Dealers of Lightning (30 page)

Considerable fine-tuning was necessary to keep this complicated sys­tem humming along. Assembling the hardware and synchronizing the
components was like getting a herd of cats to sing in unison. Since the
polygonal disk spun at 10,000 revolutions per minute (the original glass
prototype was soon replaced by aluminum), even the way the facet
edges "paddled" the air produced measurable resistance. The laser
itself had to be modulated up to fifty million times a second by a "shut­ter" fashioned from a polarizing filter driven by a $10,000 piezoelectric
cell. And because it had to conform to the speed of the copier, Stark­weather's laser apparatus had to mark more than 20 million dots on a
page every second.

Still, the most troublesome problem was not electronic. Instead it fell
squarely within the domain of traditional optics. Starkweather knew that
if
the mirrored facets were even microscopically out of alignment, the
scan lines would be out of place and the resultant image distorted or
unintelligible, for the same reason a wobbly tape deck makes an audio-cassette warble as though recorded under water. To produce clean
images, he calculated, the facets could not be out of vertical alignment by
more than an arc-second
—
a microscopic variance. In visual terms, the
mirrors could not be off by more than the diameter of a dime as viewed
from a mile away.

Disks fabricated to such an exacting standard would cost at least
$10,000 each—assuming this were technically possible, which Stark­weather doubted. It was true that there existed servo-mechanical and
optical devices that could quite effectively redirect an errant scan back in
place. But they were even more expensive and, as a further drawback,
meant adding another complicated and failure-prone component to his
printer. Starkweather understood that the tolerance issue was critical. If
he could not solve it, he would have designed a machine that could not be
cost-effectively manufactured.

For more than two months he wrestled with the puzzle. "I would sit
and write out a list of all the problems that were difficult. One by one
they would all drop away, but the mirrors would still be left."

One day he was sitting glumly in his optical lab. The walls were painted
matte black and the lights dimmed in deference to a photoreceptor dram
mounted nearby, as sensitive to overexposure as a photographic plate.
Starkweather doodled on a pad, revisiting the rudimentary principles of
optics he had learned as a first-year student at Michigan State. What was
the conventional means for refracting light? The prism, of course. He
sketched out a pyramid of prisms, one on top of another, each one
smaller than the one below to accommodate the sharper angle of neces­sary deflection. He held the page at arm's length and realized the prisms
reminded him of something out of the old textbooks: an ordinary cylin­drical lens, wide in the middle and narrowed at the top and bottom. "I
remember saying to myself, 'Be careful, this may not work. Its way too
easy.' I showed it to one of my lab assistants and
he
said, 'Isn't that a little
too simple?'"

It
was
simple. But it was also dazzlingly effective. Starkweather's
brainstorm was that a cylindrical lens interposed at the proper distance
between the disk and the photoreceptor drum would catch a beam
coming in too high or low and automatically deflect it back to the
proper point on the drum, exactly as an eyeglass lens refocuses the
image of a landscape onto a person’s misaligned retina.

"I ran to the phone and called Edmund Scientific, my supply house,
gave them my credit card, and bought ten bucks' worth of war surplus
lenses," he recalled. "I could hardly sleep the two days before they
arrived. But then they came, I put them in, and sure enough they
worked." The lens scheme was foolproof. It involved a simple physical
relationship, so it could never fail. It had no moving parts, so it could
never malfunction. And it permitted the polygonal disks to be stamped
out like doughnuts—not at $10,000 apiece, but $100.

"The mirrors no longer had to conform by the diameter of a dime at
a miles distance," Starkweather recalled. "They could be off by the
diameter of a tabletop, which was a standard anyone could meet. I
made a lot of discoveries building that machine, but it was the cylindri­cal lens that made me say 'Eureka!'"

Starkweather’s finished printer was a large, bulky machine. His open
arrangement of plump black-tubed lasers, mirrors, and wires sat atop
the clean but stolid Model 7000 copier like a ridiculous hat on a dowa­ger aunt. He christened the machine SLOT, for "scanning laser output
terminal."

"I would have called it the scanning laser output printer," he said,
"but that wouldn't have made a very good acronym."

Building the SLOT solved only half the riddle of how to convert dig­ital images to marks on paper—the back end, so to speak, of how to
apply toner once the image was delivered to the laser beam. The front
end involved translating the computer's images into something the
laser could actually read.

That half was solved by the invention of the so-called Research Char­acter Generator (RCG), another healthy piece of iron and silicon, by
Lampson and a newly hired engineer named Ron Rider. The RCG,
which stood several feet high and nineteen inches wide, and housed 33
wire-wrapped memory cards holding nearly 3,000 integrated circuits,
was a sort of super memory buffer, spacious enough to accept a digital file
from a computer, evaluate it scan line by scan line, and tell the printer
which dots to print at which point. This generated on paper an image cre­ated by pure electronics.

Today this procedure is trivial. Memory is so cheap that the computer
and printer both come with enough to hold several pages at a time. As a
page comes in from a word-processor program, it is fitted into a print
buffer the way craftsmen of the old
printing
trades clamped lines and
columns of leaded type into rectangular frames. Once in memory, the
page image can be manipulated in an
almost
infinite number of ways.
It
can be fed to the printer narrow or wide
end
first, backwards, upside-
down, or wrapped around a geometrical design. The most unassuming
desktop computer can store character
sets in
dozens of font styles and
sizes, any of which can be summoned at will and applied to
a
document
as a paintbrush swipes color at a wall.

Nothing like this was simple in 1972
because
of the cost of memory.
Nor was it enough for Rider’s machine
to
generate only the bland stan­dardized
ASCII
text of conventional line printers. The
RCG
had to
incorporate a large number of
custom typefaces
that were to be drawn
by hand, converted into digital bits, and stored somewhere in memory
until needed, as if on an electronic shelf.

This meant an exponential increase
in the
complexity of the task.
ASCII
characters were all the
same size and
each fit into the same
squared-off shape. The only formatting
a
conventional document nor­mally required was a command instructing
the
printer when to move to
the next line. By contrast, the custom-designed characters
PARC
desired
to print would be proportionately
spaced:
some fat, some thin, some
reaching above the print line, some dangling below; some roman, some
italic,
some
BOLD.

Finally, the character generator had
to adapt
to the Model 7000’s sys­tem of feeding in pages wide-edge-first, which moved paper through the
machine at a faster rate. For copiers this posed no problem—one simply
aligned the originals along the same axis. For
a
printer, however, it was a
horror. The image coming from the computer would somehow have to
be rotated before it could be printed out. Instead of printing a page in
prim linear order like a typewriter, SLOT would have to reproduce the
characters in vertical slices, somehow keeping its place on twenty or
thirty lines of print per page.

Rider ultimately came to see the proliferation of complications as a
blessing in disguise. "It forced you to think about the problem of printing
in a much more generalized fashion, so the solution turned out to be
much more robust." Despite its name, the research character generator
was less about delivering images character-by-character than about trans­mitting digitized images in whatever form the computer dictated. Like so
much PARC developed in those first few years, this turned out to be the
answer to a multitude of questions no one was yet even asking.

Starkweather and Rider worked together on coordinating the SLOT
and character generator until early 1972, when they were stymied not
by a technical obstacle but one entirely man-made. This was the relo­cation of more than twenty of PARC's seventy scientists up the hill to a
building newly rented from the Singer Company and known as Build­ing 34 (because its address was 3406 Hillview). The Computer Science
Lab, including Rider, got bundled off to the new quarters while every­one else, including Starkweather, temporarily stayed behind on Porter.
The move separated the two by a kilometer of real estate
—
too far to
string an overhead line and, with the four-lane Foothill Highway in the
way, impossible to link via a ground cable.

"The administrators said, 'Don't worry. You'll be back together in
another year,'" Starkweather recalled.
"I
said, 'Great, what are we sup­posed to do in the meantime?'"

But one Sunday afternoon shortly after the move Starkweather got a
brainstorm while sitting at home. He immediately jumped in his car,
drove to Porter Drive, and mounted a stairwell to the roof. Just as he
had thought, he could take line-of-sight aim from where he stood to
the rooftop of Building 34. He might not be able to span the distance
by cable or wire
—
but he could do it by laser beam.