Skip to main content
European Commission logo print header

European Doctorate in Image Sensors with Optical Nanotechnology at Glasgow and Awaiba

Final Report Summary - EDISON-GA (European Doctorate in Image Sensors with Optical Nanotechnology at Glasgow and Awaiba)

EDISON-GA is an European Industrial Doctorate project funded under the Marie Curie Action. It provides doctoral training to 5 early stage researchers (ESR) in an international, interdisciplinary and intersectoral industrial-academic environment. The research within project started in October 2013, with recruitment of first 3 ESRs. The fourth ESR joined the project in May 2014 with the fifth joining in October 2014. Four aspects of digital image sensor design were explored in this project – the pixel, readout, filters and application. Major outcomes from each of these are being presented here.

A. Pixel
A1. A new pixel design for low leakage current
CMOS pixels, the predominant pixels in digital cameras, have high leakage current which leads to poor low light performance. Using the physics of a photodiode, a new circuit has been designed which provides almost zero dark current. A patent application for this pixel has been filed. It was also experimentally verified by designing and manufacturing a chip in a 90 nanometer CMOS imaging process.
A2. A new pixel layout
An image sensor pixel consists of a photosensitive part as well as a non-photosensitive part. The goal of layout is to maximise the photosensitive part so as to collect as much light as possible. Furthermore, pixel layout in an image sensor is typically in a rectangular or square grid pattern. During the course of this project, it was discovered that the photosensitive part of the pixel can be further increased if pixel layouts is in a staggered fashion rather than regular grid. Two image sensor chips were designed and manufactured using this staggering technique. In both, it was observed that the photosensitivity has increased. More importantly, by using this staggered pattern, we were able to double the obtained resolution without increasing the number of pixels. This also helped significantly in edge detection.
A3. Low leakage in standard pixels
The principal source of leakage and hence dark current in a pixel has been the isolation mechanism for photodiodes in a classic CMOS manufacturing process. This either includes Local Oxidation of Silicon (LOCOS) or Shallow Trench Isolation (STI), both of which induce stresses at the silicon-insulator boundary leading to local leakage centres and hence dark current. By close cooperation with the manufacturing foundry, one of the ESR was able to design and manufacture a pixel, wherein these isolations were removed in three out of four sides of a rectangular diode. This has been shown to reduce the leakage current, yet has no effect of the performance of the circuit.
A4. A new pixel design for linear and logarithmic pixel
CMOS pixels have typically linear response. Special purpose pixels with logarithmic response have also been designed. In this project, by using a simple modification to linear pixels, logarithmic response for high intensities have been observed even in linear pixels.

B. Readouts
B1. Complete understanding of noise in readouts
Readouts in CMOS image sensors historically consist of a ramp type analogue to digital converter circuit. Herein, an increasing ramp voltage is compared to the analogue input and a counter is incremented in parallel. The counter’s output, when the ramp voltage becomes higher than the analogue input, is stored as the digital representation of the analogue input. This makes it a dynamic system, wherein the noise analysis becomes very difficult. Historically, the noise has been analysed as a random Gaussian process. However, noise in the ramp signal itself becomes a Weiner signal due to the integration of a Gaussian noise over time and hence appears as a random walk. This project led for the first time, a complete analysis of this noise leading to firm limits on the time scale of the ramp signal.
B2. A new readout for Ramp ADC
Traditional ramp readouts are unable to provide same level of performance at higher resolutions or at high speeds. To solve this problem, we have proposed a new ADC consisting of two-parallel conversions. One is the classic ramp, while other utilises consecutive time-domain measurement of the quantization error between comparator toggle and clock edge using an interpolating time to digital converter (TDC). Such TDC’s are historically quite complex and sensitive to process variations. We have developed a new scheme by utilising open-loop, local per-128 column delay lines with added TDC code redundancy, alongside a digital TDC gain correction method. Hence, we were able to design a 1 micro-second ramp time 12-bit column-parallel flash interpolated single-slope analogue to digital converter.
B3. A new readout for Sigma Delta ADC
Classic ramp ADCs, however are unable to provide higher resolutions required for next generation of image sensors. Hence, in parallel to improving the ramp ADC, we also developed a new sigma-delta ADC for column-parallel implementation. We again provided two stage – one course and another fine. The first stage utilized a 5-bits 1st-order Incremental Sigma Delta converter. which is well known for its low power requirements and tolerance to non-idealities in analogue components. The Extended Counting technique is then used to convert the residue in the remaining 9 bits using a fast cyclic conversion. More importantly, we reused circuit components between the two stages, leading to a very compact ADC.

C. Filters
C1. A new design of plasmonic colour filters for use in CMOS image sensors
Typical digital cameras use organic filters for separation of red, blue and green colours. However, these are insufficient for a large number of applications and are prone to colour cross-talk as the pixel dimensions shrink. The project has led to invention of a new filter design, which can provide for very narrow pass band in the spectrum. More importantly, it provides us an ability to design filters at any pass band. A patent application has been filed and the filters have been realised in our industrial strength foundry.
C2. Reconfigurable filter design
During the development of the plasmonic filter, a desire was expressed by industrial partner for electronic tenability of filters. Currently filters are designed at the manufacturing time with no abilities to change them during operation of the camera. Typically, these filters are red, blue and green, matching the primary colours. We have now developed a new filter using novel function materials wherein by applying an electrical voltage, we are able to change the optical transmissivity of the material. By doing so, we can convert a red filter to a blue filter and so on. This is very promising early result and we are filing a patent for it.

D. Applications
D1. Real time Steresoscopic Imaging
Stereoscopic imaging requires depth measurement. However, algorithms for undertaking these are quite complex, with difficult real time implementation. We have implemented real time dense stereo matching on FPGA accelerated embedded devices for biomedical vision systems. In addition, we have also extracted distance and speed from monocular images useful for transportation applications. These have been published as full papers in SPIE Europe.
D2. TDI imaging
We have developed a new image sensors providing multiple rows time-delay integration imaging (TDI) for very high speed imaging, Historically, TDI imagers have very few rows (2-8) and large number of columns. By utilising the staggered layout (A2) and the new ramp ADC (B2), we have been able to develop an 128-row 1024-column TDI imager. It is currently being tested. Noise analysis of such imager has been published as a full paper in SPIE Europe and further test results will be published in future.
D3. A new technique to undertake compressed sensing
Compressed sensing is a recent technique to extract information from small number of measurements. This has shown promising results in imaging. This project has been active in development of sigma-delta analogue to digital converters for image sensors. A chanced discovery during the project was the ability to use these in an innovative way to undertake compressed sensing.
D4. Radiotherapy image sensor design
Typically, stitched image sensors are used for x-ray detection and this project has envisioned a similar design for very high intensity radiation. However, in the early part of project, a new design was developed which utilises a large array of small pixels with optical coupling for radio-imaging. This innovative approach reduces the need for stitching and gaps in images formed by typical silicon detectors. A prototype was designed for small arrays imaging.

The project has a website at www.eng.ox.ac.uk/edison