As an image sensor IC designer, I had to design a CMOS vision sensor with some image processing embedded in the pixel. Adding signal variations, coming from different pixels, was identified one of the most useful operators from image processing. The main constraint was that the design had to be able to use very simple electronics, because the pixel had to be small. This Idea for Design was used to demonstrate the feasibility of this idea, even with discrete components.
The complete imager was successfully designed, fabricated, and published (see N. Massari, M. Gottardi, L. Gonzo, D. Stoppa, and A. Simoni, "A CMOS Image Sensor With Programmable Pixel-Level Analog Processing," IEEE Trans. on Neural Networks, Vo. 16, No. 6, p. 1673-1684, Nov. 2005). The pixel architecture was patented. One year later, we designed a new imager, improving the peak-accumulation electronics (see N. Massari, M. Gottardi, "A 100dB Dynamic-Range CMOS Vision Sensor with Programmable Image Processing and Global Feature Extraction," IEEE J. of Solid-State Circuits, Vol. 42, No. 3 647-657, March 2007).
The newer solution uses a simple inventer in sub-threshold, replacing the amplifier, two NMOS and PMOS source followers, and two capacitors to implement the peak and accumulation operations. This solution is much more efficient than the first one. In this case, the accumulation does not depend on the signal. Therefore, the efficiency is fairly constant along the entire dynamic range. More detailed information is available in the papers.
Although its main application was in image sensor processing, this solution could be carefully used in phase detectors, measuring the phase shift of a received signal with respect to a reference one.