The Lost History Of the Transistor

+
TRANSISTOR
SEMICONDUCTORS
THE LOST HISTORY OF THE
How, 50 years ago,Texas Instruments and Bell Labs pushed
electronics into the silicon age BY MICHAEL RIORDAN
44 IEEE Spectrum | May 2004 | NA
he speaker’s words were at once laconic and electrifying.
“Contrary to what my colleagues have told you about the
bleak prospects for silicon transistors,” he proclaimed in
his matter-of-fact voice, “I happen to have a few of them here
in my pocket.”
Silicon transistors? Did he say silicon transistors?
Yes—among the few in the world at that moment. It was
10 May 1954.
A long and till-then uneventful session on silicon devices had
been winding down at the Institute of Radio Engineers (IRE)
National Conference on Airborne Electronics, in Dayton, Ohio.
There, a parade of engineers and scientists were lamenting the
sobering challenges of developing and eventually manufacturing silicon transistors. Amid the torpor, scattered attendees were
stifling yawns, glancing at watches, and nodding off. But that was
before Gordon Teal of Texas Instruments Inc. made his surprising announcement—and jaws dropped in disbelief.
“Did you say you have silicon transistors in production?”
asked a stupefied listener about 10 rows back in the audience,
which now began to perk up noticeably.
“Yes, we have three types of silicon transistors in production,”
T
Teal replied, pulling several out of his pocket to the general amazement and envy of the crowd. Then, in a bit of quaint but effective razzle-dazzle, he cranked up a record player, which began
blaring out the swinging sounds of Artie Shaw’s big-band hit,
“Summit Ridge Drive.” Amplified by germanium transistors, the
music died out instantly as Teal dunked one into a beaker of
hot oil. But when he repeated his demonstration immersing a silicon transistor instead, the music played on without faltering.
As his talk ended, Teal mentioned that copies of his paper on
the subject, innocuously titled “Some Recent Developments in
Silicon and Germanium Materials and Devices,” were available
near the rear door. A crowd stampeded back to get them, leaving
the final speaker of the session without an audience. Minutes later,
a Raytheon engineer was overheard in the lobby shouting into a
telephone: “They’ve got the silicon transistor down in Texas!”
IN THE BEGINNING: Gordon Teal [left] directed the development of the silicon transistor at
Texas Instruments. William Shockley [middle] led the team at Bell Telephone Laboratories that
developed the very first transistor, which was made of germanium. TI’s silicon device with its
three long leads became famous, making the Texas upstart the sole supplier of silicon transistors
for several years in the 1950s. Morris Tanenbaum [right] at Bell Labs actually made the first
silicon transistor, but he felt “it didn’t look attractive” from a manufacturing point of view.
May 2004 | IEEE Spectrum | NA 45
At the time, the silicon transistor seemed to be one of the first
major breakthroughs in transistor development not to occur at Bell
Telephone Laboratories in Murray Hill, N.J., where physicists John
Bardeen and Walter Brattain had invented the transistor in December
1947. Their device featured two closely spaced metal points jabbed
delicately into a germanium surface—hence its name, the “pointcontact” transistor. They called one point the “emitter” and the other
point the “collector,” while a third contact, known as the “base,”
was applied to the back side of the germanium sliver. A positive electrical bias on the emitter enhanced the conductivity of the germanium just beneath the collector point, amplifying the output current
that flowed to it from the base.
Bell Labs achieved a long string of firsts
in the years following that momentous
invention, which it announced six months
later at a 30 June 1948 press conference in
New York City. Among its major advances
was the so-called junction transistor, first
conceived the previous January by William
Shockley, who led the group that included
Bardeen and Brattain. He figured that much
better transistor performance and reliability could be realized by eliminating the fragile point contacts and instead forming the
emitter, base, and collector as a single semiconductor sandwich with three different layers [see sidebar,
“Transistors 101: The Junction Transistor”]. Current flowing from
emitter to collector in Shockley’s device could be modulated by an
input signal on the base.
Teal (then working at Bell Labs) and his fellow physical chemist
Morgan Sparks successfully fabricated the first working junction transistor from a germanium crystal in April 1950. But—partly
because the frequency response of early junction transistors was
inferior to that of point-contact devices—Bell Labs held off
announcing this achievement for over a year, until 4 July 1951. Five
years later, Bardeen, Brattain, and Shockley shared the Nobel Prize
for inventing this revolutionary solid-state amplifier.
Their brilliant pioneering work has overshadowed much of the
subsequent development years of the transistor, including the
crucial change from germanium to silicon in the mid-1950s. That
shift in semiconductor material proved essential to the device’s
glorious future as the fundamental building block of virtually all
of today’s integrated circuits. For germanium, to put it simply, was
just not up to the task.
The material does have advantages: it is far less reactive than
silicon and much easier to work with because of its lower melting
temperature. And current carriers—electrons and holes—flow through
germanium more rapidly than through silicon, which leads to higher
frequency response. But germanium also has serious limitations. For
example, it has a low band gap (0.67 electron volts versus 1.12 eV for
silicon), the energy required to knock electrons out of atoms into the
conduction band. So transistors made of this silvery element have
much higher leakage currents: as the temperature increases, their delicately balanced junctions become literally drowned in a swarming
sea of free electrons. Above about 75 °C, germanium transistors quit
working altogether. These limitations proved bothersome to radio
manufacturers and especially the armed services, which needed stable, reliable equipment that would perform in extreme conditions.
Nowhere were these concerns appreciated more than at Bell Labs,
which led the way into silicon semiconductor research during the
early 1950s. Working in its chemical physics department with technician Ernie Buehler, Teal grew single crystals of silicon and “doped”
them with tiny impurities to make solid-state diodes in
February 1951, publishing the results a year later. He added specific
impurity atoms to the molten silicon to alter the electrical properties of crystals drawn from it. Elements from the fifth column
of the periodic table—arsenic or antimony, for example—create an
excess of electrons in the tetrahedral crystal structure, yielding
n-type silicon. Elements from the third column, such as boron or
gallium, create a deficit of electrons (usually regarded as an excess
of holes), yielding p-type silicon. By adding first one kind of impurity and then the other to the molten silicon from which they slowly
withdrew the crystal, Teal and Buehler formed transition regions
called pn junctions between the two types of silicon. Small bars cut
across these junctions act as diodes when
a potential is applied across them through
electrical contacts on the two ends.
Meanwhile, Calvin Fuller was beginning
experiments in an adjacent lab on diffusing impurity atoms from hot gases into the
germanium or silicon surface—one of the
major technology milestones on the road to
the integrated circuit. By December 1953
Fuller was so successful that Shockley
started building a new research team to
attempt to fabricate silicon transistors using
the technique. And early in 1954, Fuller and
Gerald Pearson formed pn junctions by diffusing a thin layer of boron atoms into a wafer of n-type silicon, making a hole-rich p-layer on its surface. These large-area diodes generated substantial current when sunlight fell on them. On 25 April,
Bell Labs trumpeted this achievement as the “solar battery,” the
first photovoltaic cell operating at efficiencies near 10 percent.
“They’ve got
the silicon
transistor
down in Texas!”
46 IEEE Spectrum | May 2004 | NA
By then TI had made its first silicon transistor—under Teal’s
general direction. Back at Bell Labs, he had become homesick for his
native Texas, where he had grown up a devout Baptist in South Dallas
and pursued his undergraduate studies in mathematics and chemistry at Baylor University, in Waco. Restless in Murray Hill, N.J., and
looking for more responsibility, Teal responded to an ad in The
New York Times for a research director at TI. He met with TI vice
president Pat Haggerty, who offered him the position. He began there
on 1 January 1953, bringing with him his vast expertise in growing
and doping semiconductor crystals.
Under Haggerty’s leadership, TI was moving aggressively into
military electronics, then burgeoning with the Cold War in full
swing. The Dallas company had been founded during the 1930s as
Geophysical Services Inc., developing and producing reflection
seismographs for the oil industry. During World War II, it snagged
a U.S. Navy contract to supply airborne submarine-detection
equipment; afterward it continued to expand its activities in military electronics, reorganizing itself as Texas Instruments Inc. in
1951. By the time Teal arrived, the firm had almost 1800 employees and was generating about US $25 million in annual sales.
The company was also beginning to manufacture what were called
grown-junction germanium transistors under the direction of engineer Mark Shepherd. He had attended a 1951 Bell Labs symposium on
transistor technology with Haggerty, where he listened to a Teal workshop on growing semiconductor crystals. In early 1952, after much
wheedling and cajoling by Haggerty, TI purchased a patent license
to produce transistors from Western Electric Co., AT&T’s manufacturing arm, for $25 000. By the end of that year, it was already manufacturing and selling them under Shepherd’s leadership.
Early the next year, Teal was back in Dallas organizing TI’s
PREVIOUS SPREAD: TEXAS INSTRUMENTS (TEAL, SHEPHERD, AND TRANSISTOR); MORRIS TANENBAUM
+++
TRANSISTORS 101: THE JUNCTION TRANSISTOR
JOHN MACNEILL
To understand how a transistor works, first consider the lowly
diode. It is a simple union of the two most fundamental kinds of
semiconductor, known as n-type and p-type. Both conduct
current, but the n-type does it with electrons, while the p-type
depends on electron deficiencies, better known as holes.
Joining these two types of semiconductors forms what is
known as a pn junction at their boundary. This is the core of a
semiconductor diode, which conducts current in one direction.
Connect a battery’s positive terminal to the n-type material
[figure A, top] and electrons are attracted to that terminal, while
holes in the p-type material move toward the negative terminal. In
other words, charge carriers stream away from the junction,
expanding a barren volume, aptly called the depletion region. The
diode is said to be reverse-biased, and hardly any current flows.
Now reverse the battery connections [figure A, bottom].
Electrons in the n-type material move toward the junction and are
constantly replenished by the battery. Meanwhile, holes in the
p-type material stream toward the junction, repelled by the positive battery terminal. The depletion region shrinks tremendously
as holes and electrons combine at the junction, neutralizing one
another, as more stream in on either side from the battery. The
diode is said to be forward-biased; current flows easily. Thus, a
diode can control the direction of current, but not how large it is.
A transistor, on the other hand, can control how much current
goes through it and also act as an amplifier. The simplest transistor has three parts: an emitter, a base, and a collector. Think of
the transistor as a sandwich of two pn junctions back to back
[figure B] in either npn or pnp order; they operate similarly.
In an npn transistor, for example, the n-type emitter has
many extra electrons, the relatively thin p-type base has a small
number of holes, and the n-type collector has a moderate number of electrons. (Junction transistors are also known as bipolar
devices because, in the emitter, holes and electrons flow in
opposite directions.) A transistor amplifier takes a small, varying
voltage—an input signal—between the base and the emitter, and
uses it to control a larger current flowing from the emitter to the
collector. That’s the output. The key agents in this amplification
are the depletion regions. With two pn junctions in the device,
there are two depletion regions: one between the emitter and the
base, the other between the base and the collector.
First, the emitter-base diode is forward-biased by a voltage
source [left in figure B]. Electrons flow from the emitter into the
base. The base-collector diode, on the other hand, is reversebiased, so that holes will not flow into the base, which would
intercept any electrons coming across from the emitter and
therefore block current from flowing through the device.
With this setup, the current through the transistor, from emitter to collector, is controlled by the depletion region around the
emitter-base junction. When it is thick, the current is choked off;
when it is thin, lots of current flows through the device. But hold
on—when it is thin, and electrons shoot across the emitter-base
junction, aren’t they blocked by the fat depletion region around
the base-collector junction? No—the base is narrow, so the
momentum of the electrons pouring in from the emitter brings
them close to that junction. From there, the positive voltage at the
junction then sweeps most of the electrons into the collector.
Only a few are lost in the base as they move into the vacant holes.
The transistor is designed so that the flow of electrons from
emitter to collector is very sensitive to the current into the base.
This is done by making the base very thin (so electrons don’t
have far to go before reaching the collector) and using low doping (electrons cannot easily find vacant holes to fill). The voltage
across the base-emitter junction provides the electric field that
drives electrons from the base into the collector.
With the emitter-base junction forward-biased, a varying voltage put on top of it—an input signal—varies the depletion region,
which in turn varies a relatively large current flowing through the
device. Add a load resistor in the collector circuit, and that small
varying input produces a much larger varying collector voltage:
the transistor amplifies the signal at the base. Depending on the
circuit, the result will be current, voltage, or power amplification.
Although bipolar junction transistors have been surpassed
for many applications by various forms of field-effect transistors,
bipolars remain popular for applications involving highfrequency signals. They’re found in countless modern electronic
devices, including broadband Internet modems, set-top
boxes, DVD players, and CD-ROMs.
—Alfred Rosenblatt
+
FIGURE A
-
Wide depletion region
n material
p material
Holes
Reverse-biased
pn junction
Electrons
Very low current flow
-
+
Narrow depletion region
n material
p material
Forward-biased
pn junction
C
FIGURE B
C
+
l
n material
B lB
V
p material
-
+
V
Hole
flow
Electron
n material flow
B l
V
E
Base
+
lE
Collector
-
V1
Hole
flow
Electron
Emitter flow
lE
E
May 2004 | IEEE Spectrum | NA 47
research department. Haggerty had hired him to build a team of
scientists and engineers that could generate enough ideas and technologies to keep the firm poised at the leading edge of the exploding semiconductor industry. Teal was up to the challenge. He was
introverted and difficult to work with, but also smart and stubborn.
These qualities had served him well at Bell Labs, where he pursued his crystal-growing research in the late 1940s, working doggedly
after hours with almost no support from management. Perhaps
most important, this pioneering research had made him a minor
celebrity in the fledgling industry, which would prove crucial in hiring bright young people for a group he had to create from scratch.
“We could never have attracted the stable of people that we did
without him,” Shepherd admitted in a 1993 interview. “And we got
some really outstanding scientists in those days.”
Among his new hires was Willis Adcock, like Teal a physical
chemist with a Ph.D. from Brown University, in Providence, R.I. He
had been working for a natural gas company
in Oklahoma and joined TI early in 1953.
Adcock began leading a small research group
focused on the task of fabricating “grownjunction silicon single-crystal and smallsignal transistors that would meet military
environmental conditions,” according to Teal,
who viewed this as the principal short-term
goal for his new research department.
It was no easy task at the time. Because
of a high melting temperature of 1415 °C
and its great reactivity, the molten silicon
from which crystals are drawn interacts
with just about any crucible that can contain it. Even fused quartz slowly reacts
with the melt, contaminating it with oxygen and other impurities that subsequently find their way into the silicon crystal, degrading its electrical performance. And most of the silicon samples then
available from suppliers came with substantial impurities.
Unlike germanium, which could be purified using zone-refining
techniques so that impurities could be reduced to about one part
per billion, the purest silicon available in those days had much higher
levels. And while silicon pn junctions had been fabricated for more
than a decade, ever since Russell Ohl first achieved this feat at
Bell Labs in 1940, making a successful npn or pnp junction transistor from this semiconductor material was far more difficult. [See
“The Origins of the pn Junction,” IEEE Spectrum, June 1997.] The
main problem was the extinction of so-called minority carriers
(electrons in p-type or holes in n-type layers) due to impurities in
the base layer. Electrons will easily “recombine” with holes at any
impurity centers within the base. Consequently, too few of these
minority carriers could survive while crossing this daunting bridge
between emitter and collector to achieve sufficient current gain, or
amplification, in silicon. The only solution to this problem, other
than struggling to purify the silicon, was to make the base layer
extremely thin so that the minority carriers would have some chance
of making it from one side to the other.
Adcock, Teal, and their team wrestled with these problems for
over a year. Then, in April 1954, using a special, high-purity silicon
purchased from DuPont at $500 a pound, they managed to grow a
suitable npn structure with an emitter region carefully doped to
enhance current gain and a p-type base layer about 1-mil (25 micrometers) thick. Cutting a half-inch (1.27-centimeter) bar from this crystal and attaching
electrical contacts on the morning of
14 April, Adcock’s group prepared to test
it. Soon Haggerty got an excited call from
Teal urging him to come see a demonstration. A few minutes later, “I was observing
transistor action in that first grown-junction
transistor,” Haggerty recalled at TI’s
25th-anniversary celebrations in 1979. It was
a defining moment for the budding semiconductor company. Realizing that another
company might well achieve the same breakthrough, Teal hurriedly wrote a paper for
presentation at the Dayton conference. And
held his breath after Bell Labs announced
the silicon solar battery later that month.
Smart, stubborn, and introverted, Gordon
Teal was up to
the challenge
+++
TRANSISTOR FIRSTS: Bell Labs’ junction transistor, of germanium, was fabricated in 1950 [left]. Texas Instruments’ commercial silicon transistor came four years later.
48 IEEE Spectrum | May 2004 | NA
LEFT: LUCENT TECHNOLOGIES INC./BELL LABS; RIGHT: TEXAS INSTRUMENTS
Another company, in fact, had already fabricated a working silicon transistor a few months earlier. In January 1954, Morris
Tanenbaum made one while working as a member of Shockley’s
research group at Bell Labs. But the world’s dominant semiconductor company kept this achievement under wraps, while the
Texas upstart rushed to announce it.
Tanenbaum had come to Bell Labs in June 1952 after earning degrees
ating it at up to 500 megahertz. Tanenbaum spearheaded the effort
to duplicate this device in silicon, succeeding on 17 March 1955,
with an npn transistor that worked at up to 120 MHz.
Thus, there was little enthusiasm for the rate-grown silicon
transistors that he had developed, and Bell Labs made no effort to
publicize the achievement. Tanenbaum presented his results at the
IRE Solid-State Device Research Conference in June 1954. During
the question-and-answer session afterward, he recalls, Teal mentioned similar work that had been done at TI—but was cagey about
specifics. Later that year Tanenbaum submitted a paper about his
research on rate-grown silicon transistors to the Journal of Applied
Physics, where it was finally published in June 1955.
By then the semiconductor industry was on the verge of yet
another fundamental shift. At the 1955 Solid-State Device Research
Conference held that same month, few people mentioned rategrown transistors. Everyone there was talking excitedly about the
newest breakthrough: diffusion. And Shockley was getting ready to
leave Bell Labs to start his own semiconductor company focused
on silicon transistors.
SILICON PRECURSOR: Gordon Teal (then at Bell Labs) [left] and fellow physical chemist Morgan
Sparks successfully fabricated the first working junction transistor from a germanium crystal.
AT&T ARCHIVES
in chemistry and physical chemistry at Johns Hopkins University,
in Baltimore, and Princeton University, in New Jersey. He started out
in the chemical physics department, growing large single crystals of
various semiconductors and testing their properties. In late 1953
Shockley invited him to join the team being formed to push toward
silicon transistors. Tanenbaum continued working with Buehler, Teal’s
former technician, whom he describes as “a master craftsman in
building apparatus and growing semiconductor crystals.”
Buehler was working on a technique known as rate growing. The
rate at which impurity atoms (such as gallium and antimony) are
incorporated from the melt into the crystal depends to a great extent
on the crystal’s growth rate—on how rapidly it is being pulled from
the melt. Both impurities are present in the melt simultaneously,
but the rate at which either one crystallizes out depends on the
pulling speed. This process enabled the team to make much narrower base layers, just 13 to 25 micrometers (µm) thick, which proved
to be crucial in limiting the extinction of minority carriers.
Tanenbaum cut a half-inch bar from one high-purity silicon crystal that Buehler had grown using special samples from DuPont; then
he attached an aluminum lead to the narrow base layer and carefully
reheated the silicon to restore the layer’s p-type behavior. On
26 January 1954, according to his logbook, he achieved sufficiently
high electron current and hence amplification in an npn silicon transistor. “I believe these were the first silicon transistors ever fabricated,” says Tanenbaum, savoring the moment in an interview nearly
half a century later.
“When we made these first [silicon] transistors,” he continues, “we thought about patenting the process but determined
for two reasons that it wasn’t worth the effort.” For one, others
had developed and used similar techniques. And he really did not
like the rate-growing process, which had already been patented
by General Electric Co. “It just wasn’t controllable,” he adds. “From
a manufacturing point of view, it just didn’t look attractive.”
At the time, Shockley’s group was concentrating on adapting
the new diffusion process pioneered by Fuller to the fabrication
of germanium and silicon transistors. Diffusion seemed much
more promising—as indeed it proved to be—because it was substantially more controllable and could yield much narrower base
layers, just micrometers thick, and hence transistors that work at
higher frequencies. In July 1954 Charles Lee made a successful germanium transistor at Bell Labs using diffusion techniques, oper-
+++
It is hardly surprising that the silicon transistor was invented
twice, in two seemingly independent achievements just months apart.
By 1954 the crucial underlying technologies of silicon purification and
crystal growth were at a point where the silicon transistor was perhaps inevitable, given the market demands—which were quite different for the two companies. TI was focused on military markets for
transistors as replacements for the bulkier and far more fragile vacuum tubes. The U.S. armed services, among its biggest customers,
were willing to pay a big premium for transistors that performed uniformly and flawlessly over a wide range of conditions. Bell Labs’ largest
“customer” was AT&T’s Bell Telephone System, which needed rugged,
long-lived semiconductor switches that were truly “off” when they
were supposed to be off. Because of high leakage currents, especially at elevated temperatures, germanium transistors simply could
not satisfy either of these important customers.
It is also obvious that the two achievements had common technological roots reaching back to the pioneering crystal-growth
research of Teal and Buehler at Bell Labs from 1949 to 1952. Teal
brought this expertise with him to TI, although perhaps not the
rate-growing techniques developed a bit later by Buehler. The two
groups both benefited from the fact that DuPont saw a growing
market for high-purity, “semiconductor-grade” silicon and was
beginning to supply small samples of the stuff in 1954. In both
cases, the road to the silicon transistor had to cross a narrow, highpurity bridge made of the element.
Amidst all else that was happening at Bell Labs in the early 1950s,
the first silicon transistor may not have seemed important enough
to merit the same public attention given earlier transistors and the
solar cell. At the time, the managers were likely looking ahead eagerly
to what they considered the real breakthrough: transistors fabricated
using diffusion that operated at over 100 MHz. And overconfidence
may have played a role, too. Bell Labs had habitually kept mum for
months after its earlier breakthroughs, thereby permitting its scientists and engineers to work out most of the patentable ramifications before going public.
Whatever the case, the delay allowed fledgling Texas Instruments
to leap forward and claim victory in this race. And it stood alone as
the first company to manufacture silicon transistors in volume. Thanks
to its foresight and aggressiveness, TI had the silicon transistor market essentially to itself for the next few years—and started down the
■
road to becoming the international giant we know today.
May 2004 | IEEE Spectrum | NA 49