SK Hynix was developing a new CMOS image sensor (CIS) that uses neural network technology, TheElec has learned.
The South Korean memory giant is planning to embed an AI accelerator into the CIS, sources said.
The accelerator itself is based on SRAM combined with a microprocessor __ also called in-memory computing.
The AI-powered CIS will be able to recognize information related to the subject of the image, while the image was being saved as data.
For example, the CIS will be able to recognize the owner of the smartphone when it is used on a front camera.
Most current devices have the CIS and the face-recognizing feature separate.
Having the CIS do it on its own can save time and conserve the power of the device.
SK Hynix has recently verified the design and field programmable gate array of the CIS.
The company is also planning to develop an AI accelerator that uses non-volatile memory instead of the volatile SRAM.
SK Hynix is a very small player in the CIS field. According to Strategy Analytics, Sony controlled 44% of the market during the first half of the year followed by Samsung’s 30%. Omivision had a 9% market share. The remaining three companies, which include SK Hynix, controlled 17% together.
SK Hynix is currently supplying its high-resolution CIS to Samsung; last year it supplied a 13MP CIS for the Galaxy Z Fold 3. It is supplying 50MP CIS for the Galaxy A series this year.
However, CIS companies are focusing on strengthening other features of the CIS besides resolution.
They are reaching the limits of making the pixels smaller. Pixels absorb less light and the signals are smaller when they become too small, obscuring the resolution of the images.