Mortars and Mass Production

Jay Lathrop was trying to fit a working radio inside an artillery shell. The shell was an infantry mortar round only a couple of inches across at its widest. The radio was a proximity fuze, a transmitter and receiver that would bounce a continuous wave off the ground as the round arced toward its target and detonate the warhead at the optimum height above it for fragmentation. The fuze had to survive the shock of being fired out of a tube. It had to fit, with batteries and antennas, into a volume smaller than a coffee cup. And every component had to be reliable enough that an infantry sergeant would trust his life to it. It was 1957, and Lathrop was working out of a low brick building at the corner of Connecticut Avenue and Van Ness Street in northwest Washington.

The Army had been chasing this problem for years. Proximity fuzes were one of the secret weapons of the Second World War, the radio-pinging devices that turned naval anti-aircraft fire from a hopeful spray into a precision instrument and broke up the German Ardennes counterattack with airbursts over forest canopy. After the war, the engineers who had invented those fuzes settled into a converted laboratory on the old grounds of the National Bureau of Standards. In 1953, the Army formally took the lab over, renamed it the Diamond Ordnance Fuze Laboratory after Harry Diamond, the radio engineer who had led much of the original wartime work, and pointed it at the next problem: shrinking the wartime fuzes by an order of magnitude so they could ride inside infantry mortars and field artillery, not just naval shells.

Lathrop had arrived at the lab in 1952, fresh from MIT with a physics doctorate on microwave gas discharges under Sanborn Brown. He was twenty-five, from Bangor, Maine, and intellectually trained on plasmas, not weapons. The Army assignment was practical to the point of grubby: take a circuit, make it small. He was paired with another young engineer, James Nall, who lived in Silver Spring with a young family and had a habit of staring at problems until he found the seam where they could be split open. The two men worked the kind of slow, fiddly bench science that does not appear in heroic histories. They wired up transistors. They cracked open germanium. They tried again.

The transistor at this point was less than a decade old, and the methods for making one were artisanal. To shape the semiconductor crystals at the heart of a transistor, manufacturers covered a small block of germanium with wax, scratched a line through the wax with something like a phonograph needle, and dipped the assembly in acid. The acid ate the germanium where the wax had been removed and left the rest. The geometry of the finished device was the geometry the wax happened to take. Wax under heat oozes. Wax does not draw fine lines. By 1956 a typical transistor was the size of a pencil eraser, and the engineers building one understood, in the unspoken way of bench workers, that there was a wall coming. You could not get a usable transistor down to the size of a pinhead by drawing on it with hot wax.

The wall mattered because the demands on Lathrop’s lab were absurd. A 1956 internal sketch of what an electronic mortar fuze ought to contain ran to dozens of components: oscillator, mixer, audio amplifier, thyratron switch, batteries, safe-and-arm mechanism. The available volume was a few cubic centimeters. A handful of well-funded competing programs were trying to solve the same packaging problem in mechanical ways. Project Tinkertoy, started in 1950 at the Bureau of Standards under a Navy contract, had assembled tiny components onto five-eighths-inch ceramic wafers stacked into modules that could be plugged together like the children’s construction toy whose name it borrowed. The Army Signal Corps and RCA were running a successor program, the Micro-Module, built around 0.36-inch ceramic squares with notched edges that could be soldered into stacks. Both programs were essentially mechanical: better packaging of conventional parts. Neither tried to change how the parts themselves were made.

Lathrop and Nall’s idea, when it came, came sideways. Nall noticed something about a chemical that had nothing to do with semiconductors. In aircraft factories, mechanics laying out the rivet patterns on metal wings used a photosensitive coating, made by Eastman Kodak, that hardened when exposed to ultraviolet light. The unhardened parts could be washed away with solvent, leaving the hardened pattern as a stencil. Kodak called the family of products photoresists, and they were sold in two flavors, KPR and KMER, for shop applications no more glamorous than printing nameplates. If the resist could mask the boundary of a rivet hole on aluminum, Nall thought, it could mask the boundary of a contact on a piece of germanium. The acid would not eat where the resist had hardened. The geometry would be the geometry of the light.

The trick was getting the light to draw something small enough to matter. A photographic mask, hand-cut from rubylith or photographed onto glass, could carry a pattern at a few millimeters’ resolution. A few millimeters was useless. Lathrop and Nall needed features at a few thousandths of an inch. There was no obvious way to print at that scale. They had a piece of laboratory equipment, though, that could do the opposite of what it was designed for. A standard biological microscope takes a small object and projects an enlarged image of it into the eye of the user. If you ran the optical path backward, sending light down through the eyepiece and out the objective lens onto a stage below, the same optics would shrink an image from large to small. The microscope they had on hand could reduce by a factor of around fifteen. A pattern drawn at a comfortable size on a glass mask could be projected, in miniature, onto whatever sat on the stage.

The first test version was almost embarrassingly homemade. Lathrop and Nall took a microscope, mounted it upside down with its objective lens pointing at a small stage, and put a coated germanium wafer where the slide normally went. They placed the mask, drawn at about fifteen times final size, where the specimen would have been. They aimed an ultraviolet lamp through it. The light passed through the open windows in the mask, was bent by the microscope optics, and converged onto the wafer in a sharply reduced image. Where the light hit, the photoresist hardened. Everywhere else it stayed soluble. Solvent rinsed away the unhardened resist. Acid ate the exposed germanium beneath. The geometry of the wax had been replaced by the geometry of light.

What came off the stage when the process worked was, by 1957 standards, miraculous. Transistor features that had been scratched at hundredths of an inch could now be defined at a few hundred-millionths of a meter. Lathrop and Nall built mesa transistors directly into rectangular openings on a small ceramic substrate the size of a postage stamp, with silk-screened resistors and conductors already on it, and used the same photoresist technique to evaporate gold and aluminum leads exactly where they needed to land. The result was a hybrid integrated circuit, two-dimensional rather than stacked, in which the transistor was no longer a discrete part to be soldered in. It was an integral part of the printed circuit itself.

Bell Labs had been moving in a parallel direction since 1955. Jules Andrus and Walter Bond, working with the silicon dioxide masking developed by Carl Frosch and Lincoln Derick, had adapted the photoengraving methods used to make printed circuit boards to define windows in oxide on silicon wafers. Andrus would receive his own photolithography patent in 1964. Bell’s contribution was real and earlier than Lathrop’s, and it would shape how silicon devices were eventually made, but Andrus and Bond were doing contact printing: pressing a one-to-one mask directly against the wafer surface, the way printers had laid down circuit boards for years. What Lathrop and Nall added was the optical reduction step. The mask did not have to be the size of the chip. It could be drawn large, where humans could draw, and shrunk in the optics. That single change was what made photolithography a tool of miniaturization rather than just replication. As feature sizes fell over the decades that followed, the mask kept being drawn at scales engineers’ hands could manage, and the optics did the work of getting smaller.

Lathrop wrote up the technique with Nall as a paper called “Photolithographic Fabrication Techniques for Transistors Which Are an Integral Part of a Printed Circuit.” It arrived too late to make the regular program of the IRE Electron Devices Meeting, so the conference accepted it as a late paper, slotted in for Friday, November 1, 1957, in Washington. The audience was small. Most of the chip industry, such as it was, did not yet exist as a self-conscious industry. Lathrop and Nall presented; the paper was published in the IRE Transactions on Electron Devices the following April. The patent application, filed on October 31, 1957, the day before the talk, was titled simply “Semiconductor Construction.” It listed Nall and Lathrop of Silver Spring, Maryland, as inventors, with the assignee the United States of America, represented by the Secretary of the Army. It issued on June 9, 1959, as U.S. Patent 2,890,395. The Army awarded the two men a $25,000 prize for the invention in 1958. Lathrop, with characteristic frugality, used his share to buy a station wagon.

He had also, by then, made the move that mattered more. In 1958, Texas Instruments was hiring. Pat Haggerty, the TI executive who had spent the decade pushing the company from defense electronics into the new world of silicon, had been watching the Diamond Ordnance work. He understood, as most senior people in the field did not yet understand, that the bottleneck on the next decade was not transistor physics. It was manufacturing. Whoever could turn the artisanal work of bench transistor-making into a printable, repeatable, projectable industrial process would own the field. Haggerty offered Lathrop a job. Nall took a parallel offer at Fairchild. The two co-inventors went to opposite ends of the small new industry and brought their technique with them.

Lathrop arrived in Dallas in the summer of 1958, just as Jack Kilby, also new to TI, was working through his summer alone in a quiet lab on the Tyranny of Numbers problem: the impossibility of hand-soldering enough discrete components to build the circuits the Air Force and the Strategic Air Command wanted. On September 12, 1958, Kilby demonstrated a small slab of germanium with five components etched and wired together as a working oscillator, the first working integrated circuit. The demo was beautiful as a proof of principle. As a manufacturing process, it was hopeless. The “flying wires” Kilby had soldered between regions of the germanium were laid by hand, under a microscope, with tweezers. No factory could make millions of those. What Kilby had shown was that the circuit could exist. What he had not shown was how anyone could ever build a million of them.

Lathrop was the answer to the second question. Within months of Kilby’s demo, TI was using Lathrop’s projection-photolithography to lay down photomasks for diffused resistors and capacitors, to pattern the oxide layers that defined where dopants entered the silicon, and to evaporate metal leads through patterned openings. By the end of 1958, TI’s pilot line had begun to produce integrated circuits with components defined not by tweezers but by light. The yields were abysmal at first. The masks were drawn by hand and full of defects. But the process was photographic. It could be repeated. Each step was done in parallel across a whole wafer rather than serially across one device at a time. That was the point.

Within four years, TI’s photolithographic process was being used to manufacture the first commercial integrated circuits, including the chips that flew on the Minuteman II missile guidance computer. At Fairchild, Nall and others were doing the same work in California, eventually combined with the planar process Jean Hoerni would invent in 1959 to give silicon chips a flat, oxidized surface that played to photolithography’s strengths. The mortar fuze project at Diamond Ordnance Fuze Lab, which had set Lathrop on this path, was overtaken by its own offspring. Within a decade, the lab’s transistorized mortar fuzes were being assembled with the same photolithographic methods that had been invented to make them.

The competing approaches did not survive the encounter. Project Tinkertoy and the Micro-Module Plan kept producing demonstration hardware into the early 1960s, won contracts, and quietly disappeared. Their assumption, that miniaturization was a packaging problem, turned out to be wrong. Once a circuit could be drawn with light, packaging it was almost beside the point. The difference between Lathrop’s photoresist and the Micro-Module’s notched ceramic squares was the difference between an industry and a museum exhibit. The Smithsonian today displays Tinkertoy modules behind glass. Photolithography is the operating principle of a $600 billion industry.

What Lathrop had done was simpler and stranger than he would have claimed. He had taken three pieces of off-the-shelf technology, photographic resist from a film company, an inverted laboratory microscope, and an ultraviolet lamp, and assembled them into a process that turned circuit design from sculpture into printing. Once it was printing, it scaled. The mask could be redrawn at any scale the optics could handle. The chemistry could be improved. The light could be made shorter and shorter in wavelength. Every feature shrink in the next sixty years, every doubling of transistors per chip that the industry would later rationalize as Moore’s Law, depended on this basic move: a pattern drawn at a size humans could draw, projected through optics onto a substrate at a size no human hand could manage. Mercury arc lamps would give way to deep ultraviolet, then to extreme ultraviolet generated by vaporizing tin droplets with lasers. The mask material would change. The reduction ratio would change. The basic operation would not.

Lathrop himself stayed at TI for a decade, becoming Director of Advanced Technology in the company’s Semiconductor Components Division. In 1968, he left for an academic appointment at Clemson University, where he taught electrical engineering for the rest of his career. He gave oral histories late in life, to the IEEE and the Computer History Museum, but never wrote a memoir. He died in 2022, at ninety-five, in Asheville, North Carolina. The Wikipedia entry was short. Most of the people working at the leading edge of chip manufacturing in 2022 had no idea who he was.

The mortar shell had won. In trying to fit a working radio into a volume the size of a coffee cup, an Army lab had stumbled into the manufacturing process that would define the next half-century of computing. The Soviets, the Japanese, and eventually the Chinese would all build chip industries on top of it. The competition over photolithography would, decades later, become one of the highest-stakes contests in geopolitics, fought over machines that could project light at thirteen-and-a-half nanometers onto silicon wafers spinning in vacuum. None of that was visible in 1957, in the brick lab on Connecticut Avenue, where two young engineers were trying to keep the wax from oozing.