Chapter 149: The Third Colour
Indian Semiconductor Manufacturing Corporation, GorakhpurCulminating: 21 March 1974
The notebook had no title on the cover.
It was an ordinary bound notebook — the kind sold in every stationery shop in India, blue cover, lined pages, one hundred and sixty pages — and it sat on Dr. Ramesh Chandra's desk on the morning of the fourteenth of January 1973 like any other notebook. But it was not any other notebook. Karan had given it to Ramesh at the start of the meeting, had set it down between them on the desk without ceremony, and had said: read it before I explain anything. Ramesh had read it.
It had taken him forty minutes.
The notebook contained forty-one pages of writing in Karan's precise handwriting, and those forty-one pages described, in technical detail that Ramesh found immediately alarming, a complete research programme for the development of a blue light-emitting diode using gallium nitride as the semiconductor material. Alarming not because the goal was impossible — Ramesh had been following the academic literature on wide-bandgap semiconductors and understood the theoretical basis for GaN as a blue light emitter — but because the level of specificity in the notebook went far beyond what the published literature supported in January 1973.
The notebook specified:
The substrate material: c-plane sapphire, two-inch diameter, four-hundred-micrometre thickness The buffer layer: aluminium nitride, twenty nanometres, grown at 550 degrees Celsius before ramping to growth temperature The n-type layer: unintentionally doped gallium nitride, four micrometres, silicon background doping from the reactor The active layer: indium gallium nitride, precisely fifteen nanometres, indium fraction between 0.15 and 0.20, growth temperature 750 degrees Celsius — significantly lower than the GaN layers, a detail that was critical and non-obvious The p-type layer: magnesium-doped gallium nitride, two hundred nanometres, magnesium precursor flow at a specific rate per minute The contacts: the exact metal stacks, the deposition conditions, the photolithography approach
And then, on page thirty-seven, the step that Ramesh read three times before putting the notebook down and looking at Karan:
After growth and contact deposition, anneal the complete device in pure nitrogen atmosphere — not air, not forming gas, nitrogen specifically — at 700 to 750 degrees Celsius for fifteen minutes. This step is not optional. Without it, the magnesium doping will not produce p-type conductivity. The reason is hydrogen passivation: during MOCVD growth, hydrogen from the ammonia precursor bonds to magnesium atoms and prevents them from contributing electron holes. The nitrogen anneal provides sufficient thermal energy to break these magnesium-hydrogen bonds and drive the hydrogen out of the crystal. Confirmed: this step is what the field has been missing.
Ramesh had looked at Karan.
"Confirmed," he had said. "That word."
"Yes," Karan had said.
"The literature does not support this conclusion," Ramesh had said. "There is one paper from Pankove at RCA in 1971 that achieved some p-type behaviour in GaN using magnesium doping, but the result was inconsistent and could not be reproduced. No one in the field has identified the hydrogen passivation mechanism. This is not in the literature."
"I know it's not in the literature," Karan had said.
"Then how—" Ramesh had started.
"Dr. Chandra," Karan had said, quietly, "I need you to trust two things simultaneously. First: the information in that notebook is correct. I am certain of it. Second: I cannot explain to you with complete honesty how I am certain of it." He had paused. "I can tell you that the foundation is real physics, not speculation. The hydrogen passivation mechanism is consistent with everything that is known about GaN growth. It is a conclusion that could have been reached by rigorous analysis of the available data. It was reached by a different route. The route does not change whether the conclusion is correct."
Ramesh had been quiet for a long time.
He was fifty-three years old. He had been a semiconductor researcher for thirty years. He had a specific quality that long careers in research produced in people who survived them honestly: he could distinguish between the confidence of someone who understood something deeply and the confidence of someone who was performing confidence to sell something. The two felt different. This felt like the first kind.
"The annealing step," Ramesh had said finally. "If you are correct about the hydrogen passivation mechanism, then the annealing step is the key that the entire field has been missing. The reason GaN has not produced reliable p-type material is not because the material cannot support it. It is because everyone who has attempted it has grown the material, measured the electrical properties, found them unsatisfactory, and concluded that magnesium doping in GaN does not work. Without understanding why, no one has tried to fix it."
"Correct," Karan had said.
"And if the annealing step works," Ramesh had said, "we have a p-n junction in GaN. And a p-n junction in GaN emits blue light."
"Yes," Karan had said. "Deep blue. Around 460 to 470 nanometres, depending on the indium fraction in the active layer."
Ramesh had looked at the notebook.
"I will need a team," he had said.
"You have a budget and full authority to hire," Karan had said. "I want the programme running within two months."
"What efficiency do you expect from the first devices?" Ramesh had asked. This was the question of a researcher who had learned to calibrate ambition against reality — who knew that first demonstrations were always humble.
Karan had looked at the notebook. "The first devices, before optimisation, will probably be in the range of one to two percent external quantum efficiency. That sounds low. The first red LED demonstration in 1962 achieved approximately 0.1 percent. We will be a factor of ten to twenty ahead of where that technology started."
"And with optimisation?"
"Ten percent within two years," Karan had said. "Possibly higher."
Ramesh had been quiet again. Ten percent external quantum efficiency in a blue LED was not merely a research achievement. Ten percent was the threshold at which solid-state lighting became commercially interesting. It was the number beyond which the economic case for replacing incandescent bulbs with LED alternatives became compelling.
"You are describing the end of the incandescent light bulb," Ramesh had said.
"Eventually," Karan had said. "In India first, because India's electricity situation makes the efficiency argument more urgent here than anywhere. Then everywhere."
Ramesh had closed the notebook.
"Two months," he had said. "The team will be assembled. I want Suresh Iyer from IIT Bombay for the doping and diffusion work — I know his research, he is the right person. And Vikram Malhotra, who finished his doctorate at IIT Kanpur last year on MOCVD growth. And Ashok Kumar for the crystal growth fundamentals. I want three more for the characterisation and device processing work."
"You have full authority," Karan had repeated.
Ramesh had looked at the notebook one more time.
"The MOCVD reactor," he had said. "The specifications in the notebook for the reactor conditions — the ammonia flow rates, the trimethylgallium partial pressures, the trimethylindium temperatures for the active layer — these are very specific. Our reactor will need to be capable of holding these conditions precisely."
"The reactor you are currently operating for the GaAs work handles these conditions," Karan had said. "You will need to rebuild the gas handling system for higher flow rates in some regimes, and the substrate holder rotation will need to be improved for the uniformity the active layer requires. The engineering work is approximately three months. I can give you the specifications."
He had pulled another set of pages from his folder.
Ramesh had taken them.
"You have also specified the reactor modifications," he had said.
"Yes," Karan had said.
Ramesh had looked at him for a long moment.
"I will begin," he had said.
The programme began in March 1973.
What Ramesh had assembled was a team of six — himself, Suresh Iyer, Vikram Malhotra, Ashok Kumar, Dr. Harish Mehta who had worked with Ramesh at the Tata Institute and who brought deep knowledge of optical characterisation, and Rajiv Sharma, a twenty-five-year-old research associate who was, as Ramesh noted in his programme records, the kind of person who caught numerical errors that everyone else missed and who was therefore worth considerably more than his junior status suggested.
The reactor modification took four months rather than three — Kumar, who led the gas handling work, discovered that the trimethylindium delivery system required a complete redesign because the existing bubblers could not hold the precursor at the required temperature stability. The bubbler temperature affected the indium vapour pressure, which affected the indium fraction in the active layer, which determined the emission wavelength. Temperature instability of even half a degree would shift the wavelength enough to move out of the target range.
Kumar rebuilt the bubblers from scratch, using a design that Karan had sketched in a supplementary note and that Kumar improved further through four prototype iterations. The final design held the trimethylindium temperature within 0.1 degrees Celsius — stability that exceeded even what the note had specified.
"You improved on the specification," Karan said, when Kumar presented the result.
"The specification was the starting point," Kumar said. "I understood why the stability mattered. When you understand why something matters, you find ways to do it better than you were told to."
The first growth runs began in July 1973.
The team ran twelve wafers in July. All twelve failed. Not random failure — the failures were informative, each one producing electrical measurements that told the team something about where the process was off. The buffer layer growth was inconsistent, producing crystal quality that was too variable. The growth temperature ramp from the buffer layer to the GaN was too fast.
Vikram kept a chart of failures on his wall. Not to document defeat but to document convergence — each failure that was understood was one fewer unknown in the system.
The team ran fourteen wafers in August.
On August 22nd, 1973 — seven months into the programme — Suresh sat at the probe station with Wafer 26 and engaged the current source.
The junction measurement showed forward bias turn-on at 3.1 volts.
Suresh ran the measurement again.
Same result.
He called Ramesh.
"Come to the probe station," he said.
Ramesh came. He looked at the I-V curve on the monitor. The clean diode characteristic — the exponential rise in current as voltage increased through the forward bias region, the flat blocking in the reverse bias region. P-n junction behaviour. Unambiguous.
He looked at Suresh.
"The annealing step," he said.
"I followed the protocol exactly," Suresh said. "700 degrees Celsius. Fifteen minutes. Pure nitrogen. No air."
Ramesh called the team.
They gathered around the probe station. The measurement was run again with everyone watching. Same result. Then again. Same result.
Ramesh dimmed the lights.
He engaged the current.
On the monitor, at the junction area, a blue glow appeared.
Not faint. Not marginal. A cool, deep, unambiguous blue — the specific blue of high-energy photons near the violet edge of human vision. The blue that sat at 465 nanometres, at the high-energy end of the visible spectrum, the blue that tungsten filaments and phosphor-coated tubes had never produced efficiently.
The room was quiet for a moment.
Then Vikram said, in a voice that was attempting calm and not quite achieving it: "It works."
"The annealing step works," Suresh said. He said it with the precision of a man separating confirmation of the hypothesis from enthusiasm about the result, because those were two different things and both deserved separate recognition.
Ramesh went to the spectrometer. He ran the optical measurement.
Peak emission: 466 nanometres.
"Deep blue," he said. "Exactly where it should be."
He turned to look at Karan, who was standing near the door.
Karan had been in the lab since early morning. He had said almost nothing for six hours. He had watched Suresh prepare the probe station, watched the measurement setup, watched the readout as the I-V curve appeared. He had said nothing when the diode behaviour confirmed.
Now he was looking at the blue glow on the monitor.
His expression was not the expression of a man seeing something unexpected.
It was the expression of a man seeing something he had known was coming and who had, nonetheless, been waiting to see it for a long time.
"External quantum efficiency," Karan said.
Vikram set up the measurement. The optical power output. The electrical power input. The ratio.
"1.3 percent," Vikram said.
He read it twice.
"1.3 percent," Ramesh confirmed.
In the world of semiconductor research, 1.3 percent external quantum efficiency from the first working device of a new material system was an extraordinary starting point. Not world-record territory — that existed in well-optimised mature materials. But for a material that the published literature had never successfully made into a working p-n junction, 1.3 percent on the first confirmed device was a result that told you the physics was right, the material system was viable, and the path forward was engineering rather than discovery.
"The first red LED was 0.1 percent," Ramesh said. He was not comparing to the red LED with pride. He was making a calibration point. "We are thirteen times ahead of where that technology began."
"Because we started with a better recipe," Vikram said. He looked at the chart on his wall — all twenty-six wafer attempts and their failure modes. Then he looked at Karan. He had the expression of a researcher who was doing a specific mental arithmetic and had arrived at a specific conclusion that he was not yet saying aloud.
"Yes," Karan said, as though Vikram had asked the question.
Vikram nodded once. He filed it where he filed things that were true and required thinking about and that he would probably never fully explain.
He went back to his instruments.
August to December 1973.
What followed the first working device was what always followed first working devices in a well-run research programme: systematic investigation of every variable that affected performance, conducted in a sequence designed to understand each variable in isolation before combining improvements.
Vikram's growth recipe was the first target. The indium fraction in the active layer — the ratio of indium to gallium in the InGaN — affected both the emission wavelength and the efficiency. Higher indium meant longer wavelength, shifted toward green. Lower indium meant shorter wavelength, shifted toward violet. The target was deep blue, around 460-470 nanometres, which required an indium fraction between 0.15 and 0.20. But the fraction also had to be uniform — patches of higher or lower indium created emission at different wavelengths and degraded the device performance.
Vikram ran sixteen wafers varying the growth conditions for the active layer. He mapped the emission wavelength and efficiency against temperature, against indium precursor flow, against growth rate. He produced a two-dimensional map of the parameter space that showed, clearly, where the high-efficiency region was.
The high-efficiency region was narrow. Changing the active layer growth temperature by ten degrees shifted you from 1.5 percent efficiency to 0.6 percent. This was why the earlier literature had struggled — GaN growth at the wrong temperature range gave poor results and the field had not systematically mapped what the right range was.
By October, Vikram had the active layer recipe optimised.
First device, August: 1.3 percent. After active layer optimisation, October: 2.8 percent.
Suresh's work was on the contacts. The metal contacts on the device surface had to make good electrical connection to the semiconductor without introducing high contact resistance, which would generate heat and reduce efficiency. The published literature used simple metal schemes. Suresh suspected that a different metal stack, with a thin interlayer that matched the work function of the GaN surface more closely, would reduce contact resistance and improve current injection into the device.
He ran twelve contact variations.
The best was a nickel-oxide interlayer under the gold contact. The nickel oxide, formed by depositing nickel and then briefly annealing in air, created an interface that matched the GaN's electron affinity more closely than nickel alone. Contact resistance dropped by a factor of four.
After contact optimisation, November: 4.1 percent.
Kumar's work was on the buffer layer. The buffer layer — the thin AlN that was grown first to prepare the sapphire surface for GaN — set the crystal quality of everything that grew on top of it. If the buffer layer was imperfect, the threading dislocations that formed at the buffer-GaN interface propagated upward through the entire structure, reducing the efficiency.
Kumar had been growing buffer layers by habit — the conditions from the literature, which were adequate. He approached it instead as a design problem: what conditions minimised dislocation formation at the sapphire-AlN interface? He ran a systematic study, varying temperature, pressure, and growth rate through the buffer layer step, measuring crystal quality with X-ray diffraction after each run.
The optimised buffer layer reduced threading dislocation density by a factor of eight relative to the standard conditions.
After buffer layer optimisation, December: 5.9 percent.
On the last day of 1973, Ramesh wrote in his programme notebook:
31 December 1973. Best device efficiency: 5.9% EQE. Growth yield on optimised process: 87%. Emission wavelength: 462-468nm, consistent run to run. Device has been operating in the continuous test chamber for 41 days. No degradation in optical output. Lifetime extrapolation: >50,000 hours.
Programme has exceeded all initial milestones. First working device achieved in month 7 of programme. By end of year 10 (month 10), efficiency has reached 5.9%. This is world-leading performance in GaN LEDs — ahead of any published result by a substantial margin.
Q1 1974 target: demonstrate white LED. Demonstrate packaged device. Begin productionisation planning.
He signed and dated it.
The productionisation question was the one that Karan had been thinking about since August.
A research device — a bare semiconductor chip on a probe station in a dimmed laboratory — was not a product. The path from the probe station to the consumer's hand required packaging: encasing the tiny chip in a protective housing, bonding wire connections to the chip's contact pads so it could be connected to a circuit, and providing a means of coupling the emitted light out of the device efficiently.
LED packaging technology existed in 1973 — the packaging for red and green LEDs, which had been in production at ISMC since 1971, used a standard approach: mount the chip in a metal cup that acted as a reflector, bond thin gold wires to the chip's contacts, encapsulate in clear epoxy that was shaped into a dome to act as a lens. The dome shape helped direct the emitted light forward and also protected the chip from mechanical damage.
For blue LEDs, the same package would work — with one addition. For white light, a phosphor material would be incorporated into the encapsulant epoxy. Blue light from the chip would excite the phosphor, which would re-emit yellow light. The eye would receive both the transmitted blue and the re-emitted yellow, integrating them into the perception of white.
The phosphor Karan had specified was yttrium aluminium garnet doped with cerium — YAG:Ce, a yellow-emitting phosphor with strong absorption in the blue. Mehta had synthesised it by end of November: yttrium oxide, aluminium oxide, and cerium nitrate fired at 1400 degrees Celsius, ground to fine powder, suspended in optical-grade epoxy at ten weight percent.
The first white LED was demonstrated on the 14th of January 1974 — exactly one year after the notebook meeting.
Mehta had packaged the device — ISMC's standard LED cup package, the standard wire bonding, the dome formed with phosphor-loaded epoxy. He brought it to the test chamber, which was set up with the calibrated photodetector and the spectrophotometer.
He engaged the current.
The device produced white light.
Mehta measured the spectrum. Two peaks: the transmitted blue at 465nm, the phosphor emission at 555nm — yellow-green. The combination, integrated across the detector bandwidth, produced a colour temperature of 5,800 Kelvin — cool daylight white. Colour rendering index above 75. External quantum efficiency of the white device, accounting for the phosphor conversion losses: 4.2 percent.
Four percent white light efficiency. An incandescent bulb was 5 percent overall efficiency but produced most of that as heat, not visible light — the luminous efficacy was about 15 lumens per watt. A fluorescent tube achieved about 60 lumens per watt. The white LED, at current performance, was approximately 45 lumens per watt.
Not yet better than fluorescent. But the optimisation was not finished.
"Continue the efficiency work," Karan said. "The target is 10 percent EQE on the blue device. At 10 percent, the white package will produce over 80 lumens per watt. That beats fluorescent and the economic argument for replacement becomes clear."
Ramesh made the note.
"Timeline to 10 percent?" he asked.
"Six months," Karan said. "If Vikram's multi-quantum-well structure works as expected."
Vikram had been working on this for two months. A single active layer was the simplest device architecture — one layer of InGaN between the n and p layers, one region where electrons and holes could meet and recombine to produce light. But the single layer had limitations: a thin layer captured only a fraction of the injected carriers before they passed through without recombining. Multiple thin layers — quantum wells — created multiple capture regions and improved the probability that injected carriers would produce photons rather than passing through.
The multi-quantum-well structure was known to work in red and green LEDs. Vikram had been adapting it for the GaN system, which required careful control of the InGaN well and GaN barrier thicknesses at the scale of a few nanometres.
He ran the first multi-quantum-well wafers in February.
Three wells, two nanometres thick each, separated by GaN barriers of ten nanometres.
Efficiency: 7.3 percent.
Five wells.
Efficiency: 8.9 percent.
"The saturation is starting," Vikram reported. "More wells doesn't linearly increase efficiency — there are limits to how efficiently the outer wells are filled when the inner wells capture preferentially. The optimum is somewhere between three and five wells."
"Four wells," Suresh said. He had been modelling the carrier distribution. "Four wells with asymmetric barriers — the barrier on the p-side slightly thinner than the n-side — should improve electron injection into the outer wells."
Vikram ran four-well asymmetric barrier structures through March.
Wafer 847 — which would become the programme's landmark — was the culmination of this work. Four quantum wells. Asymmetric barriers. Optimised buffer layer. Optimised contacts. The complete best-practice recipe assembled from seven months of systematic improvement.
21 March 1974 — 09:15
Suresh positioned the probe needles on Wafer 847's contacts. The microscope showed the gold pads in sharp focus. His hands were steady.
Ramesh dimmed the lights.
The current source was set to ten milliamps.
Suresh engaged it.
The device produced blue light.
On the spectrophotometer monitor: a peak at 465 nanometres. Clean. Narrow — the full-width at half-maximum was 22 nanometres, narrower than any previous device, indicating better active layer uniformity.
"Running the efficiency measurement," Vikram said.
The calculation took two minutes.
"10.4 percent," Vikram said.
The number sat in the room.
Ramesh wrote it in his notebook without a pause — the reflex of a man who had been recording data for thirty years and who wrote first and felt second because feeling while writing introduced errors. He wrote: Wafer 847. Peak emission 465nm. FWHM 22nm. EQE 10.4%. Date: 21 March 1974.
Then he set down the pen.
Then he looked at Karan.
Karan was standing near the back of the room, where he had been standing since the probe engagement. He was looking at the spectrophotometer monitor. At the peak at 465 nanometres. At the number in the efficiency readout.
"10.4," Karan said.
"10.4," Ramesh confirmed.
10.4 percent external quantum efficiency in a blue LED. In a world where, in 1974, no research group anywhere on the planet had published a confirmed working GaN p-n junction, ISMC had a device at 10.4 percent efficiency — a device that would, in the other timeline Karan remembered, not exist until 1993, built by Shuji Nakamura at Nichia Chemical Industries in Japan after a decade of extraordinary individual effort. That device had achieved 2.7 percent. This one, built by a team working from a precise roadmap that Karan had brought to them fourteen months ago, was at 10.4.
The gap between 2.7 and 10.4 was not just a number. It was twenty years of optimisation that the team had not needed to discover because Karan had already carried the knowledge of where the optimisation led.
"The white LED package," Karan said. "Built from this device."
"I can have it packaged by tomorrow," Mehta said. He was already writing a list. "The YAG:Ce phosphor stock we made in November. The standard cup package. The wire bonder is free this afternoon."
"Tomorrow morning," Karan said.
"Tomorrow morning," Mehta confirmed.
The white LED test happened on the morning of 22 March.
Mehta had stayed late to package the device himself — this was not a job he would delegate, not this one. He had mounted the chip from Wafer 847's best die in a standard LED cup, bonded the gold wires with the precision of a man who had bonded ten thousand wires in his career and who knew the difference between a good bond and a mediocre one at the tactile level. He had prepared the phosphor-loaded epoxy — YAG:Ce at twelve percent by weight, slightly higher than the November demonstration, tuned for better colour balance — and formed the dome over the chip and the cup, shaping it carefully to maximise light extraction.
He had cured the epoxy overnight at 80 degrees Celsius.
At eight in the morning on the 22nd, with the full team gathered, he placed the packaged device in the test chamber.
Ramesh dimmed the lights.
The current engaged.
White light.
Not a glow requiring careful examination in a darkened room. A clearly visible, unmistakable point of white light from a component the size of a small button, running on forty milliamps of current.
Mehta ran the spectrophotometer measurement. The blue peak at 465nm. The phosphor emission band centred at 558nm. The colour temperature: 5,600 Kelvin — the slightly cool white of an overcast daylight sky.
Then the efficacy measurement: lumens per watt. The calibrated detector, the calibrated measurement, the arithmetic.
"86 lumens per watt," Mehta said.
He said it in the tone of a man reading an instrument that is reporting a number he has to check twice.
He checked it twice.
"86 lumens per watt," he confirmed.
Fluorescent tubes — the most efficient widely deployed lighting technology in India in 1974 — produced approximately 60 to 70 lumens per watt. The white LED at ISMC, on the morning of 22 March 1974, produced 86 lumens per watt from a solid-state device with no glass envelope, no mercury, no fragile filament, no transformer required.
Rajiv was the youngest in the room and he said the thing the younger person always said in these moments, the thing that was true and that the older people were thinking but had learned to not say first:
"We just made the best light source in the world."
Nobody corrected him.
Because he was right.
The production conversation happened at ten in the morning.
Karan called it: he, Ramesh, Suresh, Vikram, Mehta, Kumar, and Rajiv, around the lab's central bench, with the packaged white LED in the test chamber still running behind them. Still producing its 86 lumen per watt output. Still at 465nm. Still exactly what it should be.
"The device is ready for production," Karan said. "The question is timeline. I want commercial product available by October 1975. Nineteen months from now."
The room absorbed this.
Ramesh was the first to speak, and he did it carefully — the specific carefulness of a researcher who had learned to separate what was possible from what was wishful and who took the responsibility of not confusing the two seriously.
"The device is ready," Ramesh said. "The process is reproducible — we have demonstrated that over the last seven months of systematic runs. The yield on the optimised process is above 85 percent. The packaging is straightforward, using existing infrastructure."
"The productionisation challenges," Suresh added, "are known problems. Scaling the MOCVD reactor to handle larger-diameter wafers — two-inch to three-inch — to improve output per run. Automating the wire bonding, which is currently manual. Improving the phosphor coating uniformity for consistent colour temperature across devices."
"Three-inch wafers," Karan said. "The MOCVD reactor — what is the modification timeline?"
"Four months," Kumar said. "The susceptor redesign and the gas flow modelling for the larger substrate. I already have drawings."
"Begin immediately," Karan said.
"The phosphor coating uniformity," Mehta said. "Manual application gives run-to-run variation in colour temperature of about 300 Kelvin — perceptible to a trained eye. For consumer products, we need 100 Kelvin variation or less. That requires a dispensing system with precise volume control."
"We have the precision dispensing equipment in the compound semiconductor packaging line," Vikram said. "The same equipment we use for epoxy application on the GaAs photodetectors. It dispenses to 0.01 microlitre precision. That should be adequate."
"Adapt it," Karan said. "Timeline?"
"Six weeks to adapt and validate the settings for the phosphor epoxy viscosity."
"Good," Karan said.
He looked at the team.
"The market we are entering first," he said, "is industrial and commercial lighting — factories, offices, hospitals, government buildings. Not residential. Residential follows in 1976 and 1977 as production scale brings cost down. The initial price target for the industrial product is competitive with fluorescent tubes on a total cost of ownership basis — purchase price plus electricity cost plus maintenance cost over five years."
He wrote on the whiteboard:
Fluorescent tube: Rs. 8 purchase + Rs. 420 electricity over 5 years (at Indian rates, 40W, 10 hours/day) = Rs. 428 total. Need to replace 3 tubes in that period — another Rs. 24. Five-year cost: Rs. 452.
White LED (target): Rs. 35 purchase + Rs. 60 electricity over 5 years (7W, same light output) = Rs. 95 total. 50,000 hour life, no replacement. Five-year cost: Rs. 95.
"The numbers are approximate," Karan said. "But the direction is clear. The LED costs five times more to buy and uses one-fifth the electricity. Over five years, the customer saves more than Rs. 350 per light point. For a factory with five hundred light points, that is Rs. 175,000 in five years."
"The Rs. 35 purchase price," Suresh said, "requires a manufactured cost below Rs. 22 at the chip plus packaging level. Can we achieve that?"
"By October 1975, at three-inch wafer scale with automated packaging, yes," Karan said. "The chip cost scales down as wafer diameter increases — same number of reactor runs, three times the chips per run. The packaging cost is dominated by the wire bonding step, which at automated rates is below Rs. 3 per device."
"Who is the customer for the first commercial units?" Ramesh asked.
"Shergill Industries facilities," Karan said. "We install the first production run in the Gorakhpur manufacturing complex — the new factory buildings being constructed for the Phase Two expansion. That gives us a controlled environment to validate the product in real operating conditions while simultaneously reducing the complex's electricity consumption." He paused. "Estimated consumption in the Gorakhpur complex at full operation: approximately 2.4 million lighting hours per year. At 85 percent replacement with LED, the saving is roughly 35 megawatt-hours per day. At current industrial electricity rates, that is approximately Rs. 80 lakhs per year in savings."
The team looked at him.
"We are the first customer," Vikram said.
"We are the best-placed first customer," Karan said. "We control the environment, we collect the performance data, we demonstrate the result, and we save money in the process."
He set down the marker.
"The programme has fourteen months of work behind it," he said. "The device is real. The path is clear. The question for the next nineteen months is execution — scaling up what works in the lab without losing what makes it work." He looked at each person in the room. "You have done the hard part. The hard part was not knowing whether it was possible. The engineering phase is difficult but it is a different kind of difficult — it is solving known problems rather than searching for unknown solutions."
Ramesh said: "The paper."
"The paper," Karan confirmed. "We submit within sixty days. Physical Review Letters. The complete story: the hydrogen passivation mechanism, the annealing solution, the device performance data. All of it documented, reproducible, and submitted to the scientific record."
"The patents," Mehta said.
"The patents are already being drafted," Karan said. "The annealing process — the specific conditions, the nitrogen atmosphere requirement. The multi-quantum-well architecture. The phosphor package. We file in India, in the United States, and in Japan. The patent portfolio protects the productionisation timeline from the point at which the paper is published and other groups attempt to replicate."
He looked at the test chamber.
At the packaged white LED still running.
At 86 lumens per watt.
"This is a world-first," he said. "Not just for India. For the world. The first solid-state white light source with efficacy exceeding fluorescent, produced anywhere, by anyone. The paper will say so because the data says so, and the data is unambiguous."
He picked up the packaged device from the test holder — holding it carefully, as you held the first of something — and looked at it for a moment. A small cylinder of white epoxy, two contacts on one end, no bigger than the tip of his thumb.
"India will light itself with this," he said. "Then it will sell this to everyone else."
He set it down.
"Ramesh," he said, "the continuous operation test on Wafer 847 — begin it today. I want 10,000 hours of continuous operation data before we submit the paper. The lifetime claim must be backed by data."
"That is over a year," Ramesh said.
"Then start today," Karan said. "The paper will note the test as ongoing and will be updated when the 10,000-hour data is available. The initial lifetime data at one hundred hours, five hundred hours, one thousand hours will be sufficient for the submission."
He looked at the team one more time.
"Good work," he said. "All of it."
He meant it precisely: not as a formula, but as a specific technical and personal recognition of what fourteen months of systematic, rigorous, disciplined research had produced. They had started with a notebook full of specific instructions that some of them had found alarming and that all of them had decided to trust. They had executed those instructions and had, through that execution, discovered the specific difficulties that separated laboratory instructions from laboratory reality. They had solved those difficulties — the bubbler redesign, the buffer layer optimisation, the contact work, the multi-quantum-well architecture — through their own intelligence and competence and discipline.
Karan had given the direction. The team had walked the distance.
Both things were true.
Later that afternoon, alone in his office, Karan sat at his desk with a blank sheet of paper.
He thought about the thing that had happened.
In the other timeline he carried, the blue LED had been solved by one man — Shuji Nakamura, working at Nichia Chemical Industries in Japan, largely alone, against the scepticism of his management, through the early 1990s. Nakamura had discovered the annealing solution, had developed the multi-quantum-well architecture, had achieved 2.7 percent efficiency and then improved it rapidly. He had received the Nobel Prize in Physics in 2014.
The work was genuinely extraordinary. One man, with limited resources, discovering from first principles what this team had executed from a known roadmap.
Karan had not felt the discovery. He had provided the discovery and had watched others execute it. That was not the same thing. He was honest with himself about that.
But the outcome — the outcome was different. In the other timeline, the blue LED was solved in 1992 and white LED lighting became widespread in the early 2000s. Thirty years of continued incandescent dominance. Thirty years of the efficiency gap. Thirty years of energy wasted in heat from tungsten filaments in a country where energy poverty was real and grinding and measurable in children not having light to read by.
In this timeline, the blue LED was solved in August 1973 and white LED lighting would be commercially available in October 1975. The nineteen-year gap became two years.
For India, with its electricity shortage, with its 50 million households that had electrical access and its hundreds of millions that didn't and that needed every megawatt of generation to support electrification expansion — that nineteen-year compression was not abstract. It was megawatts that could be redirected from lighting to irrigation pumps. It was hospital corridors that could be lit adequately without loading the grid beyond its limits. It was the specific, grinding, daily problem of electricity rationing made slightly less grinding slightly sooner.
He wrote on the blank page:
21 March 1974.The blue LED problem is solved.EQE: 10.4 percent.White LED: 86 lumens per watt.Commercial target: October 1975.
He looked at what he had written.
Then he wrote one more line:
This is what the knowledge is for.
He had thought about the transmigrant's condition often, in the four years since Shergill Industries began. The knowledge he carried — the future he remembered — was not a gift in the simple sense that gifts were gifts. It was weight and it was responsibility. Each time he used it, he was making a choice about which future would exist instead of which other future, and those choices compounded, and the compounding was not always predictable.
The blue LED was a clear case. The knowledge was specific, the direction was unambiguous, the outcome was better for more people in more ways than the original timeline. He had no doubt about the choice.
But he was aware that not all cases were clear. He was aware that the knowledge he carried was partial — he remembered the shape of the future, not its every detail, and the shape changed as his actions changed it, and some of what he remembered was already obsolete because he had already altered the conditions that would have produced it.
He was aware that he was, in the most direct sense, playing God with the future of a country. That the confidence this required was enormous. That the margin for error was not zero.
He set the pen down.
Outside the window, the Gorakhpur evening was settling into itself — the streetlights coming on, the shift change audible from the factory, the ordinary life of an industrial city moving through its routines. In Lab Seven, behind a steel door and a cleanroom airlock, Wafer 847 was mounted in its test chamber and still producing its blue glow. Ten point four percent of the electrical energy it consumed was becoming light at 465 nanometres. The rest was becoming heat, which the test chamber's cooling system was absorbing and carrying away.
Tomorrow Mehta would start the long-term test. One hundred hours. Five hundred. One thousand. Ten thousand. The device proving itself not in the acute moment of the measurement but in the sustained, daily, ordinary work of continuing to function.
That was what production devices did. They functioned. Not spectacularly — not with the drama of a measurement that produced a world-first number on a single morning. Steadily. Reliably. Hour after hour after hour.
That was what the programme had been building toward.
Not the number. The device that produced the number.
Not the achievement. The product.
He picked up the production reports that had been waiting since morning.
He began to read.
End of Chapter 149
Blue LED Programme — Summary Status, 21 March 1974
Programme Director: Dr. Ramesh Chandra Programme Site: ISMC, Gorakhpur Programme Start: March 1973
Milestone Timeline:
March–June 1973: Reactor modification, team assembly, initial growth runs July–August 1973: First working device (Wafer 26, August 22, 1973) — EQE 1.3%, peak 466nm August–October 1973: Active layer optimisation → EQE 2.8% October–November 1973: Contact optimisation → EQE 4.1% November–December 1973: Buffer layer optimisation → EQE 5.9% January 1974: First white LED demonstration (January 14) — 4.2% white device, 45 lm/W January–March 1974: Multi-quantum-well architecture development March 21, 1974: Wafer 847 — EQE 10.4%, peak 465nm, FWHM 22nm March 22, 1974: White LED from Wafer 847 — 86 lumens per watt
Key Innovation: Identification and solution of hydrogen passivation in magnesium-doped GaN. Thermal annealing at 700-750°C in nitrogen atmosphere activates p-type dopants by breaking Mg-H bonds and driving hydrogen from the crystal.
Device Architecture (final): Sapphire substrate / AlN buffer (20nm) / GaN n-type (4μm) / InGaN/GaN multi-quantum-well active region (4 wells × 2nm wells, 10nm barriers, asymmetric) / GaN:Mg p-type (200nm) / NiO/Au p-contact / Al n-contact
White LED Package: Standard LED cup, gold wire bonds, YAG:Ce phosphor in optical epoxy (12 wt%), dome encapsulant. Colour temperature: 5,600K. CRI: >75. Efficacy: 86 lm/W.
Production Roadmap:
Reactor scale-up to 3-inch wafers: April–July 1974 Automated phosphor dispensing system: May 1974 Pilot production run (Gorakhpur factory installation): October 1974 Commercial product launch (industrial/commercial sector): October 1975 Residential market entry: 1976–1977
First Market: Shergill Industries Gorakhpur complex — 2.4 million lighting-hours/year, 85% LED replacement target, projected annual electricity saving: Rs. 80 lakhs
Paper: Submitted Physical Review Letters by May 1974. Authors: Chandra, Iyer, Malhotra, Kumar, Mehta, Sharma, Shergill.