Re: MISC-d Digest V99 #44

[Mail account for MISC mailling list <misc-list@xxxxxxxxxxxxxxxxxxxxxxxxxx>]

Lisbon, 21 April 1999

Jeff,

How small are the transistors in those ultra high ferquency counters?
Are they about the size of the geometry (.8 u)?
And is Chuck using traditional counters, built of digital gates made from
normalized circuits or has he optimized the electric circuit for a digital
counter directly made from small transistors (as can be done and was done
in the electronic's early days)?

I suspect that is the case but would like to know some more details if
possible so I could understand the circuits involved.

If I know anything about Chuck I think he wouldn't build a counter from
logical gates if he could do it directly with a whole circui t simple made
from transistors as active and passive components.
Working with gates is very usefull in forms of helping designing circuits
fast and that work regardless the electrical characteristics of the
electronic components used but as a side effect, they have many
transistors that are not used for the logical gate function but to assure
that the logical levels are always right regardless the fan-in and the
fan-out in every node.
If someone bypasses these transistors in a higly optimized circuit, than
can avoid many many delays. And, in this way, achieve higher counting
rates.

I know Chuck uses to build 'pattern' generators in his chips so he can
deduce electric characteristics for the internal transistors built. Are
those circuits 'spectrum generators' or are they just ultra high
ferquency digital counters?

On Wed, 21 Apr 1999, Jeff Fox wrote:

> >>I did not realize that the counter resolution was from an analog integrator.
> >
> >Jeff never states this explicitly. At current state of the art there is no way 
> >to make a 75 MHz f/f clock at 2 GHZ. No amount of magic dust will make it so.
> >
> >That leaves analog.
> 
> The echo timer web page details the fact that there are in fact digital
> counters running at 10gigahertz inside.  The echo timer is a digital circuit
> counting at ten gigahertz.  The analog I/O coprocessor has two ramp
> encoders that count through 256 digital steps in each analog sample.  At
> 40 msps that means that counter is also running at ten gig. 
> 
> Under the hood each instruction executes in 2ns.  Each instruction takes
> about twelve steps.  So if you break the chip up into large function 
> units they execute in less than .2ns.  Small circuits like counters can
> run in .1ns steps.   OKAD simulates at 15ps intervals.
> 
I supose he doesn't use only syncronous circuits syncronized to a GHz
internal clock, does he?
That involvs a number of problems and would require (IMO) to many gates
just for syncronization purpouses...
Does he use local clock generators for each sub-unit or mainly
combinatorial logic which will be syncronized globally in terms of higher
level functions?

> The literature at MOSIS states that .8u is fast enough for a 75mhz
> flip flop.  It is truely remarkable that Chuck is building 10ghz
> counters in this technology.  I told him not to talk about it because
> first of all people just wouldn't believe it and I worried that
> he would just get press for another quirky keyboard rather than
> the electronic circuit to drive it.
> 
Chuck, I think, deserves a high respect for the acheivements he has
acomplished throughout the years and I personally believe that the best
way for his technology survive all the dificulties that has been exposed
to is to share his knowledge in the theoretical field so that others can
also try to experiment with it if they like it.
I would but I do not have any experience yet in microcircuit design for
the much I would like to have. But it doesn't imply I so not understant
how circuit work at the lower level and want always to learn more...

Antonio Jose Gomes