On Wed, 2007-05-30 at 16:05 -0700, Andrew Lentvorski wrote:
> Christoph Maier wrote:
> 
> > But to generate an AC waveform, you need to start with active devices,
> > i.e., 
> > MOSFET switches. A MOSFET ring oscillator is straightforward at 700mV, 
> > a RAIL-TO-RAIL MOSFET oscillator that can source substantial current
> > isn't. 
> 
> Huh?  Why?  I think you are thinking BJT, not MOSFET.  Or, perhaps, you 
> are thinking of an ultra-fast ring oscillator composed of differential 
> stages.  Too high tech.  Think lower tech, lower frequency, old school.
> 
> 
> First, MOSFET inverters go rail-to-rail even when they are stuck doing 
> it slowly.  It's almost impossible to prevent them from doing so. 
> Consequently, a ring oscillator composed of an odd (normally prime) 
> number of inverters will go rail-to-rail if the chain is long enough. 
> That's the beauty of CMOS technology.  Or, as I like to put it:
> 
> "CMOS is a foolproof technology.  The proof is how many fools use it."
> 
> 
> Second, I don't even need a ring oscillator for the task.  A standard 
> single MOSFET, class-D oscillator with an external crystal (extremely 
> high Q) can blow out a VLSI MOSFET if you don't clamp the pins with 
> diodes.  This is the standard circuit for the battery driven, 
> time-of-day oscillator in your computer or your wristwatch.
> 
> 
> Third, a CMOS inverter can be ratioed in order to take an analog signal 
> and convert it to rail-to-rail.  Combine that with the MOSFET w/ crystal 
> and you have a rail-to-rail oscillator for exactly 3 MOSFETs.
> 
> 
> Fourth, unlike bipolar transistors, MOSFETs are very happy to be ganged 
> in parallel for more current drive.  They don't have a thermal runaway 
> problem where one transistor winds up taking more and more current the 
> hotter it gets.
> 
> 
> Now, I do agree that doing this with BJTs would be murder.  And, you 
> would probably have to find a nice 300-500mV threshold MOSFET.  The 
> standard discrete, metal-gate 1V MOSFETs might not cut it (but it might, 
> I'd have to do the simulations.  If you get even a *slight* amount of 
> post-linear, saturation behavior it would probably work).
> 
> -a
> 

Unlike digital watch logic, which consumes next to no quiescent current, 
a voltage multiplier that drives a LED draws substantial current. 
So the rds(on) of the MOSFETs is a serious problem, and you can't just
use commercial off the shelf low voltage CMOS inverters. 
Of course, it's always possible to design a little full custom CMOS ASIC
for a science project. After all, you shouldn't spare any expenses if it
is for your children's education. 

Christoph



-- 
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list

Reply via email to