CMOS Imagers : Today and Tomorrow

Transcription

1 CMOS Imagers : Today and Tomorrow IISW2015 Highlights Albert Theuwissen Harvest Imaging (B)

2 Image Capturing Circle the Netherlands Belgium Harvest Imaging Germany France copyright 2010 Albert Theuwissen

3 Image Capturing Circle TU Delft Arvoo CycloMedia Philips Medical Systems Grass Valley TNO Teledyne DALSA 100 km IMS Fraunhofer CMOSIS Adimec Bosch Caeleste ON-Semi KU Leuven Melexis Harvest Imaging Image Engineering Softkinetic imec C-CAM XenICs PDPC Aphesa copyright 2010 Albert Theuwissen

4 Agenda Highlights of International Image Sensor Workshop 2015 : Co-chaired : Albert Theuwissen (Harvest Imaging) & Johannes Solhusvik (OmniVision), Technical Program Chair : Pierre Magnan (ISAE), General trends, Large Area Sensors for DSC Stacked Image Sensors Conversion Gain Engineering Ultra-Low Noise Technologies

5 General Trends IISW2015 Deep trench isolation (DTI) between the pixels is introduced in production, to lower the optical and electrical cross talk, partially and fully through the silicon, front-dti and back-dti. Samsung S5K2P2XX (partial) DTI

6 General Trends IISW2015 Deep trench isolation (DTI) between the pixels is introduced in production, to lower the optical and electrical cross talk, Buried CFA : building walls between the colour filters. These walls limit optical and spectral cross-talk, VisEra Technologies

7 General Trends IISW2015 Deep trench isolation (DTI) between the pixels is introduced in production, to lower the optical and electrical cross talk, Buried CFA : building walls between the colour filters. These walls limit optical and spectral cross-talk, Very thin optical stacks, down to 1.5 um for BSI sensors,

8 General Trends IISW2015 Deep trench isolation (DTI) between the pixels is introduced in production, to lower the optical and electrical cross talk, Buried CFA : building walls between the colour filters. These walls limit optical and spectral cross-talk, Very thin optical stacks, down to 1.5 um for BSI sensors, Incorporation of focus pixels for auto-focusing purposes.

9 General Trends IISW2015 Deep trench isolation (DTI) between the pixels is introduced in production, to lower the optical and electrical cross talk, Buried CFA : building walls between the colour filters. These walls limit optical and spectral cross-talk, Very thin optical stacks, down to 1.5 um for BSI sensors, Incorporation of focus pixels for auto-focusing purposes. Stacked imagers are more and more introduced : 2 layers of silicon,

10 General Trends IISW2015 Deep trench isolation (DTI) between the pixels is introduced in production, to lower the optical and electrical cross talk, Buried CFA : building walls between the colour filters. These walls limit optical and spectral cross-talk, Very thin optical stacks, down to 1.5 um for BSI sensors, Incorporation of focus pixels for auto-focusing purposes. Stacked imagers are more and more introduced : 2 layers of silicon, The incorporation of W pixels continues, in more recent devices, up to 50 % of the pixels are W pixels, OmniVision OV10820 RGBC

11 Large Area Sensors for DSC SAMSUNG : BSI 28M-pixel APS-C sensor First large area CIS with BSI technology, 2 rows are being readout at the same time

12 Large Area Sensors for DSC Micro-lens with AR-layer Metal grid 5 metal layers = 5.5 Gb/s

13 Large Area Sensors for DSC Canon : Phase-Difference Detection AF in All Pixels IN-focus = information left side of the lens and information right side of the lens coincide in the same pixel, Principle = compare left side of the pixel to right side of the pixel,

14 Large Area Sensors for DSC 2PD/pixel, 8T/pixel, 4 column busses,

15 Large Area Sensors for DSC 2 diodes/pixel could allow different exposure times for the 2 diodes, Option to extend the dynamic range, More complex timing/pixel needed,

16 Large Area Sensors for DSC Sony : 4M pixel full PDAF CIS 45 o rotated Bayer pattern, Diamondshaped pixels in a quincunx organization, 2PD/pixel,

17 Large Area Sensors for DSC Teledyne DALSA : Large Area CCDs for Professional Applications Presented 32M, 60M and 256M pixel CCDs, Ultra low dark current : 2 pa/cm 2, 16 parallel analog outputs, resulting high bandwidth of the amplifiers and consequently larger level of temporal noise, 1 hour exposure, Uncooled, ISO140,

18 Stacked Image Sensors Olympus: 3D CIS with 16Mpixels using 4M interconnects 4-shared pixel : 1.75T/pixel, Hole detection pixel, all p-type transistors, Common CDS for 4 pixels, One storage node/pixel, GS pixel with full CDS option,

19 Stacked Image Sensors

20 Stacked Image Sensors Sony : 20 Mpixel BSI Stacked CIS with Multiple Sampling 4-shared pixel : 1.75T/pixel, 1 interconnect/column 2 ADCs/column, allowing multiple sampling, 2 samples/pixel, noise reduction by sqrt(2)

21 Stacked Image Sensors

22 Conversion Gain Engineering Tohoku University : Analysis of Floating Diffusion Capacitance

23 Conversion Gain Engineering

24 Conversion Gain Engineering

25 Conversion Gain Engineering Images taken at room temperature, average number of signal electrons is : 3 electrons 10 electrons 30 electrons

26 Conversion Gain Engineering Dartmouth School of Engineering: Multi-Bit Quanta Image Sensors Detection of a handful of electrons seems to be possible, Conversion gain of 242 V/e -, This result was not published, but only shown during the presentation.

27 Conversion Gain Engineering CEA-Leti : Low Light CIS Detection of a handful of electrons seems to be possible, Conversion gain of 242 V/e -, This result was not published, but only shown during the presentation. n-type source-follower and select transistor are replaced by p-type, Also applicable to small pixels? [here : 7.5 m]

28 Conversion Gain Engineering 258 V 31.5 V ETH Zurich TU Delft 74 V Noise expressed in electrons depend on the noise floor AND on the conversion gain, Product of an image sensor is a number of electrons, so noise-equivalent number of electrons makes sense!

29 Ultra-Low Noise Technologies Caeleste : CIS with Low Noise by Inversion-Accumulation Cycling 1/f noise is generated by means of traps in the oxide that modulate the hole current, If the traps are filled (with electrons) they do no longer interact with the holes.

30 Ultra-Low Noise Technologies Noise reduction strongly depends on cycling frequency, temperature, number of oversampling, Noise level down to 0.33 e - were reached.

31 Ultra-Low Noise Technologies ON Semiconductor : Electron Multiplying CCD Excess noise due to the multiplication can be limited by using many small gain steps, Gain strongly depends on the gate voltages and temperature, Aging effect : due to hot electrons.

32 Ultra-Low Noise Technologies No gain applied, Gain of 30 applied by means of EM, Bright regions have an average signal of 2 e -.

33 Summary

34 Summary There s More To The Picture Than Meets The Eye (Neil Young, 1978)