Silicon Burns !
Carl Frosch and the Discovery of Oxidized Silicon
a remembrance by Mason Clark
A key process in the building of silicon devices is the step of burning
an oxide layer onto the silicon wafer. The layer is the start of the
photographic processes that create patterns on the surface. These patterns
make up the devices and the interconnections of the integrated circuit.
The oxide layer is a mask for localizing diffusion and metal deposition.
In the words of Andy Grove in his 1967 book, "an exceedingly
important factor in the development of planar technology." It was the
over-throwing of a silicon dogma: "silicon burns". I'll
tell its story from my own observation at the Bell Telephone Laboratory.
I was developing germanium power transistors using a process called
"alloying." It was difficult by alloying to produce flawless,
flat p-n junction surfaces over large areas. A foreign particle or stain
on the surface would cause a defect. An alternative was needed.
It was also time to switch from germanium to silicon. Alloying p-n
junctions on silicon would have this large-area defect problem --
especially troublesome in power transistors. An alternative to alloying
might be the diffusion of dopants, the impurities to create "n"
or "p" type layers. The first report of the alloying method was
mistaken by some at the Schenectady General Electric laboratory to be
diffusion, so the notion of diffused junctions, but not the reality, was
introduced by this error. Research on diffusion was being done by
To perform diffusion into silicon it was necessary to heat the silicon
in a protective atmosphere, preferably vacuum or hydrogen, to avoid simply
burning up the silicon. Carbon (you might say "coal") and
silicon are first cousins in the periodic table. Each will surely burn if
heated in air. So Ditzenberger and his assistant, Fuller, carefully sealed
wafers of silicon in a quartz capsule together with the dopant and vacuum
or hydrogen. By this means they provided data on the diffusion of boron
and phosphorus into silicon and opened the path to large-area p-n junction
perfection. But the technique was awkward to perform. Carl Frosch sought
an alternative by heating the silicon in an open quartz tube, using a flow
of hydrogen to prevent the silicon from burning.
When I went looking for diffusion techniques for power transistors I
found Carl in the fourth-floor attic of the Murray Hill laboratory. He
didn't have a proper lab for his furnaces and bottles of hydrogen. His
furnaces were the laboratory furnaces available in 1956, using two rods of
silicon carbide as heating elements. In his make-shift laboratory he used
hydrogen tanks and rubber tubing. He was just beginning to do diffusion by
adding impurities to the hydrogen stream. The technique had promise of
being more manufacturing-suitable than Ditzenberger's sealed quartz.
The next time I visited Carl he had moved to a proper laboratory, a
move that had historic importance. He was excited to show me his
discovery. His silicon wafers had turned purple--a beautiful, uniform
purple. What had happened? He and his assistant had hastened to inspect
their system and correct the problem. Carl told us of the inspiration he
had while he was driving home the evening of their purple disaster. He
remembered that when they re-installed their furnace after the move there
had been a problem with leaky rubber tubing in the hydrogen line. Air,
i.e. oxygen, must have gotten into the furnace. So the next day they
heated a silicon wafer without the flow of hydrogen -- in air. And there
was the oxide layer on which the integrated-circuit industry rests. Highly
pure, single-crystal, silicon forms a protective layer of oxide.
Frosch quickly learned to selectively remove the oxide, using
hydrofluoric acid, and reported, "Surface Protection and Selective
Masking During Diffusion in Silicon," J. Electrochem. Soc., 104, 547
It was the accident of leaky tubing and the alertness of Carl Frosch
that overcame the dogma that "silicon burns." K. D. Smith was
puzzled by Carl's discovery and took out of his ample desk drawers a
sample of the silicon used to make microwave diodes in World War II. This
he placed in a furnace, in air, and confirmed that "silicon
burns." The sample survived but looked like a particle of moss --
overgrown with dendrites of silicon and oxide. It was this early
experience with polycrystalline silicon, along with the science of the
periodic table, that had established in our minds that silicon must not be
heated in oxygen. When I moved to Pacific Semiconductors a year later, a
very competent doctor of chemistry warned me that I must not attempt to
heat silicon in an open-tube furnace lest it simply burn up. "Look at
the periodic table," he scolded. Fortunately, Carl Frosch had just
published his paper proving otherwise.
An oxidized silicon wafer with aluminum dots for evaluating the purity of
the oxide layer. The oxide is colorless. The color seen is caused by
interference between the reflections from the silicon and from the outer
oxide surface. The color is used to measure the thickness of the oxide,
which is about the wavelength of light.
Mason Clark was a member of the technical staff of the Bell Telephone
Laboratories 1952 to 1957. He was supervisor of power transistor and
microwave diode development. He became product development manager at
Pacific Semiconductors (PSI) and later at Hewlett-Packard Associates. His
web site covering his retirement interests is at