[cryptography] a Cypherpunks comeback

[Riad S. Wahby]

Bill Stewart <bill.stewart at pobox.com> wrote:
> Absolutely. I mean, has anybody actually used JFETs in recent years?  :-)

Well played, sir. :)

By the way, the answer is in most cases no, sadly.

Most vanilla CMOS processes don't have high quality JFETs available. On
older nodes maybe you can get away with turning an N-well and a P+
diffusion into a JFET, but that doesn't work very well in more modern
processes because the N-wells have strongly retrograde doping, which
makes it hard to pinch off the "bottom" of the channel. Of course, even
at older nodes where it might be possible, the fabs don't bother
characterizing it for you. Sure, you can characterize it yourself, but
if the fab isn't supporting the device that implicitly means they're not
monitoring the quality of that device with their PCM structures, so good
luck with manufacturability long-term.

JFETs are pretty easy to make in high quality bipolar processes because
the base diffusion makes a decent JFET body. Doesn't add much/any cost
to have them in this case. Of course, if you have a BiCMOS process, then
you already have devices with high impedance gates, but for high
performance analog design a JFET beats the hell out of a MOSFET, since
the latter brings along with it a shitload of 1/f noise.

One place I've recently seen JFETs is in really high voltage processes.
Think like a mostly normal 0.18u CMOS process with a 600V (Vds) JFET
available. Haven't actually worked in such a beast, but you can imagine
that compared to MOSFETs, JFETs don't make such great power devices---
who ever heard of a depletion-mode power switch?

-=rsw