1. Home
  2. Electronics Design
  3. pHEMT leakage

pHEMT leakage

I'm working on an absurdly ambitious current probe for a Far Eastern customer, who wants to measure 1 nA at the shot noise in a 100 MHz bandwidth. (1 nA for 10 ns is 62 electrons.)

I'm using a Skyworks SKY65050 pHEMT cascoded by three BF862s in parallel. Once I managed to get it to stop oscillating at 8 GHz, it looks pretty nice--250 ps edges and 20 nA gate leakage at V_GS=-0.2V. I'll be spending most of next week on it, so we'll see how far it gets. I'll be very interested to see what the noise spectrum of the gate leakage looks like, and whether cooling reduces the leakage enough to make it worthwhile.

In some of the papers I've seen, the gate leakage goes to zero quadratically at I_DSS, so it's worth making some measurements.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net
Reply to
Phil Hobbs
Loading thread data ...

20nA sounds like a gusher when compared to these:

formatting link

I hope this job is on a time and material basis and not "payment when it all works" :-)

--
Regards, Joerg

http://www.analogconsultants.com/
Reply to
Joerg

Forgot to mention, Future has stock but only a few hundred and they are $4.75 a pop.

--
Regards, Joerg

http://www.analogconsultants.com/
Reply to
Joerg

But the phemts have insanely low capacitances and huge transconductances.

They probably do have low-frequency weirdnesses, which may not matter to Phil here.

John

Reply to
John Larkin

What happens to a phemt if you pull the drain up a couple of volts and float the gate? I'd suspect the gate would go positive a few tenths of a volt, until the g-s diode current equals the d-g leakage current. And I'd still expect a fair amount of gate current noise. But I'm just guessing... haven't tried it.

John

Reply to
John Larkin

Hi,

What about using a fast low input bias current opamp (ie 200fA max) buffer with unity gain feedback to buffer the input signal before the pHEMT? I found the LTC6241 opamp with 200fA input bias current, but it is only 18MHz or so.

I am interested in making a low frequency (1kHz max) nanoamp current meter, I tried this simple circuit but think its messed up as the output voltage is choppy.

formatting link
formatting link

cheers, Jamie

Reply to
Jamie

If you know a pHEMT whose leakage is in the picoamps, I'm all ears. Most of them are specified in the microamps, so for low noise stuff I might as well use a SiGe BJT. I'm looking for a corner of the biasing space where the gain and noise are adequate and the leakage current is minimal. Time to dust off my old Keithley 410 micro-microammeter. (That one is great fun, so I use it instead of an LMC660 and a big resistor, which would actually work much better.) If this were a product rather than a special, it would be worth using a uC to do the searching.

Re Si JFETs: 20 nV noise and 8 pF input capacitance is a total non-starter, though. The pHEMT is almost 40 dB quieter and has more than a factor of 10 less capacitance.

The project is done on a progress basis--I send them a report and an invoice each month, and they can decide whether or not it looks like it'll eventually work well enough. I'm pretty sure I can use the knowledge gained for a lot of other stuff, as well.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net
Reply to
Phil Hobbs

Completely out of the performance envelope, like 60 dB out. This one is about the hardest amplifier design I've ever done, which is what makes it such fun. Doing cu-clad protos with SC70 packages is pretty fiddly--see

formatting link
, which is a photo of a partially completed breadboard.

The fine traces are made by putting 3 Dremel discs on one arbour, with #2 nylon washers as spacers. (The Dremel gooseneck is a pretty cool device.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net
Reply to
Phil Hobbs

You are running the opamp single supply but applying a bipolar signal - so the opamp cannot maintain feedback with a positive input signal. A negative input current would work.

By the way the lowest input bias opamp I found was LMC6042/LMC6044, 2fA typical bias current!

--

John Devereux
Reply to
John Devereux

That is an interesting idea to use multiple spaced dremel discs to make traces. That gives a new definition for a manually routed PCB!

cheers, Jamie

Reply to
Jamie

Hi,

I updated the file and it works with bipolar input current and outputs a +-5mV signal proportional to the input current (max +-500pA or 1nA total). The max full scale input bandwidth is only 10kHz with this opamp though at the current +-5mV output.

formatting link
meter/nanoamp meter2.asc
formatting link

cheers, Jamie

>
Reply to
Jamie

Hi,

What would be a good way to make a bipolar precision current source with similar specs of +-500pA at 10kHz, that is voltage controlled, ie

+-5mV or 0 to 5V input. I was looking at "operational transconductance amplifiers" as a start but not sure if that is the way to go.

formatting link

cheers, Jamie

Reply to
Jamie

These ones have some gross huge I_DSS like 80 mA, which is a lot in an SC70 package, so it might very well run away under those conditions.

I'm grounding the source and putting external bias on the gate via a 100 meg resistor, with feedback to keep I_D constant. I haven't figured out how to get a flat transresistance over that huge frequency range, yet. It'll probably be a T network made of parallel RCs, chosen so that the main feedback RC can be a single 0603 resistor and its intrinsic parallel capacitance.

Even with 0.25 nV/sqrt(Hz) noise, the input capacitance needs to be on the order of 0.1 pF to make the differentiated voltage noise equal the shot noise of 1 nA. I might be able to hit 0.3 pF in real life, which would make me pretty happy.

Depending on the 1/f noise, I might need two parallel signal paths, but that's always hard to get really right.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net
Reply to
Phil Hobbs

A nano amp at 1 kHz is not too hard. Just keep the capacitance on the input down. I replaced your LTC opamp with some generic version (run from +/-15 V) and it worked just fine. Maybe keep the minus power lead a bit below ground?

George H.

Reply to
George Herold

Hi,

I think a current output DAC is the way to go for this:

formatting link

cheers, Jamie

Reply to
Jamie

DRAMs use a small number of electrons to tell a 1 from a 0. What you need is a really, really small mosfet, bare silicon die, with the gate directly exposed to your signal. Packaging alone would trash such a fet, if you could buy it.

John

Reply to
John Larkin

I've had CCD envy before this. ;)

I can get reasonably close with a 1-Hz noise of 0.25 nV and 0.3 pF of capacitance, which is probably roughly right for these small pHEMTs, if I basically use the gate pad as the probe tip. I've switched to cascoding the input device with another of the same, because the noise and transconductance are both better than the BF862, at least in the flatband (whatever that is with these things).

I spent most of today SPICEing it, and tomorrow and Wednesday I'll build a couple of copies and a test jig to measure the noise and hopefully the impulse response. I'm using those beasty THS3091 op amps that you like, and a completely ridiculous FB network that looks like a 10 us lowpass driving 100 megohms // 0.1 pF back to the pHEMT gate.

We'll see!

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net
Reply to
Phil Hobbs

Some of the NEC parts, especially the ones you can't get, have Spice models. For instance, NE34018 is a SOT-343 that shows 0.1 pF Cgs plus another 0.1 pF of package capacitance.

You might consider buying chips and wire-bonding up a hybrid some day.

Why cascode? How about just dumping the drain into the opamp inverting input?

There are some super-low noise broadband amps on the market. I suspect they may be distributed amplifiers.

John

Reply to
John Larkin

Thanks, that's a help. 0.3 pF may actually be attainable, then, with some ridiculously small pad size. Fortunately these pHEMTs seem to work OK riding cutouts in a single-layer board. The oscillation I was seeing before was quite amusing--I could measure the frequency by watching the bias shift as I moved my hand in and out perpendicular to the board. It went up and down by several tens of millivolts as I moved my hand by an inch or so. (I was also able to measure the frequency by injection locking it with my very shiny Tek 11802 with SD24 TDR head and watching it start up.)

I've considered that. It would be pretty nice to be able to put the chip right on the probe tip. Of course since the application has liquid water running around, there would be some passivation issues.

It would need about 10 GHz GBW to get the same reduction in Miller capacitance. I need a large voltage gain on the first stage to keep the low noise.

Yes, the best ones I know about are from Miteq, with T_N ~ 25 K at room temperature and 10 K at LN2. Pretty neat, but only a decade or so of BW iirc.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net
Reply to
Phil Hobbs

V_GS=3D-0.2V.

gets.

output

=20

Slick. I'll remember that one.

?-)

Reply to
josephkk

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.