Can MOSFET gate driver directly drive small capacitive loads?

[Jinhao]

Hello,

background: I'm trying to build a transmitter/beamformer for Capacitive Micromachined Ultrasound Transducer (CMUT) array. Actuating CMUT requires high voltage, high frequency square wave (roughly 20V@20MHz at different phases). Each CMUT element is essentially a very small variable capacitor less than 0.5pF, meaning a small average power and small peak current draw at a reasonable slew rate (100-1000MV/sec). Since the whole system is highly digital, I plan to use teensy4.1 as square wave generator (3.3V GPIO) and gate drivers as high voltage buffers with very short rise/fall time.

Question: I am unsure if I can use gate driver as MOSFET itself to directly drive those CMUT elements. Since I only need high voltage with little requirement on current, power MOSFET seems unnecessary. For example, this gate driver (MAQ4124https://ww1.microchip.com/downloads/en/DeviceDoc/MAQ4123.pdf) already satisfies all my requirement. Or perhaps, there are better ICs that are more suitable than gate drivers for this project?

I appreciate any advice and comments.

[tszaboo]

In general, yes, you may drive an ultrasonic transducer with a MOSFET driver. About a decade ago my employer made a patent of this idea of mine. Fortunately, they only made a limited patent which is only registered in one single country, so it was only for bragging rights. They also don't have layers, and patents without layers is useless.
I was using a MOSFET driver to drive 40KHz ultrasonic transducers.

I think you will run into issues at your frequency though. I don't think MOSFET drivers will not work reliably at 20MHz. You see, they are for driving MOSFETs for circuits, like SMPS. SMPS doesn't work with those frequencies, so it is safe to assume the drivers wont work either.
You will likely run into slew rate limitations, or the drivers internal anti cross conduction circuit will delay your control signals. The driver you linked has 50ns propagation delay and your clock signal is only high for 25ns.

Maybe drivers for GAN FETs could work.

[gnuarm]

I would not assume the gate driver won't work at 20 MHz.  They are designed to drive high currents into the FET gate to produce a fast transition time.  If the transition time is short enough, it will work at high frequencies. 

[filssavi]

You can try, however I doubt it will be easier than just building a driver out of discrete transistors as the off the shelf parts are usually designed to work at much lower frequencies, and it might be explicitly built to avoid it as in the typical application those frequencies are only reached in very unpleasant circumstances (self oscillation)

[Marco]

IXYS used to have gate drivers specifically for RF, IXRFD630 ... but they've become unobtainium between the current shortages and the breakup of IXYS.

[Marco]

You can try, however I doubt it will be easier than just building a driver out of discrete transistors

With the tiny capacitances I'd just use some fast small signal PMOS like PMZ1200UPE at high side a with capacitively coupled gate, low side n-channel MOSFET and put two 50 Ohm resistors in between ... who cares about a little shoot through and burning some power in resistors.

[T3sl4co1l]

Heh, it would be nice to have like a CD4069BE, but actually usefully fast, a bit stronger in drive strength, and still the same voltage rating.  I'm not sure offhand what would offer that functionality though.  Also, level shifting from TTL...

So, a gate driver.  These certainly fit the bill, but are made for much heavier loads (peak currents of ~amperes), and most for much lower frequencies as mentioned.  Most struggle above a few MHz; you'll need to shop for special ones capable of both the speed and voltage.  They'll be intended for SiC MOSFET driving, most likely.

Note the MAQ4124, on page 9, supply current vs. frequency at 0nF load, it stops at only 1MHz.  (At a glance, I didn't see a maximum drive frequency, but it's not likely to do much when driven faster than the propagation delays.)  And with a 20V limit, it's not really made for this kind of voltage range either.

Doing it discrete is a pain, and you'll spend a lot of supply current driving capacitances and/or bias currents making the thing work.  It's certainly within the reach of say MMBT3904/6, or anything faster (e.g. MMBTH10/81, at least until they go completely obsolete..), but they'll need several to tens of mA per channel.

Really, the supply current cost is true of anything but an integrated solution: there's a lot of transistor and wiring capacitance to drive, and I guess, very little actually in the transducer.  The only solution to that is a driver integrated with these micro structures, which sound like they're maybe fabbed silicon, anyway?  (Like, is this a research thing, and just not quite far enough thought out to integrate that on the same chip/assembly?)

Also, is this square or sine wave, and how much does it really matter?  (The slew rate given, seems to imply a pretty sloppy square wave, if square at all, which is encouraging, to that end.)  I can think of some cool ways to approximate a square wave at lower power cost (potentially fewer parts too), subject to limitations on power handling, efficiency, bandwidth, etc.  Does any of that have the possibility of fitting?

Tim

[Marco]

supply current driving capacitances and/or bias currents making the thing work.

The fact that it's a squarewave of fixed frequency simplifies things a lot, having to worry about near DC effects is generally what makes high side driving a pain. Here you can just tie a high side PMOS gate to the upper rail with a resistor and a diode and capacitively couple a driver. Even with miller effect a PMH1200UPE is still only around a 50 pF load ... a 74LVC buffer/inverter should be able to handle that.

[T3sl4co1l]

Yeah, and that's a nice fairly-small transistor, also rated to low drive voltages so 3.3V logic right off the MCU will even do; well maybe not quite that easy, 1nC in 10ns is 100mA peak, but yeah a few LVC buffers in parallel definitely will.  (If only they made P-ch RF types, it would easily work! )

Tim

[Jinhao]

Hi Tim, thank you for your advice and comments.

• Apologies for the vague square wave requirement. The desired CMUT driving voltage waveform is sine waves. But, given that CMUT elements are typical 2nd order mechanical systems behaving like a low pass filter, square waves generated from the microcontroller is a simple and effective approach for me.
• The current vs. frequency plot on page 9: Yes, it definitely makes me less confident using this gate driver. Thank you for mentioning this; I didn't read the datasheet carefully.
• I want to have as few components as possible because I need to solder/reflow them myself. The phased CMUT array has 32 channels, so the motivation of using gate drivers is to reduce component number to save troubles.
• Indeed, there are some papers focused on CMUT beamformer IC design. Unfortunately, they are not commercially available. The commercially available development boards (like TX7332EVM) are mostly designed for PMUT (P for piezoelectric) which operates at a fairly lower frequency, meaning a very limited phase level/resolution at high-frequency. So they are not very suitable. The CMUT/MEMS fabrication process generally needs compromisation to integrate with IC fab process. The CMUT I use is made out of polymer SU-8, so it's even more difficult for the fabrication integration on a same die

Lastly, I had to start over and switch to other gate drivers because the design requirement got changed possibly due to the change of CMUT structure by the other lab colleague . Now it requires 50V(DC)+5V(AC) @ 12MHz. Currently, LTC4440-5 https://www.mouser.ca/datasheet/2/609/44405fb-1272005.pdf looks appearing as its output is lifted to BOOST supply level (DC bias), so a 45V-55V square wave can be easily generated. I ran a quick simple SPICE test and now start to worry about the variation in propagation delay time of different array channels. The phase control of square wave can be totally meaningless if the delay variation between channels is too large. The pressure field at the image plane will not be as designed if the phase of each channel is mistuned. Maybe this channel delay time can be manually measured and offset

Thank you for everyone's advice and comments. I appreciate them.

[T3sl4co1l]

So hey, this is just a microscopic ESL (electrostatic loudspeaker), huh?  Operates strictly on polarization, no specific material properties needed (electrostriction, piezoelectricity)?

Is there a common electrode ("ground"), and the rest need to be driven independently (well, not really independent, but some desired phase between them, at whatever common frequency is used, I guess)?  Surely you can simply bias the AC ground to some DC offset, and drive the rest direct from the Teensy or a small level shifting buffer?   (74LVC is still an option to boost drive strength, or 74ABT has 3.3V/LVCMOS compatible inputs with 5V operation.  GaN gate drivers are also an option -- these use 5-6V supply, not much higher but the clock frequencies should be available at least.)

Or at worst, use a bias tee -- a large series resistor to the high voltage, and a small coupling capacitor from the signal source.

This gets me an author position on the paper, right?

Tim

[Marco]

He needs 10V peak to peak, so 4000 series logic.

[Zero999]

He needs 10V peak to peak, so 4000 series logic.

Some level shifting will be required, which can be done with the CD4504. If more current is required, parallel more channels, or use a slightly higher voltage and add some emitter followers.
https://www.ti.com/lit/ds/symlink/cd4504b.pdf

If a -5V supply is available, then the 74HC(T)4053 is another option.
https://assets.nexperia.com/documents/data-sheet/74HC_HCT4053.pdf