 is possibly obsolete and not recommended for new designs. The through hole variant is not available on both digikey and mouser. would the IRS2153DPBF or IR21531PBF suffice?

IR2153(D)(S) &(PbF) DATASHEET

IRS2153D DATASHEET

Mouser Search Results

Digikey Search Results

I am guessing I could use the IRF100P218AKMA1 Ash had in his schematic in place of your 60V 100A plus MOSFETS?


I see the diodes in the schematic are Surface mount. I can find substitutes for the Schottkey diodes pretty easily... but I don't know about the TVS diode. you have any suggestions there? I don't have the tooling to solder surface mount unfortunately.

I don't think you are planning to start volume production of that, so the fact it is "not recommended for new designs" is not that much relevant for you.
The IRS2153 has the UV+ shifted higher up to 12V and that would be a problem: It may not start with just a 12V car battery.
The 9.9V max limit on the IR2153 is just on the limit of what acceptable for your application (in fact for standard car electronic acceptance even that won't be sufficient, normally 8V on the battery should suffice the operation, most car makers even do not accept anything above 6..7V).

Even the UV- limits are rather high for a car application, the older IR2153 will stop at 9V, the newer IRS at 10V. And that means about 11V at the the battery input (the 1V will easily be at the reverse protection diode and the regulator), is under the halogen's load.

A solution could be using some voltage boosting in front of the IC (like MC34063 based booster to lets say 12.5..13V; if the Vbat will be higher, the booster will just inherently stay inactive, for lower battery it will easily maintain the 12.5V on the IRS2153 input even with the battery going down to 5V during engine starting).

For MOSFETs yes, or you may use multiple smaller (higher Ron) in parallel...

Like Medved said, no problems with NRND parts for a single project outside of somewhat elevated possibility of getting fakes.

For SMD diodes look, you can solder them as well if you have a double-sided plated hole 0.1 inch step prototype board, and still get a nice tidy circuit. Actually, you can even use 0805 SMD resistors on such a board conveniently. If you insist on through-hole, you can use P600KE30CA or P1.5KE30CA or 36V versions, too. SS14 are equivalent to the ubiquitous 1N5819.

If you can't get DIP8 version, a solution is to use SO8 to DIP adapter board. SO8's are still very easy to solder, get a syringe of flux paste and also knife tip soldering iron works the best.

As for MOSFETs mentioned, looks these will work well, with rather high margin by the current, the only thing I'd be a bit cautious, from the datasheet they dislike to stay in the linear region for long, having heavily limited safe operation area for DC. That may mean that if trying to slow switching
down by increasing gate resistors (to reduce interference etc) you may run into a nasty surprise of blown transistors even if drain current is much lower than transistor maximum by quite a large margin! (209A vs 20A).  510 Ohm gate resistors will give better than 20 microseconds time constant, though, quite far away from troubles. Probably something like 1-2K Ohms at gates is still OK.

I would not expect any issue other than the total average power dissipation. Because we either have no switching transients and permanent 20A load, or 100Hz switching and the 20A flowing just half of the time, for the total power dissipation won't be that different it means the thing to be designed for full ON (so the 1mOhm at 25degC, so about 2.5mOhm at maximum Tj means 1W at the maximum Tj, so not requiring any complex cooling for a pair or triplet of TO220 or the SMD variant cooled into the copper plate on the PCB). And the switching then allows transients to take up to 100's us, so still not that hard to keep overshoots in check.

With MOSFET the normally published "drain current rating" by itself rells not much. It has no direct meaning, it is just a calculation, what current would need to flow through the drain to make the silicon temperature to reach the maximum Tj rating while thecase is held by a mystical ideal infinite heatsink at exactly 25degC. In real life you will never have anything close to an ideal heatsink and the temperature would be anything but the 25degC.
So you need to reevaluate the thermal loading for your case anyway. And for low frequency switching, the practical usable limits use to be in the 1/10 of that...

That is why for your 20A DC application you need to look for about 1mOhm total, so the power dissipation will be in the 0.5W range, so you won't need any huge heatsink. With most modern switching elements in switching duty you can easily parallel connect them and count on the loading to spread among them rather evenly, as the ON state drop uses to be dominated by resistances of the silicon structure, which tend to have a positive temperature coefficient, so tend to spread the power dissipation evenly among the elements silicon area and across multiple parallel devices.

For the SOA and switching speed: At this 14V the SOA is practically limited by just the power dissipation, the mass of the device and the maximum junction temperature. With 1mOhm on a 60V (that has the implication how big the silicon generating the heat is, so what is its mass) device with 20A resistive (the inductance effects become small at those long times) loading at 14V battery you are easily in the 1..10ms area, so with that your problem won't be single pulse SOA, but to keep the overall switching power dissipation in check.
The practical slopes for this 100Hz I would see in the 10..200us ballpark, there we are talking about 0.25W on switching transient related dissipation, so within 0.5W total power dissipation per device, manageable without any heatsink.

Here is an update to my circuit with some changes :

The hysteresis loop in the protections comparators - It took me a full night's sleep and look at the circuit the next day, to give myself a facepalm. Indeed, i made the feedback to the slow signal instead of to the reference. This is corrected now

Notice that the logic polarity is inverted, so the extra inverting transistor is now on the other comparator



Driving the MOSFETs directly without a gate driver IC :

The higher resistance in the MOSFET gates does slow down somewhat the current rise and fall times, reducing the transients and interference generated

However, in such case care must be taken to never drive the MOSFETs at some intermediate voltage, which will cause huge power dissipation. In the case of the MOSFETs i listed, any running in a high resistance mode (V ~few V, I ~20A) is out of the SOA (Safe Operation Area) even for just milliseconds

Such condition may happen when unintentional resistances in the circuit make a voltage divider with the resistors in the gate drive circuit

In this circuit, a risk comes from the "Full power" switch. It is a mechanical switch, and oxide layer may appear on the contact someday. If it happens to be between few 100s Ohms to few kOhms, it will form a divider with the resistors "to ground" (in my implementation) or "to the IR2153 output which is off at the moment" (in RRK's circuit)

If a gate driver IC is used (like in my first circuit), the IC guarantees proper driving of the MOSFETs in clear on or off, even when it gets some intermediate voltage at its input

In this circuit provided, with the gate driver removed (not because it is a good design choice, but to show how it would be implemented in this case), i pass the switch signal through a comparator to clear it



In the lamp circuits, some dampers for transient suppression are added

The concern there is what happens if the capacitors (connected to a very high current circuit) fail shorted, which is what capacitors normally do

In circuits built with SMD components, the main mechanism of failure is when the PCB is bent, the capacitors crack and single layers of the capacitor short out, creating a short circuit that can get the capacitor glowing hot but not always pull sufficient current to blow a high rating fuse

Some SMD capacitors have been developed which have somewhat flexible terminals, and some where an intended cracking region is provided where only one pole is present (Vishay VJ OMD series, see description in their datasheet)

The circuit i provided is fully Through Hole, so is insensitive to PCB flexing. However, components can and do fail for any number of other reasons too

Medved's suggestion is 2 capacitors in series. This reduces the chances of fault (except maybe in the case of SMD, where a single PCB flexing event may just crack both capacitors), though there always is a possibility of both failing. (Especially so because if one failed there won't be any signs of it, so the circuit can remain in use for years until the other fails at another time)

I put an RC instead, with the R being a fusible resistor. Such resistor does both limit its own fault current, and blows cleanly with minimum collateral damage, as long as it is overpowered by high enough margin to "blow" and not "cook". (Many other resistor types will also blow fairly cleanly if overpowered by high enough margin, but this resistor type is designed for this). In this case the circuit will remain without damping, but i dont think this is a safety issue



Undervoltage protection have now a faster shutdown for fast falling supply voltage. I added this seeing that the SOA for the MOSFETs at intermediate state (for the load of those lamps) is in the ms range or less

Overvoltage protection is added too



The protections circuits i provided can be used with any of the circuits shown in this thread including mine, Medved and RRK's. (With changes as needed in each case)



I might build a real life test of the circuit i designed to test if you ask, though i will only have time for this probably a week or two away from now at best

The full power mode requires an undervoltage protection too, that was the point of the RRKs trick. In place of that switch I would just use a small pnp transistor, controlled via a schmitt trigger and filter from a low power control circuit coming from the switch itself.
The RRKs circuit could also be modified to use two PNPs, schottky diodes in their emitters and base resistors - so each PNP feeds transistor B from predriver A and vice versa. A diode less voltage drop in the gate circuit, otherwise the same.

For the schmitt trigger, I don't think there are necessary so many extra inverters, just flipping the comparators and using the feedback directly from the output would do the same job (with the hysteresis feedback to be one stage shorter, which is always a good thing).
With this fault logic, it is advantegous to utilize the open collector nature of the comparators. The fact that one common output controls multiple hysteresis loops is not any problem - just the one that is actually toggling needs to have the hysteresis effective, the rest does not matter once the output is overruled by something else.