Speaker A
Every photo you took today almost certainly passed through a piece of Sony silicon. Not the processor, not the memory, the part that actually sees.
Sony revolutionized image sensors with breakthroughs in global shutter and stacked sensor tech, reshaping cameras and machine vision.
Key Takeaways
- Sony transformed sensor design by innovating back-illumination and stacked sensor architectures.
- The global shutter breakthrough enables distortion-free, high-speed imaging previously thought impossible.
- Dynamic range improvements reduce reliance on computational HDR, improving image quality in phones.
- Sony's collaboration with TSMC signals a strategic move to dominate future machine vision markets.
- Sony sensors power not only cameras but also emerging autonomous and industrial applications.
What the video covers
- Sony produces roughly half of the world's image sensors, crucial chips that convert light into data.
- The company pioneered back-illuminated sensors in 2008, flipping sensor design to improve light capture.
- In 2012, Sony introduced stacked CMOS sensors, separating pixel and logic layers to optimize performance.
- Sony's stacked sensor architecture enabled high-speed slow-motion video and industry dominance in smartphone and camera sensors.
- The first full-frame global shutter sensor was announced by Sony in 2023, eliminating rolling shutter distortion.
- Global shutter was made possible by stacking circuitry separately from photodiodes, preserving sensitivity.
- In 2026, Sony launched the A7R6 with a fully stacked sensor offering over five times faster readout at 60+ megapixels.
- Sony also advanced dynamic range with lateral overflow integration capacitors, allowing one-shot HDR without ghosting.
- Sony formed a joint venture with TSMC in 2026 to further develop advanced sensor technology and machine vision.
- These innovations position Sony as the key supplier of 'eyes' for future computing, impacting phones, cameras, robots, and autonomous systems.
Chapters
- 00:00Sony's dominance in image sensors
- 01:44Challenges in sensor design and physics
- 03:30Back-illuminated sensor breakthrough
- 04:58Stacked CMOS sensor innovation
- 05:58Global shutter technology breakthrough
- 06:52High-resolution flagship sensor advancements
- 08:42Small pixel sensors for security cameras
- 10:51Sony and TSMC joint venture
- 14:40Machine vision and autonomous systems
- 20:12Future outlook and industry impact
Full Transcript — Download SRT & Markdown
Speaker A
Sony makes roughly half of all the image sensors on Earth, the chips that turn light into data. And in 2026, the company did something with that position that almost nobody in the semiconductor industry saw coming. It cracked a string
Speaker A
of engineering problems that had been written off as dead ends for decades. It signed a deal with the most powerful chipmaker on the planet that breaks the rules that company normally plays by, and it quietly repositioned itself as the
Speaker A
supplier of eyes for the next era of computing. This is the story of how a company most people still file under televisions and PlayStations built what the chip world said could not be built and why the consequences reach from the
Speaker A
phone in your pocket to the robots that are about to walk into factories, warehouses, and hospitals. To understand why any of this qualifies as impossible, you have to understand why image sensors are among the cruelest problems in all of
Speaker A
semiconductors. A processor deals in clean abstractions, a one or a zero, a voltage high or low with enormous margins for error baked in. An image sensor deals in physics at its rawest. Light arrives as individual photons. And in a dim room, a single
Speaker A
pixel might catch only a handful of them during an exposure. The sensor's job is to catch each photon, convert it into an electron, store that electron without losing it, measure the charge without corrupting it, and do all of this
Speaker A
millions of times in parallel, dozens or hundreds of times per second on a sliver of silicon that also has to be cheap enough to put inside a $400 phone. Every improvement fights every other improvement. Make the pixels smaller to
Speaker A
fit more of them, and each one catches less light. Add circuitry to read them faster and you steal area that should be catching photons. Push for more dynamic range and you drown in noise. For most of the industry's history, sensor design
Speaker A
was a game of choosing which flaw you could live with. Sony's answer going back nearly two decades was to stop playing that game and start rearranging the chip itself. In 2008, it announced Xmor R, the first mass-produced back
Speaker A
illuminated sensor. The idea sounds almost silly when you say it out loud. A traditional sensor has its wiring on top, so light has to fight its way past a thicket of metal before it reaches the photodiodes, like looking at the sky
Speaker A
from the bottom of a well. Sony flipped the whole structure over, ground the silicon down from the back until it was almost absurdly thin, and let light hit the photodiodes directly. Everyone knew this would work in theory. Nobody had
Speaker A
managed to manufacture it at scale because thinning a wafer to a few micrometers without destroying it is the kind of process that ruins yields and bankrupts production lines. Sony shipped it anyway, and within a few years back illumination went from exotic to
Speaker A
mandatory across the entire industry. Then in 2012 came the move that set up everything happening right now. Sony announced Xmor RS, the world's first stacked CMOS image sensor. Instead of putting the pixels in the processing circuit side by side on one chip, it
Speaker A
built them as two separate chips and bonded them face to face, pixels on top, logic underneath. The pixel layer could be optimized purely for catching light.
Speaker A
The logic layer could be optimized purely for speed. Later versions wedged a layer of DRAM memory in between, giving the sensor its own private buffer so it could dump an entire frame into memory in a flash and read it out at
Speaker A
leisure. This is the architecture that made slow motion video on phones possible. And it is the architecture that turned Sony from a strong player into the company whose sensors sit inside iPhones, inside most flagship Android cameras, inside the mirrorless
Speaker A
cameras of its own rivals, and inside a growing share of the cars on the road.
Speaker A
When you hear that Nikon's best cameras run on Sony fabricated sensors, that is not a rumor. That is just how the industry works now. Which brings us to the first thing nobody thought was possible. For as long as digital cameras
Speaker A
have existed, almost every consumer sensor has used what is called a rolling shutter. The sensor does not capture the whole image at once. It reads the frame line by line, top to bottom, like a scanner sweeping down a page. The sweep
Speaker A
is fast, but it is not instant. And during those milliseconds, the world keeps moving. Photograph a golf swing and the club bends like rubber. Film out of a train window and buildings lean drunkenly. Fire a flash at the wrong
Speaker A
moment and only part of the frame catches it. The fix has a name, global shutter, and the concept is trivial.
Speaker A
Capture every pixel at exactly the same instant. Hold each pixel's charge in its own little storage site, then read everything out afterward. The problem is that the storage site has to live somewhere. And in a conventional sensor, it lives right next to the photodiode,
Speaker A
stealing the very real estate that is supposed to be catching light. On a big high resolution sensor, the penalty in sensitivity and noise was so brutal that global shutter stayed locked inside industrial and scientific cameras with big ugly pixels and small resolutions.
Speaker A
Full-frame high-resolution consumer-grade global shutter was the thing veteran sensor engineers would tell you over a drink was simply not going to happen in their careers. In November 2023, Sony announced the A93, the first full-frame camera in history with a global shutter
Speaker A
sensor. And the trick was hiding in plain sight, stacking. If the storage and readout circuitry no longer has to share a floor with the photodiodes because it lives on an entirely separate chip bonded underneath, then the sensitivity penalty that killed every
Speaker A
previous attempt largely evaporates. The result is a camera with no rolling shutter at all in any mode, shooting 24 megapixel raw images at 120 frames per second with continuous autofocus, syncing with flash at any shutter speed up to 1/16,000th of a second. Numbers
Speaker A
like that used to be physically meaningless. Press photographers at the Olympics are now using this thing to capture the exact millisecond a pole vaulter's fingers leave the pole with zero distortion, and the entire camera industry spent the following year coming
Speaker A
to terms with the fact that its roadmap had just been rewritten by a single chip. Sony kept pulling on the same thread. In May 2026, it launched the A7R6, its high-resolution flagship built around a fully stacked sensor that reads out
Speaker A
more than five times faster than its predecessor, which means a 60-plus megapixel camera now behaves with the responsiveness that used to require sacrificing most of your resolution. The second impossible thing is less photogenic and arguably more important
Speaker A
because it lives in your phone. It is called dynamic range. The gap between the darkest and brightest things a sensor can record in one shot. Your eye handles a sunset behind a friend's face effortlessly. A small phone sensor
Speaker A
historically could not because a tiny pixel storage well fills up almost instantly in bright light, like trying to measure a rainstorm with a thimble.
Speaker A
The industry's workaround was computational: take several exposures at different brightness levels and merge them in software. It mostly works until anything moves, and then you get ghosting, tearing, and those weirdly smeared highlights every phone reviewer complains about. The hardware solution
Speaker A
has been sitting in academic papers since the early 2000s under the name LOIC, lateral overflow integration capacitor. Give every pixel a second, much larger bucket, and when the main photodiode fills up, let the overflow spill into the capacitor instead of
Speaker A
being lost. One exposure, no merging, no ghosts. The reason you have never heard of it is that fitting a meaningful capacitor inside a pixel without making the pixel enormous was consi...
Speaker A
own. In 2026, Sony industrialized it. In March, the company announced a 4K security camera sensor with lowfic pixels measuring 1.45 micrometers, the smallest ever built with that structure, aimed at cameras that have to read a license plate and headlight glare at
Speaker A
midnight. Then in June it announced a loic equipped sensor for smartphones delivering 100 dB of dynamic range in a single exposure while cutting power consumption which matters because the other dirty secret of computational photography is that all that frame
Speaker A
merging burns battery and heats the phone. 100 dB means the brightest recordable detail is 100,000 times more intense than the darkest. That is a sensor that can hold a neon sign and the shadowed alley beneath it in the same
Speaker A
frame without software stitching anything. Within a couple of years, night photos and backlit video on ordinary phones are going to quietly stop being a compromise and most people will never know what happened because a capacitor got small enough and Sony was
Speaker A
not done shipping firsts in 2026. The same June, it announced the LY TIA 610, a 64 megapixel phone sensor with a pixel arrangement called RB2x2 on chip lens. The first of its kind ever mass-produced, which finally resolves one of mobile imaging's oldest
Speaker A
trade-offs. Phone sensors do autofocus by comparing light arriving from different angles, which normally means sacrificing either focusing speed or fine detail. Because the pixels doing the comparing are not available for pure resolution. The new structure paired with a remosacene algorithm designed
Speaker A
specifically for it delivers more than 20% higher definition than Sony's previous equivalent sensor while keeping autofocus performance across the whole frame. Same month again, a completely different domain. The IMX 7-Eleven, an X-ray sensor for industrial inspection that detects X-rays directly, converts
Speaker A
them to signals proportional to their energy, and runs at 26,100 frames per second. The fastest in the industry with sensitivity fine enough to work at the level of individual photons.
Speaker A
One company, one month, pushing the state-of-the-art in your pocket, in a security camera, and in a machine that inspects battery welds with radiation.
Speaker A
That breadth is the tell. Sony is not iterating on a product line. It is methodically converting its stacked architecture into a platform that can see anything anywhere at any speed. Now, here is where the story turns because all of that engineering brilliance was
Speaker A
running into a wall. And the wall had a name manufacturing. Remember that a stacked sensor is two chips in one. And only the top layer, the pixels, is classic Sony territory.
Speaker A
The bottom layer is logic, ordinary digital circuitry. And the more intelligence you want to push into a sensor on chip AI, faster readout, loic control, real-time processing for cars and robots, the more that logic layer starts to look like a proper advanced
Speaker A
processor. Advanced logic is made on cutting edge process nodes. And cutting edge process nodes are the near exclusive kingdom of one company. Taiwan semiconductor manufacturing company, the foundry that builds chips for Apple, Nvidia, AMD, and essentially everyone else who matters. Sony's own fabs are
Speaker A
superb at Pixels and mature nodes. They were never going to catch TSMC at leading edge logic, and buying capacity from Taiwan like a regular customer leaves you waiting in line behind the AI boom with everyone else. For years,
Speaker A
analysts assumed this was simply Sony's ceiling. On May 8th, 2026, Sony and TSMC announced a memorandum of understanding that made the industry sit up. And on August 11th, they signed the binding agreement that made it real. The two companies are forming a joint
Speaker A
venture called Advanced Vision Semiconductor Manufacturing Corporation based in Koshi City in Japan's Kumamoto Prefecture. Built inside a brand new fab Sony has just constructed there. The venture will develop and massproduce next generation image sensors for smartphones using advanced process
Speaker A
technology with volume production starting in 2029. The headline investment is 747 billion yen, about $4.7 billion, with Sony contributing roughly 465 billion yen in cash and assets, including that new fab and TSMC putting in about 282 billion. And reports from
Speaker A
NIK suggest the full program could run as high as a trillion yen. The ownership split lands around 60/40 in Sony's favor with Sony as the sole controlling shareholder. The venture consolidated into Sony Group and Sony appointing its leadership while both companies discuss
Speaker A
possible backing from Japan's Ministry of Economy, Trade, and Industry. If you do not follow the semiconductor industry closely, let me explain why people who do were stunned by the shape of this deal. TSMC does not do this. TSMC is the
Speaker A
most sought-after manufacturing partner in human history. And its entire model is built on being the neutral foundry that everyone trusts precisely because it never gets into business with its customers. It builds its own fabs. It controls its own operations. And when it
Speaker A
does plant a flag abroad, as it did with its existing Kumamoto facility, it runs the show. Here, TSMC is taking a minority stake in a venture controlled by someone else, embedding its crown jewel process knowledge inside a Sony
Speaker A
subsidiary on Sony's land under Sony's direction with Sony leading core technology development, product planning, and design. That is not a supply agreement. that is the world's best chip manufacturer deciding that Sony's position and vision is so valuable and so durable that it is worth
Speaker A
bending its own rules to be structurally tied to it. Sony's chief executive Hioki Tutotoki described the partnership as a move to lock in the company's global leadership in image sensors and for once the executive framing undersells it.
Speaker A
This is the sensor king and the foundry king merging their strengths into a single production line. And the stated ambition goes well beyond better phone cameras. Both companies flagged that the partnership will explore what they call physical AI, which is the industry's new
Speaker A
umbrella term for intelligence that operates in the real world. Autonomous cars, humanoid robots, drones, factory automation, surgical assistance. Every one of those systems lives or dies by perception. And perception starts at the sensor. A robot that misreads grabs the
Speaker A
wrong object or the wrong person. A car that misreads at night in rain against low sun is a lawsuit at best. The sensors for that world need exactly the cocktail Sony has spent 2026 demonstrating. Global shutter because motion distortion is unacceptable when
Speaker A
the image feeds a control loop instead of an Instagram feed. extreme single exposure dynamic range because the real world does not wait for bracketing on sensor processing because shipping raw video to a distant computer adds latency that a machine moving at speed cannot
Speaker A
afford. Sony has already shipped pieces of this future. The first automotive sensor with a built-in MIP a fi interface for long cable runs inside vehicles. The smallest precision LAR depth sensor in its class. And in April 2026, its AI division published work in
Speaker A
the journal Nature on an autonomous system running on Sony sensors. The pattern is not subtle. The company is assembling the perception stack for machines end to end. It is not doing it alone on the industrial side either. In
Speaker A
July 2026, Sony and Mitsubishi Electric agreed to form a joint venture called Advanced Vision Solutions, scheduled to begin operating this very month, October 2026, combining Sony's AI vision sensors with Mitsubishi's dominance in factory automation equipment. The pitch is
Speaker A
cameras that do not just record a production line, but understand it. Recognizing defects, guiding robotic arms, monitoring manual work with the intelligence running at the edge, inside or right beside the sensor. Japan has a shrinking workforce in some of the most
Speaker A
automated factories on Earth, which makes it both the perfect laboratory and the most motivated customer. And zoom out one more level and you can see the national story wrapped around all of this. Kumamoto is becoming Japan's silicon heartland with TSMC's Fabs,
Speaker A
Sony's new plant, and now this venture clustering together, backed by a government determined to rebuild domestic chipm after decades of decline.
Speaker A
Japan's semiconductor device market is sitting near $59 billion in 2026 and is projected to pass 73 billion by 2031.
Speaker A
and Vision Silicon is one of the few segments where a Japanese company is not chasing the leaders but is the leader everyone else is chasing. There is a wonderful irony in the timing because the very same autumn, Sony was signing
Speaker A
the deal of the decade, hobbyists were finally defeating one of its oldest chips. In September 2026, after four years of effort, a Canadian reverse engineer going by Disco Star Slayer cracked the Meccaon security chip inside the original PlayStation 2, a part
Speaker A
dating to 1999 that had guarded its secrets for 26 years, surviving every attack the modding community ever threw at it. Getting in required chemically dissolving the chip's packaging to expose the bare dye, photographing its circuitry under microscopes, reconstructing its logic from the
Speaker A
silicon itself, and eventually discovering a software exploit that let the firmware be pulled out cleanly. 26 years, one hobby chip. That is how seriously Sony has always taken silicon.
Speaker A
even when the silicon in question was just there to check whether your game disc was legitimate. And it is a reminder that this company's chip pedigree runs deeper than most people realize. This is after all the company whose engineer Ken Curagi secretly
Speaker A
designed the audio chip inside Nintendo's Super Nintendo before PlayStation existed. and the company that partnered with IBM and Toshiba in the early 2000s to build the cell processor. A genuinely radical multi-core design clocked past 4 GHz with eight synergistic processing units.
Speaker A
Hailed at the time as a supercomput on a chip. Cell was arguably too ambitious, famously difficult to program, and the industry went another way. But the lesson Sony took from it was not to stop building audacious silicon. It was to
Speaker A
pick the battlefield where audacity compounds instead of punishing you. It picked vision. The PlayStation thread has not gone quiet either, and it loops right back into this story. Just today, October 14th, 2026, fresh industry reports describe Sony planning not one
Speaker A
PlayStation 6, but a whole family of hardware. A flagship home console pushing hardware, ray tracing, and machine learning upscaling alongside a genuinely capable handheld designed to run next generation games natively rather than streaming them. All developed in close partnership with AMD
Speaker A
on a power efficient architecture. Set that beside the sensor empire and a shape emerges that did not exist 5 years ago. Sony is becoming a company that commands serious differentiated silicon across the two things that will define the next decade of computing. Machines
Speaker A
that see the world and machines that render one. Very few companies on earth can claim real leverage in either. Sony is building positions in both at once.
Speaker A
Now, honest ledger because a good story deserves its risks. Stated plainly, the TSMC venture does not produce a single chip until 2029, and 3 years is geological time in this industry.
Speaker A
Samsung has been pouring money into its own sensor division for years and would love nothing more than to catch Sony flat-footed during the transition.
Speaker A
Omnivision has been winning real design slots in Chinese flagships. The smartphone market still the cash engine of Sony's sensor business is mature which is exactly why the pivot toward cars and robots matters so much and exactly why it is risky since automotive
Speaker A
qualification cycles are long, brutal and unforgiving. Concentrating next generation production in Kumamoto also concentrates exposure to earthquakes, to geopolitics, to anything that disrupts one prefecture in southern Japan. And joint ventures between giants have a long history of dying from a thousand
Speaker A
committee meetings. Any of these could slow the story down. None of them though changes the underlying physics of the situation which is that the demand for machine vision is about to explode by orders of magnitude and the company with
Speaker A
the deepest pixel technology on earth just bolted itself to the company with the deepest manufacturing technology on earth to meet it. Think about what the demand side actually looks like. A modern premium car already carries a dozen or more cameras, and the count is
Speaker A
rising every model year. A humanoid robot needs stereo vision, depth sensing, and wide dynamic range just to walk across a cluttered room without stepping on the cat. Warehouses are filling with machines that navigate by sight. Surgical robots, agricultural
Speaker A
drones, delivery bots, inspection systems, security networks, smart retail. Every single one of them begins with a chip that converts photons to electrons. And every hard requirement they have, no motion distortion, no blown highlights, no lag, low power, built-in intelligence,
Speaker A
maps one to one onto the exact problems Sony spent 2026 solving in public. Global shutter answers the distortion requirement. Loic answers the dynamic range requirement. Stacked logic answers the intelligence requirement. And the TSMC venture exists to make that logic
Speaker A
as advanced as anything in a flagship phone. The sensors in your future camera phone are almost a side effect. The real product is perception itself sold by the billion units into every machine that needs to understand what it is looking
Speaker A
at. So let me bring this home. 20 years ago, Sony was a consumer electronics company in visible decline, bleeding money on televisions, watching Apple and Samsung eat its lunch and its boldest chip project sell was being remembered mostly as a cautionary tale. The
Speaker A
comeback did not arrive through a hit gadget. It arrived through a decision to go deeper into physics than anyone else was willing to go. flipping sensors upside down when that was considered a manufacturing fantasy. Stacking chips when that was considered a yield
Speaker A
nightmare. Then spending a decade compounding those bets until the impossible things. A full-frame camera with no shutter distortion. A phone pixel with a 100 dB of range in one shot. An X-ray imager counting photons at 26,000 frames a second started
Speaker A
falling one after another in a single year. And then holding all of that, Sony walked into a room with TSMC and came out controlling a venture that puts the world's most advanced manufacturing underneath the world's most advanced pixels aimed squarely at the moment
Speaker A
machines everywhere open their eyes. Nobody thought Sony could build a global shutter full-frame sensor. It shipped one you can buy. Nobody thought loic would fit in a phone. It ships this year. Nobody thought TSMC would ever take the junior seat in someone else's
Speaker A
fab. The contract is signed and the concrete is already poured in Kumamoto. The quiet company that makes the part of your phone you never think about has spent 2026 proving over and over that the word impossible in this industry
Speaker A
mostly just means nobody has stacked the right two chips together yet. The first sensors from the new venture arrive in 2029. And when they do, they will not just be looking back at you from a phone. They will be looking out at the
Speaker A
world from everything. If this breakdown gave you a new way to see the chip war, you know what to do. and I will see you in the next
Topics:Sonyimage sensorglobal shutterstacked CMOSback-illuminated sensormachine visioncamera technologydynamic rangeTSMCsemiconductor innovation
