IXYS High Voltage current regulator

Only for MOST receiving tubes, for consumer electronics. Some, like the 12AT/AX/AU7 had a center tapped filament for 6.3 volts OR 12.6 Volts. Some rectifiers had a 6.3 volt tap, and could be used at either voltage. The two, three or four digit numbers on industrial tubes have absolutely nothing to do with their filament voltage. Even more confusing is that some consumer tubes had an industrial counterpart with a different filament voltage, like the 6KD6 at 6.3 volts, and the 8950

12.6 volt industrial counterpart.

--
Service to my country? Been there, Done that, and I\'ve got my DD214 to
prove it.
Member of DAV #85.

Michael A. Terrell
Central Florida

The original standard governing the tube numbers instituted by the Radio Manufacturers Association or whoever designated the filament voltage like I said they did, and more importantly it applies to the tubes in question. When in doubt consult the datasheet.

There was no apostrophe on "tubes" in my original post. If you want to play that game...

Michael A. Terrell wrote:

The first audio and radio tubes made didn't follow any standard numbering system, other than some manufacturers adding a 1 or 2 as a prefix to make it appear they weren't duplicating someone else's design. The RMA came years later. There was also RETMA, EIA and several other "Standards" groups.

--
Service to my country? Been there, Done that, and I\'ve got my DD214 to
prove it.
Member of DAV #85.

Michael A. Terrell
Central Florida

I wonder if bootstrapping the zener from the reg output reduces stability.

John

I'm happy to try it, with a 9 or 10-volt zener, but I'd expect no effect at all, as long as the LM317 has enough voltage to operate.

Thanks for the kind words, Fred, but it was a bit of a disappointment seeing what it took to tame the MOSFET solution several of us have thought about in the past.

Indeed, using one of the IXYS depletion-mode MOSFET-based IXC current regulators, or even a depletion-mode MOSFET all by itself, should be far more simple than the NMOS + LM317 kludge. Although, as you pointed out, Digikey still has 19,506 of the IXCP10M45S regulators in stock (down to 19,476 after I ordered 20 and someone else 10 more), it's a bit unsettling to create new designs with discontinued parts.

One alternate is Supertex DN2540N5 depletion-mode MOSFET in a power TO-220 package, in production, and stocked by Mouser,

They don't show a very detailed Id vs Vgs plot, but I put some of these on the bench to get the data. One point: unlike the sloppy JFETs of old, these modern MOSFETs seem to have much less Vgs spread from part to part.

positive ------, 50mA high-voltage current source | d dn2540n5 ,-- g | s --/\\/\\--+---- negative | 43 | '---------------'

At low "sub-threshold" currents the dn2540's drain current rises exponentially, just like a BJT transistor, but with half the slope, 130mV/decade compared to 60mV. -2.75 volts gives about 100uA. Above 1mA the rising-current plot tails off, and we get to our 50mA goal at -2.15 volts, hence the 43-ohm current-setting resistor in the drawing above. In taking these measurements, I was surprised to find the '2540 really wanted to go into RF oscillation in the linear mode, even at rather low currents like 5mA and low voltages like 2.5 volts, so I was forced to solder a 0.1uF Vgs cap to the part. Likewise, the self-bias circuit above oscillated so badly above 25 volts that my DVM made faulty readings.

V I --- ---- 2 38.0mA 3 50.0 4 50.8 5 50.9 10 51.2 20 52.0 25 oops!

Adding a 1nF cap across the resistor seemed to stop this, but I ran out of time last night to take more measurements. One concern, at high voltages and high power dissipation with the junction temperature rising, will the Vgs change be enough to render the simple dn2540 50mA current source unusable?

We can hope the ixcp10m45s will shine in this regard, if its internal voltage reference, about which we know little, is any good. But there's one thing on the IXYS datasheet that gives me pause,

and that's figure 4, showing Vgs changing from -1.0 to -3.5V as a function of current. Ahem, did I say Vgs? In other IXYS literature on the IXC regulators they show an internal voltage reference and an error amplifier feeding the MOSFET gate, are we to conclude this "reference" changes by a huge 2.5 volts? It may be this part will perform no better than the dn2540 mosfet.

Thanks for the kind words, Fred, but it was a bit of a disappointment seeing what it took to tame the MOSFET solution several of us have thought about in the past.

Indeed, using one of the IXYS depletion-mode MOSFET-based IXC current regulators, or even a depletion-mode MOSFET all by itself, should be far more simple than the NMOS + LM317 kludge. Although, as you pointed out, Digikey still has 19,506 of the IXCP10M45S regulators in stock (down to 19,476 after I ordered 20 and someone else 10 more), it's a bit unsettling to create new designs with discontinued parts.

One alternate is Supertex DN2540N5 depletion-mode MOSFET in a power TO-220 package, in production, and stocked by Mouser,

They don't show a very detailed Id vs Vgs plot, but I put some of these on the bench to get the data. One point: unlike the sloppy JFETs of old, these modern MOSFETs seem to have much less Vgs spread from part to part.

At low "sub-threshold" currents the dn2540's drain current rises exponentially, just like a BJT transistor, but with half the slope, 130mV/decade compared to 60mV. -2.75 volts gives about 100uA. Above 1mA the rising-current plot tails off, and we get to our 50mA goal at -2.15 volts. A 43-ohm current-setting resistor would do the trick.

positive ------, 50mA high-voltage current source | d dn2540n5 ,-- g | s --/\\/\\--+---- negative | 43 | '---------------'

In taking these measurements, I was surprised to find the '2540 really wanted to go into RF oscillation in the linear mode, even at rather low currents like 5mA and low voltages like 2.5 volts, so I was forced to solder a 0.1uF Vgs cap to the part. Likewise, the self-bias circuit above oscillated so badly above 25 volts that my DVM made faulty readings.

V I --- ---- 2 38.0mA 3 50.0 4 50.8 5 50.9 10 51.2 20 52.0 25 oops!

Adding a 1nF cap across the resistor seemed to stop this, but I ran out of time last night to take more measurements. One concern, at high voltages and high power dissipation with the junction temperature rising, will the Vgs change be enough to render the simple dn2540 50mA current source unusable?

We can hope the ixcp10m45s will shine in this regard, if its internal voltage reference, about which we know little, is any good. But there's one thing on the IXYS datasheet that gives me pause,

and that's figure 4, showing Vgs changing from -1.0 to -3.5V as a function of current. Ahem, did I say Vgs? In other IXYS literature on the IXC regulators they show an internal voltage reference and an error amplifier feeding the MOSFET gate, are we to conclude this "reference" changes by a huge 2.5 volts? It may be the IXYS part will perform no better than Supertex dn2540 mosfet. We'll see.

All the MOSFET structures have so many voltage dependent parameters I doubt a single device regulator will be much good over a 10:1 VDS range. The trick for both circuits is probably to lock the current control MOSFET Vds with a second device which absorbs the Vds variations. I don't know why a simple enhancement node FET w/431 combo like this will not work: View in a fixed-width font such as Courier.

. . . . . + . | . .------+ . | | . / \\ | . | |1mA| | . v \\ / | . | d . +--- g . | s . --- | . / \\-----+ . '431 --- | . | [Rcs] . | | . | | . '------+ Vref . | Io= ---- + 1mA . | Rcs . (-) . . .

Thanks, Fred, I've been enjoying this conversation. Actually, as I see it "Vds variations" are vanishingly small for these MOSFETs, even over a 100:1 ratio, in the sub-threshold region. Part-to-part Vgs variations may be an issue (although so far my measurements show other wise), or Vgs vs. temp may be an issue (the datasheets indicate otherwise, but measurements are certainly in order). g_m could also be considered an issue, but it's rather high in the 50mA region for these MOSFETs.

As for your suggestion of trying a TL431 variant, that appeals to me; I will try it. Let's hope in the end that the required parasitic components don't kill the idea. We do still have to deal with a high-voltage MOSFET's propensity to oscillate.

Winfield,

I sent you an email this afternoon and recieved an error message. I just resent it. Please let me know if you got it. My email is a little squirly once in a while.

Paul P.

Maybe a ferrite bead in the drain and a largeish capacitor between G-D...

Chances are, you need to put the loss on the source or gate to have a much effect.

60V-300V ------+----------+------ ! ! \\ ! / 100K ! \\ ! ! !!- TO-220 +----+--!!- N Channel ! ! !!- ! ! ! 20V /-/ --- +----- Vout ^ --- ! ! ! --- ! ! --- 0.1u 2 places ! ! ! GND GND GND

Removing the 0.1 in parallel with the zener made the oscillation stop. Putting a bead on the drain didn't. Putting the bead on the source or gate did. I think the combination of the highish impedance of the drain and the capacitance from the tab to the heatsink prevented the bead on the drain from working.

Sorry, didn't see anything.

Zener diodes have a fair amount of capacitance, which is sometimes enough to cause trouble. One FET manufacturer recommends a small resistor in series with the gate. A ferrite bead should do the same thing.

Walt Jung has written a detailed lab report on all of these circuits:

Yes indeed, Walt's two audioXpress articles are masterpieces of low-voltage and high-voltage current-source exploration, as they might be used in audio amplifiers. I get that little magazine and saw his article when it arrived, but didn't give it the attention it deserved.

Walt also suggests reading through the work of John Broskie, on the

website. That tube crowd really goes at it.

Yesterday I received a shipment of 20 IXYS depletion-mode mosfets, the IXCP 10m45s in a TO-220 package.

When my tests went above 35 volts I experienced the same oscillation problem as I had with a Supertex dn2450. Walt features both of these parts in his article (apparently he hasn't found any other TO-220 depletion-mode power MOSFETs available right now either). I see he uses a 100-ohm gate damping resistor (fig 13C), so I'll try that. I wonder if he added that after experiencing RF oscillation, or simply from long experience with linear power MOSFET circuits.

Walt also ignores IXYS' naming of the 10m45s pins: K and A, instead calling them D and S, thereby treating this part as just another MOSFET, rather than a power IC with a built-in voltage reference and an error amplifier driving the gate, as shown in the IXYS block diagrams. The 10m45s datasheet plots stop at 1mA, but I plan to explore at low currents; a quick look last night down to 100uA showed the Id-vs-Vgs plot continuing as one would expect for an ordinary MOSFET.

I always include mosfet gate resistors, even on switchmode fets. It's good insurance against the roughly 80% probability the fet will oscillate, and it also lets you slow down switching edges if they get too radical.

I always use base resistors on emitter followers, too, although emitter followers are rare beasts these days.

GaAs fets, grounded source, seem to be stable, probably because of their impressively low Cdg.

John

I did a cursory Google Patents search on patents assigned to IXYS to try to get to the bottom of the 10M45 and found nothing obvious unless it's US5629536 and/or US5751025 both High Voltage Current Limiter and Method for Making, but they're both just cross-sectional doping profiles and from the 1997/1998 time frame. Nothing earlier seems to match, but then again I was brain dead at the time.