Efficient High Voltage Smoothing for Negative Power Supplies - Tips & Tricks"

[Ravaner]

Hi. I've built a (3 in fact) negative HV power supply continuouly adjustable from 0 to - 8 kV. It uses a CCFL inverter followed by a voltage multiplier and a regulation loop. In order to eliminate regulation peaks I would like to smooth final voltage by using a 10 nF / 15 kV cap (connected to ground) , but ... I not completely sure but it seems that I already did that in the past and the result was the destuction of several diodes ad caps of the voltage multiplier. Is there a condition to fill in order to properly realize this smooting ?

 

[yungman]

Put a current limiting HV resistor in series with the output, then take the voltage feedback from the output side of the resistor. The resistor serve as the current limiting and feedback from the output side of the resistor will compensate the voltage drop across the resistor. You just have one more pole to deal with but it is easy. The filter works even better with the resistor. We did a lot of HV supply design, never ran into any problem. Just pay attention on the closed loop feedback.

 

[cmb]

A conventional voltage multiplier already has such capacitance from output to ground, so the addition of further capacitance should not be the cause of blowing up components.

You say '..and a regulation loop..'. Care to expand on this? Maybe this is where the problem lies.

 

[yungman]

I don't think extra cap at the output will cause it to burn. But if the charge of the internal caps is being discharged through the diode without current limiting, that spell trouble. Those high voltage diodes are not as rugged as the regular diode. HV discharge can be very high current even though the capacitance are very low because of the amount of energy store due to the voltage. You are talking about hundreds of amp discharge in nano seconds. Even Transorbs need to have current limiting resistor. Notice I specified HV resistors that have long body to avoid creepage. If you use normal resistors, an arc will cause the charge to jump over the resistor body and you loss the current limiting effect of the resistor. This is very important in HV design.

Also one very important thing, you need to check to make sure there is no conductive materials( metal) left floating. Every bit of metal has to be tie to a known potential. Loose metal will be charged up in HV environment and start arcing to the next door. This is very hard to trouble shoot. Some PCB design actually ached rev number in copper and that is deadly in HV environment. You have to scrape them off or tie it to ground. Once it arc, all bets are off.

 

[cmb]

Incidentally, what diodes are you using? 'Regular' diodes (1N40XX) aren't quite quick enough for 40kHz off CCFLs. They just about do it, but may begin to overheat, so they'll look like they are working with no load, but flunk out with any further loading on the stack.

I've used them before in this application, rectifying CCFL, but for a teeny current (uA - feeding Geiger tubes), but it's not a clever idea if you need it to be robust.

 

[yungman]

High voltage diodes are very long, nothing of the 1N40XX!

When a high voltage arc, it take out big diodes. 1N4001 or 02 etc. is only 1A diodes, I am talking about burning 5W diodes from HV arc! We burned transorbs like P6KEXX or bigger! You don't let current avalanche...period. Everything is about current limiting in HV design, component do not survive arcing. Discharging a 5' RG59 ( 30pF per foot) is enough to burn a lot of things if inrush current is not limited. Take my word for this. Do a calculation to discharge a 500pF cap charged to 10KV in 1nS and you'll appreciate the amplitude of the instantaneous current.

 

[cmb]

Discharging a 5' RG59 ( 30pF per foot) is enough to burn a lot of things if inrush current is not limited. Take my word for this. Do a calculation to discharge a 500pF cap charged to 10KV in 1nS and you'll appreciate the amplitude of the instantaneous current.

Not sure I agree with the wording - to 'burn' implies heating to some 'hot' degree, but 10kV on a 500pF is 25mJ. For sure, voltage breakdown can be very damaging - and even more so if there are inductive effects that push the voltage higher still, but 'burn' implies more than just a few mJ!

 

[yungman]

Not sure I agree with the wording - to 'burn' implies heating to some 'hot' degree, but 10kV on a 500pF is 25mJ. For sure, voltage breakdown can be very damaging - and even more so if there are inductive effects that push the voltage higher still, but 'burn' implies more than just a few mJ!

In 1nS! When You put all the power in such a short period of time, power rating does not reflect the current handling capability as heat has no time to transfer out and ALL localized in one spot. Burn is burn, I don't really care about whether it smoke or not, they usually shorted out...still cold! You never see burn in form of smoke like the normal sense that it get hot, smoke and die.

I am telling you from years of experience designing 15KV power supply and high power HV supply and I learned all these the hard way. You'll be surprised that 5' RG59 can burn some serious stuff when arc.

 

[cmb]

In 1nS! When You put all the power in such a short period of time, power rating does not reflect the current handling capability as heat has no time to transfer out and ALL localized in one spot. Burn is burn

I respectfully disagree/don't understand your point. Whether nS or ms, X mJ of energy will raise whatever it is heating to the the same temperature. The specific heat capacity of a thing is its specific heat capacity.

Thermal conductance away from some wire bond/junction/whatever won't happen much faster or slower on that timescale, if we are talking mJ. That's why if you look at any component sheet it says that derating/thermal issues are related to pulse energy rather than current, once below a ms or so.

 

[Bob S]

Discharging a 5' RG59 ( 30pF per foot) is enough to burn a lot of things if inrush current is not limited. Take my word for this. Do a calculation to discharge a 500pF cap charged to 10KV in 1nS and you'll appreciate the amplitude of the instantaneous current.

Actually not. Discharging 5' of RG-59 (Z = 75 ohms, v/c = 66%) will produce a 15 ns long pulse. If it is charged to 10 kV, the pulse current is 67 amps. RG-59 has a delay length of about 7.5 ns, so the pulse length will be 15 ns long at 5 kV when discharged, and the output impedance is 75 ohms, so the instantaneous current is 5 kV/75 ohms = 67 amps.

See for example pages 7 and 8 in
http://web.missouri.edu/~kovaleskis/ApplEMandEP/Lectures/Lecture-5.pdf

Bob S

 

[yungman]

I respectfully disagree/don't understand your point. Whether nS or ms, X mJ of energy will raise whatever it is heating to the the same temperature. The specific heat capacity of a thing is its specific heat capacity.

Thermal conductance away from some wire bond/junction/whatever won't happen much faster or slower on that timescale, if we are talking mJ. That's why if you look at any component sheet it says that derating/thermal issues are related to pulse energy rather than current, once below a ms or so.

Instantaneous current will raise the temperature at one local spot. Heat takes time to transfer out.

If you work in IC design and testing, you'll see. In pcb, we cut trace with exato knife. But in IC, you can't do that. What we do is use needle probe and stick on both end of the trace that needed to be cut on the die and then just discharge a small cap. You can see the trace under microscope just pop. The trace supposed to be able to conduct a few amps but it will pop by the cap. The reason is it does not have enough time to transfer all the heat out at one time even the pulse is very short by high current.

As I said, whether you agree or not, this is real life. As I said, I have been in this field and anyone in the field know exactly what I mean. Go download schematic from Bertan or other HV supply manufacturers and you'll see they ALWAYS have a resistor to isolate the bridge or other rectifiers from the output and take the voltage to feedback to regulate the voltage.

Particular if you have voltage multiplier that have caps in between diodes directly, you discharge the cap through the diode, you burn the diode almost every time.

 

[yungman]

Actually not. Discharging 5' of RG-59 (Z = 75 ohms, v/c = 66%) will produce a 15 ns long pulse. If it is charged to 10 kV, the pulse current is 67 amps. RG-59 has a delay length of about 7.5 ns, so the pulse length will be 15 ns long at 5 kV when discharged, and the output impedance is 75 ohms, so the instantaneous current is 5 kV/75 ohms = 67 amps.

See for example pages 7 and 8 in
http://web.missouri.edu/~kovaleskis/ApplEMandEP/Lectures/Lecture-5.pdf

Bob S

You might be right, it might be only 7.5nS as it is not a round trip. but as you can see even the 5' coax give 67amp for 15nS and that is enough to burn a lot of things. Discharging a cap is in hundreds of amps.

 

[Ravaner]

Many thanks for all your replies. To give you more information the 3 voltage multipliers used 10 stages composed by 4.7 nF / 2 kV and double 1N4007 ( 2 serialized diodes ). One of the multiplier has to produce 15 µA and the 2 others something in the order of 1 µA. Frequency is 32.5 kHz and amplitude at the output of CCFL is around 800 V.

 

[cmb]

1N4007 ( 2 serialized diodes ). One of the multiplier has to produce 15 µA and the 2 others something in the order of 1 µA. Frequency is 32.5 kHz and amplitude at the output of CCFL is around 800 V.

Sounds like you may be suffering the same 'issue' I ended up addressing. I replaced 1N4's simply because they weren't fast enough. First-pass inspection looks like they work, but actually they are struggling.

Is that 15uA into the storage caps of the rectifier, or 15uA from the whole circuit - remember losses in the caps will suck up load, which you may be adding to by adding further external capacitance, taking the diodes over what they can cope with, as they will be struggling with the switching rate.

For such applications, I now tend to use 8kV diodes from a Chinese ebay seller [look up '8kv diodes'] that does 10x Trr=150ns 8kV diodes for $8 (or 80 for $25), shipped. Use those, or similar.

You have to think about Trr when rectifying above a few kHz...

 

[yungman]

I don't think Trr is a factor using in rectification purpose. Trr is in nS, which is much faster than the transformer and operating frequency of 32KHz. The Trr is only important in for arc protection. That is the reason we only use either Transorbs or Schotky diodes for clamping.

 

[yungman]

Many thanks for all your replies. To give you more information the 3 voltage multipliers used 10 stages composed by 4.7 nF / 2 kV and double 1N4007 ( 2 serialized diodes ). One of the multiplier has to produce 15 µA and the 2 others something in the order of 1 µA. Frequency is 32.5 kHz and amplitude at the output of CCFL is around 800 V.

For this low current, the more reason to put a HV resistor in series with each diodes to limit the current.

I never heard of HV supply burning diodes because of the filter caps, When things burn in HV supply, it usually cause by arcing. I don't know the detail of your design and how it is implemented, so I can't exactly pin point the cause. Read the post #4 where I explain things that can really blow stuff. Without arcing, HV and normal supply work the same, even with multiplier, they all work.

We designed 10mA 6KV supply from ground scratch and even wind our own transformer. We ran at much higher frequency ( hundreds of KHz) in order to lower the number of turns. We tested the ruggedness by putting as spark gap and let it spark and make sure it survive continuous arcing at the output. That is some power.

Remember the trickiest part is the floating metal around the circuit, they arc every time. The way to find those are to arc the output in the dark, you will see those floaters arc also. I even gave them the name "sympathetic arc"! Get rid of all the arcing, the HV supply is every bit as rugged as any other switchers.

Most people don't appreciate the amount of energy radiated out in such a short time of an arc, just look at all the computer freeze from the large EM radiated out.

 

[cmb]

I don't think Trr is a factor using in rectification purpose. Trr is in nS, which is much faster than the transformer and operating frequency of 32KHz. The Trr is only important in for arc protection. That is the reason we only use either Transorbs or Schotky diodes for clamping.

Show me a datasheet with the Trr for 1N4X, then. They are so slow it's not even quoted! These things are intended for 50Hz mains rectification, not multi kHz. You need 'fast' diodes for this. Not 'ultrafast', but certainly quicker than mains rectifiers.

 

[yungman]

PN junction by nature is at least in MHz region. No diode are too slow for 50KHz! I think you better show me the physics to back up why you think it would be too slow. I am not an expert in transit time, but it is much shorter than 20uS which you are implying about less than 50KHz. Even if you consider it as square wave that has 1,3 5...harmonics. The 3rd harmonics is only 5X32KHz=160KHz which is over 5uS period!

This is really the first time I even read question about diodes not fast enough in power supply application.

FYI, the limitation of a switcher is the core. You are limited to around 100KHz to 200KHz. If I have my way, I would have them run in 2MHz to cut the turns of the coil. . We can get over 10 volts per turn with higher frequency. The kind of diode has never been an issue. You ever design a switching or HV supply as a real product? I don't count experiment of voltage multiplier in school or read a few books as real experience!

 

[cmb]

PN junction by nature is at least in MHz region. No diode are too slow for 50KHz!

See http://parts.digikey.co.uk/1/1/516461-rectifier-1a-400v-do-41-1n4004-tp.html.

Trr quoted as 2μs.

So with a 50kHz frequency, 20μs period, switching at a rate that takes up 10% of that cycle would become problematic.

...Now put two in series!...

 

[cmb]

You ever design a switching or HV supply as a real product?

I've made a patent application for a prototype of a stackable supply module design for cheap manufacture that cost about $20 in parts and can drive up to 5mA at 20kV (variable, according to input power voltage), which I have been using for my HV work for several months now, showing no signs of any failure or fatigue, and can also withstand short-circuit, and also has a floating output so either side can be grounded, or lifted high to stack them.

I've also built a more complex HF PSU that can drive 4 channels simultaneously in quadrature at up to 6MHz and 500V, with a nominal pulse rating of 50A between them. It's not easy to do that, components just don't work like the data sheets say they should over 2MHz in power applications. Looks easy on paper, but the learning exercise is steep. You should try it out rather than leaving it as a hopeful wish, because the challenge is wholly different to running a few 10's kHz. I run 200kHz switcher circuits with BJTs! But MHz requires MOSFETs, and they need to be coaxed into being driven that fast by tuning the gate. Switching at zero load is straightforward enough with signals, but try it with a power/reactive load.

If we are comparing notes, what are your HV PSU achievements and what makes them unique?

 

[yungman]

I don't do 50A supply at 500V but this is not in this subject, we are talking about 8KV 20uA low power supply. I designed a 40A magnet controller that settle to 5ppm within one second and I appreciate how high current designs are as a lot of common sense gone out the window, even the sensing resistors have to be 4 terminals, we had to have water tubes solder onto the back of a copper plate to which we mount the big MOSFETs. Even the piece of material in between the bottom of the MOSFET and the copper block has to be special metal that is so soft it will conform to the shape of the surface when you torque the transistor down.

I don't design HV supply per say, I was a manager of EE in Charles Evans and Assoc and later bought out by Physical Electronics manufacturing various Mass Spectrometers including Time of Flight. I have a power supply engineer working for me, I came up with idea on a 10 mA 6KV supply that can recover in 100nS from a 10+volt dip and can withstand continuous arcing using a spark gap. I design the high speed compensation part and more on the transformer and the closed loop feedback of the supply.

Also I came up with idea of designing various 20KV isolation DC to DC converters that transport up to 200Watts up to +/-15KV deck to power a total computer control system floating on top. This I think would be very similar to your stackable supply. I use this to put a 24volt supply onto HV deck, I can easily make it a HV supply stack on top of another one. What I did was to sent 24volt to the top of the HV deck to power up an embedded system controlling a few PMT supplies all floated on top and stack the PMT onto the main HV supply. Idea is very similar to yours. We used RS232 to communicate to the HC11 floating on top to which it control the various HV supplies. The version I designed in 2004 used CAN bus fiber optic. We do closed loop regulation of the converter through fiber optic link and V to F and F toV link.The challenge part is again the transformer. I designed with very few turns and use 35KV high voltage cables as secondary winding. We did not apply for a pattern so you are safe if you do something similar. All the above were designed in the mid 90s.

I came up with design of very high voltage opamps that drive from -2500V to +2500V using MOSFETs and optical feedback. We used to buy so many Bertan supply, they even gave us the whole set of schematic of their 605 and 606?? and PMT supplies schematics and I did studied them. We opened Glassman supplies to look at their design also. I also designed a 5KV DC pulsing circuit with about 100nS rise and fall time and can operated at all duty cycle and frequencies. This is the one that cause the droop of the 6KV supply because I switch from ground to 5KV and a lot of instantaneous current it takes to charge the long coax cable.

I own a pattern of a special 32 channels detector design that incorporate passive HV components into a pcb that can detect sub nano second pulses. The challenge is to combine HV ( large spacing) with RF and ultra high vacuum compatible:


United States Patent 7,561,438

Liu July 14, 2009
________________________________________
Electronic device incorporating a multilayered capacitor formed on a printed circuit board
Abstract
An electronic device. The device comprises a printed circuit board, a multilayered capacitor formed on the printed circuit board, and a conductive strip disposed on a top surface of the printed circuit board. The conductive strip interconnects to the multilayered capacitor. The multilayered capacitor includes a plurality of capacitance plates and a plurality of dielectric layers wherein each dielectric layer is disposed between two of the capacitance plates. The printed circuit board further comprises ground plated sidewalls disposed about the printed circuit board. Each of the ground plated sidewalls extends from a top surface to a bottom surface of the printed circuit board.
________________________________________
Inventors: Liu; Yungman (Daly City, CA)
Assignee: Revera Incorporated (Sunnyvale, CA)
Appl. No.: 11/023,271
Filed: December 22, 2004
________________________________________
Current U.S. Class: 361/766 ; 361/760
Current International Class: H05K 1/16 (20060101)
Field of Search: 361/306.1-306.3,766,761-765,780-785,760 257/296-298 250/372,336.1,305,370.13

I published two papers in American Institute of Phisics, Review of Scientific Instruments:

http://rsi.aip.org/rsinak/v71/i11/p4144_s1?isAuthorized=no

You will never burn a diode by the drive of those HV transformers as the current is so low and the internal impedance is so high. It is the arcing that destroy components. You design all that what you claimed but you never run into what I detailed in the last few posts? Those are the major failure that I have seen and all the trouble shooting were exclusively done in the dark to find the arcing and the "sympathetic arcing" of the floating metal piece. Those are the ones that burn components and lock up the systems. You put a HV resistor in series with the output so if the output arc to ground, you limit the current surge. You should know exactly what I am talking about. This is so common in HV field and all the people that know HV know exactly what I am talking about!
I might not be doing every detail design on the HV supply as I am more the high speed pulsing, high voltage pulsing, RF, embedded and FPGA designer, but I came up with the HV supply idea and design the custom transformer to get ultra reliability and winding pattern to get total balance waveform between the push pull sides.

I am familiar and use 1N4007 diodes, NEVER heard of a silicon diode that can not support to this frequency. But as I said, one major thing I pushed is to raise the switching as high as possible, I was only limited by the core loss( loss of mu)...at least in the 90s. Diode has never been an issue. Our 6KV supply was running over 100KHz so I can wind Teflon insulated 26 gauge wires onto a 2" by 2" size core for 6KV supply. This was to guarantee safety and pass the CE. We never use BJT, always MOSFET.BTW, I went and looked up the data sheet of 1N400X from FairChild and another name. There is no spec on Trr. BUT they tested the capacitance of 15pF at 1MHz! You don't trust Digikey, it is only a parts vendor! I would not use 1N4007 because the body is too short and even if you put two together in the multiplier, the creepage across the body might present a problem. I would use a longer body diode.

 

[cmb]

yungman, I respect your background but I am left wondering if you've actually ever measured the behaviour of 1N4X diodes, or just presumed it.

It just so happens that I have on my bench a switching circuit I am currently preparing and testing (to drive a 300V, variable 1-100kHz, 4kW square wave generator design I am working on). So, hey, let's not argue about this, let's measure it like good engineers!

So I set the switcher up to drive a 40V pk-pk square wave (50% cycle) across a 500 ohm load, ±20V, and rectified it, half-cycle with a single diode. I then 'scoped the voltage directly across the rectifying diode.

I picked a 1N4005 straight out of a new bag of components, and the same again for a UF5408.

Here is the behaviour of the 1N4005 with the switcher set to run at 55kHz:

So you can see that it is not only very 'slopey', but if you count off the divisions you'll see that the 'off' time is 10us, whereas the slopey 'on' time is 8us. So that is a us of missing rectification, as well as the slow time to fully 'on'.

If we push higher still, the effect becomes very obvious. Below is comparing the UF5408 to the 1N4005 directly. With the fast diode you can see the clear, square shape of the desired rectified pulse, and you can also see consequent ringing (these were loose on the bench, on the ends of leads, so you should expect to see this if the switching is quick enough).

Whereas the poor 1N4005 doesn't even reach 20V during its 'rectification' period!...

Now consider feeding a rectifier with a 1N4005 like that above, but this time with 800V. How much heating in that diode is going to suffer, all because the diode isn't switching cleanly and fast enough? If that diode is half open and dropping 400V 10% of the time, how quickly will the junction and wire bond inside the package take to overheat?

So, the traces above show that the 1N4005 is taking some 6μs to get to fully 'on', and especially if you're rectifying 800V with any sort of load on, then 6μs of 'partially on' time 50 thousand times a second is just waaay to slow to avoid thermal damage.

I trust I have presented the case clearly now. If you think I have done anything wrong in this measurement, then you should try measuring this for yourself as I am confident you will see the same results and [I suspect only] then will you be convinced.

 

[yungman]

No I did not go that deep into this particular diode. I did not use 1N4007 for reason stated that the body is too short and I worry about the creepage over the body. I just never think the diode can be this slow. You got the picture you got the proof. Good for you.

 

[cmb]

Then I was in error... it looks like I can convince you!

Mutual conclusion for the OP - use a long bodied multi kV diode with Trr<200ns.

 

[yungman]

One question is why is it that bad to have a slow diode? One problem with all the switcher is when the load is too light or no load, it tends to go out of the stability region, the width get to narrow. With a slow diode, it will prevent the PWM from having too narrow of a pulse. This no load condition is very hard to tame, most switcher manufacturers spec the minimum load current.

We were not making power supply to sell, we did all that to cut cost and simplify the design as there were nothing close to like that on the market. If there were products like what you are doing, we would have bought instead of designing. There was nothing on the market, if you want two HV, one 10KV, the other 11KV tracking, we had to buy two 12KV supplies and set up the two voltage, both errors will add. We use a lot of this kind of tracking voltages in our systems and you should look at the size of the rack that contain the Bertan supplies. That was the motivation for me to for once did the high isolation DC to DC converter and throw the whole system up onto the the HV deck. I do think you have the market for the stackable HV supply.

 

[yungman]

Then I was in error... it looks like I can convince you!

Mutual conclusion for the OP - use a long bodied multi kV diode with Trr<200ns.

I am strongly opinionated, but I can be convinced. This issue just never came up for me. In high voltage, you don't really want to stack component for this purpose if possible. It will work a lot better if you are going to pot the whole thing to eliminate the creepage. But that itself is another can of worms!