The camera in your phone exists because a NASA engineer wanted to shrink spacecraft imaging onto a single chip, while much of his field doubted it could ever match the sensors already flying. Those chips now ship in the billions every year.

In 1993 a NASA engineer published a paper whose title amounted to a playground taunt. Active Pixel Sensors: Are CCD’s Dinosaurs? argued that the imaging technology that had just conquered astronomy, broadcasting and much of electronic imaging was living on borrowed time. Its author, Eric Fossum, predicted the successor would take over most applications by the turn of the century.

Much of his field doubted it could ever match the sensors already flying.

Fossum had standing to make the claim. The Jet Propulsion Laboratory had hired him in 1990 as a specialist in the very devices he was writing off, a detail recorded in NASA’s Spinoff publication. The paper was an inside job.

Why spacecraft cameras were so bulky

A charge-coupled device collects light as electrical charge in each pixel, then shuffles that charge across the chip, pixel to pixel, like buckets passed along a line of firefighters, until it reaches a single amplifier in the corner. Lovely physics. Brutal logistics. Every bucket has to arrive almost perfectly intact, which means high voltages, a hungry power budget and support electronics that will not fit on the same piece of silicon.

Workable for a spacecraft built like a bus. Hopeless for the smaller, cheaper missions NASA wanted in the early 1990s. Fossum’s brief, as he described it to the National Inventors Hall of Fame, was to take a camera system roughly the size of a bread box and shrink it to something you could hold like a coffee cup.

The trick that made it work

Give every pixel its own amplifier and the bucket brigade disappears. Each pixel reads out its own signal, the charge never travels, and the voltage requirements collapse along with it. The catch was noise, which had sunk earlier attempts at the same idea.

Fossum’s fix borrowed a technique straight from the CCD playbook: measure each pixel twice, before and after exposure, and treat the difference as the real signal. He explains it with a deli counter. Weigh the empty container, weigh it again with the food in, and what is left over is what you came for.

Once the noise was handled, the timing circuits, the converters and the image processing could all move onto the same chip. Power consumption dropped by as much as a factor of one hundred against a comparable CCD system, by NASA’s accounting. A camera stopped being a system and became a component.

The decade nobody believed him

“It was a long, uphill fight to prove that to everybody,” Fossum told the Hall of Fame, recalling how firmly his field resisted the prospect of CMOS displacing the CCD.

The sceptics had a case. Early active pixel sensors really were noisier and clumsier, and Fossum’s own paper conceded that the incumbent would take years to dislodge. Parity came later, through refinements built around a structure called the pinned photodiode, developed with partners including Kodak. Only then did these chips start beating CCDs on picture quality as well as on cost and power, as Draper’s announcement of the engineering prize he later won sets out.

Roughly a decade separates being right from being believed.

Dentists got there before phone makers

In 1995 Fossum and four colleagues left JPL to found Photobit, which took out a licence on the technology and became the first company to sell CMOS image sensors commercially. The early adopters were obscure. One was a three-person dental imaging outfit in Long Island City, and as the NASA feature on the imager dentists use daily tells it, Fossum kept warning the founder that the chips were immature and he could not yet make the pixels small enough. The founder did not care. He wanted them anyway.

The original brief came good too. Perseverance carries 23 cameras across the rover and its entry vehicle, and nine of its surface engineering cameras share a global-shutter CMOS detector that cancels noise with the same double-sampling trick, according to the mission’s camera paper in Space Science Reviews. A camera that costs almost nothing in mass or power is a camera you can afford to bolt on two dozen times.

What seven billion a year looks like

Every phone. Most cars, several times over. Doorbells, endoscopes, webcams, barcode scanners, the camera watching a production line for defects. Yole Group, an analyst firm tracking the industry, put CMOS image sensor revenue at 23.2 billion US dollars in 2024 and forecasts 30 billion by 2030, with smartphones still driving most of the demand.

In January 2026 the National Academy of Engineering awarded Fossum the Charles Stark Draper Prize, describing his invention as the core technology behind roughly seven billion cameras produced each year. It carries half a million dollars, and joins a Queen Elizabeth Prize, a National Medal of Technology and a Technical Emmy.

Which would make a tidy ending, except that Fossum is now at Dartmouth building the Quanta Image Sensor, a design that counts individual photons rather than pooling them. Same move as before. Find the assumption the incumbent cannot function without, then check whether silicon has quietly outgrown it.