IEEE Solid-States Circuits Magazine - Summer 2022 - 16

SHOP TALK: WHAT YOU DIDN'T LEARN IN SCHOOL
Chris Mangelsdorf
The Strangest ADC Ever Built
I
Long Ago and Far Away . . .
It is early in 1981. Cambridge is just
starting to thaw out. The harsh, leafless
landscape makes for an intolerable
glare and gives the reluctant
New England spring a postapocalyptic
cast. The campus looks more
depressing than ever.
A lone graduate student is huddled
over a metal desk, wondering
for the nth time why he did
not accept Berkeley's invitation to
California. It's more than just the
weather. Analog circuit design is not
in favor at this gray and forbidding
institution. The faculty has been
seduced away by the glamor
of younger digital technology
and its sexy cousin,
computer science. In fact,
there is talk across the
street of creating " artificial
intelligence, " and,
if you believe them, it's
poised to take over any
day now.
But none of that matters
to our young graduate
student. He is
butting his head against an arcane
technology that promises-but rarely
delivers-huge advances in integration:
merged-function design.
The Lost Dream of a
Former Generation
What you need to understand is
that, in 1981, transistors are huge.
(Some fab workers wear steel-toed
shoes in case they accidentally drop
Digital Object Identifier 10.1109/MSSC.2022.3184440
Date of current version: 25 June 2022
16 SUMMER 2022
an IC.) And wafers are small. The
state-of-the-art diameter is still 4 in,
or about the length of a Twinkie. Single-die
ICs are referred to proudly
as monolithic, even though a strict
translation of that term would mean
" from a single large stone. " Minimizing
transistor count and passive
components to save area is still a
primary goal.
While there is much excitement
Our student
is butting his
head against
an arcane
about ever-higher levels of integration,
the deep thinkers among us
wonder if maybe we've missed something.
It all seems too " flat. " Is this
2D network of transistors the best
we can do? The 3D
integration we know
today is still out of
reach in 1981, but what
if we somehow merged
technology that
promises-but
rarely delivers-
huge advances
in integration:
merged-function
design.
the function of one or
more devices to do a kind
of " superintegration " ?
That sounds simplistic
and naive today, but in
1981, it seems like a good
idea. For example, designers
working with lateral
PNPs quickly discovered that you
could incorporate multiple collector
diffusions and produce multiple
current source outputs from a single
device. And these outputs could
be weighted by the relative size of
the collectors. If you connect one of
the collectors to the base contact,
you have a current mirror. (See Figures
6 and 7 in [1] for examples.)
In other words, you can make this
complete functional block in one
well and use only a little more space
than a single transistor would conIEEE
SOLID-STATE CIRCUITS MAGAZINE
sume. This is surely a paradigm
worth pursuing.
(Aside:
Is all
this bipolar stuff
is putting you off? Don't let it.
Just think of the bipolar transistor
as a MOSFET with a leaky gate.
FIGURE 1: Believe it or not, this is the
layout of a 4-b ADC (without the metal).
It needs only four comparators and eight
current sources to complete the function. It
uses an old bipolar process, but it can keep
up with the fastest converters of its day,
and it consumes only a small fraction of the
die area and power of its competitors.
IEEE Solid-States Circuits Magazine - Summer 2022

Table of Contents for the Digital Edition of IEEE Solid-States Circuits Magazine - Summer 2022
