Fig. 1-1 illustrates the COBRA detector roadmap conducted from 2002 to 2007 to produce both space qualified 2D and multi linear arrays, with high Spectral Detection Efficiency, for GEO and LEO Earth observation missions. Four main goals were reached during these years: high electro-optical performances, linear swing improvement, low noise readout by pixel Correlated Double Sampling and conversion gain adjustment.
COBRA2M: A 2 MILLIONS PIXELS DETECTOR FULLY SPACE QUALIFIED IN 2007 FOR AN OPERATIONAL GEOSTATIONNARY OCEAN COLOUR MISSION
Geostationary Ocean Color Mission 
The detector COBRA2M has been developed for the Geostationary Ocean Color Instrument (GOCI), see Fig. 2-1, implemented on the COMS spacecraft to provide multi-spectral data to detect, monitor, quantify, and predict short term changes of coastal ocean environment around Korea, see Fig. 2-2, for marine science research and application purpose, in eight spectral narrow bands, see Fig. 2-3.
A prototype of 1 million pixels was developed in 2002-2004 to confirm UMC 0.35µm excellent radiometric performances  in term of Spectral Detection Efficiency (SDE) and Modulation Transfer Function (MTF), see Fig. 2-4 and 2-5.
A new prototype COBRA1MP with various ionic implantations was tested in 2005 to maximize output voltage swing , offering improved charge handling capacity necessary for photonic noise dominated application as Geostationary Ocean Colour Imaging
Experience gained in the frame of LOLA mission  with the development of a 750x750 airborne APS  was used to define the COBRA2M package, using the same concept of Interstitial Pin Grid Array (IPGA) connected to a PCB Flex with multiple flexible parts, see Fig. 2.6, anticipating a Focal Plane Array with the detector back glued on a SiC part to offer excellent thermal control.
From the radiometric performances extrapolated from COBRA1M and the increased performances of the readout circuit in term of output voltage swing, tested on COBRA1MP, accurate modelling of the COBRA2M performances was possible, especially in term of elementary integration time and the number of cumulated images necessary to meet specified SNR of 1000 for each of the eight spectral bands.
For the band of maximum flux and maximum Spectral Detection Efficiency, a minimum elementary integration time of 100msec was necessary to collect 100ke-. The frame period for this value is 100msec and was provided by four outputs with a pixel rate of 5.3 MPixels/sec.
Row and column decoders, already developed in the frame of COBRA1M, were implemented, see Fig. 2-7.
Demonstrator Model (DM) manufacturing and test phase
A small number of devices were packaged by e2v and tested by Astrium.
Readout circuit performances were confirmed with respect to COBRA 1M_P test results, see Fig. 2-8, with a output voltage swing greater than 1.1V for a 3.3V power supply and a Readout noise in darkness of 500µV rms due to a√(2kT/C) hard reset noise and a estimated 200µV rms noise for sampling circuit.
MTF performances measured on COBRA1M were confirmed, using a specific Astrium test bench, see Fig. 2-9
At the end of the DM phase, a COBRA2M device was soldered to a PCB Flex and connected to the Front End Electronic (FEE) to test compatibility, see Fig. 2-10.
Electro-optical tests were repeated to validate FEE behaviour.
Flight Model (FM) manufacturing and qualification phase
DM test results presented at the Critical Design Review (CDR) authorized the FM manufacturing and qualification phase. A FM test bench was implemented in an Astrium clean room, see Fig. 2.11 to screen the 32 devices packaged and delivered by e2v and select Qualification models, Flight model and Flight spares, see Fig. 2.12. The qualification test results demonstrated only Dark Current increases with Total Ionizing Dose (5krad) and protons irradiations (2 1010 p+/cm² 60MeV) as depicted in figure 2.13.
COBRA2M development lessons learnt
A learning curve is always necessary when using a new foundry and that some prototypes are necessary before producing flight models. Developing an efficient fully automated test bench greatly assists the conduction of testing during an FM production phase, securing both schedule and quality of test results, especially for the repeatability of test before and after qualification tests. The selected Flight Model was delivered in June 2007, (two years after project kick off), to be soldered on the Flight PCB Flex and then glued onto the Focal Plane Array Sic part, before to be mounted on the Instrument telescope.
S2 VNIR: A MULTI LINEAR SENSOR OF THE MULTI SPECTRAL INSTRUMENT (MSI) FOR THE LOW EARTH ORBIT SENTINEL 2 MISSION 
Sentinel 2 mission, instrument and VNIR focal plane
Sentinel 2 is a LEO Earth observation mission in the frame of Global Measurement Environment and Security (GMES) program. The instrument is equipped with VNIR and SWIR Focal Plane Arrays, Each Focal Plane Array is made of twelve detectors in staggered configuration, see Fig. 3-1. The VNIR detector offers tens spectral bands with 10m, 20m and 60m resolution, and the SWIR detector offers three spectral bands with 20m and 60m resolution, see Fig. 3-2.
For each spectral band, a SNR corresponding to a reference flux and the maximum integration time is specified. A maximum flux is also specified for each spectral band. The detector sensitivity has to be adjusted band per band through Charge to Voltage conversion Factor (CVF) adjustment to meet SNR specification for a reference flux with avoiding saturation for maximum flux.
To help meeting SNR requirement, a Correlated Double Sampling readout circuit is implemented to eliminate photodiode reset noise and provide total readout noise lower than 200µV
A multi linear array prototype COBRA NxK providing two lines with 10m resolution and nine lines with 20m resolution associated to two outputs was developed and tested in 2006 and beginning of 2007, see Fig. 3-3.
This prototype validates the Correlated Double Sampling readout circuit with the expected readout noise reduction, due to photodiode reset noise suppression.
Specific test vehicles (COBRA CVF) were developed and tested to validate conversion gain adjustment.
Test bench and method heritage
The detector development took also advantage of know-how gained in the frame of the COBRA2M development, in term of test bench architecture, definition and implementation, see Fig. 3-4.
The same test methods and processing were mastered, providing high accuracy in test results, see Fig. 3-5.
Black coating heritage
As for COBRA2M, e2v works on the S2 VNIR back end manufacturing. Due to VNIR reflectivity requirements, a post process step is needed in order to strongly reduce the die global reflectivity. A black coating will be deposit on the non photosensitive area of the die to meet reflectivity requirements. This process has been still validated on COBRA NxK wafers, although black coated COBRA NxK devices have been packaged by e2v, see Fig. 3-6. Fig. 3-7 illustrates reflectivity measurements. Very low reflectance is achieved.
S2 VNIR Detector development status
End of Design and validation phase
Today the design and validation phase is nearing completion, concluded by a Preliminary Design Review (PDR)
Fig. 3-8 presents the retained architecture:
• Pixels lines associated to the different spectral bands are read through three outputs at a sample rate of 4.8MHz enabling to read the 10m resolution bands in less than 1.51msec
• Each group of lines associated to one output are selected by a row decoder. Readout circuit sampling capacitances are addressed by shift registers
Fig. 3-9 shows a breadboard of the S2 VNIR device using COBRA NxK die to demonstrate required photosensitive flatness better than 10µm.
EM and FM phases
Procurements of different parts: die wafers, package parts, see Fig. 3-10 and manufacturing of jigs for assembly and tests, will be initiated after PDR release. Adaptation of test bench software and test bench acceptance are in progress, as definition of manufacturing and testing flows is established to meet schedule and quality requirements. EM phase ends with CDR in April 2009.
FM qualification and delivery is foreseen for January 2010.
Thanks to the use of 0.35 µm CIS process, a first generation of CIS has been developed by Astrium and ISAE/CIMI, enabling the response to the needs of space optical instruments for GEO and LEO Earth observation. Our team is today investigating several ways in order to further improve the performances of monolithic CIS, while retaining all their functional and technological advantages.
Sensitivity improvement is foreseen by thinning and backside illuminating CIS (Fig. 4-1)  or by use of micro lenses coupled to front side illuminated devices. For the NIR part of the spectrum; improvement is expected using silicon wafers with thicker epitaxial layers.
Larger format 2D arrays are in progress (target being about 10 millions pixels), the use of smaller nodes (e.g. 0.18 µm process) allowing the design of smaller pixel pitches while keeping the fill factor constant. Dynamics improvement is targeted via two ways:
• Increasing the output video signal voltage swing, which can be reached thanks to low threshold voltage transistors available through modern CMOS processes.
• Reduction in the read out noise obtained by using 4T photodiodes instead on 3T ones.
Finally, on chip signal processing (within or outside the pixel) is a field that can strongly improve CMOS performances for specialised applications, such as lightning imagers, requested for future meteorological space missions.