A Custom CMOS Sensor for Pyramidal Adaptive Optics System Kuldip N. Modha, Ian M. Stockford, Roger Light, Matt Clark, Mark Pitter and Barrie Hayes-Gill SiOS Labs, Applied Optics Group, School of Electrical/Electronic Engineering, University of Nottingham, Nottingham NG7 2RD, UK
Abstract— This paper reports a custom CMOS light sensor which can be used in a pyramid wavefront sensor based adaptive optics system for eye imaging. A pyramid sensor produces four pupil images and therefore, this CMOS light sensor consists of 4 arrays of 32x32 active reset pixels with synchronous analogue outputs. In an eye imaging adaptive optics system the light available for wavefront sensing is of the order of nW. Accordingly the photodiode capacitance size within the pixel is carefully chosen to detect this low light. The pixels were layed out with the ﬁll factor of 56% and the chip consumes 56 mW of power during the read out. The custom CMOS sensor with an area of 12.6 sq. mm was fabricated using an Austria Micro Systems 0.35 µm C35B4 CMOS process. The sensor is mounted on a PCB and packaged in a box of dimensions 6 cm x 5.5 cm x 5 cm, which makes it feasible to install in an optical system. A dedicated high speed camera reading system has been constructed that can read upto 1000 frames per second. The realised complete camera system clearly detects the four pupil images created by a pyramid wavefront sensor. Index Terms— CMOS sensor, pyramid wavefront sensor, adaptive optics system, reset pixels.
Fig. 1. An incident beam on the apex of the pyramid wavefront sensor which produces four pupil images a, b, c and d. . A pyramid can be easily obtained by crossed prisms.
I. I NTRODUCTION In imaging systems, the turbulent medium can cause wavefront deformations which produces a distorted image of an object. An adaptive optics system (AOS) can measure these wavefront deformations through a wavefront sensor and corrects them using a deformable mirror. The Shack-Hartmann (SH) wavefront sensor based AOS are widely used in astronomical applications where the atmosphere causes image aberrations. The Shack-Hartmann sensor essentially consists of a lenslet array which forms sub-apertures that samples the wavefront and focuses on the light sensor. Based upon the position of the spots on the light sensor, the tilt at the sampling aperture can be computed. Recently, the AOS has found new applications such as retinal imaging, refractive surgery and line of sight communications. These new applications require the AOS to operate at high speed, be cost effective and ﬂexible in operation. The Shack-Hartmann devices are expensive and their sensitivity is limited by the ﬁxed lenslet arrays, which upon modiﬁcation results in alignment problems. This has motivated the investigation of a pyramid wavefront sensor (PWS)  based AOS. Most importantly, the PWS has architectural advantages over SH in designing integrated light sensors. A PWS as shown in Fig. 1 has four faces, and when the incident light is exactly focused on the apex of the pyramid, four balanced pupil images are produced. When the light exhibits a tilt and the focus shifts to one of the faces of the
pyramid, the four pupil images would no longer be balanced. Using the intensities of these four pupil images, the vertical and horizontal curvatures can be calculated through Equations 1 and 2. Based upon these curvatures calculated at the pixel level, the deformable mirror can be re-shaped to produce a corrected image. The Ia , Ib , Ic and Id are the intensities of the pupil images on the light detector. Curvaturex =
(Ia + Ib ) − (Ic + Id ) Ia + Ib + Ic + Id
(Ia + Ic ) − (Ib + Id ) Ia + Ib + Ic + Id
The pyramid wavefront sensors are highly sensitive, because when the light exhibits a tilt the incident light focuses on one of the faces of pyramid and suddenly one of the pupil images will be very brighter than the others. The sensitivity of the pyramid can be modiﬁed by oscillating the pyramid, so that the incident light ray spends a fraction of time on every face and illuminates all the four pupil images . By increasing the rate at which the pyramid is oscillated the sensitivity can be reduced. Although, the PWS which is low cost and ﬂexible was originally proposed for telescopes , their performance in
Fig. 2. Layout of the iWFS1 chip showing the four pixel arrays, operational ampliﬁers, column multiplexers and address decoders. Each array dimension is 1 mm by 1 mm and array pitch is 1.5 mm.
ophthalmic imaging  has been promising. One such system has been demonstrated by Chamot et al.  that uses a CCD light sensor. A CCD suffers from high power consumption, high cost, poor readout frame rate and inability to integrate with CMOS signal processing circuitry leading to this research on custom CMOS camera for PWS based AOS. In this paper a custom CMOS sensor for the PWS based AOS is proposed . The rest of the paper is organised as follows: section II describes the design of the CMOS camera including the reset pixel design and enclosing the sensor in a box; section III describes the high speed camera reading system; section IV discusses the measured results and section V summarises this work. II. CMOS C AMERA D ESIGN The proposed custom CMOS image sensor  is composed of four arrays of 32x32 active reset pixels to detect the four pupil images generated by the PWS. The common addressing of the digital address decoders and multiplexers is implemented which enables to generate synchronous analogue outputs from the four arrays. In addition, the arrays were made to share the reset and shutter lines. Each array is divided into four groups of 8 rows of pixels that allows to carry out rolling shutter and reset of the individual arrays. The layout of the custom CMOS image sensor is shown in Fig. 2. This is the ﬁrst generation of the integrated wavefront sensor chip and therefore, is named as iWFS1. The arrays are named as NE, SE, SW and SW corresponding to their location on the chip. The chip has been fabricated using an Austria Micro Systems 0.35 µm C35B4 CMOS process with a p-substrate, four metal and two poly-silicon layers. Each array has an on chip operational ampliﬁer conﬁgured as buffer, shared by the pixels within each individual array to drive the chip reading circuitry.
A. Reset Pixel Design The schematic diagram of a common 4T reset pixel is shown in the Fig. 3. The reset pixel consists of a pMOS reset
Fig. 3. Schematic of a reset pixel, showing the crossection of the nwell photodiode. The n-well and p-type substrate junction exhibits a junction capacitance.
transistor, an n-well over p-substrate photodiode, a shutter switch (pMOS) transistor, a source follower (nMOS) and a row select transistor (nMOS). Using a pMOS reset transistor allows to achieve a hard reset and the diode capacitance gets charged upto the rail level. As a result the well capacity of the pixel has improved however, at the price of increased pixel area due to the layout size of a pMOS device. In each array the pixels were biased using a column level current mirror. The row select transistor allows the selection of the entire row of an array of pixels during the readout. In the AOS the amount of light available is of the order of tens of nW, out of which, around 80% is used for imaging leaving only 20% for wavefront sensing. Consequently, the camera exposure time is made the same as the frame time of the whole system (around 10,000 µs) to get access to maximum amount of light. Depending upon this available light level (few nW) the size of the photodiode capacitance was chosen such that it does not discharge completely or saturate by the end of the frame time of the system.
B. Reset Pixel Operation The pixel operation begins with the turning ON of the reset transistor which couples the photodiode to VDD and the charge builds up on the photodiode junction capacitance. As soon as the reset transistor is turned OFF the photodiode capacitance starts to discharge until the required exposure time due to the reverse photodiode current induced by the light incident on the pixel. At the end of the exposure time the shutter switch opens and the pixel output is held to a constant value at the gate capacitance of the source follower, which can be read as voltage at any time through the column out. The pixel is layed out such that the pixel pitch is 30 µm and the ﬁll factor of 56% was obtained. The pixel array was surrounded by a ring of dummy pixels so that all the pixel in the array can achieve an indentical performance.
Fig. 4. Photographs of the boxed iWFS1 camera a) Side view. (b) Back panel view. The camera can be controlled through the digital IO connector and the analogue outputs can be accessed through the female SMA connectors.
C. Camera Design The iWFS1 CMOS sensor chip along with the digital drivers and the operational ampliﬁers conﬁgured as buffers were mounted on different PCBs that are joined together and packaged into a box as illustrated in Fig. 4. The digital buffers ensures the correct high and low voltage levels required by the digital decoders and multiplexers within the chip. The boxed camera with a compact size of 6 cm x 5.5 cm x 5 cm makes it feasible not only for testing purposes but also makes it easy to install in an optical system. The boxed camera has digital IO connector at the back panel and female SMA connectors which provide the analogue outputs of the four arrays. A DC power supply (6-10 V) socket is ﬁxed on one of the side panels of the iWFS1 camera. The camera has an M4 threaded hole at the bottom panel which allows it to mount it on a post.
III. DATA ACQUISITION S YSTEM The communications interface of the iWFS1 camera has to be compatible with the other components used in the our ongoing research on the development of an AOS such as a linux workstation. In order to read the iWFS1 camera an analog to digital converter (ADC) is required. The modern day ADCs provide digital IOs however, are software paced and cannot gaurantee the generation of digital signals at precise timings. Therefore, the ADC digital IOs cannot be used to reset the iWFS1 camera and generate high speed digital addressing signals. This led to the choice of a dedicated FPGA system to produce the reset, shutter and high speed digital addressing signals to control the iWFS1 camera. The block diagram of the camera data acquisition system is shown in Fig. 5. The readout system comprises a high speed 12 bit, 10 MHz (continuous multiple channels) Measurement Computing PCI DAS4020/12 ADC installed on a linux workstation. Linux control and measurement device interface (COMEDI1 ) was conﬁgured on that workstation. COMEDI device drivers provide a set of common functions which can be used to control the ADCs made by different manufacturers. A low cost Spartan 3 FPGA from Xilinx was programmed to generate signals which can control the iWFS1 camera. The FPGA also feeds an external clock and a trigger to the ADC. 1 www.comedi.org
Fig. 5. Block diagram of the iWFS1 camera reading system. Reset, shutter and addressing signals generated by the FPGA, are fed into the camera using the digital IO connector shown in Fig. 4(b). The camera analogue outputs are connected to the input channels of the ADC.
The camera reading operation begins with a start reset issued by the ADC upon the user request through COMEDI. As soon as the FPGA receives a start reset it then resets the iWFS1 camera, closes the shutter switch (see Section II-B) and allows all the pixels to discharge until the set amount of exposure time which can be chosen via the FPGA controller. At the end of the exposure time the shutter switch is opened and all the pixels are held to a constant value corresponding to the amount of light incident on the pixels. At this moment the FPGA selects the pixel 0 and triggers the ADC to start acquiring the analogue data. Now the FPGA without any inﬂuence from the ADC produces the digital addressing signals to select individual pixels of the array and meanwhile ADC continuously reads them, completing one frame.
IV. MEASUREMENT RESULTS AND DISCUSSION To test the operation of the iWFS1 camera, it has been installed in an imaging system which consists of a pyramidal prism lens realised using two roof to roof Fresnel biprisms  as shown in Fig. 6. The camera is aligned such that the four pupil images are positioned in the centre of the array as close as possible. Using the developed software a single frame was read after an exposure time of 8 msec and the shutter switched opened. Fig. 7 shows the image of four spots with one in each array. Fig. 7 is obtained by plotting the pixel output voltage against the location of the pixel. The pixel output voltage is measured with respect to ground and therefore, in the high light regions which was 200 nW for this particular
A roof to roof Fresnel biprism used as a pyramidal prism lens.
experiment the pixel voltage have decayed (blue in the Fig. 7) more than the low or no light regions on the arrays of the iWFS1 camera. Although, this is quite a lot of light compared to the few nW available during wavefront sensing this was used to demonstrate functionality of the camera.
The iWFS1 camera is designed to detect low light level and therefore requires a full noise analysis. The pixel suffers from a calculated reset noise of 90 e rms. Initial noise measurements suggests that the camera exhibits a read noise of 3 mV and a 3% variation of the pixel output voltage i.e. a ﬁxed pattern noise across the array. The features of integrated wavefront sensors chip (iWFS1) are summarised in Table I.
Fig. 7. Four spots corresponding to the pupil images that are generated using a roof to roof prism are captured by the iWFS1 camera. The pixel output voltage is plotted against the location of the pixel. The pixels exposed to pupil image are saturated and are shown in the form of blue spots.
a high speed commercial data acquisition card and a low cost FPGA has been built. The camera system is expected to be used in an eye imaging adaptive optics system with a pyramid sensor. ACKNOWLEDGEMENTS
The authors would like to thank the Engineering and Physical Science Research Council, UK for providing the funding to carry out this research work. In addition we would like to thank the School of Electrical/Electronic Engineering of University of Nottingham, UK in providing research facilities. R EFERENCES
V. C ONCLUSIONS A custom CMOS sensor for pyramid wavefront sensor based adaptive optics system has been designed, fabricated and tested. The sensor has been packaged in a box making it suitable for use in optical systems. The camera successfully detects the four pupil images produced by the pyramid lens. A camera reading data acquisition system operating at 1000 frames per second, consisting of a dedicated linux workstation, CHARACTERISTICS Technology Chip Size Pixel Pitch No. of Pixels Fill Factor Power DC Supply Frame Rate Reset Noise Fixed Pattern Noise
COMMENTS 0.35 µm C35B4 CMOS AMS 3.38 mm x 3.37 mm 30 µm 4096 56% 56 mW 3.3 V 1000 90 e rms 3% TABLE I S UMMARY OF THE FEATURES OF THE I WFS1 SENSOR CHIP.
 R. Ragazzoni, “Pupil plane wavefront sensing with an oscillating prism,” Journal of Modern Optics, vol. 43, no. 2, pp. 289–293, 1996.  R. Ragazzoni, E. Diolaiti, and E. Vernet, “A pyramid wavefront sensor with no dynamic modulation,” Optics Communications, vol. 208, pp. 51– 60, 2002.  R. Ragazzoni, S. Esposito, A. Ghedina, A. Baruffolo, M. Cecconi, E. Diolaiti, J. Farinato, L. Fini, E. Marchetti, A. Puglisi, M. Tordi, and E. Viard, “The pyramid wavefront sensor aboard [email protected] and beyond: a status report,” in Proceedings of SPIE, vol. 4494, 2002, pp. 181–187.  I. Iglesias, R. Ragazzoni, Y. Julien, and P. Artal, “Extended source pyramid wave-front sensor for the human eye,” Optics Express, vol. 10, no. 9, pp. 419–428, 2002.  S. R. Chamot and C. Dainty, “Adaptive optics for ophthalmic applications using a pyramid wavefront sensor,” Optics Express, vol. 14, no. 2, pp. 518–526, 2006.  I. M. Stockford, K. N. Modha, M. P. Matt Clark, Roger A. Light, and B. Hayes-Gill, “Prototype low cost, high speed pyramid wavefront sensors using a custom cmos sensor,” in The EOS Conference on Frontiers in Electronic Imaging is part of the World of Photonics Congress, the 18th International Conference on Photonics in Europe., 2007.