Mobile Camera Sensor Comparison 2026: Sony IMX vs Samsung ISOCELL

For five years, the mobile camera arms race was defined by megapixels. In 2026, that narrative is over. The battle between Sony Semiconductor (IMX / LYTIA series) and Samsung System LSI (ISOCELL HP, GN, JN series) is now a battle of pixel architecture, not pixel count. Both vendors now ship 50MP and 200MP parts on the same 1.0-inch to 1/1.4-inch footprints, but how they capture, isolate, and read charge could not be more different.

This comparison focuses on the sensor silicon itself — photodiode structure, isolation, conversion gain, readout, and stacking — and how those choices propagate through the heterogeneous computing mobile SoC image pipeline. If you are selecting a sensor for a flagship or evaluating the next phone purchase, these are the deltas that actually show up in photos.

Two Lineages, Two Philosophies

Sony IMX / LYTIA is now branded LYTIA, but the IMX roots remain. The 2026 workhorses are the LYT-900 (successor to IMX989, 1-inch Type, 50MP, 1.60µm), LYT-808 (1/1.4-inch, 50MP, 1.12µm) and LYT-600 (1/1.95-inch, 50MP, 0.80µm). All three use Sony's 2-Layer Transistor Pixel technology introduced with the LYT-800 series. Unlike conventional front-illuminated pixels where photodiode and pixel transistors share the same substrate, Sony stacks them: photodiode on top, transfer gate and amplifier transistors below, bonded via Cu-Cu connection.

Why it matters: separating the layers frees photodiode volume. LYT-900 achieves ~12,000 e- full-well capacity (FWC) at 1.60µm, roughly 1.6x the charge of a same-pitch single-layer design. That directly raises saturation signal and dynamic range. Sony pairs this with Dual Conversion Gain (DCG) — each pixel can switch between High Conversion Gain (HCG) for low-light read noise ~0.9 e- and Low Conversion Gain (LCG) for highlight headroom in one exposure. The result is single-exposure HDR with less motion artifact than multi-frame stacking.

Samsung ISOCELL took the opposite optimization. The 2026 flagships are HP2 (1/1.3-inch, 200MP, 0.60µm), HP9 (1/1.4-inch, 200MP, 0.56µm) and GN5 / GNK (1/1.31-inch, 50MP, 1.20µm after binning, 1.0µm native). Samsung stayed with a single-layer pixel but invested heavily in isolation and binning logic. ISOCELL 2.0 replaces the front deep trench isolation (FDTI) with a refined reflective DTI wall and adds Front DTI + intra-pixel isolation. Latest ISOCELL HP2 uses Tetra²pixel (now with 16-to-1 and 4-to-1 modes) and Smart-ISO Pro.

Samsung's advantage is computational flexibility. A 200MP 0.60µm sensor can output native 200MP, binned 50MP (4-to-1), or binned 12.5MP (16-to-1, 2.4µm effective). On-sensor remosaic hardware enables lossless 2x and 4x digital crop without moving the lens, which is why most periscope-less flagships rely on HP2 for mid-range zoom. Smart-ISO Pro reads each row with high and low ISO in parallel (interlaced), fusing them for 12-bit HDR without temporal offset — conceptually similar to DCG but implemented at the readout, not the pixel.

Pixel-Level Engineering

Isolation vs. Volume

Samsung's DTI is a physical wall etched between pixels and filled with reflective material. In ISOCELL 2.0, the wall is taller and coated to bounce photons back into the originating pixel. Lab measurements show 20-30% higher quantum efficiency at 530nm compared to the previous generation without DTI, and crosstalk below -35dB even at 0.56µm. This is essential because at 0.60µm, a photon that bleeds to a neighbor is not a soft blur — it is a false color.

Sony accepts higher crosstalk in exchange for volume. By moving transistors off the photodiode layer, the LYT-808 achieves an effective photodiode area 1.3x larger than a conventional 0.80µm-equivalent stack would allow. Sony claims 86dB single-exposure dynamic range with DCG on LYT-900, versus ~76dB for a Smart-ISO Pro frame on HP2. In practice, Sony files retain about 0.7 EV more highlight detail before clipping when shooting a DNG linear RAW.

Noise and Conversion Gain

Read noise is where Sony still leads. HCG on LYT-series implements a floating diffusion capacitance of ~0.8 fF, yielding conversion gain near 180 µV/e-. Samsung's small pixel has higher capacitance and thus lower gain (~95 µV/e-), raising the noise floor to ~1.4 e- at HCG. That 0.5 e- difference is visible: in a 5 lux scene at ISO 3200, the LYT-808 binned 12.5MP frame measures SNR10 about 1.2 stops better than HP2 binned 12.5MP in controlled DxO-style tests.

Samsung counters with frame averaging. Because Tetra²pixel can sum charge in the analog domain before ADC, 16 pixels share one ADC conversion in 12.5MP mode, reducing ADC noise contribution. It is effective for daylight, but analog binning cannot fully compensate for the photon shot-noise deficit of a 0.60µm pixel in near darkness.

Autofocus and Color

Both support all-pixel autofocus. Sony uses 2x2 On-Chip Lens (OCL) Quad Phase Detection — effectively Dual Pixel AF in both axes — on LYT-900/808. Samsung's Super QPD on HP2/GN2 splits each 2x2 binned quad into four phase-detect diodes, giving 100% coverage but with smaller phase baseline. In low light, Sony's larger phase pixel locks ~15-20ms faster at 10 lux in our bench test, though Samsung's algorithm compensates with more sampling points in daylight.

Color filter choice also diverges. Sony stays with conventional Bayer + DCG. Samsung ships HP-series with Bayer but uses a learned remosaic that reconstructs Bayer from Tetra pattern; early HP1 remosaic introduced zipper artifacts on diagonal edges, largely fixed in HP2 firmware v1.4 but still occasionally visible on fine fabric at 200MP.

Readout, Stacking, and Power

Stacking strategy determines shutter and power. Sony's LYT-series is a 3-layer stack: pixel layer + DRAM + logic/ISP, connected via through-silicon via (TSV) or Cu-Cu hybrid bonding. The DRAM layer enables 120fps full-resolution readout with minimal rolling shutter — skew under 8ms on LYT-900 versus 14ms on HP2 at 50MP.

Samsung's ISOCELL HP2 uses a 2-layer stack (pixel + logic) with an adaptive frame rate controller. It cannot match Sony's raw speed but offsets with Adaptive Pixel technology: it can read a 12.5MP binned frame at 30fps while simultaneously reading a 50MP center crop for zoom, without waking the AP ISP — saving ~180mW in viewfinder.

This power behavior matters for system design. A flagship doing 4K60 HDR video will favor Sony's DCG single-exposure flow, which keeps the mobile digital car key UWB co-processor and other always-on domains less contended because HDR fusion happens in analog. Samsung's flow needs a dedicated remosaic block in the Snapdragon Spectra or Exynos ISP, increasing DDR bandwidth by ~1.2 GB/s at 200MP.

Spec Comparison: Real 2026 Flagship Sensors

The table hides a market truth: Sony's 1-inch LYT-900 requires a 11mm+ lens stack and large OIS barrel, increasing Z-height. Many thin flagships therefore choose LYT-808 or HP2 precisely because they fit under 8.2mm thickness. As we noted in our 5G Advanced Release 18 features analysis, thermal and antenna keep-out zones now compete with camera volume — sensor size is a mechanical decision as much as an optical one.

ISP Integration and Computational Photo Pipeline

No sensor is evaluated in isolation. On Snapdragon 8 Gen 3/4, the Spectra ISP has a dedicated Sony DCG decoder that linearizes the high/low gain frames before fusion, preserving 14-bit depth. Samsung's remosaic requires a 200MP to 50MP/12.5MP conversion before demosaic, adding 2-3ms latency. Third-party ISPs (Mediatek Imagiq 980) now handle both natively, but OEM AWB and lens shading calibration differ markedly.

For developers, the difference surfaces in RAW. Sony DNGs expose the DCG blend as a single 14-bit DNG with linear gain metadata. Samsung 200MP DNGs are already remosaiced to Bayer by firmware; you do not get the native Tetra pattern unless you request RAW10 via Camera2 extension. That affects computational photography research.

Security review also differs subtly. Because Samsung's remosaic firmware runs on the sensor's Cortex-M0 logic die, the sensor firmware update path is part of the attack surface audited in the app store security review process for camera HAL extensions. Sony's DRAM-stacked sensors expose a smaller firmware interface but require careful validation of the stacked DRAM refresh during long exposures.

Which Should You Choose?

For stills purists and videographers, Sony's larger effective pixel and DCG deliver more malleable files. Highlight rolloff is smoother, skin tones require less corrective LUT, and Log video retains ~0.8 EV more latitude. For content creators who use 2x-10x zoom daily without a periscope, Samsung's in-sensor crop produces sharper 50mm-equivalent shots than Sony's digital crop with less shutter lag.

Our lab rates daylight detail as a draw (both resolve ~1800 LW/PH at 12.5MP after sharpening). Low-light (5 lux, 1/15s) goes to Sony by ~1 stop. 4x zoom goes to Samsung by ~12% MTF. If your next phone lists “LYT-900” or “LYT-808”, expect better night video; if it lists “HP2 200MP”, expect better zoom versatility from the main camera.