Wire bonder suggestions?

I have a friend who has a hybrid company, and he owes me.

Ask for Mike. He could do stuff for you, or give you advice. In fact, I gave him a copy of your book, after he saw mine, so he'd be delighted to hear from you.

I understand that wire bonding has a learning curve. I've never tried it myself.

John

Thanks, I'll do that. It's one of these pHEMT-on-a-stick ideas. I wonder if he can thread a ferrite bead on the bond wire. ;)

I'm getting a whole pile of different low-noise microwave FETs to see which ones have the lowest gate leakage. If I can get the gate leakage down, the noise might just be limited by the differentiated voltage noise,

i_N = e_N Omega C.

For 0.5 pF and 250 pV noise in 1 Hz, that's actually pretty quiet if it works. I'd be pretty happy with gate leakage below 50 nA. Of course there are liable to be high frequency contributions to the input noise current. we'll see. Microwave transistor datasheets are truly horrible.

Cheers

Phil Hobbs

He does a lot of weird stuff. He also works with Alex, on mostly-fiber electro-optical assemblies.

Right, you're lucky to get Idss, much less a transfer curve. I have some homebrew data on a few parts, like NE3508/9, some of the SOT89s, and CLY2-type mesfets.

I don't have a good feeling about 50 nA. I recall microamps of gate leakage. There may be a magical gate bias voltage where leakage is close to zero, but that may be "external" leakage, seen at the pin, which is just a standoff between internal leakages. Bias the drain, float the gate, and see where it lands.

Share your data, please, if you can.

John

Will do. I try to structure NDAs so that my work product isn't confidential unless it's part of the client's product or is in the process of being filed as a patent. That way I get to use it again.

Most pHEMTs seem to work fine at V_DS ~ 1 V, so since I don't need to go faster than 100 MHz or so, I can trade off speed for lower leakage, while keeping the nice low noise. We shall see.

Cheers

Phil Hobbs

Oh, one other comment, which you may know about already.

Compound semis are dirty, full of "trapping states." One effect is that their gain varies with frequency, with the transition in the sub-MHz range. Another is that their DC offset shifts on a daily basis, probably (?) in super-popcorn-like jumps, something in the millivolts or maybe 10s of millivolts sort of range. It varies from part to part off the same reel: some are stable, some jump around.

We had to trash our first-generation NIF laser modulator design because of phemt threshold shifts. It would be working great one afternoon, and all gone to hell the next morning.

I haven't tried phemts as followers, but grounded-source, they are very HF stable. You probably won't need ferrite on the gate lead.

John

Thanks.

I can believe that about the popcorn noise. Their 1/f noise curves (those that even give them) are pretty horrendous, and the two often go together. Mikko Kiviranta has some data on his website too.

Fortunately the events these are made to watch will be all over in a few hundred nanoseconds, and AC coupling covers a multitude of sins. ;)

I'm probably going to use the pHEMT running straight into the summing junction of a CFB amp, and run a DC feedback loop to keep the output still. Sounds like it'll need a baby-scale (i.e. two-speed) loop to control the popcorn. (Hopefully I won't have to use a pilot tone.)

They may want DC coupling, in which case I'll have to use a two-path approach, but that's always a mess when it's a gain stage with no AC feedback.

Cheers

Phil Hobbs

Luis Cupido has created a yahoo list on wirebonding, you might want to contact him. The list was _very_ low volume, I can't remember when the last post has been there.

regards, Gerhard

Phil Hobbs wrote: : >>>>>>

: >>>>>> Do any of you folks have ones that you like? (I can supply my own : >>>>>> microscope.)

We have a Delvotek which I occasionally use to bond my samples. The work is so nerve taxing, however, that I mainly prefer to leave it to our technicians. Delvotek seems to do its job OK, at least I feel more comfortable with it than with the Kullicke&Soffa I used before.

: >>>>> I understand that wire bonding has a learning curve. I've never tried : >>>>> it myself.

Wire bonding is a delicate process indeed. If evereything goes well it is not difficult at all, but when the wire does not stick, you need a good understanding of the process in order to troubleshoot.

: >>>> I'm getting a whole pile of different low-noise microwave FETs to see : >>>> which ones have the lowest gate leakage. If I can get the gate leakage : >>>> down, the noise might just be limited by the differentiated voltage noise,

I agree with John in that the gate leakage of HEMTs tends to be very large (I only have experience on some Avago types, though, and even that was years ago). A nice thing is that the leakage freezes when the device is cooled to LHe. You can very good performance that way, like Nikolay Ukhanski did in his Rev. Sci. Instr. paper (vol 74 no 2 p 1145).

SQUID readout guys have explored to (mis)use microwave transistors for low-frequency low-noise usage to some extent. Another top-notch guru in this field is Dietmar Drung. I seem to recall that Roger Koch also constructed some pretty good amplifiers in the past.

: >>>> i_N = e_N Omega C. : >>>>

: >>>> For 0.5 pF and 250 pV noise in 1 Hz, that's actually pretty quiet if it : >>>> works. I'd be pretty happy with gate leakage below 50 nA. Of course : >>>> there are liable to be high frequency contributions to the input noise

I first read this as 250pV/rtHz at 1Hz (which would have been astonishing) but surely you mean 250pV rms over 1Hz BW in the white region?

: > Oh, one other comment, which you may know about already. : >

: > Compound semis are dirty, full of "trapping states." One effect is : > that their gain varies with frequency, with the transition in the : > sub-MHz range. Another is that their DC offset shifts on a daily : > basis, probably (?) in super-popcorn-like jumps, something in the : > millivolts or maybe 10s of millivolts sort of range. It varies from : > part to part off the same reel: some are stable, some jump around.

This is what Nikolay tells, too. I have not that much experience. GaAs (and related compound) devices have a reputation of having their 1/f noise corners in the range of MHz's or even 100's of MHz, because of the wealth of trap states. That's why I am more interested in the SiGe bipolars, which are cleaner. Some discrete BJTs have their beta to go as high as 20 000 at 4.2K, which would allow you to easily bias them at nA-range base current - unfortunately you cannot use cryogenic stuff I suppose.

: (those that even give them) are pretty horrendous, and the two often go : together. Mikko Kiviranta has some data on his website too.

Do I ???

Regards, Mikko

noise,

Phemts have noise temps in the 50K sort of range when used as RF amplifiers, but that's at high frequencies and with optimal tuned input networks.

John

noise,

Thanks, good leads. Roger was a good friend of mine, who unfortunately died in 2008--he had a bicycle crash, and it wasn't clear whether the crash caused his death or whether it was e.g. a heart attack that caused the crash. He was a fine physicist and an even finer human being.

I thought so--I suppose it's the SiGe stuff I was thinking of.

Nitrogen would probably be okay in this situation, but these folks aren't set up for LHe. Realistically a beta of 1000 would probably be okay--do you know which types work (at least sometimes)?

Thanks

Phil Hobbs