Re: HERE IS: published circuit of non-monotonic output driver

Sorry for the late reply.

The note from Mark Johnson motivated me to do some digging and I found
the following...

The Petrovik(spelled Petrovick) paper refers to CMOS 4S/Hydra drivers,
which were not the subject of my IBIS study. I have not seen their
DC output IV curves. I suspect they would cause non-monotonic behavior
like Phoenix.

OMNI and Phoenix were designed after these drivers, and this compensation
scheme was improved upon. There is a paper in 1993 CICC digest
of papers(section 25, paper 5), that describes the driver compensation
scheme used in Phoenix...

    Makoto Ueda, Allen Carl, et al, "A 3.3V ASIC for Mixed
    Voltage Applications with Shut Down Mode", IEEE
    1993 Custom Integrated Circuits Conference,
    Digest of Technical Papers ("CICC-93"), pp. 25.5.1.

I would like to point out that the nodes on the reference stacks
between devices T11 and T12 are not connected(this caused problems).
T12 Drain is tied to Vdd and T11 Drain is tied to GND. The shut off
mechanism is provided by an extra transistor in the stack whose gate is
fed from the output node via a half latch.

In summary, the non-monotonic curves are created by a circuit of
the form

            Vdd
             |
            --- IO
         | | | PAD----------
   A O---| | P | | |
         | | | | |
            --- --- |
             | Vgate | | | |
             O-----------------O-----------------------| | N | |
             | | | | | |
            --- | Vdd --- |
         | | | | | | |
   A O---| | N | | --- GND |
         | | | | | | | |
            --- | | P | |---O A |
             | | | | | |
            GND | --- |
                               | | |
                               | | |
                               | | |
                               | --- --------- |
                               | | | | | Delay | |
                               | | N | |-----| Buffer |----------
                               | | | | | |
                              --- --- ---------
                             | | | |
                             | N | |----O Vref
                             | | | |
                              --- ---
                               | | | |
                              GND | N | |---O Vdd
                                      | | |
                                       ---
                                        |
                                       GND

where A is a control signal such as (data NOR user_enable_not). Just
the simplified pull-down is shown above. During the transition,
Vgate < Vdd
is defined by the voltage stack reference voltage Vref. This voltage
is larger at high Vdd and fast process conditions to minimize the
driver performance window. This sets up the low Vgs at hi Vds curve in the
IV curves for IBIS. When the transition is finishing(the worst
di/dt region is past) the feedback delay buffer shuts off the
stack and Vgate = Vdd(now Vds on big output decvice is small). This
is the hi Vgs at low Vds curve. The IV curve also shows the transition region
between the curves. The success of this scheme is highly dependent
on the tuning of the transistors involved. As mentioned, care must
be taken such that the stack shut off does not cause an impedance mismatch
on the line that generates new reflections.

Variations of this scheme were used in OMNI and Phoenix. So the answers
to the $64K questions are 1.) yes and 2.) yes.

John Brennan

************************************************************************
John Brennan Internal: jbrennan@btv
IBM Microelectronics Internet: jbrennan@vnet.ibm.com
862C / 1000 River Road Tel: 802-769-6982 (Tie-Line: 446)
Essex Junction, VT 05452 Fax: 802-769-5882 (Tie-Line: 446)
************************************************************************

------------------ Referenced E-Mail ----------------------------

Date: Fri, 01 Sep 1995 17:22:58 PDT
To: ibis@vhdl.org, si-list@silab.Eng.Sun.COM
From: mjohnson@netcom.com (Mark Johnson)
Subject: HERE IS: published circuit of non-monotonic output driver
Message-Id: <199509020022.RAA23496@netcom2.netcom.com>
The schematic diagram of an output buffer which displays
non monotonicity (measured at low frequencies, for
example in a direct current I-V plot), is published in
the following paper:

    J. Petrovik et al, "A 300k-Circuit ASIC Logic Family",
    1990 International Solid State Circuits Conference,
    Digest of Technical Papers ("ISSCC-90"), pp. 88-89.

Figure 6 of that paper shows a remarkably clever output
driver circuit which compensates for process variations,
supply voltage variations, and temperature variations.
It performs this compensation better than any other circuit
I've ever seen. (Quiz: but there's a cost; find the
cost. Hint: it starts with the letter "p")

If you put this output buffer on your curve tracer
to measure the I-V curves, AND if you took more than
10 nanoseconds to move from one point on the curve
to another point (i.e. if the A/D converter inside
your curve tracer needs more than 10ns to settle),
then you would see the exact I-V curve that Arpad
Muranyi drew:

>
> *
> * *
> * * * * * * * * *
> *
> *
> *
>

The region where the slope of the I-V curve goes negative
(an incremental *negative resistance*, not usually felt to
be the greatest thing in the world to hook up to a
transmission line), is caused by the compensation circuitry.
As Arpad hypothesized, there is indeed feedback in this
output driver circuit. And, oh by the way, the circuit
is symmetric; it has negative resistance in both its
IOH-VOH curve and also in its IOL-VOL curve.

Take a look at Figure 6 of the paper. Better yet, put it
in your local SPICE simulator and simulate it "at speed"
(transient) and also simulate a DC transfer curve. Don't
believe me, I'm some random guy you've never met. Don't
believe this message, it's just phosphor dots dancing on
your CRT. Be self sufficient, try it yourself and
make up your own mind.

There are two $64,000 questions:

   1. Does anybody actually _use_ circuits like Figure 6
       in real products?

   2. When you run the thing at full speed (transient
       analysis), does it work OK?

Final remark: the authors of this paper were from
"IBM General Technology Division, Essex Junction,
Vermont, USA".

--Mark Johnson
Received on Fri Sep 15 14:18:52 1995