Friday, February 24, 2012

Sony Announces 18MP/24fps BSI Sensor, 1080p/30 CCD, 4K Super-Resolution Video Processor

Sony Cx-News vol. 67 has few interesting announcements:

The IMX118CQT is 1/2.3-inch 18.47MP 24fps-fast sensor for consumer DSCs. The sensor is based on 1.26um BSI pixels, made in 90nm 1P5M process. Sony says that by improving "pixel configuration" (layout? PD sharing?) it "managed to increase the electron count ratio of saturation signal level per unit area by 35%", although the exact number is not specified.

Compared to the 12.4MP IMX078CQK announced a year ago, the new sensor is said to achieve an 80% speed increase of the internal circuits. The column-parallel ADC was improved to reduce power consumption by 35% over the previous generation product (in all-pixel scan mode at 12 bits). In live view when 0.67M effective pixels are output at about 30fps the sensor's power consumption is 136 mW, which is said to place it among Sony's most energy-efficient sensors.

Another nice Sony achievement is 2.83MP CCD capable of 1080p/30 mode, the ICX687ALA (B&W) / ICX687AQA (color). Th1 1/1.8-inch CCD is based on 3.69um pixels clocked at 54MHz. To achieve 30fps in 1090p mode the CCD has two parallel outputs. In full 2.83MP resolution mode the CCD can output 25fps. The pixel fill factor is said to be improved to maintain the similar PD size as the previous generation 4.4um pixel:

Yet another interesting Sony announcement is "database-type Super-resolution" Signal Processing that generates realistic 4K (4,096 × 2,160 pixels) video signals from full HD (1,920 × 1,080 pixels) resolution video.

"Database-type super-resolution" refers to real-time video analysis and pattern classification by referencing a database to enable perfect super-resolution processing of any video...

Newly-developed CXD4736GB ...has evolved the pattern classification procedure of this unique "database-type super-resolution" technology by adding a "learning function" that improves the classification function and dramatically enhances the picture quality of the generated video.

The "learning function" groups patterns produced by multidimensional features more efficiently in classifying the numerous characteristics that make up an image into a number of patterns and makes possible pattern classification according to the dynamic changes caused by input signal characteristics. This makes it possible to appropriately classify input video (or image) depending on its characteristics to enable optimum super-resolution processing.


  1. Holy cow. I never thought that a DSC pipeline can be made to this complicated. Pattern pattern recognition, database and learning function? Plus all those other functions that is glimpsed over... i m shocked.

  2. It's for upscaling projectors. Probably too large/power-hungry for a camera right now.


All comments are moderated to avoid spam and personal attacks.