HFU HF Underground

Technical Topics => The RF Workbench => Topic started by: OgreVorbis on October 13, 2019, 1749 UTC

Title: NCP driver sucks
Post by: OgreVorbis on October 13, 2019, 1749 UTC
I'm on my 4th revision of my PCB and the first one was the best. On the first one, I used TC4452 because I didn't know about the IXDD614 at that point. After that I found out about the NCP and thought it would be better than the 614 and so I built my last three PCBs based on it. It was a big mistake and I wasted so much time on it and now if I want to use the 614s, I have to basically start over at square one. So I am giving up for at least a year. It pisses me off because I'm not the type to be able to put something like this down, but it's not worth it to get angry over something like this.

BTW: The waves going in and out of them are square as can be (as square as the crystal source which is connected with very short straight traces).

So what's wrong with them?
1. They get dangerously hot at 12V and above - they may do 15V with a heatsink on top, but they were emitting a smell like they were about to pop (and I did pop several).
2. The datasheet (especially the truth table) is either just really confusing or the design is just wrong. Based on the truth table, there is no way to use the inverting input as the output will always be low. I found this to be the case in practice also.

So yeah, I think it will work with GaN cause they only need about 8V, but if you got it working with SiC at higher voltages, then you must have magical powers.

So for now, I'm just going to use my original design board that works good below 5 MHz. I guess I won't be on the 'fun' band any time soon lol  :)

EDIT:

Here's the datasheet: https://www.onsemi.com/pub/Collateral/NCP81074-D.PDF (https://www.onsemi.com/pub/Collateral/NCP81074-D.PDF)

Truth table page 5.
Title: Re: NCP driver sucks
Post by: Azimuth Coordinator on October 13, 2019, 2154 UTC
I think the issue is the frequency your trying to use the device at..  looks like max Ft is 2Mhz Should be fine for AM 550Khz to 1700Khz  But not for 6Mhz +
Title: Re: NCP driver sucks
Post by: redhat on October 13, 2019, 2323 UTC
For the record, all these parts are spec'd to 2 MHz...that is all.  In the early days of using power conversion parts, I quickly found that the DIP package parts smoke driving anything over 1 MHz.  I switched to tabbed devices and all my problems went away...sortof.  It did take a bit of fiddling with different parts until I settled on the '614's.  in most applications, you can use the TO-263 version up to about 9 MHz or so, depending on what you are trying to drive.  My fets are about 1000pf each and the 614's will happily drive them to about that, maybe more with decreased performance and more heat.  I'm not sure if there is much of a performance difference between the 4452's and the '614's but the heatsink tab on the latter is tied internally to VSS, which makes layout and thermal management easier.

Fall is here, and 4 MHz historically has worked much better for me than 6 MHz when the leaves start to fall  8)

+-RH
Title: Re: NCP driver sucks
Post by: Stretchyman on October 14, 2019, 0608 UTC
The NCP's are fine to beyond 14MHz driving GaN.

The issues arrive from driving Qg and the SiC's I use, C3M0280090D have a Qg of 9.5nC. (GaNs have Qg of 6nC)

Power required to drive FET is QgxfxV, so 9.5e-9x7e6x12 = 800mW so they get a bit warm but are fine.

Apart from one op blowing the sh*t out of the whole Tx by powering it with A.C. I've never had one blow.

Are you using the same SiC FETs?

Str.
Title: Re: NCP driver sucks
Post by: OgreVorbis on October 14, 2019, 1147 UTC
Thanks for the replies!

The NCP's are fine to beyond 14MHz driving GaN.

The issues arrive from driving Qg and the SiC's I use, C3M0280090D have a Qg of 9.5nC. (GaNs have Qg of 6nC)

Power required to drive FET is QgxfxV, so 9.5e-9x7e6x12 = 800mW so they get a bit warm but are fine.

Apart from one op blowing the sh*t out of the whole Tx by powering it with A.C. I've never had one blow.

Are you using the same SiC FETs?

Str.

Yes, that's the one I'm using.

There must be some small difference about my PCB that's making it not work, but it's just too marginal. 800mW in that SOIC-8 seems like a lot. In my situation they are also hand soldered, so the dissipation through the pins is probably not as good as if I had used solder paste.

If I was going to continue this, it wouldn't be hard to make a GaN version of the board because it would just require swapping the drain and source pins. Then the NCPs would run easily with a lot of headroom.

So then the only problem I would have is getting the inverted inputs on them to work.

Stretchy: Have you used the inverting inputs on these? How do you wire it up? I tried connecting the inverted input to the crystal and grounding the non-inverted input, but the output stays low. I have the two OUTH and OUTL tied together, but I think it's wrong. How should it be connected for inverting? Like I said the datasheet is no help and the truth table doesn't make sense. There is no diagram of how to use the inverted input and clearly it's not just a matter of connecting the input to it.

For the record, all these parts are spec'd to 2 MHz...that is all.  In the early days of using power conversion parts, I quickly found that the DIP package parts smoke driving anything over 1 MHz.  I switched to tabbed devices and all my problems went away...sortof.  It did take a bit of fiddling with different parts until I settled on the '614's.  in most applications, you can use the TO-263 version up to about 9 MHz or so, depending on what you are trying to drive.  My fets are about 1000pf each and the 614's will happily drive them to about that, maybe more with decreased performance and more heat.  I'm not sure if there is much of a performance difference between the 4452's and the '614's but the heatsink tab on the latter is tied internally to VSS, which makes layout and thermal management easier.

Fall is here, and 4 MHz historically has worked much better for me than 6 MHz when the leaves start to fall  8)

+-RH

Eventually some day when I have more time, it's clear to me that the 614s combined with GaN are probably the best, most durable and fool proof way. I love how the tabs are grounded on the 614s and the transphorm GaN. All fets should be that way.
Title: Re: NCP driver sucks
Post by: Stretchyman on October 14, 2019, 1228 UTC
The design I use uses the AD9850 PCB's on Ebay and they O/P opposite phases already so I've no need to invert.

I'll have a look at your 'problem'....

.................................................

OK, tie the O/P's together, ground the I/P you don't want to use, drive the other.

That should give opposing O/P's.

Str.
Title: Re: NCP driver sucks
Post by: OgreVorbis on October 14, 2019, 1456 UTC
The design I use uses the AD9850 PCB's on Ebay and they O/P opposite phases already so I've no need to invert.

I'll have a look at your 'problem'....

.................................................

OK, tie the O/P's together, ground the I/P you don't want to use, drive the other.

That should give opposing O/P's.

Str.

Right, that's what I did and none of the inverted ones are working. They are always low. I checked with a scope right on the output of them.

If I fix this, maybe it will work  :o Even with only one side of the amp firing, I am getting almost exactly 50% efficiency, so I think if I can make the other side work it should be pretty good.
I found that 10V may actually be enough for the drivers. It doesn't seem to lower the output/efficiency until I go to 9V or less.

I don't know if you checked the image of the board, but here it is. You can see the subtle change on the drivers for each side.

Thanks for the help stretchy.

(https://i.imgur.com/9qJcx8p.png)
Title: Re: NCP driver sucks
Post by: redhat on October 14, 2019, 1541 UTC
Maybe you have to tie the positive input pins high to get the inverter function to work?

+-RH
Title: Re: NCP driver sucks
Post by: OgreVorbis on October 14, 2019, 1548 UTC
Maybe you have to tie the positive input pins high to get the inverter function to work?

+-RH

Now that I look at it again, yeah, I think you're right.
Title: Re: NCP driver sucks
Post by: Stretchyman on October 14, 2019, 1639 UTC
Yes indeed!

If you 'translate' the truth table.....

Drive the '-' I/P on both sides but pull up the '+' I/P on one side and down on the other side.

What RH said!

Sounds like your nearly there....

Str.
Title: Re: NCP driver sucks
Post by: JimIO on October 14, 2019, 1812 UTC
It's called a truth table because it is.    8)

~
Title: Re: NCP driver sucks
Post by: OgreVorbis on October 15, 2019, 0016 UTC
Drive the '-' I/P on both sides but pull up the '+' I/P on one side and down on the other side.

Sounds more complicated than necessary. Do I need to do it that way?
Alright the modification has been made. I just connected the non-inverted input to 5V and drive the inverted input only.

(https://i.imgur.com/Cc0gxE6.png)
Title: Re: NCP driver sucks
Post by: Stretchyman on October 15, 2019, 0502 UTC
On one side only..

Ok, does it work??

Str.
Title: Re: NCP driver sucks
Post by: OgreVorbis on October 15, 2019, 1112 UTC
On one side only..

Ok, does it work??

Str.

Well, the non-inverted side already works fine (with less than 12V). After reading the truth table again, I don't see why this wouldn't work. Unfortunately, the only way to find out is to send for the PCBs again. I'm just going to go for it. It's only $35 for 6 of them and it's not something I can hack onto my current board because those pins are grounded, so I can't just connect a wire there; I have to get new PCBs. And I shouldn't need pullups or pulldowns because that pin should always be high.
Title: Re: NCP driver sucks
Post by: Monophonia on October 15, 2019, 1354 UTC
I have used the NCP up to 7.290 in the past. Inverting and non-inverting. What keeps them from frying is a few ohms between the driver and the SiC gate. 
Title: Re: NCP driver sucks
Post by: OgreVorbis on October 15, 2019, 1444 UTC
I have used the NCP up to 7.290 in the past. Inverting and non-inverting. What keeps them from frying is a few ohms between the driver and the SiC gate.

I thought that was only for protection if the mosfet blows out, so it reduces the current consumption of the driver with no side effect?
Title: Re: NCP driver sucks
Post by: redhat on October 15, 2019, 1616 UTC
Per my post in a previous thread...

"I would leave a small amount of R in series with the gates for two reasons;

1. it helps dampen any tendency of the network and other strays to ring (Better drive waveforms).

2. should a transistor failure occur, it prevents the output of the driver from going complete short circuit.  I have lost a few drivers due to transistor failure,  No more once a little R was added.  I'm using three 10 ohm 1206 resistor in parallel to make a wider low L resistor, and also spread the dissipation."

+-RH
Title: Re: NCP driver sucks
Post by: ChrisSmolinski on October 15, 2019, 1653 UTC
Per my post in a previous thread...

"I would leave a small amount of R in series with the gates for two reasons;

1. it helps dampen any tendency of the network and other strays to ring (Better drive waveforms).

2. should a transistor failure occur, it prevents the output of the driver from going complete short circuit.  I have lost a few drivers due to transistor failure,  No more once a little R was added.  I'm using three 10 ohm 1206 resistor in parallel to make a wider low L resistor, and also spread the dissipation."

+-RH

Quite a few years ago I designed a dual 400 watt MOSFET PSU based switcher for x-ray tubes (final output was up to +/- 200 kV).  I also found that a small resistance in series with the gate drive was good protection.

The frequency was a bit lower than 43 meters  ;D
Title: Re: NCP driver sucks
Post by: jFarley on October 15, 2019, 1843 UTC
Per my post in a previous thread...

"I would leave a small amount of R in series with the gates for two reasons;

1. it helps dampen any tendency of the network and other strays to ring (Better drive waveforms).

2. should a transistor failure occur, it prevents the output of the driver from going complete short circuit.  I have lost a few drivers due to transistor failure,  No more once a little R was added.  I'm using three 10 ohm 1206 resistor in parallel to make a wider low L resistor, and also spread the dissipation."

+-RH

Quite a few years ago I designed a dual 400 watt MOSFET PSU based switcher for x-ray tubes (final output was up to +/- 200 kV).  I also found that a small resistance in series with the gate drive was good protection.

The frequency was a bit lower than 43 meters  ;D

X-ray tubes...I hear those things are dangerous!
Title: Re: NCP driver sucks
Post by: ChrisSmolinski on October 15, 2019, 1936 UTC
Per my post in a previous thread...

"I would leave a small amount of R in series with the gates for two reasons;

1. it helps dampen any tendency of the network and other strays to ring (Better drive waveforms).

2. should a transistor failure occur, it prevents the output of the driver from going complete short circuit.  I have lost a few drivers due to transistor failure,  No more once a little R was added.  I'm using three 10 ohm 1206 resistor in parallel to make a wider low L resistor, and also spread the dissipation."

+-RH

Quite a few years ago I designed a dual 400 watt MOSFET PSU based switcher for x-ray tubes (final output was up to +/- 200 kV).  I also found that a small resistance in series with the gate drive was good protection.

The frequency was a bit lower than 43 meters  ;D

X-ray tubes...I hear those things are dangerous!

I actually just re-found the videotape (yes, videotape, it's from 1994) of the CEO of the company I worked for at the time testing out the prototype. With a long lucite rod and a ground wire. Many arcs were made.  Digitizing it for posterity.
Title: Re: NCP driver sucks
Post by: Stretchyman on October 16, 2019, 0524 UTC
On one side only..

Ok, does it work??

Str.

Well, the non-inverted side already works fine (with less than 12V). After reading the truth table again, I don't see why this wouldn't work. Unfortunately, the only way to find out is to send for the PCBs again. I'm just going to go for it. It's only $35 for 6 of them and it's not something I can hack onto my current board because those pins are grounded, so I can't just connect a wire there; I have to get new PCBs. And I shouldn't need pullups or pulldowns because that pin should always be high.

Lift the grounded pin and wire high.

Put a sharp blade under the leg, heat and bend up. I wouldn't got to any extra expense until it's proven.

Str.
Title: Re: NCP driver sucks
Post by: Stretchyman on October 18, 2019, 1038 UTC
Just fired up my 'Older' design with 4 SiC FETs (I use GaN now) and Yep, 93% efficient.

382W @ 48V 8A55.

Running a 100W carrier at 24V eff is around 96%.

The whole amp is passively cooled and barely gets warm at carrier, so cool in fact I wonder where the power is coming from!

The whole system is calibrated so I know it accurate to a few %, DC and RF wise.

12V I is around 410mA and that's for 4 NCP's and the AD9850 (which gets a bit warm).

So I rekon around 80/90mA per driver?

The drivers do warm up but I can just keep my finger on them (for a while!)

I run the NCP's at 12V and noticed ONLY a 0.3% increase in eff moving to 15V, so 12V is fine.

So not sure why your having the heat problems.

This is my layout, super simple......


(http://i.imgur.com/bnzWlTl.png) (https://imgur.com/bnzWlTl)

(http://i.imgur.com/xWHl6rN.png) (https://imgur.com/xWHl6rN)



If youre still having trouble I can send you a boxed generator so you know I'm not making this up and you can test it yourself!

Regards

Stretchy.
Title: Re: NCP driver sucks
Post by: OgreVorbis on October 19, 2019, 0534 UTC
Just fired up my 'Older' design with 4 SiC FETs (I use GaN now) and Yep, 93% efficient.

382W @ 48V 8A55.

Running a 100W carrier at 24V eff is around 96%.

The whole amp is passively cooled and barely gets warm at carrier, so cool in fact I wonder where the power is coming from!

The whole system is calibrated so I know it accurate to a few %, DC and RF wise.

12V I is around 410mA and that's for 4 NCP's and the AD9850 (which gets a bit warm).

So I rekon around 80/90mA per driver?

The drivers do warm up but I can just keep my finger on them (for a while!)

I run the NCP's at 12V and noticed ONLY a 0.3% increase in eff moving to 15V, so 12V is fine.

So not sure why your having the heat problems.


Your layout isn't that different from mine except that I am trying to use the built-in inverted inputs.
It's too hard and kludgy to try and lift those pins, so I just ordered a new board. Waiting for it to arrive now.

It could be because I put too much thermal paste under the fets and it bulged out and bridged the gate and source. It's not conductive and I've verified that, but it could be a little capacitive. That's my only guess. I'll be careful not to do that when I install the fets on the new board. They do seem to work alright at 10V, so that's what I am going to use. 12V and I can't put my finger on them for more than a second. I may be able to find a type of heatsink like they use on motherboard voltage regulators if I decide to run it at 12V.
Title: Re: NCP driver sucks
Post by: Stretchyman on October 20, 2019, 1609 UTC
Eff will suffer at 10V and FETs wont be fully on and dissipate more heat.

Don't know why you can't just whip off the drivers with a hot air gun, bend the pin and reseat or just lift the pin with a scalpel whilst heating with an iron. I'd always breadboard first.

However you should be ok with the design now. Don't use any grease and just use the greasless insulators (orange ones).

Str.
Title: Re: NCP driver sucks
Post by: OgreVorbis on October 21, 2019, 1846 UTC
Eff will suffer at 10V and FETs wont be fully on and dissipate more heat.

Don't know why you can't just whip off the drivers with a hot air gun, bend the pin and reseat or just lift the pin with a scalpel whilst heating with an iron. I'd always breadboard first.

However you should be ok with the design now. Don't use any grease and just use the greasless insulators (orange ones).

Str.

My PCB company is so fast that it's not worth the effort. The board arrived yesterday and I finished installing everything and tested it.
The inverted inputs work!!!
The efficiency (87%) is not too good, but I know the reason why and I think I can fix it.
The way I connected my balun, I think is a little unbalanced  :)
I need 270pf on one side and nothing on the other to produce the 87%, so I've got a balance issue. I am going to fix the balun because with the way it connects, the one side is a little longer.
That should help and then I am going to try 12V with some little heatsinks and see if I can get into the nineties.

At 12V 4 of my NCPs draw 430mA (operating at 10V through regulators), but that's along with the crystal oscillator and voltage regulators which get a little warm.
Title: Re: NCP driver sucks
Post by: Stretchyman on October 21, 2019, 1945 UTC
That sounds about right!

Looks like you're on your way...

I must have had mine finished for over 2 years now but stuffing the thing in a commercial available box with the modulator and filter + making it unblowupable has been a challenge.

I think building one of anything is relatively easy. However building 25 and making it easy to do has taken me over 2 years!

I reckon I'll sell 2 a year, so really a complete waste of time one might say but I don't mind, keeps me busy!

I really pleased at the progress you've made.

May I ask about your modulation scheme?

Regards.

Stretchy.
Title: Re: NCP driver sucks
Post by: OgreVorbis on October 24, 2019, 0833 UTC
That sounds about right!

Looks like you're on your way...

I must have had mine finished for over 2 years now but stuffing the thing in a commercial available box with the modulator and filter + making it unblowupable has been a challenge.

I think building one of anything is relatively easy. However building 25 and making it easy to do has taken me over 2 years!

I reckon I'll sell 2 a year, so really a complete waste of time one might say but I don't mind, keeps me busy!

I really pleased at the progress you've made.

May I ask about your modulation scheme?

Regards.

Stretchy.

Sure. I'm going to use a Greek PWM board. I have it already, but have not tested it yet. It's a spruced up version of this kit (using bigger FETs): https://www.s9plus.com/puwma%20draft4.pdf (https://www.s9plus.com/puwma%20draft4.pdf)

So I've done a bit more experimenting. I made it to 92% @ 6.8 MHz with 12V drive.  :)
Something is a little weird though. I need exactly 470pf on only one side of the amp. Like I said before, I think I need to change my balun back to the original wire wrapped one I was using. This new one uses coax and it's a bit strange. Could I somehow be fooling the bird meter by using a cap on one side only? If not, then it's got to be my balun. I'm going to put the old one back in tomorrow and see. The new one was actually 1% less efficient or in the noise.

I should never need more capacitance on one side, right? Each side of the amp is exactly the same, so I am guessing it has to be the balun.
Title: Re: NCP driver sucks
Post by: Stretchyman on October 24, 2019, 0920 UTC
Measure the voltage peaks at the drains with a scope. Should be Vcc X 3.5 and have the nice class E rounded shape. Both sides should look the same.

I started by copying the PUWMA and have moved on. You can integrate the opto into and driver (FOD3180) for a start. Any half bridge has the high side floaty thing built in already ..

Str.
Title: Re: NCP driver sucks
Post by: redhat on October 24, 2019, 1416 UTC
Be careful using the '3180's as they have very limited drive capability.  The datasheet only specs them to about 63 KHz for full current output, and I have smoked a few of them running larger fets, see page 10 of the ON/Fairchild datasheet.  If you are using a half bridge type arrangement (totem pole) you can use a bootstrap type supply to derive VCC for the upper driver.  This doesn't work very well on single ended (non push-pull) arrangement however.

I wound up using a HCPL7723 driving a TC4452 and this has been rock solid into anything I've tried.  The only downside is the photocoupler requires 5V (on both sides, mind you), the driver closer to 12V for good modulator efficiency.

+-RH
Title: Re: NCP driver sucks
Post by: Stretchyman on November 04, 2019, 0720 UTC
Indeed....

However whilst so many stick with what they know and use them old things; 11N90's & IRF260's etc with many nF's of gate C there be many modern devices which are much easier to drive.

Mind you there's much better devices to drive the FETs directly SI8234 for example..

I'll be trying those next...

Str.
Title: Re: NCP driver sucks
Post by: Charlie_Dont_Surf on November 01, 2023, 0047 UTC
Yes, the NCP81074 does indeed suck for our purposes. I'm not planning to use it ever again.

It is simply not suitable for use at >6 MHz on anything with a Qg >~8 nC; you will blow the NCP81074. The reasons are:
1) The NCP81074 runs a high supply current at these frequencies, which dramatically increases its power dissipation.
2) It does not have an exposed paddle (called an "EPAD" in industry lingo) under its "belly" to allow you solder it to the PCB and get the heat out, which is absolutely necessary. (Most heat will come out of the belly since that is where the silicon die is attached to the package leadframe. Adding a heatsink to the top is not going to help much.)

You can get it to work and stay alive at lower Vdd (< 9 V) depending upon the transistor but then you can't drive many LDMOS, GaN and especially SiC transistors) to full on (lowest Rds_on), because the Qg needed to get the gate up to ~12 V far exceeds what the driver can deliver and not burn up. The demands upon the gate driver to source current (and thus dissipate heat) are significantly higher at 12 V than 6 V for example. Then when modulation of the final FET is added to the mix, the current demands at the gate increase again too. The NCP81074 can drive 4-6 A but it can't dissipate the heat that this causes when trying to do this at 6+ MHz, 50% duty cycle, Vdd = 12 V.

Keep in mind that the NCP is not characterized to work beyond 2 MHz. Obviously you can try it - we all do this - but understand that you may suffer unpredictable performance. Every time I have blown one of these up and soldered a new one down, I get a somewhat different supply current than the previous during operation. This tells me everything I need to know. You don't know what you are going to get the next time and the next time, etc, because you are working in conditions that are marginal and unintended for the chip. You likely won't see this at < 2 MHz because that is where it is spec'ed and intended to operate.

Instead, I have been using the Renesas ISL89163 dual driver or the TI UCC27614. Both have EPADs, can drive 6+ A and are plenty fast. (The ISL89163 can be paralleled for ~12A, the UCC27614 can do 10 A.) I haven't explored the TI chip much but I had the ISL89163 running at 18 MHz on a 28 nC gate Transphorm Cascode FET the other day and it looked like I still had extra pulse width to spare, i.e., the edges weren't so rounded that the pulse was horribly distorted. No self-destruction so far. 8) The ISL89163 has roughly half the supply current of the NCP81074 at the same frequency. Of course, these aren't spec'ed to operate at 6-7 MHz either, but so far I have not seen the same limitations as the NCP81074.

TI does make a FET driver that is intended to be used up to 50 MHz for LIDAR (LMG1025-Q1) but it can't go above 5 V.
Title: Re: NCP driver sucks
Post by: Stretchyman on November 01, 2023, 1934 UTC
Used them to 15MHz without issue.

Driving 6nC GaN @6V.

Str.
Title: Re: NCP driver sucks
Post by: Charlie_Dont_Surf on November 11, 2023, 0612 UTC
Used them to 15MHz without issue.

Driving 6nC GaN @6V.

Str.

Let us be clear about the situation we are discussing.

The GaN Systems transistors that seem to be your transistor of choice these days appear to be quite nice based upon review of the datasheet and I'm sure they work well in reality. However, the drive requirements are somewhat light compared to the other 99.99% of transistors on the market. (I chose one of the three of their current production FETs with Qg ~6 nC to show below as an example but the exact model number matters little.)

I have no interest in continuing to use a FET driver with a somewhat niche application (and freaks out when presented with anything beyond the lightest of loads like a GaN Systems FET) when I can easily obtain other drivers that can drive your GaN Systems transistor to full Rds_on with ease and also handle most or all low, medium and high- power transistors that I or any other hobbyist would want to use.

I'm actually using the ISL89163 in a work application, providing what amounts to continuous synchronization pulse distribution/fan-out driver for hundreds of loads though many tens of meters of wiring harness and PCB traces - with all the parasitic L and C that this represents - and it doesn't argue with me, doesn't do weird shit and doesn't burn up. Case closed.

(https://i.imgur.com/5pZEUhZ.png)
(https://i.imgur.com/WysWT0j.png)