CHAPTER 1
ADVANCEMENTS IN SMALL-PIXEL, VIDEO-RATE, BACKSIDE-ILLUMINATED CHARGE-COUPLED DEVICES
John R. Tower
Samoff Corporation CN 5300 Princeton, NJ 08540-5300
1 ABSTRACT
Sarnoff Corporation has developed a CMOS-CCD process technology that has enabled the development of a new generation of backside-illuminated charge-coupled devices (CCDs). The new devices achieve high quantum efficiency (QE), combined with high modulation transfer function (MTF) performance, at pixel sizes down to 6.6 microns. This new generation of CCDs also provides excellent noise performance at video readout rates. To permit the realization of large-format devices, photocomposition (stitching) has also been demonstrated within this new process technology.
2 INTRODUCTION
Sarnoff, formerly RCA Laboratories, began work on CCD development in 1971. The backside-illuminated CCD work at RCA dates back to 1978. The backside-illuminated process that has now been employed for over twenty years utilizes whole wafer thinning with backside lamination to a glass support substrate. The back surface is implanted and then furnace annealed to provide stable, high quantum efficiency from < 400nm to > 1,000 nm.
Over the past few years, Samoff has been moving toward a more unified processing approach across our imaging products. Recently, we have integrated aspects of our double polysilicon spectroscopic CMOS-CCD process with our backside-illuminated, triple polysilicon CCD process. We have also moved the majority of our products to Canon 5X lithography to achieve tighter design rules. This paper will summarize the present state-of-the-art at Sarnoff.
3 PROCESS TECHNOLOGY
The new generation of backside-illuminated devices is processed in a triple polysilicon, single-level metal process technology with aluminum metallization. The process technology supports full CMOS circuitry in a twin well configuration. Standard oxide thicknesses are employed for pixel sizes > 12 microns, and scaled, thin oxides are employed for pixel sizes < 12 microns. The CMOS/CCD process supports CMOS-quality electrostatic discharge (ESD) pad protection.
A number of process options have been incorporated into the process flow. Pixel or horizontal register buried blooming drain structures can be implemented to achieve anti-blooming. Buried contacts can be realized to reduce the floating diffusion capacitance. This floating diffusion stray capacitance reduction increases the output sensitivity and reduces the amplifier equivalent noise.
To achieve high-speed clocking of vertical register gates, metal-to-poly strapping contacts have been demonstrated for pixel sizes down to 8 microns. These small geometry contacts are the enabling technique for achieving 1 MHz vertical clock rates on CCDs with > 50-mm-long gates at an 8-micron pixel pitch.
To produce very long linear array devices, Sarnoff has developed photocomposition (stitching) capability on the Canon 5X stepper. Employing stitching, Sarnoff has demonstrated > 60-mm-long devices. Figure 1 shows a long linear array produced with stitching. Figure 2 shows the excellent registration and feature size fidelity achieved with stitching. The figure indicates the stitch boundary location. The poly 2 gates at the top of the figure are 4 microns in width. The stitch boundary is difficult to detect, with typical panel-to-panel misalignments of 0.1 micron.
4 IMPROVEMENTS IN QUANTUM EFFICIENCY
The major advantage of backside illumination is the high quantum efficiency that can be achieved, particularly at wavelengths below 550 nm. With 100% optical fill factor and no gates to absorb short wavelength photons, backside-illuminated devices can approach ideal silicon quantum efficiency. Figure 3 indicates measured QE for the best non-AR coated devices peaking at 80%. With the AR coatings now being developed, the peak QE should be > 90%. Furthermore, as indicated in Figure 3, the AR coatings can be tailored to peak the QE in the band of interest.
5 IMPROVEMENTS IN DYNAMIC RANGE
As the pixel size is reduced, the dynamic range is compressed due to reduction in full well capacity. To maximize the dynamic range for small pixel devices we have 1) increased the area charge capacity (e/µm2) of the pixel, and 2) decreased the amplifier noise floor. The measured improvement in full well will be discussed first. Two small pixel designs have been fabricated with standard buried channel implants and enhanced, high capacity implants. The measured full-well results are indicated in Table 1. The criteria for full well is the maximum charge that can be transferred without extended-edge-response charge-transfer efficiency (CTE) degradation. The Nova 8-micron pixel results indicate that > 1.5 X improvement in full well can be achieved moving from the standard implant to the high capacity implants. The charge density is > 9,000 e/µm2 for the improved Nova device. The Mark V 6.6-micron pixel design does not exhibit charge densities as high as the Nova design. By layout changes, the design can be optimized further to provide higher charge capacity. However, as is, this charge capacity is excellent for such a small pixel device.
To achieve a full 12-bit dynamic range (> 4096:1) at video clock rates, a new generation of output amplifiers was developed for the small-pixel CCDs. The first of the new designs was a two-stage amplifier employing a buried contact. The measured performance of the amplifier as demonstrated on the 6.6-micron pixel, 1k x 1k Mark V imager is shown in Table 2. The amplifier signal was processed with an off-chip correlated double sampling (CDS) analog processor for these measurements.
The output sensitivity of 15 µV/e is reflected back through the measured voltage gain of 0.54 V/V to a floating diffusion sensitivity of 27 µV/e. The predicted noise for cooled operation at -IOC is < 7 e RMS at a 5 MHz clock rate and < 5 e RMS at a 1 MHz clock rate. The predicted maximum clock rate for this amplifier is 12 MHz, consistent with single port, 640 x 480 format, RS170 video rates.
6 IMPROVEMENTS IN MODULATION TRANSFER FUNCTION
A major challenge in achieving small-pixel, backside-illuminated CCDs is the attainment of high modulation transfer function (MTF) at practical silicon thicknesses. A rule of thumb has been that the silicon thickness should scale with the pixel size, with the pixel dimension no smaller than the silicon thickness dimension. The driver for this guideline was minimizing thermal diffusion of carriers outside the depletion edge for short wavelength illumination. Following this guideline the 6.6-micron pixel Mark V CCD would require a silicon thickness of roughly 6 microns to achieve reasonably high MTF. This silicon dimension is not practical from a manufacturing perspective. Furthermore, the thinner the silicon, the lower the broadband quantum efficiency.
Sarnoff has developed an approach to backside-illuminated iinagers that provides high MTF for small-pixel devices at standard silicon thicknesses. Figure 4 shows the measured and modelled MTF for 500-nm illumination for the 6.6-micron pixel at a silicon thickness of 10 microns. At the Nyquist spatial frequency of 75.8 cycles/mm, the MTF is 46% (@ 500 nm). The graph indicates the Nyquist geometrical limit of 63% for the 100% fill-factor pixel. The graph also shows the predicted Nyquist performance of only 23% (@500 nm) for Samoffs historical, "large-pixel" process. The measured performance represents a 2X improvement in MTF compared with conventional backside-illuminated approaches.
7 SUMMARY
Sarnoff has developed a new generation of small-pixel, video-rate, backside-illuminated CCDs. Devices have been demonstrated at 8-micron x 8-micron pixels and 6.6-micron x 6.6-micron pixels. This new generation of backside-illuminated CCDs complements our other imaging device technologies. Table 3 provides an updated summary of Sarnoff s process technologies for imaging applications.
CHAPTER 2
NEW DEVELOPMENTS IN THREE-DIMENSIONAL IMAGING WITH OPTICAL MICROSCOPES
Craig D. Mackay
PerkinElmer Life Sciences Abberley House Great Shelford, Cambridge, England
1 CONFOCAL IMAGING: INTRODUCTION
Optical microscopes are capable of extremely good depth resolution. A small change in the focal position can bring features sharply into focus. The performance of optical microscopes, however, is substantially degraded because of scattered light from other parts of the sample that are out of focus. In order to see how this happens, consider using not the normal uniform source of illumination but a single small pinhole illuminated from behind.
The objective of the microscope produces a converging beam that will produce a bright spot of illumination on one plane within the sample. If the eyepiece is also focused on the same plane where the hole is in focus, then the user will see a bright spot of light in the sample, surrounded by a halo of faint illumination. This halo arises because the column of light converging towards the bright spot is also illuminating the sample with its out-of-focus beam on either side of the focal plane. The level of illumination in out-of-focus planes is considerably smaller than that found in the focussed spot, but nevertheless the general low-level halo from the spot is significant. When a sample is illuminated with a continuous uniform field, then every element that is illuminated also gives rise to a similar halo around it and what the user sees is a summation of all the illuminated spots and their halos. For this reason, when a conventional microscope is focused on a specific plane within the sample, the background scattered light level can be very high indeed.
One elegant solution to the problem is to use the confocal arrangement. Think again about the case where a single pinhole illuminates the sample. The optics can be arranged so that only light actually emitted by the brightly illuminated spot will be passed onto the detector system, and faint halo light will be suppressed by a second pinhole, also in a plane focussed on the illuminated spot. This is the confocal arrangement. The most common arrangement for confocal microscopy is to generate a bright spot of light by using a scanning laser beam, ensuring that the laser beam illumination spot is always confocal with a pinhole scanned in synchronism with the laser. This way it is possible to suppress very greatly indeed the out-of-focus illumination to give remarkably clear and sharp pictures. The disadvantage of this arrangement is that only one element of the sample is illuminated at once, and obtaining an output image of reasonably good resolution may require many seconds of scanning to create a single image. A further difficulty is that in order to get an adequately strong signal, the laser beam must be very bright, which can often lead to photobleaching with fluorescence microscopy.
2 HIGH-SPEED CONFOCAL SOLUTIONS
There are many applications in optical microscopy where it is important to be able to take many images of the sample per second. As we have seen above, with a conventional scanning laser microscope it is not possible to do this without using exceptionally high light levels. An elegant solution to this problem is to employ a white light source, rather than a laser beam, to illuminate a mask comprising a large number of pinholes. The confocal mask also has pinholes, which are scanned in synchronism with the illuminated mask. A large number of pinholes may therefore be scanned across the sample at once, reducing the scanning time dramatically and also allowing confocal microscopy to be carried out with non-laser sources such as tungsten lamps, with appropriate colour filters as required. It also allows genuine white light confocal imaging, which may be convenient in applications where the sample has its own colour. This scanning disk, known as a Nipkow disk, has been used for a number of years to provide high-speed confocal imaging. The disadvantage of the multiple pinhole approach is that in order to ensure that the general background illumination does not get too high, the pinholes must be well spaced and therefore the overall light transfer efficiency through the pinhole mask becomes relatively poor. Tungsten lamps are not able to produce the higher brightness obtainable with the laser, and therefore Nipkow disk confocal microscopes suffer from relatively low sensitivity, particularly in applications such as fluorescence microscopy.
3 FAST, EFFICIENT REAL-TIME CONFOCAL IMAGING
A more recent development by research workers in Japan has been commercialised by Yokogawa. In their system, the overall light gathering efficiency of the Nipkow disk is improved by providing a second disk before the pinhole disk, consisting of a large number of tiny lenslets, each of which is focused on a corresponding pinhole. The amount of light that passes through the pinhole is greatly increased because each tiny lens is able to capture the illuminating light across the aperture of the lenslet rather than simply across the aperture of the pinhole. The lenslets and the pinholes are arranged in a spiral pattern and the disk is rotated to give approximately 360 images per second.
Recently, a commercial system has been developed by PerkinElmer Life Sciences and is now being sold by them worldwide. The system is integrated with a high-speed digital CCD camera and a multiple-wavelength laser. Images may be obtained very rapidly indeed. One viable mode of operation is to take an image at one focus setting and then to step the stage in the same direction by a small amount and repeat the image capture. This procedure is repeated many times, effectively building up three-dimensional data cubes of images, allowing genuine three-dimensional confocal image data sets to be acquired in real time. The system presently marketed by PerkinElmer Life Sciences uses these data sets with a sophisticated software package to generate 3D images of the sample. These images may be displayed on a computer screen and rotated and zoomed in three dimensions so that the microscopist may inspect the sample in whichever orientation and in whichever way is most appropriate.
4 DIGITAL (NON-CONFOCAL) DECONVOLUTION
There is another approach that allows one to record three-dimensional images without the complication of using a confocal microscope. Images are taken at a set of stepped focal positions in the sample, just as with the fast scanning confocal microscope described earlier. As previously explained, the difficulty here is that each image contains the desired in-focus image plus a considerable contribution of light from the out-of-focus images. It is now possible to purchase powerful computers able to sort out in-focus from out-of-focus images. This process is known as deconvolution. In principle, what it does is as follows. The programme looks in one of the image planes for a feature that is very compact and is essentially unresolved with the optical configuration used. By having good knowledge of the system's optical configuration (and a detailed description is essential for the system to work as well as it can), the software is able to compute the contribution from the other image planes from this element in the one plane that is in focus. It then subtracts the out-of-focus component corresponding to that single bright spot for each of the other image planes. This procedure is repeated for every bright element in the three-dimensional image data sets and the process is continued and repeated again and again until what remains is a substantially confocal image.
This method has the advantage over confocal microscopy of allowing conventional microscope systems to be used and particularly of having a very high light efficiency. The difficulty with this method is that in situations where the sample to be imaged is relatively weak and the out-of-focus images are relatively bright, the overall signal to noise that may be achieved is compromised in any one plane by noise contributions still present from other image planes. The process is in essence a complicated way of subtracting one image plane from another adjacent (and therefore similar) one and looking at the differences. If the signals are large and the differences are small images, it will be clear that the overall signal-to-noise must be compromised.
In practice the choice between digital deconvolution and scanning confocal microscopy is very much dependent on the kind of sample used. The software system provided by PerkinElmer Life Sciences with the scanning confocal system described in the last section also has software facilities that allow the digital deconvolution method to be used, allowing the customer to choose the method that happens to best suit his or her application.
