Help with Nixie partially lighting up

[Luka C]

Hi everyone,

I need help with my newly built clock. The thing is, after I turn it on and the clock runs for like 5-10mins, some of the digits on tubes become partially lit like on the picture. Now, I noticed the power supply gets very hot after running for 5-10mins and I measured the voltage on +HV (+185V) pin of the power supply while the clock is running. It seems like the voltage keeps dropping as the supply gets hotter (last time I measure it, it went down to 148V and these artifacts became showing on the tubes. The digit usually lights up normally at first, then after the cathode poisoning routing it becomes partially lit like on the first picture, then after the second cathode poisoning routine, it gets partially lit like on the second picture and so on, it shows on all of the tubes at some times. I'm using 27k anode resistors on IN-14 tubes and I guess the power supply is of the design from the picture I attached.

Thanks in advance!

Power supply design (I suppose)

[gregebert]

Sounds like the power-supply isn't able to drive all of the tubes. Have you tried running with 1 or 2 tubes to see if the supply still overheats or drops ?

[Ian Sparkes]

Hi Luka

what do you mean by "supply gets very hot", does that mean the MOSFET get very hot, or the input 12V supply?

Looking at that design, I would imagine that the reason it has "DO NOT USE" on it is because the MOSFET is turned off by the 330 Ohm resistor when the output transistor is turned off, and that does not give a very clean turn off for the MOSFET. If this is the case, you not get a very good inductive spike. Also I notice that the current sense pin isn't doing anything, so I imagine you're going to run into problems with efficiency and heat.

I think you need a better HV supply. But if you want to persist with the one you have, try heat sinking the MOSFET, or if you don't have a heatsink, pee on it every couple of minutes to keep it cool... (I'm joking)

Are you sure that is your HV power supply? It seems strange to use a design that even the designer has canned....

[gregebert]

I stick with linear power supplies, especially for the anodes (high-voltage), whenever possible. They are much simpler to design, least-likely to overheat, and very reliable. In fact, the main reliability concern is the HV filter cap. You can mitigate that risk by over-design. If anyone's interested, let me know and I'll post my best-known-methods that I use.

[Luka C]

Thank you both.

@gregebert

I would be very thankful if you could explain it or send a link here so I can see how I could implement it into my design. My clock uses 4x IN-14 tubes + 4x INS-1 tubes

@Ian

I can't say for sure since the supply is on a very small board so the whole thing gets very hot, but yes, I'd say the MOSFET is the hottest component. And, to be honest, I bought the PS from eBay from EU seller because I though Chinese ones could be problematic (such irony!) and there was only the picture I attached here. After I noticed the problems occurring in the PS, I inspected the PS and concluded it should be the one from the schematic I posted in my previous message (unfortunately).

[Jon D.]

@gregebert I'd be interested in seeing your best-known methods for power supplies.

[Ian Sparkes]

Hi Luka

these designs are very well known and there are a few variants around, basically divided into the ones that have the pull down transistor and those who don't, and those who use a NE555 (yuk) and those who don't.

The TO-92 package at the bottom left is presumably a 78L05 (guessing because of the +5V terminal on the right, and I can just see the "5" poking out, and the fact that there is a 100nF cap close to it). If that is the case, then the board basically matches the schematic.

So here are the things that I see questionable about this design:

1) The IRFD220 is under-rated
a) It has a power dissipation of 1W, and a current capability of 1A, spike 6.4A. We're going to be pushing about 12A through it at turn on, dropping to 6A when it is hot. Naughty.
b) It has a high positive temperature coefficient of Rds (resistance from drain to source). This means that it will heat quickly if you drive it too hard, and the heating effect will snowball. It doubles from 1 Ohm at 30 degrees to 2 Ohms at 130 degrees. You also can't heat sink it easily, so you are only left with the other option I mentioned earlier. This will drop the power when it gets hot. I think this is what you are seeing.
c) The thermal conductance of the package is 120 degrees/W. This means at rated load we expect the temperature to be 120 degrees above ambient. We are working outside of this normal range of this device. It will probably have a short and unhappy life, especially if you put it in a box (ambient temperature will then rise to 30-35 degrees).

2) There is no pull down on the FET gate: This is the problem I first thought. This will reduce the efficiency of the circuit. The 330 Ohm (written on the board, but look, there is a 270 Ohm resistor installed) will discharge the gate in a given time. The IRFD has a 6nC gate charge. The 270 Ohm resistor will be able to discharge the gate in 6E-9 S * 270 = 1.62 uS. During this discharge, the drain current reduces linearly, meaning that the turn if is "smooth", and not sharp like we need. This is a issue particular to the 34063: it can only source current but not sink it.

3) The 34063 is running without a current sense resistor, and this doesn't help the efficiency either.

The data is from here:

http://datasheetcatalog.com/datasheets_pdf/I/R/F/D/IRFD220.shtml

I see that the guy claims it can drive IN-18, good luck with that...

So, what does that all mean? Well, before complaining too loudly, you need to be sure that your input power supply is capable of driving this board: It needs to be 12V (not 16V or 10V) and have the capacity to drive the inrush current (up to 12A spikes for a few uS). The input smoothing capacitor (200uF) will not cover all of this.

I'm sure it can work with small tubes if you get everything just right, but it's not at all tolerant. With the addition of a pull down and a more robust MOSFET, it would work much better. We're talking maximum 30 cents difference in production price.

Regards

Ian

[taylorjpt]

2) There is no pull down on the FET gate: This is the problem I first thought. This will reduce the efficiency of the circuit. The 330 Ohm (written on the board, but look, there is a 270 Ohm resistor installed) will discharge the gate in a given time. The IRFD has a 6nC gate charge. The 270 Ohm resistor will be able to discharge the gate in 6E-9 S * 270 = 1.62 uS. During this discharge, the drain current reduces linearly, meaning that the turn if is "smooth", and not sharp like we need. This is a issue particular to the 34063: it can only source current but not sink it.

The 270 ohm pull down resistor is the only change you can easily make to the existing design.  At 1.62uS discharge time, that is more than 10% of the data sheet period with a 470pF timing capacitor:  The transistor in switching regulators is meant to operate either ON or OFF, never in linear mode.  Turning the transistor ON to charge the inductor is easy as the regulator output transistor is capable of delivering 1.5A but that resistor can't turn the transistor OFF.  Add to this the fact that when the controller wants to turn off the FET with 12V on the gate, the gate has to fall about 6V before the transistor really begins to turn off which throws off the timing of the control loop.  One last thing that happens is that when the transistor does start to turn off, the fast rising drain voltage is coupled into the gate through the drain-to-gate capacitance trying to turn it back on which results in a period of unhealthy oscillation.

Since the gate driver is capable of 1.5A of output current, you can lower the pull down resistor pretty dramatically without breaking anything except the low load efficiency (Which can't be very good with the transistor getting "Hot"!).  You will have to scale up the power rating of the resistor by replacing it with a larger, perhaps axial part, but this should improve your transistor heating.  A 100 ohm gate pull down resistor will improve your turn off time by a factor of 3 but at the cost of about 1.25W of dissipation (Use a 2W resistor). 

Another way to speed this up is to use an active pull down circuit on the gate in place of the passive pull down resistor.  Using a current sink instead of a resistor has the added benefit of drawing the same current regardless of the applied voltage vs the RC discharge curve and it is less susceptible to the drain-to-gate coupling.  Where the discharge current of the 270 ohm resistor is 44mA at the start of the gate discharge, it falls to only about 15mA at the critical 4V gate threshold voltage.  A current sink like that shown below pulls the same current regardless of the applied voltage down to it's dropout voltage of 1V which is well below the 2V minimum gate threshold voltage of the MOSFET.  The current sink at 4V looks like an equivalent 80 ohm resistor and at 2V a 40 ohm resistor, i.e. 3 to 7 times faster turn off for the same power dissipation of the 270 ohm pull down.  The 20ohm resistor does not have to dissipate a lot of power as this task is shifted to the transistor.  Increasing the current is now just a function of adjusting the 6.2K resistor.

[gregebert]

Why not just use a totem-pole pre-driver for the MOSFET ?

BTW, this touches on a challenge when designing HV drivers, whether they are for DC-DC converters or driving cathodes. Most high-voltage NMOS devices (~400V)  need 5-6 volts for solid gate turn-on, yet logic-levels on most micro-controllers and FPGAs are only 1.2 to 3.3volts.  That leaves a few options:

1. Use high-voltage logic, such as 4000-series CMOS, to drive HV NMOS. My first nixie clock does this.

2. Use NPN drivers; you can easily design constant-current drivers running from 3.3V. I've used surface-mount devices with 2 NPNs inside, and they are rated at 250V.

3. Use a level-shifter between your logic and the driver. I'm doing this on my current design with a HV5530.

4. Find a device with a low-voltage gate. Good luck on this; most logic-levels NMOS devices only handle 60V or less. If you also want low Rds (necessary for DC-DC converters), you have even fewer options.

5. Pull-up resistor. Ugghhh... It will work, but you will burn more power. Could be dangerous for a DC-DC converter if the logic/software turning off the gate goes haywire, which would lead to large current/burnout.

I advise against relying upon the voltage-drop of the nixie tube to protect the driver, though this is commonly done with designs using the 74141. I always select my driver to handle the maximum-possible anode supply, and add margin above that. I've never had a fried driver, ghosting, etc. Parasitic capacitance and leakage can stress drivers above their maximum ratings, which will lead to premature failure.

[Luka C]

Okay, so I read all the advices and thank you all for explaining the problems with this power supply. Since the board is extremely small and I will be installing it in a very thin housing, I don't really have much space to do modifications on it. So, I'll implement PS in my PCB design and solder its components on it.
I'd like to know if you could check this PS and see if it should be ok for my clock (4x IN-14 and 4x INS-1 tubes)?

Thanks a lot in advance.

[Luka C]

The picture seems to be moved, so this is the PS I'm planning to implement into my design. Do you think this one is fine for 4x IN-14 and 4x INS-1?
Thanks.

[gregebert]

I was hoping you were going to use a linear power supply.

My experience with switching power supplies is they require a fair amount of debug/optimization. I suggest you carefully prototype it and do thorough checkout with a scope. Ideally, you would do all your debugging with a PCB board, because board layout will contribute to problems. Breadboarding might work, but you'll probably go thru another round of debug with the PCB so I recommend you save time by going straight to a PCB, and plan for a respin.

[Ian Sparkes]

Hi Luka

Linear for sure has a more retro feel to it and will still be going in 25 years, but if you are already down the switching path then I think this one deals with the most immediate issues. It has a nice heatsinkable MOSFET in it, and the active pull down and the current sense resistors.

If I am not wrong, the original designer might be in this group, and I believe offers that design on Ebay. Not trying to hard sell you, just saying... You could save yourself an evening or two, and give him a nod of thanks as well.

I'm working on a switching design based around an ATTiny at the moment, but this is just because of my perverse nature...

[gregebert]

Though slightly off-topic, I had a bad experience (as in smoke, sparks, and explosion) with a microcontroller-based DC-AC inverter many years ago. It was a tad bigger than a nixie-supply because the DC supplies (+170 VDC, and -170VDC) each had 10,000uF of filtering; almost 300 joules of energy.

The microcontroller crashed, and that caused the HV transistor-stack between -170 and +170V to short-out. Pretty hefty transistors, too. (rated at 50amps / 600V). They are bolted-down, not soldered.

Lesson-learned: Add protection circuitry to prevent unexpected or "impossible" events from happening, because they do happen. At a minimum, I'd place a fuse between the DC supply and the inductor, and size it as small as possible.