Samsung Pushes Pixel Size Even Further With New Camera Sensor

Samsung has announced a new image sensor for smartphone cameras that it says has the smallest pixels in the industry. From a report: The ISOCELL JN1 is a 50-megapixel sensor with a relatively tiny 1/2.76-inch format, meaning its pixels are just 0.64um in size. For comparison, Samsung already broke records in 2019 with the slightly larger ISOCELL Slim GH1, another 50-megapixel sensor with 0.7um pixels. Conventional camera wisdom says that smaller pixels usually result in worse image quality with higher noise, so why is Samsung doing this? According to the company, it's about form factor versatility. The sensor's smaller size means it can be used in ultrawide or telephoto camera modules -- which are challenging to design when size is at a premium -- or as a way to reduce the height of the primary camera bump. As with other high-resolution camera sensors, the JN1 will make use of pixel-binning technology that combines multiple pixels into one for higher light sensitivity. In this case, Samsung says the sensor will capture 12.5-megapixel photos with the equivalent of 1.28um pixels, and the company is also claiming a 16 percent boost to light sensitivity with its ISOCELL 2.0 tech.

In before...

[RightSaidFred99]

Did I make it before some guy raised on ca. 2005 digital cameras chimes in with "but we all know the megapixel race is a scam! Who wants a 50-megapixel sensor it will be all noisy!!"

Ahh, conventional wisdom - some people just never let go of it.

Re:

[Joce640k]

Ahh, conventional wisdom - some people just never let go of it.

Laws of physics? We don't need no steenkin' laws of physics...

Re:

[RightSaidFred99]

Laws of physics, LOL! Can you point me to the law of physics that says a 4 pixel grid of smaller pixel sensors is inherently worse than a single pixel of the same overall size? I assume you lose some area due to spacing between the pixels, but that may (or may not) be made up for by the ability to do interpolation and apply various algorithms to the data. Note that this ability has nothing to do with any laws of physics that I'm aware of.

Re:

[RightSaidFred99]

OK. Now let's do some math. If you combine the light from those pixels, what's 4 * 1/4?

Wavelength, Diffraction

[Roger W Moore]

Can you point me to the law of physics that says a 4 pixel grid of smaller pixel sensors is inherently worse than a single pixel of the same overall size?

At 640 nm these pixels are now smaller than the wavelength of red light which is a hard physical limit on resolving power. However, this is rarely reached due to diffraction effects which means that the resolution limit is likely to be larger than the pixel size for all wavelengths. If your pixel size is less than the physical optical resolution of your device then it is hard to see how adding more, even smaller pixels will improve anything. Astronomers seem generally limited by the optical resolution of t

Re:

[RockDoctor]

Can you point me to the law of physics that says a 4 pixel grid of smaller pixel sensors is inherently worse than a single pixel of the same overall size?

Unless you change the sensing technology, then yes, smaller pixel area results in smaller photon counts, and for technology "X", 1 photon yields (say) 0.6 electrons to measure in your electronics. Smaller electron counts result. Which means more noise in your reading, for the same numerical reason that integer percent measures have more noise than integer

Re:

[Joce640k]

Did I make it before some guy raised on ca. 2005 digital cameras chimes in with "but we all know the megapixel race is a scam! Who wants a 50-megapixel sensor it will be all noisy!!"

The information you want is right there in the summary: "...the JN1 will make use of pixel-binning technology that combines multiple pixels into one for higher light sensitivity. In this case, Samsung says the sensor will capture 12.5-megapixel photos"

So... 12.5 megapixels in low light conditions.

Egg, meet RightSaidFred's face.

Re:

[Fly Swatter]

Soon there will be a 1,000,000 megapixel sensor, but you will only be able to capture a 15 mp image. It's a shame people fall for marketing gimmicks. Will it have better images? Sure. But people will only compare megapixel thingy counts.

Re:In before...

[RightSaidFred99]

No, that's my exact point. People are so used to reflexively guffawing at the "megapixel race" that they don't bother understanding what's going on. It's like they think Samsung must be run by morons or something and they, SlashDot Nerdy Guy, are the only ones who know about how silly the "megapixel race" is.

Only it's not silly, the people at Samsung aren't morons, and they use pixel binning to get better metrics across a broader range of scenarios using a tiny 50mp sensor and utilizing various algorithms to pixel bin (or not, in some scenarios). So that 50mp sensor is better than a 12.5mp sensor in the same form factor. The replies to my post illustrate this lack of understanding.

I wonder what their refrain will be if/when a large form factor camera maker goes with the same strategy? "But muh megapixelz war is teh silly!!".

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[Bengie]

Many years back I was reading about high density pixel cameras where a single logical pixel was actually composed of several physical pixels, each collecting slightly different information, like phase timing. This allowed them to do cool thing computationally like bringing the entire photo into focus.

As for the people talking about "it's smaller than the frequency". I'm not sure about available products, but there are micro-meter antenna in labs that can not only receive data from frequencies magnitudes l

Re:

[nagora]

So that 50mp sensor is better than a 12.5mp sensor in the same form factor.

In what way? What is the visible difference?

Re:

[Solandri]

they use pixel binning to get better metrics across a broader range of scenarios using a tiny 50mp sensor and utilizing various algorithms to pixel bin (or not, in some scenarios). So that 50mp sensor is better than a 12.5mp sensor in the same form factor.

While binning and noise reduction algorithms have helped, the main factor is simply that sensor sensitivity has improved. You need fewer photons to capture a good signal.

• 20 years ago you could barely pull a decent image from 1600 ISO with pixels 10 um w

Physical Optical Limit

[Roger W Moore]

they use pixel binning to get better metrics across a broader range of scenarios using a tiny 50mp sensor and utilizing various algorithms to pixel bin (or not, in some scenarios).

The new pixel size is smaller than the wavelength of red light and, by the time you add diffractive effects in there these pixels are almost certainly below the physical optical resolution of the device. Could you please explain how pixel binning, or indeed any algorithm, will let you improve the resolution over the physical optical resolution of the device? If they have managed this it would be genuinely interesting to hear how but given that previously Samsung made a phone than spontaneously combusted I

Re:

[theshowmecanuck]

I have a 50MP sensor now, near enough. Oh wait. It's on a 35mm format, but that is still pushing it. Ahh, conventional wisdom is still wisdom. Some people just don't have it.

Re:

[mistergrumpy]

Close but not quite...

If the readout noise is really fixed (doesn't depend on pixel size), and is uncorrelated between the four pixels, adding it in quadrature means you get twice the noise as a single pixel. There are other noise terms that go as the square root of the number of photons absorbed and the capacitance of the sensor. Optimizing for all use cases is impossible, but I bet the designers understand these issues pretty well.

Re:

[webmistressrachel]

I am liking your thoughts and wish to subscribe to your newletter.

I enclose two pounds coins, taped the form as requested. For obvious reasons, the delivery address is a PO Box, no doubt they will take those away soon....

Oh great.

[Pig Hogger]

The smaller the pixel, the more susceptible to quantum noise it is.

10 years ago, I got myself a camera with a 8.4 m pitch pixels full-frame sensor [digicamdb.com]. It’s right above the 6 quantum noise threshold, and 10 years later, I am still amazed by the pictures taken in low-light. You only start to see noise above 1600 ISO!

So, those extra-super-duper-pixel cameras must be noisy as hell, and a lot of processing power will be used to remove the noise, and with it, a bit of picture details...

Do not want.

[msauve]

" a way to reduce the height of the primary camera bump."

Or they could simply make the phone as thick as the cameras, providing room for a larger user replaceable battery, a micro SD card slot, and an audio jack. But nope, that doesn't provide the critical features of planned obsolescence, overpriced storage, and product tie-in

Re:

[account_deleted]

Comment removed based on user account deletion

Noise

[BrendaEM]

While many pixels are great, small pixels means noise.

Pixels smaller than the wavelength of the light?

[john83]

Visible light goes up to about 0.75 microns, a little bigger than these pixels. I wonder whether they have to do any processing to compensate? Of course, they're doing other post-processing anyway.

Re:

[ImprovOmega]

Worse than that, diffraction [stackexchange.com] adds more to the effective limit, so it's more around 1.6 microns, so even a square of four of these probably won't add much to the resolving power.

Re:

[Jamlad]

Camera sensors are not continuously sensitive. Look at the quantum efficiency curves and you'll see broad, but definite RGB absorption peaks that extend into the near IR [alliedvision.com]. So I expect they've either narrower bands or are just taking the sensitivity hit. Film stocks have similar spectral sensitivity curves.

Video

[dsummerbell]

The video and other information of the announcement can be found here [wavband.com]