Bernard et al. describe an artificial retina in [Bernard et al. 93a]. The main difference between this retina and other implementations is that the image is digitized at the very first stage, and processing is performed by Boolean operators acting at the pixel level. The main advantage of this digital retina over analog approaches is its programmability for performing different tasks with the same hardware.
Photodetection has been realized using photocurrent integration followed by thresholding. Therefore the signal becomes digital right at the detector level. The rest of the implementation concerns the design of digital Boolean processors in an architecture called neighborhood combinatorial processing (NCP). A partial set of implemented Boolean operations is shift-up, shift-down&left, shift-right, circular permutation and inversion, copy, inverting copy, conjunction, conjunction and inversion, writing the photodiode, and reading the photodiode. These instructions are coded into a pseudo-static digital circuit, which uses six control signals. The architecture of Bernard's digital retina indicating its pixel level interaction is illustrated in Figure 2.7. Further details about the design of this digital circuit can be found in [Bernard et al. 93a].
By combining simple instructions more complicated operations can be performed. Operations such as edge detection, motion detection, and halftoning have been successfully demonstrated using the chip.
Several versions of the chip have been designed and fabricated. One of
them occupies an area of 50 
and contains a 65 
76 array of
photodetectors and Boolean processors, using a 2 
m CMOS process.
Figure: Architecture of Bernard et al.'s digital retina.