What a Partially Stacked Sensor Changes

Picture a heron lifting off the far bank of a river, with the camera on the 180-600mm and the shutter set to electronic at 20 frames a second. In every one of those frames, the top row of the picture is recorded about 14.5 thousandths of a second before the bottom row. That gap is the readout time. It decides how far the electronic shutter can be trusted with anything that moves, and it is the number Nikon set out to shrink with what it calls a partially stacked sensor.

Nikon says the Z6III reads its sensor about 3.5 times faster than the Z6II did. Most of the camera's headline features trace back to that change: a usable electronic shutter for moving subjects, 20 frames a second, a viewfinder that stays live through a burst, and video 6,048 pixels wide at 60 frames a second.

01.A Sensor Is a Grid of Light Counters

The image sensor is a rectangle of silicon 35.9 mm wide and 23.9 mm tall (p. 997), covered in a grid of tiny light-sensitive sites called photosites, one for each pixel. The largest image is 6,048 pixels across and 4,032 down (p. 998), so $6048 \times 4032 = 24{,}385{,}536$ photosites, about 24.4 million, end up in the file.

Light striking a photosite frees electrons, which the photosite holds as charge. More light, more charge. When the exposure ends, each charge must be measured and turned into a number by an analogue-to-digital converter (ADC), a circuit that takes a voltage in and puts an integer out. In a 14-bit NEF, Nikon's RAW stills format (p. 998), each reading becomes one of $2^{14} = 16{,}384$ values. The chip is a CMOS sensor (p. 997), short for complementary metal-oxide-semiconductor, the process used for computer chips. Each photosite has transistors of its own, so the camera can make one row at a time report.

02.Why the Sensor Is Read One Row at a Time

A converter is far larger than a photosite, so there cannot be one per photosite. In the usual design the converters sit along the edge of the grid, one or a few per column, and the sensor is read in a sweep: the first row is connected and all its columns are measured at once, then the second row, and so on to the bottom. The time from reading the first row to reading the last is the readout time, written $t_r$.

Nikon does not publish the Z6III's readout time, and the RG never mentions one. Two independent measurements agree closely: Thom Hogan measured 14.4 ms for full-resolution stills with the electronic shutter, and DPReview's review measured about 14.6 ms with 14-bit readout. A millisecond (ms) is a thousandth of a second. Splitting the difference gives about 14.5 ms, roughly 1/70 s, since $1/0.0145 \approx 69$. Spread over 4,032 rows, that is $14.5\ \text{ms} \div 4032 \approx 3.6$ microseconds per row, a microsecond being a millionth of a second.

Nikon's product page puts the readout at about 3.5 times faster than the Z6II's, naming no mode and giving no absolute figure for either camera. The ratio implies $14.5 \times 3.5 \approx 51$ ms, about 1/20 s, for the Z6II, and Hogan's readout table measures the Z6II at 50.8 ms. I use 51 ms below as the slow case.

What the Electronic Shutter Does With the Sweep

With no curtain in the way, the camera starts each row's exposure by emptying it, a step called a reset, and ends it by reading it. The reset sweeps down the frame on the same schedule as the read, one shutter time ahead, so every row gets exactly the exposure you set. Each row's exposure simply happens a little later than the one above, and the bottom row's happens $t_r$ after the top row's.

Shutter speed and readout time are therefore separate numbers. At 1/16000 s, the fastest electronic speed (pp. 140, 1000), each row gathers light for $1/16000\ \text{s} = 0.0625\ \text{ms}$, yet the sweep still takes 14.5 ms, which is $14.5 \div 0.0625 = 232$ times longer. A fast shutter speed freezes motion within each row. It does nothing about the delay between rows.

A Worked Example: An Upright Bar Crossing the Frame

Take anything upright that crosses the frame's full width in a quarter of a second, such as a dancer running along the stage front at 180 mm. The speed is illustrative.

The frame is 6,048 pixels wide, so the subject moves at $6048 \div 0.25 = 24{,}192$ pixels per second. During the Z6III's 14.5 ms sweep it travels $24{,}192 \times 0.0145 \approx 351$ pixels, so the bottom of the bar is recorded 351 pixels further along than the top and the bar comes out leaning. Over 4,032 rows that tilts a vertical by about 5 degrees, since $\arctan(351/4032) \approx 5^\circ$, where $\arctan$ turns a slope into an angle. At the 51 ms derived for the Z6II, the shift is $24{,}192 \times 0.051 \approx 1{,}234$ pixels, a fifth of the width, and the tilt about 17 degrees.

In general, if $v$ is the subject's speed across the image in pixels per second, the shift $\Delta x$ in pixels between the top and bottom of the subject is

$$\Delta x = v \, t_r$$

If the subject takes time $T$ to cross a frame $W$ pixels wide, then $v = W/T$, and the shift as a fraction of the width is

$$\frac{\Delta x}{W} = \frac{t_r}{T}$$

The quarter-second crossing gives $0.0145 \div 0.25 = 5.8\%$ on the Z6III and $0.051 \div 0.25 = 20.4\%$ at the Z6II's derived readout. A subject taking a full second shifts by 1.45%, about 88 pixels, which few viewers will spot.

The Same Sweep Explains Flicker Bands

Many LED (light-emitting diode) and fluorescent lamps flicker at twice the mains frequency, 120 times a second on 60 Hz mains, where Hz (hertz) means cycles per second. If $f$ is the flicker frequency, one readout spans $n = t_r \, f$ cycles. At 120 Hz, 14.5 ms spans $0.0145 \times 120 \approx 1.7$ cycles, so a short electronic exposure can carry one or two broad bands, where 51 ms would span about 6.1 and show about six. Faster readout makes bands fewer and wider. It does not remove them. The RG's Silent mode cautions list flicker and banding (p. 846), and photo flicker reduction does not work with the electronic shutter, in silent mode or in the high-speed frame capture modes (p. 484). Shutter Types, Rolling Shutter and Flicker covers the remedies.

03.What Stacking Means

In a conventional sensor, the photosites and the circuits that read them share one slab of silicon. The grid takes the middle and the converters and timing circuits are packed into the margins, so margin space limits how many converters work at once, and with it how quickly the rows can be swept.

A stacked sensor builds the chip in layers. The top layer holds the photosites, and a layer bonded behind it holds the readout circuitry, joined to the layer above by a great many tiny vertical connections. Behind the image area there is as much room as the image area itself, so many more converters fit and work side by side. Nikon's Z8 and Z9 use fully stacked sensors, as What the Z6III Is explains.

04.What Nikon Means by Partially Stacked

Nikon's own description is brief. Its launch press release of 17 June 2024 calls the sensor partially stacked and says only that high-speed processing circuits are stacked above and below the image sensor, and the Z6III product page credits it with the 3.5-fold faster readout. Nikon publishes no cross-section, no layer count and no list of which circuits moved, and the RG says only that the sensor is CMOS (p. 997), never using the word stacked.

DPReview's review describes an additional layer of silicon along the chip's edges carrying more capable readout circuitry. Hogan's review takes "partially" to mean the sensor is not bonded to a second full chip, and suggests Nikon built a fast parallel transfer path along the top and bottom of the sensor chip itself. Both are outside inferences. My reading, also inference, is that the two describe readout work moved to the edges of a conventional image area, which fits a readout between the Z6II's and the fully stacked Z8 and Z9.

05.What the Faster Readout Changes

The Electronic Shutter Becomes a Tool for Motion

Custom Setting d6 Shutter type chooses how exposures start and end (p. 668). The Mechanical shutter uses two physical curtains, the Electronic front-curtain shutter starts electronically and ends with the rear curtain, the Electronic shutter does both with the sweep described above, and Auto switches between the first two by shutter speed. Their top speeds are 1/8000 s, 1/2000 s and 1/16000 s respectively (pp. 668, 1000). The curtains also sweep the frame, but far faster than the sensor reads, and Hogan's review shows a fan's blades bending with the electronic shutter but not the mechanical one.

With a readout near 51 ms, an electronic shutter suits still subjects. At 14.5 ms it handles most human and animal movement, and fails on subjects crossing the frame in a fraction of a second, fast pans, and wingtips or blades moving far faster than the body carrying them. The RG's Silent mode cautions list the same effects (p. 846): distorted moving subjects, a distorted frame when the camera moves, and flicker or bands under some lights.

Flash Works, Up to 1/60 s

A flash burst lasts about a millisecond at most, 1/980 s at full output for Nikon's SB-5000 by Nikon USA's specification, and its light reaches every row only if every row is gathering light at that instant. The top row starts first and the bottom row $t_r$ later, so they overlap only if each row's exposure lasts at least as long as the sweep. Writing $t_s$ for the shutter time, the condition is

$$t_s \ge t_r$$

With the electronic shutter, the fastest speed usable with flash is 1/60 s regardless of the sync setting, and auto FP high-speed sync, which lets flash work above the normal sync speed, is unavailable (pp. 417, 691, 914, 1000). The mechanical shutter and Auto, the default (p. 624), sync at 1/200 s (p. 1000). A 1/60 s exposure lasts $1000 \div 60 \approx 16.7$ ms, clearing the 14.5 ms sweep by about 2.2 ms, room for a full-power burst, while the next faster standard speed, 1/80 s, lasts 12.5 ms and would not. Nikon does not explain the limit. DPReview treats the 1/60 s electronic sync as a measure of readout speed, and these numbers agree.

Faster Bursts

A burst rate is set by the slowest step: moving the curtains, reading the sensor, or processing and storing frames. In the table, fps means frames per second, every rate is approximate and measured by Nikon in-house (p. 1000), RAW means NEF, and JPEG (Joint Photographic Experts Group) and HEIF (High Efficiency Image File Format) are the processed formats.

The mechanical best, 14 fps in Continuous H (extended), equals the Z6II's top rate in that mode, which its published specifications cut to 10 fps for 14-bit NEF. The Z6III keeps 14 fps with NEF, always 14-bit on this camera (pp. 154, 998), which suggests the curtains, not the sensor, limit that column. The electronic column is where readout shows. At 20 fps each frame has $1000 \div 20 = 50$ ms, of which a 14.5 ms readout uses under a third. A 51 ms readout would not fit at all. The fastest rates carry conditions, such as possibly uneven exposure across a Continuous H (extended) burst (p. 155), covered in Release Modes, Frame Rates and the Buffer.

The high-speed frame capture + modes, which this guide shortens to HSFC+, go further: C15, C30, C60 and C120 shoot at about 15, 30, 60 and 120 fps (p. 1000), JPEG only at the Large size, with C120 limited to the DX crop (p. 157). FX is Nikon's name for the sensor's full area and DX for a smaller crop from its centre. Hogan's readout page says Pre-Release Capture, which these modes carry, uses a video readout rather than the stills one. At C120 each frame has $1000 \div 120 \approx 8.3$ ms, and even the DX crop's 2,656 rows (p. 998) at the stills rate would take $2656 \times 3.6 \approx 9{,}560$ microseconds, about 9.6 ms, so C120 must read rows faster than stills mode does, skip some, or both. No readout time for these modes has two sources, so I give none. The slowest shutter speed is 1/60 s, or 1/125 s in C120 (p. 157), and Nikon gives no reason. The C120 limit fits its 8.3 ms frame, but C15 and C30, with about 67 and 33 ms per frame, stop at 1/60 s too.

Pre-Release Capture builds on these modes, keeping frames from just before the shutter-release button is fully pressed, and is set with Custom Setting d3 Pre-Release Capture options (p. 156). Nikon's press release credits the new sensor's readout, together with the EXPEED 7 processor, for keeping up to a second of frames at up to 120 fps. Pre-Release Capture explains its use.

The Viewfinder Stays Live Through a Burst

An electronic viewfinder (EVF) is a small screen behind the eyepiece showing the sensor's live picture, about 5,760k dots on the Z6III (p. 999). When the mechanical shutter fires, the curtains cover the sensor and break the feed. DPReview found that the viewfinder then goes dark for an instant at each frame in most modes, which is called blackout. With the electronic shutter nothing covers the sensor, and a fast enough readout keeps the picture flowing.

The RG states the result under Custom Setting d15 Release timing indicator (p. 679): in HSFC+ and Continuous H (extended), the display stays live at each release instead of going dark as in other release modes, and in Continuous H (extended) the d15 settings apply only when d6 is set to Electronic shutter. DPReview found no blackout in electronic bursts, and Hogan describes a live view at 20 fps. Nikon's press release says the viewfinder skips fewer frames at 20 fps, and the specifications mention a high frame-rate display (p. 999), switched by Custom Setting d22 High fps viewfinder display, default OFF (pp. 625, 689), but the RG names no refresh rate, so I give none.

Without blackout, d15 offers other release cues through Indicator type: Type A darkens the display at each release, which Nikon says helps when panning, Type B frames the picture with a border on all four edges, Type C uses side borders only, and Off shows nothing (p. 679). The defaults are Type B and a Type A auto restore delay of 1/6 s (p. 625). Custom Setting d14 Display on during burst decides whether the display shows the scene during a burst at all, default ON (pp. 625, 678).

Video Modes the Readout Makes Possible

Video is a stream of frames, each read in full before the next begins. If $F$ is the frame rate in frames per second, each frame has $1/F$ seconds, so the readout must satisfy

$$t_r \le \frac{1}{F}$$

At 60p, which the RG defines as 59.94 frames per second (p. 201), each frame has $1000 \div 59.94 \approx 16.7$ ms.

Video readout is faster than stills readout. DPReview measured about 9.5 ms in every full-frame video mode it tried and 6.3 ms in the DX crop, and Hogan reports 1/163 to 1/107 s across video settings, about 6.1 to 9.3 ms. Part of the gain is fewer rows, 3,402 for the largest video frame (p. 200) against 4,032 for stills, but that alone gives $14.5 \times 3402 \div 4032 \approx 12.2$ ms, so each row must also be read faster. One plausible reason, which Nikon does not confirm, is that RAW video is 12-bit (pp. 197, 1004) where NEF stills are 14-bit, and a converter resolving fewer levels can finish sooner.

Sizes are named loosely by width, so 6K means frames about 6,000 pixels wide. The largest RAW mode, $6048 \times 3402$ at 60p, gives 16.7 ms per frame for a 9.5 ms readout. It records only in N-RAW, Nikon's 12-bit RAW video format (p. 200). The 5.4K mode, $5376 \times 3024$ at 60p, records only in H.265, a compressed video standard, and is fixed to the FX area (pp. 201, 210). The 4K mode at 120p allows about 8.3 ms per frame, which the 9.5 ms full-frame readout cannot meet, and the RG fixes it to the DX area, where 6.3 ms fits. Yet 4K at 100p is DX-only too (p. 210), though its 10 ms frame would hold 9.5 ms, so readout may not be the whole reason, and Nikon gives none. Hogan's review calls 4K at 120p uncropped, against p. 210, so I weigh his video figures below DPReview's. Full HD, $1920 \times 1080$, at 240p allows about 4.2 ms per frame yet keeps about 95% of the FX view (p. 210). DPReview's 9.5 ms over 3,402 rows is $9500 \div 3402 \approx 2.8$ microseconds a row, so 4.2 ms, or 4,167 microseconds, covers at most $4167 \div 2.8 \approx 1{,}490$ rows. The camera almost certainly reads fewer rows, by a method Nikon does not describe. Hogan calls the mode sub-sampled. In the slow-motion mode that records at 30p, the sensor is read at 120p, so about ten seconds of action plays over about forty (p. 203).

The Z6II's published specifications stop at 4K at 60p and Full HD at 120p in H.264, the older standard, so the larger sizes, faster rates, internal RAW and H.265 above are new (pp. 197, 200-201, 1004). The video part of this guide covers each mode.

What the Speed Costs

DPReview's review finds slightly more read noise, as with other fast-readout sensors, so deep shadows in RAW files become unusable a little sooner, and calls it a direct trade-off for speed. Hogan links it to running the converters faster and finds Z6II files hold up better when brightened by five stops or more. Read noise is the small random error the reading circuitry adds to every measurement whatever the light, and it matters most in the darkest parts of an image at a low ISO, the camera's sensitivity setting.

06.Recommended Settings

Wildlife and birds. With the 180-600mm, I use the electronic shutter in Continuous H (extended) for perched birds and large birds with slow wingbeats, because 20 fps and a live viewfinder matter more than a lean of a few percent. I switch to the mechanical shutter for small birds with fast wings and fast pans at 600 mm. The test from the worked example: if the subject or the background crosses the frame in well under a second, expect a visible lean.

Stage and outdoor dance. With the Tamron 70-180mm on stage, silence decides it, once the setup menu's Camera sounds > Shutter sound is OFF, because its default ON plays a release sound with the electronic shutter (pp. 806, 844), as Silent Shooting, Timers and Time-Lapse explains. A dancer taking a second or more to cross leans by a percent or two. I check the first frames for bands under LED stage lighting and change to the mechanical shutter if they appear and the venue allows the sound.

Portraits and flash. With the 85mm and a flash, I use the mechanical shutter. From 1/60 s to 1/200 s is about 1.7 stops, since $\log_2(200/60) \approx 1.74$, where $\log_2$ counts doublings, and that is ambient light I can cut at the same aperture.

Street, landscape and video. In the street, the silent electronic shutter, with Shutter sound OFF, is worth having. For landscapes where I plan to lift deep shadows, the read-noise trade-off makes me expose a little brighter. For gimbal dance video, a 9.5 ms readout means a dancer crossing in one second shifts by $0.0095 \div 1 = 0.95\%$ of the width, which I accept.

07.Beginner Mistakes

Expecting 1/16000 s to cure rolling shutter. A fast shutter speed shortens each row's exposure, not the 14.5 ms sweep. Judge the lean by readout time against crossing time, and use the mechanical shutter when the ratio is large.

Using the electronic shutter with flash. It cannot sync faster than 1/60 s, and auto FP is unavailable (pp. 691, 1000). Set d6 to Auto, the default (p. 624), or Mechanical shutter, which sync at 1/200 s (p. 1000), for flash work that needs faster speeds.

Leaving Silent mode on at a flash shoot. Silent mode forces the electronic shutter and the flash will not fire (p. 846). Turn it off first.

Panning at 600 mm with the electronic shutter and blaming the lens. When the whole frame tilts in a pan, the sweep is the cause, and the RG's Silent mode cautions list the effect (p. 846). Use the mechanical shutter for fast pans, or pan more slowly.

Treating the Z6III as a fully stacked camera. Its readout still bends fast subjects, and DPReview ties the 1/60 s electronic flash limit to it. Use the electronic shutter where the numbers favour it.

08.Related Chapters

• What the Z6III Is: where the camera sits in Nikon's range.
• Shutter Types, Rolling Shutter and Flicker: the d6 options and flicker remedies.
• Release Modes, Frame Rates and the Buffer: burst modes and their display settings.
• Pre-Release Capture: keeping frames from before the full press.
• Silent Shooting, Timers and Time-Lapse: silent mode and its electronic shutter.
• Image Area and Crop: the FX and DX areas.
• Glossary: short definitions of readout and related terms.