A Proposal of PUF Utilizing Pixel Variations in the CMOS Image Sensor
2017 May30 – Jun2
2017 International Image Sensor Workshop (IISW)
In order to make the IoT (Internet of Things) a success, the information security will have to be assured and the privacy of the collected data protected . While the security of the communication channel is well developed, the identity theft can be a security hole unless security functions are integrated in the sensor device . In order to define whether images and video sequences were recorded using a specific camera, a camera identification technique which utilizes photo response non-uniformity (PRNU)  was proposed. While the technique is effective as long as the illumination level on the focal plane is uniform, the identification pattern generated with PRNU varies depending on various imaging conditions since it depends on the input optical power. A chip ID generation utilizing random telegraph signal (RTS)  is also expected to be applied to the CMOS image sensors. The repeatability of the chip ID with RTS is good because of the ID is independent of the illumination. However, each pixel has to be sampled many times to define the ”hot” pixel, during which the normal image capture is disturbed. For the image information security, we propose a CMOS image sensor with a physically unclonable function (CIS-PUF), which utilizes the pixel-to-pixel fixed pattern noise (PPFPN) as a fingerprint of each device. Though many silicon PUFs, which utilize the device matching of Latch , SRAM , ReRAM , and so on, have been presented, the PPFPN is expected as the most compatible with the CMOS image sensor because (1) a conventional circuit is used to read out the enhanced PPFPN and (2) the uniqueness of the fingerprint pattern is ensured with the large number of pixels. In Sec. II, the concept and diagram of the CIS-PUF are described. Evaluation results of the device ID of 2 Mpixel CIS are reported in Sec. III, followed by a conclusion in Sec. IV.