In a recent paper titled "A 5.94-μm Pixel-Pitch 25.2-Mpixel 120-Frames/s Full-Frame Global Shutter CMOS Image Sensor With Pixel-Parallel 14-bit ADC", Sakakibara et al. from Sony Semiconductor Solutions (Japan) write:
We present a 25.2-Mpixel, 120-frames/s full-frame global shutter CMOS image sensor (CIS) featuring pixel-parallel analog-to-digital converters (ADCs). The sensor addresses the limitations of conventional rolling shutters (RSs)—including motion distortion, flicker artifacts, and flash banding—while maintaining image quality suitable for professional and advanced amateur photography. A stacked architecture with 3- μ m-pitch Cu–Cu hybrid bonding enables more than 50 million direct connections between the pixel array and the ADC circuits. The pixel-parallel single-slope ADCs operate with a comparator current of 25 nA and use a positive-feedback (PFB) scheme with noise-bandwidth control using an additional 11.4-fF capacitor, achieving 2.66 e−rms ( 166.8 μVrms ) random noise (RN) at 0-dB gain with an REF slope of 2161 V/s. The 5.94- μ m pixel pitch accommodates 30-bit latches designed under SRAM rules in a 40-nm CMOS process. Noise analysis reveals that in subthreshold operation, the dominant noise contributors are the comparator current, REF slope, and second-stage load capacitance. The sensor delivers 14-bit resolution, a 75.5-dB dynamic range (DR), and 120-frames/s operation at a power consumption of 1545 mW. A figure of merit of 0.083 e−rms⋅ pJ/step is comparable to state-of-the-art RS sensors. These results demonstrate that pixel-parallel ADC technology can be scaled to tens of megapixels while preserving high image quality and energy efficiency, enabling motion-artifact-free imaging in battery-powered consumer cameras.
Full paper link [behind paywall]: https://ieeexplore.ieee.org/document/11219086
great data of sensor
ReplyDeleteAny plans to industrialise the sensor?
ReplyDeleteIt should be already in the Alpha 9 III camera released in 2023, IMX810, I think.
DeleteThis group photo of IISW2025 was taken by the first author Sakakibara-san, using his Alpha 9. You can even see him pressing the remote shutter!
Deletehttps://imagesensors.org/wp-content/uploads/2025/11/DSC00762_s-1.jpg
This paper is from 2025, A9iii is from 2023 - while many metrics seem similar, for example read noise at 0dB gain seems significantly lower than that of A9iii. So I'd guess they're not the same.
DeleteI am wandering if FD leakage and parasitic light sensitivity are a problem in some applications and how/if they are solved in this work. Since signal has to be stored for a full frame time on FD, how is parasitic light on FD affecting a pixel which converts early (dark) and one which converts late ( bright)?
ReplyDeleteThis is a digital pixel sensor in which the pixel information is immediately converted and stored in the digital domain. In this way FD leakage and parasitic light sensitivity are no issues.
ReplyDeleteValid point! I would like to challenge the "immediately" here though: the conversion time for digital pixel sensor (pixel ADCs) is in the order of milliseconds (~1ms in this paper), as opposed to micro seconds of conventional column readout. Signal and reset stay on FD all that time until the ADC has finished the conversion. Hence my question above about potential issues of this long storage on FD. Also, since CDS time is so long, how does this affect 1/f and RTS noise of the source follower?
DeleteIt's different. when you read an analog GS sensor, you have two exposures: one is GS and the other is RS. so you can have a lot of artefacts due their mixture. But in in-pixel ADC, these 2 rxposures are both GS, so the FD leakage adds a small extra-exposure time but will not generate artefacts.
DeleteYang Ni