ATX to TEC Power Converter and regulator build

[Uncle Jimbo]

Hehehe, good to see we've cleared all that up.

Its all good when we learn something.

So back to the converter: ya guys think there are enough 34v TECs out there that i should put that into the converter's operational range?

Glyph -
Just sticking to standard 'high power', some of the bigger TECs have Umax of 40V at 27C... that can go to 45V at 50C. The use of more elements increases the power rating at lower Imax - so a 288 element unit capable of 560W transfer draws only 20A.

Of the TECs commonly available in qty 1, 36V Umax at 20A should be plenty.

I like the idea of going to multiple pass elements - You might also get away with multiple inductors and reduce the current requirements there also. But multiphase is probably an unnecessary complication for TEC power since the power doesn't need to be all that smooth.

[Glyph]

I see, so i guess i'll stick to the 40v max operational range.

I agree that multiphase is a little design excessive. I also haven't found any multiphase synchronous boost ICs cheap enough to implement it.

[Glyph]

Here's the latest schematic i've got going.

To smooth out the voltage ripple inherent to boost topologies at these voltages i've added another Pi-filter circuit.

This is still not the final version. I'm going to add a thermistor and constant temperature control features. I'm also going to add some LEDs.... afterall, NO project is cool without some LEDs...

Current design allows for a 5v-40v operational range at 600 watts. 30 amp output max. although max current is lower at higher voltages due to wattage limitations.

Efficiency is predicted to be better than 90%

Current cost for the whole unit is $80, the extra features will probably add $5 or so.... I know the $80 total cost will turn off most people but unless i reduce the operational range or kill the efficiency $80 is the best i can do. But keep in mind this is with Canadian dollars and canadian parts loaded with our excessive canadian taxes (in case you didn't realize, i'm canadian). Individuals from other countries might be able to obtain the parts for cheaper.

extra features coming soon.

comments, criticisms (except those comparing me to ebay prices for meanwell) and design suggestions welcome.

[zipdogso]

I was going to see how much it would cost here in the UK BUT...

I need more info on:-

mosfets
IC chip
Diodes
and C13 10mf is a bit of a mystery

The rest of it only adds up to a few quid so the IC and the mosfets are the main bits.

[HDCHOPPER]

glyph it's beautiful

[zipdogso]

glyph it's beautiful

Maybe but bear in mind it is not finished !!

[Glyph]

I was going to see how much it would cost here in the UK BUT...

I need more info on:-

mosfets
IC chip
Diodes
and C13 10mf is a bit of a mystery

The rest of it only adds up to a few quid so the IC and the mosfets are the main bits.

Absolutely, once i hammer out the exact component values i'll post a complete bill of materials including serial numbers and part numbers.

Some guidelines:
The mosfets are from NXP with a part number of: PHB191NQ06LT

These just happen to be ones that i found that fit bill. you may substitute it with any N-Mosfet with at least 50v drain-source handling and an Rds(on) of less than 5 miliohms. That's 5 *MILLI*ohms. The ones i'm using are D2Pak but if ya make your own PCB i guess you can go TO-220.

the chip is from Linear Technology and is the LTC3814-5. Its only available in surface mount.

D1 is a simple fast-recovery diode rated for at least 80v. amperage about 1amp should be plenty

D2 however is more demanding. It should be a schottky diode with a blocking voltage of better than 50v and a current rating of better than 10amps.

The ripple capacitors (along the top) C4,C8,C9,C10,C11,C14,C15,C16,C17 and c12 (yes i should have labelled em better.... i'll fix that later) must be low ESR ceramic capacitors capable of taking better than 5 amps ripple current EACH. otherwise they WILL overheat. you may substitute with polymer caps. Voltage handling must be better than 50v.

Output capacitor C13 can be any aluminum electrolytic or paralleled bunch of electrolytics as long as the total capacitance is better than 10 milifarads and the esr less than 1 ohm. voltage capability must be better than 45v.

Another important note: Inductors must have a saturation current handling ability better than 100 amps. ONE HUNDRED AMPS. I strongly reccomend Large air core inductors using #10 gauge wire or molybdenum core inductors.

glyph it's beautiful

If you think its beautiful now, wonder how it'll be when its finished

[zipdogso]

The mosfets are from NXP with a part number of: PHB191NQ06LT

Hmmm....Searching n-mosfet on my recent compilation of UK electronic suppliers came up nil. A google for NXP got their suppplier as Digikey so I can order through Digikey UK (takes weeks as it's a non-stock item.) but that is only for surface mount which I am not familiar with. Part PHP191NQ06LT is the more familiar TO-220 package but the first price break is 1000 units !! over £900. Whooops....

[HDCHOPPER]

MOSFET N-CH TRENCH 55V D2PAK - PHB191NQ06LT /T3Order 'mosfet n-ch trench 55v d2pak - PHB191NQ06LT /T3' online from Digi-Key. Manufactured by NXP Semiconductors (VA). Digi-Key part number 568-2188-6-ND.
parts.digikey.com/1/parts/1506699-mosfet-n-ch-trench-55v-d2pak-phb191nq06lt-t3.html - 22k - Cached - Similar pages

[Glyph]

If you're bent on through-hole then i suppose the FDP050AN06A0 from fairchild is also a good choice.

EDIT: In a few hours, i'll post the design that includes thermistor control

[zipdogso]

HDCHOPPER

mobile phones have got you into bad habits.... read the whole post on a thread not just the first line !!

....I know Digikey supply them !!

[Glyph]

Are you able to get the fairchild part where you are?

[zipdogso]

Yeah through Digikey I can £1.23 each.

If you're bent on through-hole then i suppose the FDP050AN06A0 from fairchild is also a good choice.

Not bent on through the hole just never used surface mount and since I tend to use stripboard i'm figuring surface won't work. (Plan to do a proper board for this though - I assume you really need the width of the tracks ! fed up with soldering wire on the bottom.)

[HDCHOPPER]

[Glyph]

A new level of coolness has been reached! This design incorporates ONE WHOLE LED!!!

Actually, the LED on the left does something useful: It tells you if the unit has reached the voltage you set. This is good for diagnosing a failure somewhere or if the unit is receiving enough current.

As promised, the thermistor temperature control circuit has been added. The circuit connected to the bottom of R8 measures the temperature through the thermistor R6. R7 controls the temperature set point. The unit will then adjust the voltage up or down until thermistor matches the temperature set point. The unit will overshoot its voltage mark at first but (hopefully) will readjust and eventually settle onto the right voltage. If the rate of change is too fast causing undampened oscilation, increase the values of R13 and C18. If the change is too slow, decrease those values. Since everyone's TEC setup and feedback rate is going to be different, i can't really make a "one size fits all" value for the circuit. So be aware that R13 and C18 readjustment might be necessary. (changing R13 to a trimpot might be helpful)

If you guys don't need thermistor regulation, just cut off the circuit after R8 and ground the pin.

R5 sets the maximum "ceiling" voltage the unit will go to. the "floor" voltage depends on the mode, around 5v for low voltage mode, and around 17v for high voltage mode. This floor can't be changed without radically altering the circuit.

Therefore, the thermistor regulator will move its voltage target between the floor and the ceiling. I realize this does not give you the ability to "turn off" power to the TEC, it will always run at atleast the floor voltage. Personally, i don't think this is a big issue but if its a problem tell me. I'll see what i can do.

The value of the this thermistor regulator circuit over others is that its a true constant-temperature controller rather than a chopper or bang-bang controller. You get much less thermal-expansion/contraction stresses from such a setup and the current is drawn at a constant rate rather than suddenly spiked or loaded. This in turn is better for the ATX you're drawing it from. Also, with constant power draw, with lower than maximum loads, the converter won't be hitting its ceiling voltage and stressing its components or the TEC, this lasts longer and is more efficient. Bang-bang controllers can only alternate between ceiling and floor voltages, leading to higher stresses, lower efficiency, and shorter device life.

if you cut off the thermistor regulator then R5 will directly set the output voltage.

D3 must be a silicon diode, do NOT use a schottky or germanium diodes. Its function is to clamp the feedback voltage. Its forward voltage drop should preferably be 0.7v-0.8v

IC1 could be any rail-to-rail op-amp. just make sure it works to 12volts.

The unit is designed for a 10k NTC thermistor. This can be changed if you also change R7 and R9 to be around 10 times the thermistor resistance.


Each transistor pair (Q1:Q4, Q2:Q5, Q3:Q6) gives it 200 watts capability. So with three pairs its 600 watts. If you don't have TECs that need that much wattage, you may remove a pair or two to save costs.


I realize i never mentioned how to select the mode: Attach positive pin of the TEC to the "OUT" line. Now, for low voltage mode, connect the negative pin to the 12v rail. For high voltage mode connect the negative pin to the GND rail.

This works because the unit actually only has the high voltage mode, but we cheat by subtracting 12v by connecting it to the 12v line. This is exactly like the infamous "7v fan mod".

My thermal calculations say this unit *IS* going to heat up. By how much is anyone's guess. If you are building it, make sure you got plenty of airflow over the mosfets. Good heatsinking is reccomended.

R2 sets the device frequency, its currently around 500kHz. Increasing R2 will decrease the frequency. This will raise the efficiency a *tiny* bit and decrease heating of the mosfets but it will increase heating of the inductors and capacitors. I don't know yet where exactly to set the frequency for maximum performance.

This unit REQUIRES a load to be present. If this is run open circuit it *WILL* fry. Make sure you're drawing at least 1 amp with whatever load you're using before turning this unit on. TECs are excellent loads. For testing purposes a (very) high wattage resistor (about 10 ohms) works as well.

==================

I think I got all the features i want in it now. I'll probably spend my time refining the design and hammering out the calculations for the exact component values. The ones listed are a rough guide. I don't see anything blatantly wrong with the design but i haven't tested it yet.

Copy it if you'd like just do so at your own risk.

Cost, in Canadian dollars, comes to around $80-$90. But as i mentioned before, we pay high taxes on everything so you may be able to build it for much cheaper.

Comments, questions, or more features you'd like seen are greatly appreciated.

If you are building your own, tell me how much the parts costs in your area and which companies you're getting them from.

[Glyph]

This is the Cheapo version of the original circuit. It chucks out all "extras" and sacrifices some efficiency to give a very cheap circuit that can be built for $30. if you use homemade air-core inductors. (very easy to make btw...)

Because its only 80%-90% efficient you'll need to heatsink the mosfets and the 30-amp schottky diode. Voltage ripple is also much larger here, but if you only intend to run below Vmax (70% as UJ says) then that shouldn't be a problem and TECs can handle large ripples.

operational range is the same as the original. 5v-40v

This circuit is intended for those that pinch every penny. The idea is that you could cobble together some cheaper ATX supplies (or any high current ~12v supply you happen to have), plug them all into the converter and get the voltage range you want for your TEC. This is not as elegant as running directly off the main ATX, but cheap ATXs + cheapo converter is way cheaper than a brand new meanwell.

as with the original, each mosfet needs to be less than 5 milliohm Rds(on) and have a Vds greater than your TEC Vmax. Since each mosfet gives 200 watts capability; for additional cost savings eliminate the extra ones if you're using lower wattage TECs .

you could shave off another $6 if you directly solder the power leads to the board rather than use terminal blocks like i do. Ofcourse this ruins your ATX but if you're using extra ATXs anyway for this....

[HDCHOPPER]

exellent ! thank you Glyph

[Glyph]

Sorry it took so long to actually build and test this. But its done now.

I changed the controller to LTC3703. other than that, the basic idea is still the same.

Mosfets got hot at high power levels but the circuit didn't fail after extended use. I don't know about the longterm longevity of the circuit, but if you're worried just mount it in the path of a high air-flow region in your case.

I haven't measured efficiency... none of my meters accurately measure 30 amps . i'll get around to it someday.

the Sv1 component is an expansion header, you can remove it and connect R8 directly to ground. I put it there because i plan to add on a "dew point follower" circuit that automatically adjusts cooling to match the current dew point temperature for all seasons.

R10 is a current limiter. I didn't actually implement it in my unit since i wasn't too worried about current limits. Omit it if you aren't either. I just left it in the schematic for completeness.

Unit is very small. 8cm x 8cm x 4cm.

Without further ado, here's the circuit. Copy it as much as you like, pass it off as your own work even, i don't mind, I declare it public domain.

[memolipd]

hmmmmmm. That looks way too complicated for me and I've got a physics degree!

would you be willing to make one and send it to me in England?

also is there any chance of making a fan-header controlled version of it? that would be very useful....

[Glyph]

fan header controlled? i haven't thought of that... good idea!

back to the drawing board for me i guess.

As for making and selling these things... while the total parts cost is cheaper than a meanwell. My labor costs make the sum total cost higher than a meanwell. a meanwell is also backed by a warranty and has proven track record. i don't think people are willing to pay for an "experimental" device without warranty or remedy in case the device fails or worse... fails and causes damage to the computer its attached to.

If you want to be part of the experimental process, PM me. But in terms of a builder/customer thing where i'm selling you a product, i just can't absorb the liabilities if something should go awry. Maybe wait around a few more months when i've fully tuned the design to perfection.

I did this primarily as an experiment and proof-of-concept, to see if i could really make an ATX-TEC voltage booster/converter. It is a niche market that i'm surprised the few companies making TEC products have not yet attempted. I've proven it can be done, and it can be done cheaper than existing solutions (provided ofcourse they make more than 100 units to take advantage of economies of scale).

Hopefully some company engineer dude is reading this and decides to release the same product. (Please send a couple of free units if you are )

[memolipd]

fan header controlled? i haven't thought of that... good idea!

back to the drawing board for me i guess.

I think having the TEC voltage (high current) match the fan headers output voltage (low current) would be VERY useful. This would allow us to simple plug the control unit into a fan header which not power, but control the TEC voltage DYNAMICALLY. So with the addition of temperature sensors could turn the TEC and thus electricity bills down when the system is idling. This is much more useful in my situation where the TEC is acting as a water chiller and turns on when the water reaches a certain temp.

So thats why i'd much prefer your solution instead of a stand alone manwell. As the manwell cannot be dynamically load controlled.

[Glyph]

i see your point.

if you scroll up to post #40, i have a temperature contol circuit. It is still directly compatible with the newer converter. Its manually controlled, but once set will continually adjust power to achieve constant temperature cooling.

advantage is that even if something weird happened to your computer the temperature controller is independent and will continue to function. If you externally power the circuit then it'll give you control even when the computer is off.

Disadvantage ofcourse is that its not elegantly computer controlled

I'll design the fan header tonight. I should have a viable circuit by tomorrow.

Quick question though, do you already have power supply for your TEC? If so, it might be MUCH cheaper to go with Uncle Jimbo's controller. My converter has the voltage boost function which is why its so expensive. Uncle Jimbo's controller only cuts voltage, it doesn't boost it. If you already have a power supply then that might be the cheaper option. However his option is not fan header controlled.

I can still design for you a fan-header controller without voltage boost. you just insert it into your current TEC power supply. And it would be simple enough for you to build yourself.

[memolipd]

i see your point.

if you scroll up to post #40, i have a temperature contol circuit. It is still directly compatible with the newer converter. Its manually controlled, but once set will continually adjust power to achieve constant temperature cooling.

advantage is that even if something weird happened to your computer the temperature controller is independent and will continue to function. If you externally power the circuit then it'll give you control even when the computer is off.

Disadvantage ofcourse is that its not elegantly computer controlled

I'll design the fan header tonight. I should have a viable circuit by tomorrow.

Quick question though, do you already have power supply for your TEC? If so, it might be MUCH cheaper to go with Uncle Jimbo's controller. My converter has the voltage boost function which is why its so expensive. Uncle Jimbo's controller only cuts voltage, it doesn't boost it. If you already have a power supply then that might be the cheaper option. However his option is not fan header controlled.

I can still design for you a fan-header controller without voltage boost. you just insert it into your current TEC power supply. And it would be simple enough for you to build yourself.

at the moment my TEC is ran from my secondary atx powersupply. It is activated by my programmable fan controller which switches a relay when the water goes above a certain temp.

it would be nice if i could relate the low current fan output to a high current atx supply, dynamically of course.

Its a bit above my electronics ability as the circuits need to be seperate, i know Current mirrors exisit but im looking for a voltage mirror :S I'll check out Uncle Jimbo's one to see if i can make it dynamic.

[Glyph]

As promised, here is the fan header controller plug-in for my converter.

Just cut the "jumper" between pin 1 and 2 of sv1 on the ATX-TEC converter and plug in sv1 of the following circuit. sv2 plugs into the fan controller. since it uses an optocoupler it'll be able to link to almost any fan controller and draw very little current.

R2 compensates for the non-idealities of the various components. Adjust it so that the fan controller runs at 50% duty cycle at the desired temperature for best response.

(first set R2 to half way point, then adjust the fan controller to reach the desired temperature, then readjust R2 while measuring the duty cycle until it reaches 50%.)

Still want the "non-booster" circuit version? might be cheaper and easier to build.

[zipdogso]

As some fan headers are a bit iffy and prone to fail on some boards it might be considered wiser and perhaps even easier to take a take a 12v line straight from a molex.

'course in the end the results the same so it's personal choice.