Trends in Development of Infrared FPAs

It is well known that detector size, d, and F-number (//#) are primary parameters of infrared systems [46]. These two parameters have a major impact on both detection/identification range and the NEDT, since they depend on FX/d [37]:

where X is the wavelength, D is the aperture, M is the number of pixels required to identify a target, Ax, C is the scene contrast, // is the detector collection efficiency, Фв'^т is the background flux into a 2rt FOV, and r_mt is the integration time. The above two equations indicate that the parameter space defined by FA and d can be utilized in the optimum design of any IR system.

Most military systems today have the classical view presented in Fig. 4.21 [47], where detector size ranged from 10- to 50-pm. For long- range identification systems, high//# optics are used (for a given aperture) to reduce the detector angular subtense. On the other hand, wide field- of-view (WFOV) systems are typically low //# systems with short focal lengths as the focal plane had to be spread over wide angles. In recently published papers, it has been shown that long-range identification does not need to be limited to high//# systems and that very small detectors can achieve high performances with a smaller package [48-50].

The detection range of many uncooled IR imaging systems is limited by pixel resolution rather than sensitivity. Figure 4.22 presents a trade-off analysis of the detection range and sensor optics for a thermal weapon sight, using the NVESD NVTherm IP model, assuming a detector sensitivity of 35 mK NETD (//1, 30 Hz) for the 25-, 17-, and 12-pm pitch pixel

FIGURE 4.21 FAId space for infrared system design. Straight lines represent constant NEDT. There are an infinite number of combinations that provide the same range (adapted after Ref. [47]).

FIGURE 4.22 Calculated detection range as a function of sensor optics and detector pixel size and format, using NVESD NVTherm IP modeling, assuming a 35 mK NETD (//1, 30 Hz) for all detectors (adapted after Ref. [51]).

of uncooled FPAs. The advantages of a small pixel pitch and large-format FPAs are obvious. By switching to a smaller pitch and larger format detectors, the detection range of a weapon sight increases significantly with a fixed optic.

The fundamental limit of pixel size is determined by diffraction. The size of diffraction-limited optical spot or Airy disk is given by:

where d is the diameter of the spot, and X is the wavelength. The spot size for//# ranging from//l to//10 are shown in Fig. 4.23. For typical//2.0 optics at 4-pm wavelength, the spot size is 20 pm.

It is generally interesting to investigate pixel scaling beyond the diffraction limit, using wavelength- and even subwavelength-scale optics, that are enabled by modern nanofabrication (diffraction-limited pixel size is still relatively large compared with the feature size that can be achieved with state-of-the-art nanofabrication approaches).

FPAs of 1 cm² still dominate the IR market, while pixel pitch has decreased to below 15-pm over the past few years, now reaching 12-pm [52], 10-pm [53,54], 8-pm [55], and even 5-pm in test devices [56,57]. Leonardo DRS offers HexaBlu™ camera modules with a 6-pm pixel pitch, 1280x960 FIgCdTe MWIR FPA technology that leverages DRS’s proprietary F1DVIP® [High Density Vertically Integrated Photodiode, similar

FIGURE 4.23 Optics diffraction limit. The spot size of a diffraction-limited optical system is the Airy disk diameter.

FIGURE 4.24 HexaBlu™ cryo-cooled MW thermal camera module (after Ref. [58]).

to that shown in Fig. 4.12(b)] and detection in a small Integrated Dewar Cooler Assembly (IDCA) (Fig. 4.24). This new pixel design is enabled by HexaBlu’s miniature form factor, weighing in at under 295 g and displacing just 80 cm³. This trend toward small IDCA is expected to continue. Systems operating at shorter wavelengths are more likely to benefit from small pixel sizes because of the smaller diffraction-limited spot size. Diffraction-limited optics with low F-numbers (e.g., //1) could benefit from pixels of the order of one wavelength across. Over-sampling the diffractive spot may provide some additional resolution for smaller pixels, but this saturates quickly as the pixel size is decreased. Pixel reduction is mandatory to also achieve cost reduction of a system (reduction of the optics diameter, dewar size, and weight, together with the power, and increased reliability). In addition, smaller detectors provide better resolution [50]. Reduction of the focal plane proportionately to the detector size has not changed the detector field-of-view, so that, in the optics-limited region, smaller detectors have no effect on the system’s spatial resolution.

Figure 4.25 shows the influence of pixel shrinkage on format enlargement of Lynred’s IR arrays [59]. A catalogue of detectors with pixel pitch of 15-pm [Epsilon (384x288), Scorpio (640x512), and Jupiter (1280 X 1014)] is compared with the Daphnis 10-pm product family.

Shrinking of pixels decreases system size, weight, and power consumption (SWaP), in consequence reducing the system’s cost and increasing the operating temperature in HOT detectors (see Section 3.3). Smaller pixels enhance the value of proposition of the imaging systems and their functionality. An example of their capability is illustrated in Fig. 4.26 [47], which shows DRS’s production 640 X 480 FPAs with pixel size from 20-pm to 12-pm. Additional improvement in SWaP is achieved under HOT

FIGURE 4.25 Number of pixels versus pitch size for Lynred IR FPAs (adapted after Ref. [53]).

FIGURE 4.26 Progressive reduction of DRS’s 640x480 LWIR package as (a) a function of pixel pitch and (b) increasing of operating temperature (after Ref. [47]).

FIGURE 4.27 Pixel pitch for (a) HgCdTe photodiodes, and for (b) amorphous silicon microbolometers have continued to decrease due to technological advancements (adapted after Refs. [60, 61].

conditions. Smaller pitches are scheduled in the short term (see Fig. 4.27a) [60]. A similar tendency has been observed in the case of microbolometers (see Fig. 4.27b) [61].

The detector-limited region occurs when FA/d <0.41 and the optics- limited region occurs when FA/d > 2 (Fig. 4.21). When FA/d=0.41, the Airy disk is equal to the detector size. A transition in the region 0.41 < FA/d < 2.0 is large and represents a change from detector-limited to optics-limited performance. The condition FAId=2 is equivalent to placing 4.88 pixels within the Rayleigh blur circle. The lines presented a constant FA/d, indicating a constant range and NEDT. For a given aperture, Д and operating wavelength, A, the detection range is given by the optimum resolution condition FA/d = 2 and a minimum NEDT for a given т_ш [see Eq. (4.13)]. From these considerations, the system//# should be fixed to the pixel size to predict the potential limiting performance of IR systems.

Figure 4.21 also includes experimental data points for various classes of thermal imaging systems that have been produced at DRS Technologies, including both uncooled thermal imagers and cooled photon imagers. The earliest uncooled imagers, fabricated at the beginning of the 1990s [barium strontium titanate (BST) dielectric bolometers and VO_x microbolometers], had large pixels of approximately 50-pm pitch and fast optics to achieve useful system sensitivities. With decreasing detector size, the relative apertures remained around//1. As is shown in Fig. 4.21, as the pixel dimensions decreased over time, the uncooled systems steadily progressed from the “detector-limited” regime to the “optics-limited” ones. However, they are still far from the ultimate range capacity for//l optics.

The cooled thermal imagers include early LWIR scanning systems and modern staring systems, operating under both MWIR and LWIR bands. It has been shown that LWIR imaging systems typically approach FA/d = 2 conditions, whereas, for MWIR systems, values of FA/d less than 2 are typically employed, with lower available photon flux making it difficult to maintain system sensitivity. Only the DRS camera (HexaBlu™ MWIR) approaches this limit.

The system MTF can be cascaded by simply multiplying MTF_gptX MTF_del. Figure 4.28 summarizes different behavior of the system MTF. The transition region can be further split by setting the optics cutoff frequency to equal the detector cutoff frequency, resulting in an FAId= 1.0 [62]. When FA/d= 1.0, the spot size equals 2.44 times the size of the pixel. The “optics- dominated” region lies between the diffraction-limited region and this curve, whereas the “detector-dominated” region is located between this curve and the detector-limited curve. In the optics-dominated region, changes to the optics have a greater impact on the system MTF than the detector, and similarly for the detector-dominated region. Historically, most systems have been designed to have a resulting optics blur (to include aberrations) of less than 2.5 pixels (~ FA/d< 1.0). This is, of course, very dependent on the application and range requirements.

FIGURE 4.28 System MTF curves illustrating the different regions with the design space for various FAId conditions. Spatial frequencies are normalized to the detector cutoff (adapted after Ref. [62]).

Table 4.3 provides the required//# for FXId=2 for various detector sizes. As is shown, with//l optics, the smallest useful detector size is 2-pm in the MWIR and 5-pm in the LWIR. With more realistic//1.2 optics, the smallest useful detector size is 3-pm in the MWIR and 6-pm in the LWIR.

Host and Driggers [49] have considered the influence of optics infrared design on range approximation, which is illustrated in Fig. 4.29. When the IR system is detector-limited, decreasing the detector size has a dramatic effect on the range. On the other size, in the optics-limited region, decreasing the detector size has minimal effect on range performance. The acquisition range is reduced when atmospheric transmission and the NEDT are included.

As has been indicated by Kinch [37,63], challenges that must be addressed in fabrication of small-pixel FPAs concern:

• • Pixel delineation,
• • Pixel hybridization,
• • Dark current, and
• • Unit cell capacity.

TABLE 4.3 Required//# to = 2.0. Real Optics Usually Has//#> 1 (after Ref. [49])

The above topics are considered in more detail in [64].

Unlike visible sensors, where the pixel size has been reduced to about l-pm, the scaling of infrared pixels is much more difficult. The detector pixels are generally connected to the per-pixel electronics through indium bump bonds. As the pixel size is reduced, “bump-bonding”, ROIC, signal- integrating capacitor, and signal-to-noise ratio all become more difficult.

While bump-bond pitches of 12-|tm or larger are relatively common, pitches of less than 8-pm offer significant challenges in terms of array yield, pixel operability, and cost. It is expected that developments will

FIGURE 4.29 Relative range as a function of FA/d.

continue to extend bump-bonding technology to smaller pixels, as well as to improve manufacturability, and to reduce cost at all pixel sizes. In the past, much of the processing done on a focal plane was confined to the two-dimensional (2D) real estate directly under a given pixel. One novel research area has been the development of 3D integration technologies, providing alternatives to bump bonding [65]. Recently as many as three layers of CMOS have been stacked and vertically interconnected, offering the potential to increase the amount of processing available within a pixel footprint.

In conventional analog ROIC technology, the photocurrent generated by a detector is accumulated and stored locally in a capacitor (electron well); the maximum charge stored during an integration time is equal to the product of the total capacitance and the maximum allowable voltage across the capacitor. A simplified ROIC unit cell (pixel) circuit diagram and a typical ROIC layout are presented in Fig. 4.30. As indicated in this figure, the integrating capacitor dominates the unit-cell area usage. Thanks to multiple gain selection at the pixel level, the charge capacity can be tuned to match the targeted applications and scenario with maximum signal-to-noise ratio (SNR) and dynamic range. The gain and integration time are set, depending on the application requirements (//#, motion blur,

FIGURE 4.30 (a) Analog ROIC architecture (b) with simplified unit-cell or pixel circuit diagram and (c) unit-cell layout. As shown in (b), photocurrent Ij_et is generated by the photodiode and subsequently integrated onto the capacitance C,„, through the injection transistor M,, which also provides the photodiode bias Vh~ The signal voltage across C,„₍ can be switched onto the multiplexor bus for readout via control signal ф_$е, on switch M_sel; the signal voltage across C_inl can be reset by control signal ф_п[ using switch M_rst. A maximum 2.2-volt process (set by yj and C_jnl equal to 1850 femtofarads results in a maximum stored photocharge equal to 25 million electrons. Note that circuit capacitance as shown in (c) dominates the pixel footprint (adapted after Ref. [66]).

IR radiation flux), mainly concerning the detection range as well as the detection performance under various weather conditions.

With small pixel pitch, the photon flux is reduced and the integration time is generally increased. As a consequence, the long-range devices involving lower instantaneous field-of-view (IFOV) are often limited by the motion blur, a consequence of holder movement during the integration time. In this case, the impact on range is kept low with a short integration time (2-5 ms). Short integration times are also very useful to freeze a scene with rapidly moving objects. On the other hand, integration time could be increased (10 ms) with a stabilized system independent of the platform’s maneuvers and vibrations.

To achieve high sensitivity (e.g. below 30 mK), LWIR FPAs with 5-pm pixels require large amounts of integrated charge to be accommodated in very small unit cells. State-of-the-art storage density of conventional ROIC design rules are about 2.5 X 10⁴ e/pm². It can provide high-sensitivity values (say <30 mK) for most tactical MWIR and LWIR applications, utilizing pixel dimensions of 12-pm [62]. For a 5-pm planar unit cell, the charge capacity in standard ROIC technology is less than 1 million electrons, whereas 8 to 12 million electrons are required for high sensitivity. Therefore, small pitch IR detectors are not available today.

The challenge of charge storage in small pixels is being addressed by fabricating microelectromechanical systems (MEMS) capacitors suited to a 3D ROIC design. In recent years, Si-foundries have developed the process technology charge-handling capacities for 2 fF/pm² density metal- insulator-metal (MIM) capacitors in a 0.18-pm CMOS platform [66]. Multiple stacking of MIM capacitors results in higher densities of up to ~7 fF/pm². A redesign of the AIM Infrarot-Module GmbH AIM 640 X 512

15- pm pitch ROIC, in 0.18-pm Si-CMOS technology with MIM capacitor technology, causes improvement of NEDT values in comparison with standard ROIC (Fig. 4.31) [67].

Lincoln Laboratory has developed a digital pixel FPA with per-pixel,

16- bit full dynamic range, analog-to-digital conversion, and real-time digital image-processing capability [68]. Developed and tested LWIR ROIC MEMS technology overcomes many performance and scaling limitations imposed by conventional ROIC technology. The MEMS capacitor array can be fabricated in a separate 8-inch (200 mm) wafer. This technology yields 20 million electrons in a 5-pm unit cell. This breakthrough will pave the way for small-pitch FPAs to operate with very high sensitivity. Figure 4.32(b) shows a transmission electron micrograph picture of

FIGURE 4.31 Measured NEDT versus charge-handling capacity for LWIR 640x512 HgCdTe 15-pm pitch modules (adapted after Ref. [67]).

FIGURE 4.32 (a) Schematic illustration of the 3D integrated LWIR FPA design and (b) micrograph of the MEMS capacitor array cross section (adapted after Ref. [68]).

a portion of the MEMS capacitor array. Using the HDVIP technology, a fully functional 1280x720, 5-pm unit cell LWIR HgCdTe FPA has been demonstrated.

To maintain or increase the unit cell’s dynamic range will require employment of increasingly deeply scaled, higher-density CMOS processes. Higher-density CMOS fabrication processes can also be used to increase in-unit cell processing capacity. The extrapolated unit-cell transistor count, shown in Figure 4.33, suggests the feasibility of advanced in-pixel signal processing in smaller pixels within the next decade. As is predicted in [66], sophisticated in-unit cell processing, coupled with interpixel data communication and control structures, would enable massively

FIGURE 4.33 (a) The circuit transistor density and trend line for state-of-the-

art commercial microprocessors plotted versus the year of introduction of each microprocessor to the marketplace, (b) The maximum number of transistors that can be packed into a pixel unit cell as a function of the circuit transistor density for several pixel sizes between 5 and 30 pm. A magnitude estimate of the number of transistors required to achieve three levels of digital processing within the unit cell is also indicated. By leveraging deeply scaled CMOS processes, digital FPA technology enables designers to miniaturize pixel pitch and/or increase on-chip processing capability, depending on application-specific needs (adapted after Ref. [66]).

FIGURE 4.34 Required well capacity as a function of cutoff wavelength for BLIP operation at room temperature in the diffraction limit for a 30-Hz frame rate (adapted after Ref. [69]).

parallel computational imagers and resultant sensor systems with capabilities far beyond what is achievable today.

The ultimate desired goal of ROIC for photon detectors is to achieve BLIP operation at room temperature. It requires that the charge from the full 2n FOV, integrated during a frame time, should be effectively stored on the detector node, even as the pixel pitch is being decreased. Figure 4.34 illustrates the required well capacity for a 5-pm detector pixel as a function of the cutoff wavelength for BLIP operation at room temperature in the diffraction limit. The figure indicates that a significantly larger well capacity per ,um² is required to achieve BLIP performance at room temperature, compared with the values available with current ROIC designs.