A niche chip (LT4320) for mains powered AC -> low volt DC "linear" based psu ?

[johansen]

a zero loss diode bridge would not be very useful for high power high voltage circuitry anyway.. because you would need a high volume inductor after the bridge to keep the power factor within reason.

there are a few diode-less pfc topologies available, most produce a lot of common mode noise.

there is a topology i like but i haven't developed a controller for: see attachment

[Phaedrus]

Well right now the most common topologies in PCs go like this:

Common Mode Filter --> Bridge Rectifier --> Interleaved boost converter PFC --> PWM --> Transformer --> Secondary

The most common PWM topologies are either Double Forward (for 80%-87% efficient PSUs), or LLC Resonant Half or Full Bridge (for 87-92% efficient PSUs). There's also some high-end ZVS Phase-Shift Full Bridge units, and a popular Active Clamp Forward platform.

Anyway, efficiency is an important marketing feature, and the big efficiency certification is 80PLUS. Currently the top of the line runs 80PLUS Platinum, which requires 89-92% efficiency at 115VAC at room temperature. There are about a half dozen ODM vendors who can currently reach this and still maintain "enthusiast grade" voltage regulation and ripple.

But 80PLUS is working on a new standard, called 80PLUS Titanium. The spec isn't finalized, but word is the efficiency at 50% load will have to be either 94% or 95%. That's 2-3% that we need to gain over current designs, most of which just clear Platinum.


We've adopted the DC-DC secondary topology which netted the 1-2% boost needed to hit 80PLUS Gold back in the day. It's hard to improve that much more, considering we're already using synchronous rectification both for the main +12V rail and for the minor rails. Assume we can boost the efficiency of the DC-DC modules, call that a 0.25% gain.

The transformer and other magnetics efficiency is improving slowly but surely. We can probably get another 0.5-1% there by end of next year.

With improvements to layout, connectors, shortening power traces, using thicker copper, etc, we might get an extra 0.25% there.

So we need *at least* 1% more gain. PWM topologies are currently close to maxed out, it's unclear how much more we can gain there. So where do we have obvious losses we could cut? The bridge rectifier and the PFC diodes.

Replacing the diodes in a boost converter is a classic power electronics problem that has yet to be solved. I don't think anyone's made one that doesn't nuke itself yet.

So that leaves the bridge rectifier. The current trend is to go bridgeless PFC, which replaces the bridge rectifier with active MOSFET rectification. No one has built an adequate analog controller to do this yet (the LT4320 is the closest I've seen), so currently you have to use MCU control. This normally ends up with the entire PSU being MCU controlled, instead of analog, which is a $30-$50 increase to BOM cost.


If someone could figure a way to do it analog, without requiring a digital control loop, for $20 BOM or less, that would be a killer advantage in the market. If we could get 0.5% efficiency there that would put us in reach of 80PLUS Titanium certification at a lower cost than the competition.

[digsys]

The transformer and other magnetics efficiency is improving slowly but surely. We can probably get another 0.5-1% there by end of next year.

I find it interesting that they can't get better? We've had MPPT chargers (which have a much more difficult task) at app 97-98%
for many years. Our current design, by one of our members, runs at 98.5% at 4:1 In/Out range over most the power range.
It's a resonant synchronous design. They even have resonant controller chips out now. Mainstream doesn't seem to be interested in
efficiency unless they're forced to.
 

[Phaedrus]

The transformer and other magnetics efficiency is improving slowly but surely. We can probably get another 0.5-1% there by end of next year.

They even have resonant controller chips out now. Mainstream doesn't seem to be interested in
efficiency unless they're forced to.

Because the end consumer isn't going to want to pay for it. Look how competitive the PC power supply market is. You have a never ending flood of cheap bogus rated chinese crap going for 20 bucks or so. Thats all most consumers care about price. In order to get high efficencies you need more complex resonent topologies and sync rectification. Even for those to hit high efficencies you have to use more expensive semiconductors.

Most PC supplies I've looked at are junk. They use the cheapest fets or IGBT's, cheapest rectifiers minimal if any line filtering, heatsinking that would be woefully inadequate to deliver rated power given the crap semi's they use.

Actually there's quite a large market for quality power supplies. The gamers, enthusiasts, and Litecoin miners all want quality kit, and will pay for it. The cheap $20 crap may move in volume, but it's not particularly profitable most of the time. The $50 - $150 segment is extremely competitive and usually quite profitable. In the US, 650W, 750W, and 850W power supplies that are 80PLUS Bronze (82-85%) or 80PLUS Gold (87-90%) sell very well; total market size of 25,000-40,000 of those wattages a month. Average price near a hundred bucks.

Crack open one of our CoolerMaster V850 power supplies and tell me that's cheap crap. It may not be lab grade, but it's solid engineering and good build quality.

[M. András]

if the damn psu is as reliable as it can be like those decades old hp  lab power supplies then i will gladly pay a fair price for it, not like the cheap 50-100bucks cheap 500+watts shit what cant even deliver its rated power and when i dies kills the attached things too.

im wondering if this chip could be used with pretty low rdson fets and its power supplied externally it doesnt need much so a small bridge rectifier from the main ac feed maybe? high power losses would be minimal too

[tszaboo]

Crack open one of our CoolerMaster V850 power supplies and tell me that's cheap crap. It may not be lab grade, but it's solid engineering and good build quality.

wow, they are using double sided PCB, which is not from paper, and there are even smd components on a small board. They are only 20 years behind todays technology, not 25. To be honest, I really think that PC power supplies are lagging behind. A modern design would use planar magnetics, moderate-high switching frequency, multiple phases for DC-DC, SMD mosfets, power blocks like drMOS, and pulse skipping or other intelligent driving to increase efficiency for light loads. Also most PC power supply is standby power hog, because the lack of secondary circuits for that. Well I guess these will be standard in 20 years or so.

[digsys]

im wondering if this chip could be used with pretty low rdson fets and its power supplied externally it doesnt need much so a small bridge rectifier from the main ac feed maybe? high power losses would be minimal too 

Very unlikely. If you've ever looked at designing a FET bridge, there's a very tight relationship between the AC side and DC side.
Supplying a separate low VAC for the charge pumps, even IF you could line up phases etc is unlikely worth it.
Basically, the "control chip" watches the VAC level coming in, compares it to the DC level, and once it becomes +ve, switches the FET.
On higher currents, it waits until the onboard Shottky conducts before doing so. You can see it in the waveforms.
If you don't do this, at the instant of switch-on, depending on cap storage = app INP/OUT impedance, the VAC can suddenly DROP
again, below DCV, in which case you now have current feeding BACK into the Transformer (in this case). It's only a short time, but
can cause huge current spikes / losses. I'm about to make up a bi-dir current sense and really get a better picture.
Having made these devices in the past (using optos, external charge pumps and isolated high speed logic control), it's a really
very tricky design, which is why I was so impressed with it.
ALL it needs is a 0V side referenced shutdown pin (maybe Ver2?) and it'd then truly be AWESOME !! I've tried to email them with
suggestions, but it's impossible to find a tech feedback addy. You would have thought?
Why do we need this? O/Temp, O/Current, O/Anything SHUTDOWN ! The perfect oN/oFF control of up to 20-30A with a 10uA 3V signal !!!
woohooo heaven.

[Phaedrus]

Crack open one of our CoolerMaster V850 power supplies and tell me that's cheap crap. It may not be lab grade, but it's solid engineering and good build quality.

wow, they are using double sided PCB, which is not from paper, and there are even smd components on a small board. They are only 20 years behind todays technology, not 25. To be honest, I really think that PC power supplies are lagging behind. A modern design would use planar magnetics, moderate-high switching frequency, multiple phases for DC-DC, SMD mosfets, power blocks like drMOS, and pulse skipping or other intelligent driving to increase efficiency for light loads. Also most PC power supply is standby power hog, because the lack of secondary circuits for that. Well I guess these will be standard in 20 years or so.

It's a 300KHz LLC resonant full bridge. We use SMD mosfets for secondary rectification for the +12V rail, mounted on the bottom and heatsinked to the housing. We use pulse skipping to improve efficiency at light loads; at 10% load you can expect 85-87% efficiency (depending on if you have the 700W, 850W, or 1000W). Planar magnetics aren't cost-effective for us, and DrMOS has proven unreliable for us with few advantages over discrete components.

You could buy something like this twenty years ago--in a bespoke device costing $10k+. Now you can get it in a consumer device for $80 - $150. And still they bitch?




The snobbery of some engineers continues to astound. 

[necessaryevil]

I agree it is indeed amazing what you get for your bucks buying an ATX psu. I guess that making a more efficient ATX supply is not more cost effective (the reduction in electricity costs is not enough to pay back the higher initial costs).

But back on topic. Would it be possible to make a batch of those pcb's (which will make it cheaper)?

[tszaboo]

...

It looks like it has more features than I anticipated. Probably they did not take fully apart because of the SMD on the bottom.
May I ask why was DrMOS unreliable? I'm recently looking into the possibility use them in industrial applications. The only downside I see right now, that the internal MOSFET sizes are not optimal for my output voltage (I need about 2-3-4 volt output, not 1.2-1.5V they are designed). Can you share some background information?

[Phaedrus]

...

It looks like it has more features than I anticipated. Probably they did not take fully apart because of the SMD on the bottom.
May I ask why was DrMOS unreliable? I'm recently looking into the possibility use them in industrial applications. The only downside I see right now, that the internal MOSFET sizes are not optimal for my output voltage (I need about 2-3-4 volt output, not 1.2-1.5V they are designed). Can you share some background information?

We haven't used them in our PSU designs, but I worked for a company that used them in motherboard VRMs. The problem is that DrMOS seems to be very sensitive to reverse voltage; just a little bit and it'll let the magic smoke out. You have to be very picky with your inductors, and don't trust the peak current ratings. The motherboards using them had higher than average DOA rates, and very high returns with blown VRMs 6-18 months down the road. Correlating with higher CPU wattage of course, but the combined current of the DrMOS should have been more than adequate.

They also aren't good voltage for ATX PSUs.

[megajocke]

Why do we need this? O/Temp, O/Current, O/Anything SHUTDOWN ! The perfect oN/oFF control of up to 20-30A with a 10uA 3V signal !!!
woohooo heaven.

But, you still have the body diodes of the MOSFETs, don't forget those! Removing the gate drive from the FETs would just turn your nice low-loss bridge rectifier into an ordinary high-loss one, and the magic smoke would be released soon if the cooling is designed for the low-loss mode of operation.

To get shutdown functionality you need more switches anyways so the functionality could be made completely separate, using just a single FET on the DC side for example.

[digsys]

  But, you still have the body diodes of the MOSFETs, don't forget those!    

DOH !! :-)  I wish we could "edit" or "delete" posts :-) ..... lol

[necessaryevil]

@NANDblog, Pheadrus,

Any recommended docs/books about ac-dc topology for the novice? I'm quite interested.

[ResR]

I made the PCB design for that chip also with all through hole components, the pcb is 74x69mm in size.

[digsys]

  I made the PCB design for that chip also with all through hole components, the pcb is 74x69mm in size. 

Great. What current are you expecting to run yours at? Have you used it in a project yet?

[Phaedrus]

@NANDblog, Pheadrus,

Any recommended docs/books about ac-dc topology for the novice? I'm quite interested.

I learned from Wikipedia, Texas Instruments white papers, and talking to engineers. Sorry, can't help with books.

[BravoV]

I made the PCB design for that chip also with all through hole components, the pcb is 74x69mm in size.

Hmmm.. DIP 8 package, where did you buy that chip ? Just checked LT's shop, the LT4320IN8#PBF or LT4320IN8-1#PBF for the dip 8 still not available yet, and FindChips search results too, no luck.

Although you already have the big cap on the top at your layout there, read the datasheet at "CLoad Selection" section, its recommended to have small caps to complement that big cap above, that are placed really close to OUTP and OUTN pins, probably at bottom right of your pcb layout there, close to the IC.

Edit : Don't forget PCB mounting holes at the four corners. 

[tszaboo]

@NANDblog, Pheadrus,

Any recommended docs/books about ac-dc topology for the novice? I'm quite interested.

I'm really not an expert on that field, sorry.

[digsys]

Although you already have the big cap on the top at your layout there, read the datasheet at "CLoad Selection" section, its recommended to have small caps to complement that big cap above, that are placed really close to OUTP and OUTN pins, probably at bottom right of your pcb layout there, close to the IC.   

One thing I've definitely proven with this IC is - you can't bust it !! It works perfectly well with no / excess compensation, poor layout,
too much / too little filtering etc etc You just can't stop it working ! BUT, keep that extra safety - add noise caps etc etc

[resistor]

Could this chip be used for AM demodulation as an envelope detector?  Or would the feedback mechanism interfere with the intentional amplitude variations?

[digsys]

Could this chip be used for AM demodulation as an envelope detector?  Or would the feedback mechanism interfere with the intentional amplitude variations? 

I have very little experience in this area - but note possible limitations : 600Hz max AC (ver2), 6VAC min, 100-220uF sec min.

[BravoV]

Fyi, this chip max out at only 600 Hz, not even 1 KHz, AM modulation is way-way higher isn't it ?

[resistor]

OK, true, the frequency limitation would kill this for actual AM radio.  But could it be used as an envelope detector for a frequency that was inside its range?

[peter.mitchell]

Could this chip be used for AM demodulation as an envelope detector?  Or would the feedback mechanism interfere with the intentional amplitude variations? 

I have very little experience in this area - but note possible limitations : 600Hz max AC (ver2), 6VAC min, 100-220uF sec min.

Just a quick question, have you tried it at over 600hz?
Maybe a siggen? or if you need some oomph behind the siggen, feed it through an audio amp first.