Advances in large-format visible focal plane technology

The next generation of focal plane arrays (FPAs) for space-based remote sensing must be able to support planetary missions with 30× higher radiation levels (e.g., Jupiter) than existing earth orbit missions such as NASA's Landsat or the Moderate Resolution Imaging Spectroradiometer (MODIS). Customers want large-format FPAs that cover bigger areas to increase imaging efficiency (See Figure 1). Astronomy customers want low noise and high quantum efficiency (a measure of detector sensitivity) in the near-IR to provide the highest signal-to-noise ratio for star identification and mapping. Imaging customers will only tolerate small clusters of non-operable pixels so that imagery can be preserved without the need to interpolate (i.e., substitute in values for dead pixels) across more than a single 8μm pixel cluster.

We have found that combining silicon p-i-n (p-type, intrinsic, n-type) detectors and an 0.18μm CMOS integrated circuit foundry process results in FPAs that are radiation-hard to >200kRad. Our patented Reticle Image Composition Lithography (RICL) technology achieves a photolithography pattern by stitching various parts of the array together using a programmable stepper, enabling fabrication of a seamless FPA up to 16×16k. Our direct-bond-hybridization process technology results in operability of > 99:99% and clusters of less than two pixels. Low-noise design, including on FPA 13-bit ADCs, results in FPA noise of less than seven electrons. Selecting custom-thinned detectors enables the customer to optimize modulation transfer function or increase quantum efficiency in the near-IR.

Figure 1. Megapixel focal plane array (FPA) and image (insets) compared with a commercial image.

Recently, we achieved a small-feature ADC design that allows placing more than 110 13-bit ADCs running at 170Mb/s on the FPA. Figure 2 shows the differential non-linearity (DNL) (a measure of analog-to-digital converter performance) from a typical ADC to be <0:25. In addition, the FPA has low-voltage data signaling inputs and outputs that provide lower system noise. Other FPA features include programmable sleep modes to optimize power when circuits are unused, windowing, gain, redundancy, and a built-in self-test. These are all programmable through the serial data word, which provides command and control of the sensor chip assembly. Programmable gains allow matching the transimpedance (i.e., amplification) of the imager to the environmental conditions being imaged. An integrate-while-read snapshot mode is available to achieve the highest frame rates, and non-destructive reads are available to support commercial, tactical, and astronomy applications.¹

Figure 2. Thirteen-bit analog-to-digital converter (ADC) differential non-linearity (DNL) is 0.25.

Figure 3. Array response distribution versus ADC counts shows excellent response uniformity across a one megapixel sensor chip assembly (SCA). ph: Photons.

Figure 4. Detector dark current (i.e., undesired buildup of signal due to detector leakage current) plotted as a function of temperature.

Figure 5. FPA read noise plotted as a function of integration time for temperatures ranging from 200 to 240K.

A good measure of the operability of an individual pixel is to compare its light response to the mean response of an array of pixels. Fifty percent operability means that the pixel's response falls within a range of 50–150% of the mean (i.e., neither too low nor too high to be useful). Figure 3 shows the responsivity plot for a 1.04-megapixel 8μm-pitch FPA. Only 55 detectors are lower than 50% of the mean response, and zero are >150%; resulting in operability of 99.995%. Screening for response and noise-equivalent irradiance (NEI) operability of >20% of the background limit reveals 94 inoperable pixels. After excluding pixels that do not meet response and NEI requirements, the result is still >99:99% and there is only one cluster on the array with a size of two pixels. All other clusters are a single pixel. This performance results in worst-case interpolation between pixels of only a single 8μm width anywhere on the array.

Figure 4 shows a measured FPA leakage of less than one electron per second up to 200K. Figure 5 shows that the total noise of the FPA without incident flux is less than 10 electrons while operating up to 220K. The measured results from Figures 4 and 5 indicate that, under typical incident imaging conditions, the scene shot noise quickly dominates the electronics read noise because the contribution from the detector and electronics is very low.

Planetary exploration, remote sensing, and astronomy applications require low detector leakage, low electronics noise, and radiation-hardened large-format (>25 megapixel) FPAs. The data for a one megapixel array uses the same reticle (mask) to produce 25 megapixel arrays using RICL. The direct-bond-hybridization process has increased operability to >99:99% and reduced clusters to a point where only single-pixel interpolation is required. In the future, we plan to further improve operability as volume production streamlines processing, reduce leakage current in both the electronics and detector to increase operating temperature, reduce costs as wafer-to-wafer processing minimizes processing steps, and fabricate larger-format and smaller-pixel FPAs using our RICL and direct-bond-hybridization technologies.