THE LM1991 STORY


     LM1991Metal.jpg

Want to go where no man has gone before?
The real challenge may not be mother nature,
but rather human nature.

By the time the LM1991 Design Review was given, a standard
CMOS process had just become available to Linear Design
Engineers at National Semiconductor.

            DR_title.jpg

The design of the LM1991 was in response to Delco Electronics
asking for stereo 5 band audio equalizer chip which integrated all
the needed capacitors. At that time MF10 had just come out. Delco
wanted the
5 band audio equalizer to be all switch capacitor.

         MF10.jpg

The MF10 shown above had just come out and there was an on going
investigation of noise. A switch capacitor is just a way to
make a resistor out of a capacitor. In theory, the effective "resistor"
should have around the same thermal noise as a normal resistor
of the same resistance. The MF10 was not there yet, but on its way.

The problem with Delco's request is that the internal capacitor
required large effective resistances to work at audio frequencies.
Delco wanted noise levels far lower than what was being defined
by physics. If National Semiconductor developed this chip, Delco
would not be able to use it. While the 5 band equalizer did not
look feasible, National could integrate most of the other
audio signal processors functions into something economically
attractive. So a trip was made to Delco to offer something else
instead.

Well Delco really liked the second idea too. So the Consumer
Linear IC department (CLIC) ended up with two projects. But this
problem was quickly solved when Jim Solomon, the design manager of
the Standard Linear IC department (SLIC), found out that Tim Isbell,
the design manager of the Consumer Linear IC department (CLIC) was
even thinking about doing a switch capacitor IC. The Standard
Linear IC department (SLIC) had the charter on doing anything
involving switch capacitors. Tim Isbell with tongue in cheek
probably made Solomon a "Tom Sawyer deal" to let him have the
project. So Nello Sevastopoulos of MF10 fame was assigned the
5 band equalizer project.

         delcoCircuit.jpg
The schematic above is the actual circuit sent by Delco.
All the component values were kept pretty much the same as to
what was already inside a Delco Car Radio. By this time Delco
was putting National's 4 bit COPS microcontroller inside
their radios.


      serialproto.jpg
National's COPS group was developing a protocol which would
later be called "MicroWire". First an address would be
serially read in and then the data value.


                volumepots.jpg

For things like volume control, there was a parasitic vertical NPN
present which could be used to buffer any CMOS switches. The taps
on the volume control resistor needed to be scaled logarithmic like
what was already in the radio. Delco also added a loudness
control feature to match what they already were using. They also
needed a left and right "balance control" and a front and back
"fade control".



                     delcoCircuit.jpg
This was all fine until it came to the "tone controls".
Delco needed an Op Amp to do that function.

        CMOSnoise.jpg

The curves above shows the relative audio noise levels of
CMOS relative to BiPolar. This data was taken off of
large
CMOS geometries which have one picoFarad of gate capacitance.  
Notice that only two electrons are needed to create a third of a
microvolt across a 1pF capacitor.

Now no one had built up the Lateral NPN at this time. But
the noise of both the Vertical NPN and Lateral NPN should
be much better than CMOS. And Tim Isbell felt confident enough
to commit to Delco the development of the first Bipolar
Op Amp ever built on a standard CMOS process.

       lateralnpn.jpg
Design engineers commonly complain about Spice models.
Imagine trying to design a HiFi quality Bipolar Op Amp
using NPN transistors whose behavior could only be guessed
at. Still, it was the lesser of the evil to work on a
project which was somewhat uncertain when the alternative
was to do the stereo 5 band equalizer chip.

  
           bandgap.jpg

The only assumption that might be made was that match could
be counted on. In other words, could two Lateral NPNs be
counted on to be equally bad. The Bandgap and the Op Amps
where designed around that principle.

             toneneeds.jpg

The Op Amp also needed to be unity gain stable. Slow lateral
PNPs where pretty much the unstable elements in Op Amps. It
could only be hoped that lateral NPNs were at least better.

              OpAmp.jpg

The Op Amp really did not need any common mode performance.
And perhaps a little capacitance bypassing might allow
improve stability. The circuit above was put into the LM1991.
 
       CMOSswitches.jpg
The really big question was what would it sound like?
Fortunately it was possible to build up a breadboard using
actual CMOS transistors.


        breadboard.jpg

The breadboard shown above was built using CMOS transistors which had
no ESD protection. It was possible to do this because one quickly
learns the habit of automatically grounding oneself.

Switching the volume levels did not appear to be a show stopper.
One could hear the digital steps best with a sine wave and much
less with actual music. There needs to be voltage across the
resistor network for a change in switch position to be heard.
In the tone control circuit, there was some external capacitors
which could hold some charge. The following invention was
done to address that problem

4,523,156     Anti-distortion anti-transient tone control circuit

        invent.jpg

This invention was done before the term "clicks and pops"
came into existence because no digitally controlled
tone/volume/balance IC existed at this time.

         delcodnr.jpg
Delco also wanted to use National's new DNR IC while at
the same time save money on external capacitors. They needed
to be able to switch between several music sources which
could be then sent off the the DNR IC. And they wanted to
do it with only two DC blocking capacitors. So the circuit
continued to evolve.

          frontpanel.jpg

As far as anyone knew, Delco was perhaps National Semiconductor's
only audio customer who
was actually putting a microcontroller into
an audio product. Concern was raised as to how all the other audio
customers would still be able look at this IC. So some additional
analog ports where added which the purely analog customers could
use instead of the digital serial port.

         externalProg.jpg

The memory and serial logic needed to be inside the LM1991 anyway.
Adding the analog control required a little more CMOS logic and
an extremely crude ADC. So it was done. 

           audioInput.jpg

The digital port could be completely lab tested using just three
switches. The analog ports where designed to be tri state.

    firstSIL.jpg

Everything in the first silicon actually worked. Unfortunately,
National Semiconductor was doing a little more
trail blazing
than this. This was also the time when analog was switching from
rubys to CAD. Producing silicon was now on a painful learning
curve at many different departments. Departments were in the
process of confusing each other.

     workplatefields.jpg

The CD pattern was suppose to be a standard. But for a standard,
the figure above shows that it had become different everyplace
it was be used.

Notice the date is 1984. This is when the Macintosh first came
out. Much of the problem up to this time was the fact people
did not have an easy way to communicate using pictures.
Everything was being done using words. This was essentially
Abbott and Costello's "Who's On First" routine involving
everybody in Mask Making.

Jim Solomon was pushing hard for everything to be done on
computers. Analog engineers where pretty unhappy trading in
their breadboards for Spice simulations and their rubies
for GDSII files. The first CAD computers at this time were
Daisy workstations which still had their bugs. The GDSII
files were then shipped on to the Mask shop who were learning
how to debug Dracula on
mainframes.

Around this time period, pretty much nobody who was not in
Solomon's group was getting any circuits out of the mask shop
and on to first silicon. Solomon was collecting Berkeley CAD
people into his department who were in fact able to help
themselves out of most of this confusion. These CAD people
would later all leave National with Solomon to form what
ended up becoming Cadence.

By the time the LM1991 was finally coming out, Delco had a
better idea as to what they wanted. Even before the LM1991
had seen first silicon, the design of the LM1992 had started.

A product engineer named Randy Flatness was eager to get into
design. Even through the Mask Shop was still in a state of
major constipation, there was no shortage of design work.
So Randy got set up with everything from the LM1991 project.
And Dan Culmer,(who worked for Solomon) by this time had
actually built up the Lateral NPN. The first
silicon
of
LM1991 came out completely working. But by this time Delco
was only interested in the LM1992. This IC actually made
its way into production and spawned many other products.
And Randy Flatness gave a paper on it at the Consumer
Electronics Conference of Aug 1986.

Nello continues to work on the 5 band equalizer, even when
both he and the Delco engineers could see the noise challenges.
Ultimately the project died because the noise was just too
high. But Nello was not alone. Paul Gray, a good Berkeley
buddy of Solomon's, had recruited  Bill Jett and Milt Wilcox
from CLIC into doing a IF filter using switch capacitors.
Same story. The size limitations of internal capacitors
makes the noise far too high. But then again they gave a
very nice paper at the dec 1983 solid-state circuit conference.