What if a photo had a trillion megapixels? The physics limit

A trillion megapixels is 10^18 pixels: enough to photograph the whole Earth at 2.3 cm per pixel. Diffraction says the lens alone would be 390 metres wide, and that is before the data bill.

The Trifid and Lagoon nebulae in pink and blue clouds against a dense star field, from the Vera C. Rubin Observatory's first images
Photo: NSF–DOE Vera C. Rubin Observatory, CC BY 4.0, via Wikimedia Commons

A photo with a trillion megapixels, 1018 pixels, cannot be taken in one shot by any camera, because diffraction limits how many separate points a lens can resolve. To fill that many pixels with real detail across a normal 60° view, the lens would need to be about 390 m wide and perfect, the sensor would cover 25 hectares at today's smallest 0.5 µm pixels, and one raw file would be about 1.5 exabytes. The only way to make such an image is a mosaic built over time, like photographing the entire Earth at 2.3 cm per pixel.

Key takeaways
  • A trillion megapixels is 1018 pixels: the whole Earth at 2.3 cm per pixel.
  • Diffraction sets the real limit: a 60° view needs a perfect lens about 392 m wide.
  • At today's smallest 0.5 µm pixels, the sensor would cover 25 hectares.
  • One 12-bit raw file is 1.5 exabytes, about 139 days to read at 1 Tb/s.
  • Only a mosaic over time can do it: about ten billion 100 MP frames.

Megapixel counts have climbed fast: phone sensors now claim 200 MP, the Rubin Observatory's camera takes 3,200 MP in one exposure, and stitched panoramas have passed 365,000 MP. A trillion megapixels is another factor of about three million beyond that panorama.

It sounds like a number you could reach by waiting. Physics disagrees, and the reasons are a tour of everything that limits a photograph:

  • the wave nature of light (diffraction);
  • the graininess of photons (shot noise);
  • the cost of moving bits (storage and readout).
1018pixels in a trillion-megapixel image
~390 mlens diameter needed to resolve them over a 60° view (my calculation)
0.5 µmsmallest commercial pixel pitch, Samsung ISOCELL HP5 (2025)
365 GPMont Blanc panorama, 2015: 70,000 frames, 46 TB

What does a trillion megapixels actually mean?

A trillion megapixels is \( 10^{12} \times 10^6 = 10^{18} \) pixels, a billion gigapixels. To get a feel for it: Earth's surface is about \( 5.1\times10^{14} \) m², so \( 10^{18} \) pixels would cover the whole planet, oceans included, at

\[ s = \sqrt{\frac{5.1\times10^{14}\ \text{m}^2}{10^{18}}} \approx 0.023\ \text{m} = 2.3\ \text{cm per pixel} \]

That is enough to see every brick in every wall on Earth, in a single image. Today's records look small against it.

ImagePixelsHow it was madeTimes smaller than 1018
Flagship phone sensor (0.5–0.56 µm pixels)200 millionOne sensor, one shot5 billion
AWARE-2 multiscale camera, Duke University (2012)about 1 billion98 microcameras behind one shared lens, one shot1 billion
LSST Camera, Rubin Observatory (2025)3.2 billion189 sensors on a focal plane over 60 cm wide310 million
Mont Blanc panorama, in2white (2015)365 billion70,000 frames stitched, 15 days of shooting, 46 TB2.7 million
Trillion megapixels1018?1
The story

In 2015 the Italian photographer Filippo Blengini and a small team from his in2white project spent 15 days shooting the Mont Blanc massif from about 3,500 m. A Canon DSLR on a motorised pan-and-tilt head from the German maker Clauss swept the mountain frame by frame until they had 70,000 pictures. The source files filled 46 terabytes, and stitching them took about two months of computer time. When the panorama went online in May 2015 it was reported at 365 gigapixels, the largest photograph in the world. Look at what it took: two weeks at altitude, tens of thousands of frames and months of processing. That is the honest price of pixels, and it is still millions of times short of the number in this post's title.

New Atlas: 365-gigapixel Mont Blanc panorama (May 2015)
Pixel records on a log scale10⁸10¹⁰10¹²10¹⁴10¹⁶10¹⁸phone, 200 MPAWARE-2, 1 GPRubin, 3.2 GPMont Blanc, 365 GPa trillion megapixelsa gap of about 6.4 decades
Each tick is a factor of 100. The biggest photograph ever made sits roughly a third of the way along; the trillion-megapixel image is 2.7 million times further.
A wide panorama of the Aiguille du Midi station and the glaciers of the Mont Blanc massif above Chamonix
The Mont Blanc massif, subject of a 365-gigapixel stitched panorama in 2015. Even that record is 2.7 million times short of a trillion megapixels.Photo: Christian David, CC BY-SA 4.0, via Wikimedia Commons

Why can't you just add more pixels?

Because beyond a point, extra pixels only sample blur. Light passing through any aperture spreads into an Airy disc, and once a pixel is smaller than that disc, more pixels add file size but no detail. This is the diffraction limit, and it is the first hard wall.

The Airy disc diameter, to the first dark ring, for a lens at f-number \( N \) is

\[ d_{\text{Airy}} = 2.44\,\lambda N \]

For green light at 550 nm, an f/1.8 phone lens makes a 2.4 µm disc and an f/8 lens a 10.7 µm disc. A 0.5 µm pixel is already about five times smaller than the best phone disc, which is why phone makers bin four or sixteen pixels into one for normal photos. The chart shows the gap.

One point of light vs 0.5 µm pixelsAiry disc at f/1.8: 2.4 µmeach square: one 0.5 µm pixelthe disc spans ~5 pixels acrossso neighbouring pixels seealmost the same lightphones bin 4×4 = 16 pixelsinto one 2 µm "big pixel"
Drawn to scale at 100 px per micrometre. Once pixels are smaller than the Airy disc, adding more of them resamples the blur instead of adding detail.
Airy disc diameter at 550 nm vs the smallest pixel
See the data as a table
Size (µm)
f/1.82.4
f/2.83.8
f/45.4
f/810.7
f/1621.5
f/2229.5
0.5 µm pixel0.5

Source: Computed from d = 2.44 λN at λ = 550 nm; pixel pitch from Samsung ISOCELL HP5 (2025) · Two pixels per Airy radius is about the useful limit; beyond it, more pixels only resample the blur.

So pixel count isn't really the right measure. The right one is the number of resolvable spots a lens can deliver, called its space-bandwidth product. That depends on the lens's aperture and field of view, not on the sensor. I go through the everyday version of this in do megapixels matter?

How big a lens would a trillion-megapixel photo need?

About 390 m across, if it were diffraction-limited over a 60° by 40° view. Here is the working.

Calculated in this post from the Rayleigh criterion, θ ≈ 1.22 λ/D (Wikipedia: Airy disk), with these example inputs: 550 nm light, a 60° × 40° view, 10¹⁸ pixels and two pixels per resolvable spot on each axis.

  1. Assumptions. Wavelength \( \lambda = 550 \) nm. A 3:2 field of view of 60° × 40°, or 1.047 × 0.698 radians. A perfect lens (no aberrations), and Nyquist sampling: 2 pixels per resolvable spot along each axis, so 4 pixels per spot.
  2. Angular resolution. By the Rayleigh criterion, a lens of diameter \( D \) separates two points at angle
    \[ \delta\theta = 1.22\,\frac{\lambda}{D} \]
  3. Count the spots.
    \[ N_{\text{spots}} = \frac{\theta_x}{\delta\theta}\cdot\frac{\theta_y}{\delta\theta} = \frac{\theta_x\theta_y D^2}{(1.22\lambda)^2} \]
  4. Solve for D. We need \( N_{\text{spots}} = 10^{18}/4 = 2.5\times10^{17} \):
    \[ D = 1.22\lambda\sqrt{\frac{N_{\text{spots}}}{\theta_x\theta_y}} = 1.22 \times 5.5\times10^{-7}\ \text{m} \times \sqrt{\frac{2.5\times10^{17}}{1.047 \times 0.698}} \approx 392\ \text{m} \]
  5. Check the method. The same formula says a 200 MP image needs a perfect 5.5 mm aperture; phone main cameras have apertures of only a few millimetres, which is why their 200 MP mode is already past the limit. A 3.2 GP image needs 22 mm, and a 1 terapixel image 39 cm.
A 392 m lens is the height of a 100-storey tower, and it would have to be perfect across a 60° field.

No lens design achieves that even at 5 cm: wide fields bring aberrations that grow with aperture. Duke's AWARE-2 got round this in 2012 with a multiscale design, one shared front lens feeding 98 small cameras with 14 MP sensors each, to reach about a gigapixel. Scaling that trick by \( 10^9 \) is the honest description of the problem.

The LSST Camera's focal plane: a large circular mosaic of square image sensors in a clean room
The LSST Camera focal plane, more than 60 cm across, with 189 sensors giving 3.2 gigapixels. The biggest single-exposure camera is a tiling of ordinary-sized chips.Photo: Jacqueline Orrell/SLAC National Accelerator Laboratory/NSF/DOE/Rubin Observatory/AURA, CC BY 4.0, via Wikimedia Commons

How big would the sensor be, and could it even record the light?

At the smallest pixel pitch made today, 0.5 µm, \( 10^{18} \) pixels would cover \( 10^{18} \times (0.5\times10^{-6}\ \text{m})^2 = 2.5\times10^5 \) m², or 25 hectares, about 35 football pitches. Each of those tiny pixels can hold only a few thousand electrons, which caps how cleanly it can record light.

That cap comes from shot noise. Photons arrive randomly, following Poisson statistics, so a pixel that collects \( n \) electrons has noise \( \sqrt{n} \) and a best possible signal-to-noise ratio of

\[ \text{SNR}_{\max} = \frac{n}{\sqrt{n}} = \sqrt{n_{\text{full well}}} \]

OmniVision's 0.56 µm pixel holds 5,500 electrons, so \( \text{SNR}_{\max} = \sqrt{5500} \approx 74 \), about 6 stops of clean tone per pixel. Shrink pixels further and each one gets noisier. That is why the frontier of sensor research has moved to counting single photons.

0.56 µm-pitch CMOS image sensor for high resolution applicationChun Yung Ai et al., OmniVision Technologies · International Image Sensor Workshop · 2023

The paper describes a 200-megapixel stacked sensor built on a 28 nm process with 0.56 µm pixels, the successor to the company's 0.61 µm design. It reports a full-well capacity of 5,500 electrons with lower dark current and read noise than the previous generation, similar quantum efficiency, and a quad-pixel design that shares one microlens across four pixels because tiny pixels lose light-gathering efficiency. The limit: a manufacturer's paper, with performance measured by the maker.

Read the paper (free)
The Quanta Image Sensor: Every Photon CountsEric R. Fossum, Jiaju Ma, Saleh Masoodian, Leo Anzagira, Rachel Zizza · Sensors · 2016

Fossum, who invented the CMOS image sensor's active pixel, describes a sensor made of billions of "jots", sub-micron photodetectors that each report whether a photon arrived. Demonstrated jots at 1 to 1.4 µm pitch reached read noise as low as 0.22 electrons rms (0.29 on average), good enough to count single photons. Images are formed afterwards by summing jots over a small cube in space and time, so resolution and noise can be traded after capture. A 1-gigajot sensor at 1,000 frames per second would produce more than 1 terabit of data per second. The limit: these were test chips; the readout data rate is the hard part.

Read the paper (free)

Where would you even store it?

One raw trillion-megapixel frame at 12 bits per pixel is \( 10^{18} \times 12 / 8 = 1.5\times10^{18} \) bytes, 1.5 exabytes. Reading it out at the terabit-per-second rate Fossum's team calls the challenge of a gigajot sensor would take

\[ t = \frac{1.2\times10^{19}\ \text{bits}}{10^{12}\ \text{bits/s}} = 1.2\times10^{7}\ \text{s} \approx 139\ \text{days} \]

So a "snapshot" would take four and a half months to read. The Mont Blanc panorama, at \( 3.65\times10^{11} \) pixels, already needed 46 TB of source frames and two months of processing; scaling it up by 2.7 million gives about 126 million terabytes of source files. Compression helps only where the scene is simple, and a scene worth \( 10^{18} \) pixels is, by definition, not simple.

Calculated in this post from the 46 TB of source frames and two months of processing reported by PetaPixel (46 TB also in New Atlas), scaled by 10¹⁸ ÷ 3.65 × 10¹¹ ≈ 2.7 million.

What does an N-pixel photo cost in physics?

Is there a version that would actually work?

Yes, if you give up the idea of one camera and one moment. A trillion-megapixel image is reachable as a mosaic: many cameras or many passes, stitched. Physics allows it; logistics decide it.

The Earth as one picture

Consider the Earth at 2.3 cm per pixel. The per-frame physics is easy. A camera from 1,000 m altitude resolving 2.3 cm needs an angular resolution of \( 2.3\times10^{-5} \) rad; by \( D = 1.22\lambda/\delta\theta \), that's a lens just 2.9 cm wide (atmosphere and motion permitting).

Calculated in this post from the Rayleigh criterion (Wikipedia: Airy disk) and Earth's 510 million km² surface (Wikipedia: Earth), with these example inputs: a 1,000 m altitude and 550 nm light.

The hard part is the count:

  1. Frames. At 100 MP per frame, \( 10^{18} \) pixels is \( 10^{10} \) frames, ten billion photographs.
  2. Flying time. Even a fleet of 1,000 aircraft each shooting one frame per second would need about 116 days of flying, before clouds, oceans and night.

    Calculated in this post from 10¹⁰ frames (see above; Earth area from Wikipedia: Earth), with these example inputs: 1,000 aircraft and one frame per second each.

  3. Time itself. A real image of that size would be a mosaic from many dates, so it would not be one moment anyway.

Would there be enough light?

Light has a deeper ceiling too. Each pixel needs photons: for an SNR of 10 you need about 100 detected photons per pixel, so \( 10^{20} \) photons per frame.

By my rough estimate, a sunlit scene imaged at about f/1.4 delivers on the order of \( 10^{19} \) photons per second to each square metre of sensor, so 25 hectares would collect \( 10^{20} \) photons in well under a millisecond. For daylight scenes, brightness is not the limit. At night, it would be. This is the same light budget you juggle in the exposure triangle and in RAW dynamic range, scaled up by a billion billion.

Close-up of a small CMOS image sensor chip in its package, showing the rectangular pixel array
A CMOS image sensor. At 0.5 µm per pixel, a trillion-megapixel version would cover 25 hectares and still be limited by the lens in front of it.Photo: Phiarc, CC BY-SA 4.0, via Wikimedia Commons
Verdict: possible only with changes (a mosaic, not a snapshot)

As a single photograph, a trillion megapixels is ruled out in practice by diffraction (\( \delta\theta = 1.22\lambda/D \)): a 60° view would need a perfect 392 m lens, plus a 25-hectare sensor whose 0.5 µm pixels each top out near SNR 74, and a 1.5 EB file that takes 139 days to read at 1 Tb/s. As a mosaic it is possible: the whole Earth at 2.3 cm per pixel equals \( 10^{18} \) pixels, and each frame's optics are trivial. The deciding number is ten billion 100 MP frames, about 2.7 million Mont Blanc panoramas.

What if next?

The opposite scale, a camera whose blur is a centimetre wide, is in what if your camera were the size of a building; the phone side of the story is in how phone cameras use computation. More at Kala.

Questions people ask

What is the highest-resolution photo ever taken?

Stitched panoramas hold the pixel records; the 2015 Mont Blanc panorama by Filippo Blengini's in2white team was reported at 365 gigapixels from 70,000 frames. For a single exposure, the LSST Camera at the Rubin Observatory records 3.2 gigapixels per shot.

Is there a limit to how many megapixels a camera can have?

Not to the count, but to the useful count. A lens of diameter \( D \) resolves angles no finer than \( 1.22\lambda/D \), so the number of real details is fixed by aperture and field of view. Past that, extra pixels only resample blur.

Why do 200 MP phone photos not look 4× sharper than 50 MP?

Because a phone lens at about f/1.8 makes an Airy disc around 2.4 µm wide, nearly five times a 0.5 µm pixel. The sensor bins pixels in normal light, and the full 200 MP mode mostly adds samples of the same blur, plus more noise per pixel.

How much storage would a trillion-megapixel photo need?

About 1.5 exabytes uncompressed at 12 bits per pixel, which is 1.5 million terabytes. Read out at 1 terabit per second, that takes about 139 days.

Sources

  1. Fossum et al.: The Quanta Image Sensor: Every Photon Counts, Sensors (2016) (source for jots, 0.22 e⁻ read noise and the 1 Tb/s readout figure)
  2. Ai et al.: 0.56 µm-pitch CMOS image sensor, International Image Sensor Workshop (2023) (source for the 5,500 e⁻ full well)
  3. Notebookcheck: Samsung unveils the smallest 200 MP sensor (2025) (source for the 0.5 µm pixel)
  4. Laser Focus World: Gigapixel camera from Duke University (2012) (source for AWARE-2's 98 microcameras and 120° field)
  5. Rubin Observatory: First images from the 3,200-megapixel camera (2025) (source for the 3.2 GP camera and 189 sensors)
  6. PetaPixel: 365-gigapixel panorama of Mont Blanc (2015) (source for 365 GP, 70,000 frames, 46 TB)
  7. New Atlas: 365-gigapixel Mont Blanc panorama (2015) (source for the 15 days of shooting at 3,500 m, the Canon DSLR and the Clauss pan-and-tilt head)
  8. Wikipedia: Airy disk (source for the 2.44 λN diameter and the Rayleigh criterion)
  9. Wikipedia: Earth (source for the 510 million km² surface area)
  10. Zheng, Horstmeyer and Yang: Wide-field, high-resolution Fourier ptychographic microscopy (2014) (an example of building gigapixel images by trading time for space-bandwidth)
megapixelsdiffraction limitAiry diskgigapixel photographyspace-bandwidth productimage sensor pixel sizeshot noisequanta image sensor
Sapta

Sapta (Saptarshi Sarkar) is a mechanical engineer turned marketer in Bengaluru and a solo touring rider. He writes the seven Saptaverse blogs. Work with him or read more of Artblog.

Keep reading in Artblog

From the other blogs