This document summarizes the project's results within the project duration. The results for the different work-packages can be accessed via the links below.
Project Summary and conclusions
At the beginning of the project, the objectives were defined as:
All these observations come together in the main goal of SeeBetter, which is to merge the most recent advances in image sensor fabrication, silicon retina CMOS circuit design, and understanding of the mammalian retina into a common goal of fabricating an advanced silicon retina vision sensor. In order to meet this challenge, we must combine expertise in experimental neuroscience, theoretical modelling, semiconductor fabrication and packaging technologies, and CMOS mixed-signal circuit design.
Specifically, we propose to experimentally characterize the functional utility of the six major classes of retinal ganglion cell neurons and to mimic this functionality in a custom vision sensor semiconductor chip. These cells are: a large and transient ON cell, a small and transient ON cell, a large approach detector OFF cell, a large sustained OFF cell, a small and transient OFF cell and the small local edge detector ON-OFF cell. We intend to implement this understanding using the following advanced techniques combined for the first time:
1. The development of electronic circuits implementing the retina classes which optimally capture both temporal and spatial contrast in the scene.
2. The processing of the wafers with the electronic circuits to obtain a backside illuminated sensor which achieves the best quantum efficiency, which is an absolute requirement for optimum performance.
It became clear halfway through the project that the developments at circuit level would not be able to follow the efforts of WP1.The main reasons for this are:
- The design and manufacture of a test chip is costly in time.
- Designing and manufacturing a full wafer is costlier in money.
- For reasons 1 and 2, a meaningful architecture, based on validated existing circuitry has to be chosen, or
- A validated and reliable simulation and or emulation platform must be in place prior to the development of new architectures or circuits.
From the point of view of applications and project planning, the step from the original DVS to the DAVIS was the clearest option, even if it fell short of the advances in understanding the retinal cells achieved in WP1. Nevertheless, as the first measurements have shown, the addition of the frame readout and the color filters represent a big step forward in the development of DVS image sensors.
On the other hand, by the end of the project all the effort on WP2 has resulted in simulation and emulation platforms which allow the testing of new architectures based on more complex pixel elements. This should allow to accelerate the development of more complex architectures. Unfortunately this is only possible by the time the SeeBetter project is completed.
Also, the successful demonstration in WP4 that BSI processes can be reliably used with DVS type cameras means that the possibility exists to
increase the pixel complexity without compromising the optical performance of the array.
In conclusion, we believe that the results obtained by SeeBetter open the doors to faster development of new bio-mimetic image sensors. The issues which still need to be addressed are:
- The costs of the development (full wafer design and manufacture, plus back-side processing) are large enough that they cannot be carried out by a research institution on its own. Hence, funding such as the one available for SeeBetter (i.e. EC projects) is needed to carry out this research.
- The concept of DVS type cameras needs to be evaluated side by side with existing frame-based cameras in real-world applications, to clearly show the advantages and find the new features that need to be added.
In the publications page of this website all publications which resulted from
the project are listed.