Building the Reactor by Hand

The pressure now shifted to the concrete, physical struggle of execution. Nakamura stood in the laboratory at Nichia’s Tokushima plant, facing another failed crystal growth. The substrate held nothing usable—a film of gallium nitride that had cracked during cooling, its surface a web of fractures reflecting the overhead fluorescent lights. He adjusted the gas valve on the reactor he had built himself, turning the dial by fractions of a millimeter. The device towered over him, a cylinder of stainless steel and quartz glass wrapped in heating elements and gas lines, all assembled from parts sourced individually. No commercial reactor could do what he needed.

The two-flow MOCVD reactor represented his answer to problems that had defeated better-equipped laboratories. Conventional reactors introduced precursor gases from a single direction, flowing across a heated substrate in a pattern optimized for gallium arsenide and indium phosphide—materials that cooperated with crystal growers. Gallium nitride did not cooperate. It demanded temperatures that cracked sapphire substrates, pressures that leaked through seals, and it produced films full of defects when it produced films at all.

Nakamura’s design added a second gas flow, introducing reactive gases from above while a carrier gas flowed horizontally across the substrate. The idea was simple: create a boundary layer that would force reacting gases to contact the surface more efficiently. Simple in concept meant nothing in practice.

Every dimension mattered—the angle of the gas inlet, the distance from nozzle to substrate, the ratio between the two streams, the temperature at which gases decomposed and the temperature at which the crystal formed. Each parameter connected to every other, and Nakamura had to find the combination that would produce a usable film through trial and error.

He documented each run in a laboratory notebook. The notebooks would later become evidence in a lawsuit, their pages examined by lawyers and expert witnesses seeking to establish priority and prove the value of his work. In 1989, they were simply records of failure. Run after run produced nothing worth keeping. Gallium nitride films cracked, peeled off sapphire substrates, or grew with crystalline structures so disordered they could never function as electronic devices. Each failure told him something about the boundaries of the possible, but the space of possible parameters was vast, and he was exploring it alone.

The isolation was total. Nichia had no other researchers working on gallium nitride. The company had no research division in any meaningful sense. Nakamura was a development engineer in a chemical company that made phosphors for fluorescent lamps and LEDs that glowed red and green. Blue LEDs were a fantasy most of the industry had abandoned. Major electronics corporations had tried and failed. IBM, RCA, Philips, and dozens of university laboratories had invested years in gallium nitride research before concluding the material was fundamentally unsuitable for practical devices. The problem of p-type doping—creating a gallium nitride layer that conducted positive charge carriers—seemed insoluble. Without p-type material, there could be no junction, and without a junction, no LED.

Nakamura knew this history. He had studied every paper published on gallium nitride, tracking the rise and fall of research programs that consumed millions of dollars and produced nothing but dim, unreliable devices. The first blue-violet LED using magnesium-doped gallium nitride was made at Stanford University in 1972 by Herb Maruska and Wally Rhines, doctoral students in materials science. Maruska had been on leave from RCA Laboratories, where he collaborated with Jacques Pankove. The Stanford group obtained a patent (U.S. Patent US3819974 A). But the device was too dim for practical use, and research into gallium nitride devices slowed. In August 1989, Cree introduced the first commercially available blue LED, based on silicon carbide, but its efficiency was no more than about 0.03%. Major laboratories moved to other materials. Zinc selenide became the favored candidate for blue light emission. Gallium nitride joined the list of materials that promised much and delivered little.

The industry consensus was not irrational. Gallium nitride was genuinely difficult. The material required high temperatures to crystallize, temperatures that threatened to melt or warp any substrate. The only practical substrate was sapphire, which had a different crystal structure and thermal expansion coefficient. When the grown film cooled, the mismatch created stresses that cracked the gallium nitride layer. Researchers tried buffer layers—intermediate films that might relieve stress—but results remained inconsistent. Even when crystal grew without cracking, it contained dislocations in the atomic lattice that killed the electronic properties needed for a functioning device.

Nakamura had chosen gallium nitride anyway. His reasoning was practical. Zinc selenide was being pursued by major laboratories with resources he could not match. If he worked on zinc selenide, he would compete against teams at 3M, Philips, and Sony—teams with better equipment, more researchers, and longer track records. Gallium nitride was abandoned territory. If he succeeded where others failed, the achievement would be his alone. If he failed, he would fail in the same way better-funded researchers had already failed. The calculation was cold, and probably correct.

The calculation had convinced Nobuo Ogawa, Nichia’s founder, to fund the project. Ogawa had built his company from nothing, starting with phosphors for cathode-ray tubes and expanding into LEDs as the market for solid-state displays grew. He understood that Nichia’s survival depended on finding products that larger companies did not already dominate. A blue LED would open markets red and green LEDs could not touch. Full-color displays required blue. White light from LEDs required blue. Ogawa authorized the spending, giving Nakamura money, time, and freedom.

The freedom was real, but came without support. Nakamura had to design his own equipment because no commercial system could do what he needed. He had to source his own materials because Nichia had no supply chain for gallium nitride precursors. He had to solve every problem himself because no one else in the company understood the work. The isolation that gave him freedom also left him alone with his failures.

The two-flow reactor took shape over months. Nakamura sketched designs, ordered parts, assembled components by hand. The machine that emerged looked nothing like polished systems sold by equipment manufacturers. It was a tangle of gas lines and electrical cables, a cylinder wrapped in insulation tape, a collection of valves and gauges held together by will as much as welding. But the design incorporated specific ideas about how to solve problems that had defeated others. The dual gas flows were the key. By introducing reactive gases from above while maintaining horizontal carrier flow, he believed he could achieve better uniformity across the substrate. Better uniformity meant better crystal quality. Better crystal quality meant a chance at a working device.

The theory was sound. The practice was brutal.

Each growth run consumed hours. The reactor had to heat slowly to growth temperature, more than a thousand degrees Celsius. Gases had to flow at precisely controlled rates. The substrate had to be positioned at exactly the right distance from the inlet. Cooling had to proceed at a rate that would not crack the film. At any point, something could go wrong: a fluctuation in gas pressure, a temperature spike, contamination in the gas line. The finished film would emerge, and Nakamura would examine it under a microscope, looking for the smooth surface and even color indicating usable crystal.

What he usually found was disaster. Films cracked along crystal planes. Films delaminated from substrates. Films grew in scattered islands rather than continuous layers. Films showed the wrong crystalline structure, atoms arranged in patterns that would not support electronic function. Each failure required adjusting some parameter and trying again. The gas flow ratio. Substrate temperature. Cooling rate. Buffer layer thickness. Precursor concentration. Hundreds of variables, each connected to every other, each requiring adjustment through experiment rather than calculation.

The numbers mounted. Ten runs. Fifty. A hundred. Each run consumed materials and time and produced nothing usable. Nakamura documented everything. Notebooks filled with parameters and observations, sketches of reactor modifications, notes on what had failed and what might be worth trying. The documentation was meticulous, a record of invention, creating evidence that would establish priority if he ever succeeded. Also a record of failure, proof that success was not inevitable, that the path from idea to achievement required more than inspiration.

The isolation deepened as failures accumulated. Nichia’s management had changed while Nakamura was in Florida learning about MOCVD technology. Nobuo Ogawa ceded the presidency to his son-in-law Eiji Ogawa in 1989. The new leadership viewed Nakamura’s project with skepticism. Gallium nitride research had consumed years and produced nothing. The company had invested in equipment, materials, and travel with no return. Eiji Ogawa ordered Nakamura to suspend work on GaN, claiming it consumed too much time and money. The order was clear. The project was to stop.

Nakamura ignored it. He continued his work, spending company money on materials and equipment, running growth experiments in a reactor he had built himself, pursuing a goal his own management had declared unworthy of further investment. The insubordination was extraordinary. In a Japanese corporation of the late twentieth century, an engineer who defied direct orders from company leadership was supposed to cease to exist. The hierarchy did not permit individual initiative to override collective decision. But Nakamura had the founder’s backing, or at least the memory of it. Nobuo Ogawa had authorized the project. Money had been allocated. Equipment purchased. Nakamura continued spending the allocation, drawing down the budget Ogawa senior had approved, exercising the freedom Ogawa senior had granted.

The dynamic created a strange kind of patient capital—funding that persisted despite disapproval from current management, drawing on authorization from a founder who had stepped aside. Nichia’s size was its advantage. The company was small enough that one engineer could consume resources without triggering oversight mechanisms that would have shut down a failing project in a larger organization. The founder’s authorization gave Nakamura cover. The new management’s disapproval gave him no resources to waste. He had what he had already acquired, and he used it to continue.

The two-flow reactor evolved through the failures. Nakamura modified the gas injection system, adjusting angles and distances. He changed heating elements, seeking more uniform temperature across the substrate. He experimented with substrate preparations, trying to find surface treatments that would encourage better crystal growth. Each modification required disassembly and reassembly, hours of work before another growth experiment. The machine was never finished. It was always in flux, being adjusted, being rebuilt.

Hundreds of failed growths taught him the boundaries of the process. Critical parameters emerged from those that could vary without effect. Early signs of failure became recognizable: the color of a film that would crack, the texture of a surface that would not support electronic function. The machine itself became familiar, sounds when working correctly, vibrations indicating problems with gas flow. The knowledge was tacit, embodied in hands and eyes rather than written in textbooks. No one else could operate the reactor. No one else could interpret results. Equipment and engineer had become a single system.

The parallel work of Akasaki and Amano at Nagoya University provided a distant reference point. Professor Isamu Akasaki had pursued gallium nitride since the early 1980s, maintaining a research program when most of the field had abandoned the material. Hiroshi Amano joined Akasaki’s group in 1982 as an undergraduate student and had been researching growth, characterization, and device applications of group III nitride semiconductors since then. The Nagoya group achieved results others had not. They demonstrated that a low-temperature buffer layer could improve crystal quality, providing a template for subsequent growth. They produced the first p-type gallium nitride by irradiating magnesium-doped material with electrons, activating dopant atoms that had previously been inert. Their work proved gallium nitride could be doped, could form a junction, could theoretically produce a blue LED.

Nakamura drew on their methods. The low-temperature buffer layer became part of his process. Electron-beam activation of magnesium-doped GaN offered a path to p-type material. But Nagoya results were laboratory demonstrations, devices proving principles without achieving commercial viability. LEDs they produced were dim. Manufacturing processes were not scalable. The gap between laboratory demonstration and commercial product was wide, and Akasaki and Amano had not crossed it.

Nakamura needed to cross it. His goal was not a paper in a journal or a patent for a laboratory method. His goal was a product Nichia could manufacture and sell. The two-flow reactor was designed with manufacturing in mind. Gas flows, temperatures, and substrate sizes were chosen to be compatible with production processes. If he could grow good crystal in his homemade reactor, methods might translate to larger scale. The commercial goal shaped every technical decision.

The failures continued. Run after run produced films that could not be used. Nakamura adjusted parameters, modified the reactor, tried again. The process was iterative, each failure generating information guiding the next attempt. The information was fragmentary and ambiguous. A film that cracked might indicate problems with cooling rate, buffer layer, or gas composition. A film with poor crystallinity might indicate temperature too low, contaminated substrate, or turbulent gas flow. Diagnosis required judgment. Judgment required experience. Experience came only from more failures.

The laboratory became his world. He arrived early, stayed late. Weekends and holidays. The reactor demanded attention. Gas flows needed monitoring. Temperature needed adjustment. Substrates needed preparation. Each step required his presence, his hands, his judgment. The work was physical, repetitive, consuming. There was no one to delegate to, no one to consult, no one to share the burden.

The number of growth runs passed two hundred. Then three hundred. Each run consumed hours and grams of materials. Each ended with examination under the microscope, judgment about quality, decision about what to change next. Documentation accumulated. Notebooks filled with data. The record of invention grew alongside the record of failure.

The reactor itself became an artifact, a physical embodiment of knowledge Nakamura had gained. Modifications reflected lessons learned. The gas injection system had been rebuilt multiple times, each version incorporating changes previous failures suggested. Heating elements repositioned to eliminate cold spots. Seals replaced with materials that could withstand reactive gases. The machine was no longer what he had originally designed. It was the product of iteration, a device that had evolved through use.

The distinction between design and use was not sharp in Nakamura’s practice. He designed the reactor, built it, operated it, modified it based on results. The feedback loop between conception and execution was immediate. When a growth run failed, he could examine the machine, identify the problem, implement a fix. The fix might work or not. Either way, the next run provided more information. The cycle of design, build, operate, modify, repeat was compressed into the work of a single person.

The compression was both strength and weakness. Nakamura could iterate faster than a larger team because he did not need to communicate findings or wait for approval to make changes. He could pursue hunches and abandon approaches without justifying decisions. Autonomy accelerated the experimental cycle. But compression also meant errors were not caught by colleagues, assumptions not challenged by peers, interpretations not tested against alternative views. The isolation that gave him speed also left him vulnerable to blind spots.

Blind spots were hidden in the machine itself. Every design decision encoded assumptions about how growth worked. The angle of gas inlet assumed a particular flow pattern. Distance between inlet and substrate assumed a particular reaction zone. Ratio of gas flows assumed particular mixing behavior. If assumptions were wrong, the machine would produce failures regardless of parameter adjustments. Nakamura had to discover correct design through experiment, and experiment required a machine that embodied correct design. The loop was circular. Breaking out required insight or luck or both.

Insight came slowly. The two-flow design was intended to improve uniformity of gas composition across the substrate. By introducing reactive gases from above while maintaining horizontal carrier flow, Nakamura hoped to create a boundary layer that would stabilize growth. Theory was plausible. Practice showed something missing. Films still grew with poor uniformity. Crystal quality varied across substrate. Gas flows were not mixing as intended.

He began experimenting with flow rates. Carrier gas moving horizontally had to be fast enough to carry away reaction products but not so fast it disrupted vertical flow of reactive gases. Vertical flow had to be slow enough to allow decomposition but not so slow that gases reached substrate already depleted. Balance between the two flows determined the shape of the reaction zone, where precursors decomposed and deposited.

The balance was delicate. Small changes in flow rate produced large changes in film quality. Nakamura mapped parameter space, running growth experiments at different flow ratios, documenting results, looking for patterns. Patterns emerged slowly. Too much horizontal flow, and film grew thick at upstream edge, thin downstream. Too much vertical flow, and film grew in center but not at edges. Optimal balance was somewhere between, a narrow window where film grew uniformly across the entire substrate.

Finding the window required hundreds of runs. Each tested a different combination of parameters. Each required heating, growing, cooling. The process consumed days. Results filled notebooks. Failures outnumbered successes by factors that would have discouraged any reasonable person.

Nakamura was not reasonable. He continued.

Crystal quality remained the central problem. Even when film grew uniformly, it contained defects killing electronic performance. Dislocations, linear defects in atomic lattice, propagated from substrate through buffer layer into gallium nitride. Sapphire substrate and gallium nitride had different lattice constants, different atomic spacings. Mismatch created strain at interface. Strain relaxed through defect formation. Defects acted as recombination centers, places where electrons and holes combined without emitting light. A film full of defects would not produce a bright LED.

The buffer layer was supposed to solve this. Akasaki and Amano showed that a thin layer of aluminum nitride or low-temperature gallium nitride deposited before main growth could improve crystal quality. Buffer layer provided transition between substrate and film, a surface matching gallium nitride better than sapphire. Nakamura adopted the technique, but results were inconsistent. Some buffer layers improved quality. Others did not. Difference was not obvious. Deposition temperature, thickness, gas composition, all seemed to matter. Optimal combination remained unclear.

He experimented with buffer layers as he experimented with everything. Different materials. Different thicknesses. Different deposition temperatures. Each variation required full growth run, hours of machine time, examination of resulting film. Failures accumulated. Notebooks filled. Parameters shifted by increments.

The work was not purely empirical. Nakamura had theoretical framework guiding choices. He understood thermodynamics of gas-phase reactions, kinetics of crystal growth, physics of defect formation. Theory told him what parameters might matter, what directions to explore. But theory could not predict exact combination that would work. Complexity exceeded calculation. The only way to find answer was to try.

The trying continued through 1989 and into 1990. The reactor ran, failed, ran again. Nakamura adjusted, modified, documented. The company around him continued making phosphors and red LEDs. Management continued viewing his project with disapproval. Budget continued declining. Materials continued being consumed.

Pressure was not abstract. Nakamura was spending Nichia’s money with nothing to show. The company was not wealthy. Resources supporting his work could have been used for other purposes, for products with proven markets and reliable returns. Every growth run consumed materials costing money. Every day in the laboratory was a day not spent on development work that might produce immediate results. Opportunity cost was real, even if no one calculated it.

Cost was offset by founder’s original authorization. Nobuo Ogawa had given permission to pursue gallium nitride. Permission had never been formally revoked. New management had ordered him to stop, but Nakamura chose to continue, and his choice was tolerated if not endorsed. Tolerance was a form of patient capital, willingness to let investment play out even when returns were not visible. Tolerance was not infinite. At some point, failures would accumulate to a level demanding an end. Nakamura did not know where that point was. He continued as if it did not exist.

The laboratory became a refuge. Work was hard, but work he understood. Failures were frustrating, but failures he could learn from. Isolation was lonely, but isolation he had chosen. He had staked his career on gallium nitride. He was not going to abandon the stake.

The two-flow reactor continued to evolve. Each modification reflected a lesson learned, hypothesis tested, direction explored. Machine was physical record of experimental process, collection of decisions embodied in metal and glass. Nakamura had built it himself. He knew every joint, valve, wire. Knowledge was intimate, practical. Sounds when machine worked correctly. Feel of gas flows stabilizing. Appearance of growth proceeding as intended.

Hundreds of failed growths had taught him something, even if they had not produced usable film. They had taught boundaries of the process, combinations that would not work, approaches that were dead ends. Negative knowledge was valuable. It narrowed search space, eliminated possibilities, concentrated attention on regions that remained.

Building upon this foundation, Theodore Moustakas at Boston University patented a method for producing high-brightness blue LEDs using a new two-step process in 1991. The parallel development in America showed the field was not entirely dormant. But Nakamura was not aware of Moustakas at this point. He worked in isolation, guided only by his own failures and the distant example of Nagoya.

The search continued. The reactor ran. Notebooks filled with parameters and observations and sketches. Each entry documented a choice, an adjustment, an attempt to force gallium nitride into crystalline order. The material resisted. The machine failed. Nakamura tried again.

The stark figure quantifying the iterative struggle, hundreds of failed growths, set the stage for results that would finally emerge.