Infrared （IR） focal plane arrays （FPA） with continually improved performance are widely used to detect the radiance intensities of target scenes^［1-2］. Based on the two-dimensional spatial heat distributions，thermal imagers distinguish the target from the background through temperature differences. However，conventional radiance intensity imaging has limited recognition capabilities in complex scenarios with interferences like clouds，dust，and tree shadows，since the target-background contrast tends to be unclear^［3-5］.

For this issue，infrared polarization imaging has emerged as a viable solution. As a fundamental property of light，polarization reveals more intrinsic characteristics of the imaged object，such as surface features，shapes，and roughness^［6-8］. Using the distinct polarization signatures between manmade and natural objects，the target-background contrast can be effectively enhanced even under thermal equilibrium conditions. The target profiles can thus be highlighted and more details can be extracted^{［3，9-10］}. In order to acquire high-quality polarimetric images，it is urgent to develop high-performance polarimetric IRFPAs. Mercury Cadmium Telluride （HgCdTe） is considered to be an ideal material for its superior sensitivity and quantum efficiency^［2］.

Among various polarimeter configurations^［11］，thanks to the rapid development of micro-nano technologies，micro-polarizer array （MPA） based on division-of-focal-plane （DOFP） architecture has attracted extensive attention^［12-14］. With less hardware redundancy and smaller occupation，MPA enables miniaturized，integrated，and compact polarimetric imaging systems. The MPA is usually discretely or monolithically integrated with the FPA. For discrete integration，although the MPA fabrication process is easier，additional bonding or gluing steps are required. Such steps increase the FPA-MPA separation，resulting in more significant diffractions. The crosstalk effects can thus be severer，which leads to reduced extinction ratios （ER）^［15-16］. For instance，a discretely integrated medium wave （MW） 256×256 HgCdTe polarimetric FPA using gluing process exhibited moderate ERs around 5^［17］. A long-wave （LW） 640×480 HgCdTe polarimetric FPA implemented by Indium bump bonding process possessed ERs less than 10^［18］. For monolithic integration，with MPAs directly fabricated on the FPA surface，the diffractive crosstalks can be effectively suppressed，which enables higher performance upper limit. For example，with MPAs directly fabricated on chip，both MW and LW 256×256 HgCdTe polarimetric FPAs achieved ERs exceeding 10^［19］. Nevertheless，it is still a bottleneck to meet the stringent process compatibility requirements posed by HgCdTe FPAs while attaining desirable performance.

In addition to integration technologies，the wave band should also be carefully selected. Compared with short wave （SW），MW and LW have significant advantages in night operation^［5］. In this work，considering the manufacturing cost and technical difficulties of FPAs，the MW 640×512 （25 μm） HgCdTe FPA was chosen. The structural parameters of line gratings as well as the MPA layout have been designed to keep the balance between the performance and fabrication feasibility. An on-chip line grating fabrication process compatible with HgCdTe FPAs was exploited. Comprehensive tests on the performance of fabricated polarimetric FPAs have been conducted. In addition to performance tests，the crosstalk effect was also theoretically analyzed and experimentally characterized，which provided a better understanding about this mechanism that dominated ER.

The schematic of the proposed monolithically integrated HgCdTe polarimetric FPA is shown in Fig. 1（a）. The MPA consists of periodically distributed superpixels composed of 2×2 arranged micro-polarizers. Each pixelated micro-polarizer was formed by sub-wavelength metal line gratings. Within a superpixel，the polarization orientations of the micro-polarizers follow as a clockwise order of 0°，45°，90°，and 135°，respectively，as shown in Fig. 1（b）. For an infrared radiation transmitting through the superpixel，its intensity is modulated by quad line gratings of different orientations，correspondingly generating four response signals：I₀，I₄₅，I₉₀，and I₁₃₅. With these four measured response signals，the Stokes parameters of the incoming radiation can be calculated via the following equations^［20］：

The degree of linear polarization （DOLP） as well as the angle of polarization （AOP） can thus be calculated via the following equations：

It should be noted that the relationship between the calculated and the actual DOLP values takes the form as^［21］：

where DOLP_act denotes the actual DOLP. Clearly，the higher the ER，the more accurate the measured DOLP. For this purpose，a finite-different time-domain （FDTD） model was established to provide guidance on line grating optimizations. The boundary conditions as well as the meshing size have been properly set，as shown in Fig. 1（c）.

The structural parameters of line gratings，i.e.，thickness，period，and duty cycle，were initially set as 300 nm，800 nm，and 50%，respectively. The variation tendencies of ER with respect to a certain parameter were simulated with the other two parameters maintained unchanged. As can be seen from Fig. 2（a），when the thickness increases，ER tends to be higher. However，too thick line gratings are not practical，since the re-deposition effect existing in the etching process can be detrimental to ER as well as the transverse magnetic （TM） wave transmittance. Figure 2（b） illustrates that ER increases with the reduced grating period，nevertheless，an excessively shrunk grating period will pose a challenge to the fabrication process. Regarding to Fig. 2（c），it can be inferred that ER rises rapidly with an improved grating duty cycle，however，simply improving the duty cycle can degrade the TM transmittance. To keep the balance between the process feasibility and polarization selectivity，the thickness ranging from 300 nm to 500 nm and period from 600 nm to 800 nm are suggested，while the duty cycle is kept as 50%.

It should be addressed that the above-mentioned model corresponds to an ideal case that the line gratings are infinitely distributed，wherein the crosstalk has not been involved. Nevertheless，for pixelated micro-polarizers，the distributed areas of line gratings are limited. Therefore，the crosstalks are nonnegligible，which results in significant ER reductions^［16］. The crosstalk mechanisms can be categorized as two types：optical and electrical^［21］. The optical crosstalk can be caused by the misalignment or diffraction. For the former one，if a micro-polarizer spans two pixels due to the MPA-FPA misalignment，the crosstalk occurs. The misalignment induced crosstalk can be effectively avoided by improving the alignment precision and setting opaque regions at the pixel edges. Regarding to the diffraction induced crosstalk，it appears when the radiation diffracts through the micro-polarizer aperture. The radiation propagates through the MPA-FPA separation，and ultimately impinges adjacent pixels^［21］. Obviously，this kind of crosstalk tends to be more pronounced when the MPA-FPA separation increases^［16］. In our previous work，with the entire superpixel modeled，the variation tendency of ER with respect to the MPA-FPA separation has been clarified^［22］. With MPA directly fabricated on the FPA，the crosstalks caused by diffractions can be effectively suppressed.

The electrical crosstalk can be attributed to photo-generated carriers migrating from one pixel to the adjacent one. When the opaque regions at the pixel edges are expanded，the photo-generated carriers can be confined within the pixel central regions，thereby alleviating electrical crosstalks.

Although expanding opaque regions is effective for suppressing both optical and electrical crosstalks，this method could result in severe transmittance degradation. Therefore，a trade-off is necessary. Herein，two types of MPA layouts with different configurations of opaque regions are designed，as shown in Fig. 3. The first one，referred to as MPA layout A，follows as the design reported in our previous work，wherein the vertical and horizontal micro-polarizers have wider opaque regions than the diagonal ones^［22］. When the opaque region widths are the same，due to the inconstant line grating lengths，the diagonally oriented pixels have inferior polarization selectivity. Therefore，the opaque region widths of differently oriented micro-polarizers were modified to balance ERs of differently oriented pixels. Another design is referred to as MPA layout B. Regardless of orientations，for all the polarizers，the widths of the opaque regions were set as 4 μm to suppress crosstalks. To compensate for the reduced photosensitive area，the period of line gratings was properly scaled down. This design aimed at verifying if enlarging opaque regions can implement ER enhancements at the expense of tolerable transmittance reduction. The structural parameters of these two MPA layouts are listed in Table 1.

A simple and low-damage monolithic integration process was exploited to directly fabricate MPA on the surface of MW640×512（25 μm） HgCdTe FPA（Fig. 4（a））. The FPAs have been previously tested and the qualified ones were selected for the subsequent process.

Firstly，a 20 nm thick Ti layer and a 400 nm thick Au layer were successively deposited on the FPA by e-beam evaporation，serving as the structural layers （Fig. 4（b））. The Ti layer was employed to enhance the adhesion between the Au layer and the FPA. Subsequently，the MPA was patterned using the laser direct writing process （Fig. 4（c）） and etched using the ion-beam milling process （Fig. 4（d））.

During the deposition，the temperature of HgCdTe FPA should be strictly controlled below 80 °C. The e-beam evaporation，magnetic sputtering，and ion beam sputtering processes can all meet this requirement. All the Ti/Au layers attained by these three types of processes remain intact during the entire process flow. Given that keeping the continuities of the patterned line gratings is extremely important for attaining expected ERs，the qualities of metal layers were evaluated from the perspective of roughness. As can be clearly seen from Fig. 5，the e-beam evaporation allows for Au layers with the minimal roughness，which is selected as a preferable deposition technique in this work.

For sub-wavelength line gratings，the lithography process is a crucial step which determines if the performance of the fabricated MPA addresses expectations. Herein the maskless laser direct writing technique was employed to achieve both flexible MPA layout modifications and line width down to 300 nm. The laser intensity and the focal length are two critical parameters. To determine the optimal process condition，a two-dimensional test matrix was employed in this work. Within the matrix，each “element” had the same pattern involving a group of line grating arrays with various duty cycles and line widths，whereas each row and column corresponded to a different laser intensity and focal length，respectively. Therefore，various combinations of exposure parameters together with their corresponding exposure effects were obtained. Consequently，the optimal parameters were determined，and the corresponding MPA pattern is given in Fig. 6. Another key indicator of lithography is referred to as misalignment. By properly designing the alignment mark and carefully calibrating the equipment，the misalignment down to ±1 μm has been realized，which can be totally covered by the opaque regions at the pixel edges. Therefore，the crosstalk led by misalignment can be effectively suppressed.

As for etching，ion milling featuring ion bombardment is a physical dry etching process，making it appropriate for etching multi-layer metals^［23］. It should be noted that the particles of target materials are sputtered at different angles under ion bombardment^［23］. The sputtered particles with large angles relative to the normal direction tend to redeposit onto the sidewalls of the readily etched patterns. This redeposition is deleterious to the etching profile，especially for subwavelength structures^［24］. As can be seen from the cross-sectional profile shown in Fig. 7（a），with process conditions generally used for micrometer-scale patterns，i.e.，normal incident of ion beam，low acceleration voltage，and high ion density，strong redeposition effect occurred and the line grating gaps were nearly filled up. To solve this issue，modifications such as increasing the incident angle of ion beam，raising the acceleration voltage，and reducing the ion density have been carried out，which greatly alleviated the redeposition effect，as shown in Fig. 7（b）.

To sum up，through a series of process flow optimizations，the MPA was successfully fabricated on the FPA surface，and the damage was controlled. Figure 8 gives the scanning electron microscope （SEM） photographs of the fabricated MPA.

The performance of polarimetric FPA assemblies was characterized using the test setup as shown in Fig. 9. The black body served as a source of signal irradiation. The rotational polarizer with ER over 500 was utilized to modulate the unpolarized irradiation as the polarized one with various polarization angles. A customized test printed circuit board （PCB） was connected to the electrical interface of the assembly. The measured data could thus be transferred to the computer so that the performance indicators could be extracted.

The fabricated polarimetric FPAs with MPA layout A and B were encapsulated in standard test Dewars. The radiometric performance of polarimetric FPAs was firstly measured. In this case，the polarizer was not needed. For both two types of polarimetric FPAs，pursuant to GB/T 17444-2013^［25］，the tests were performed at the liquid nitrogen temperature under the half-well integration capability filling condition，wherein the background and target temperatures were set as 293 K and 308 K，respectively. For both two types of polarimetric FPAs，the measured data before and after monolithic integration were summarized in Table 2.

In ideal conditions，after MPA fabrication，the integration time should be twice as long as that before，since half of the unpolarized radiance intensities are reflected by the line gratings. However，due to limited line grating distribution areas together with the fabrication tolerances，there are acceptable divergences between the measured integration time and the ideal one. In addition，after the MPA fabrication，the integration time of B-type polarimetric FPA only slightly extended compared to that of the A-type one，indicating that the energy loss caused by the expanded opaque regions was not severe.

After the MPA fabrication，the response signal and noise did not worsen，indicating that the FPA damage induced by the process was controllable. According to GB/T 17444-2013，the peak detectivity can be determined via^［25］：

where G is the spectral factor，M and N are referred to as the FPA rows and columns，d and h are referred to as the dead and overheat pixels，and the sum of d and h represents the total blind pixels of the FPA. A_D and τ denote the pixel area and the half-well integration time. V_N（i，j） represents the noise of a certain pixel，and R（i，j） referred to as the responsivity can be calculated using the following formulas^［25］：

where V_S（i，j） denotes the response signal of a certain pixel，σ is referred to as the Stepan’s constant，T₂ and T₁ are the target and background temperatures，respectively. According to Eqs. （4-6），after the MPA fabrication，the peak detectivity is supposed to reduce due to the increased integration time. The measured results were in accordance with the theoretical analysis.

Regarding to NETD，it takes the form as^［25］：

Similarly，the NETD variation tendencies can be inferred according to the response signals and noises before and after the MPA fabrication. For both two types of polarimetric FPAs，the measured results matched with the expectations.

What’s more，the increased response nonuniformities for both two types of polarimetric FPAs can be mainly attributed to the ion beam etching. To improve the uniformity，the key point is to have a nearly maintained etching rate in both central and peripheral regions of the FPA. As for the effective pixel rate，upon blind pixel distribution diagrams as given in Fig. 10，for both two types of polarimetric FPAs，most blind pixels were concentrated on the FPA edges. Such a phenomenon is caused by the thick photoresist accumulated there. Faced with this issue，we have made some preliminary attempts，including modifying the spin-coating parameters of the photoresist and increasing the exposure energy at the FPA edges. With blind pixels remarkably reduced，the effective pixel rate has raised up to 98.23%，as shown in Fig. 11. To have more substantial progress，further research is ongoing.

Above all，despite expanded opaque regions，by taking appropriate compensation strategies such as reducing the line grating period，the B-type polarimetric FPA exhibited comparable radiometric performance as the A-type one.

With the polarizer incorporated into the test setup，the response signals under different polarizer rotation angles can be attained. The strongest and weakest signals，generated by TM and transverse electric （TE） incidents，are referred to as V_TM and V_TE，respectively. The ER can thus be determined via the following equation：

For both two types of polarimetric FPAs，the polarization response curves of differently oriented pixels are plotted in Figs. 12 and 13，respectively. The extracted ERs were listed in Table 3. Sinusoidal fittings have been carried out，which inferred that the measured data matched well with Malus’ Law. It is evident that the B-type polarimetric FPA possessed significantly higher ERs than those of the A-type one. It is worth mentioning that for the B-type polarimetric FPA，the ER of 0°-oriented pixels was 7.26，which was lower than those of pixels with other orientations. It was found that the cycle duty of 0°-oriented line gratings was slightly reduced，which could result from the nonuniformities induced by the laser direct writing process.

In regard to the B-type polarimetric FPA，we have further optimized the entire MPA fabrication process and have formally encapsulated it in the standard metal micro-Dewar. As shown in Fig. 14，the measured ERs ranged from 17.32 to 27.41，which were greatly enhanced compared to those measured in the test Dewar. This phenomenon was caused by different transmittance ranges between the test and formal Dewars. For the test Dewar，due to the absence of optical filter，the transmittance range was 1-5 μm. For incident lights with shorter wavelengths，the line gratings should be shrunk to the corresponding subwavelength scale to ensure sufficient polarization selectivity. However，regarding described MPA designed for 3.7 - 4.8 μm，the line grating period was too large for shorter wavelengths，thus leading to the reduced overall polarization selectivity over such a wide transmittance range. By contrast，for the formal Dewar，the optical filter was involved，and the transmittance range of 3.7 - 4.8 μm matched well with the designed MPA，thus achieving much higher ERs. Table 4 compares the described polarimetric FPA with the reported ones. The performance of this work is superior.

Above all，it has been experimentally validated that expanding opaque regions at the pixel edges is an effective methodology to improve ERs while maintaining promising radiometric performance. The ER differences of pixels with various orientations indicate that the uniformity of the whole monolithic integration process needs to be further improved，which will be a key focus in the future work.

It is critical to have an insight into the dominant mechanism that limits ER so that further optimizations can be conducted to make greater performance breakthroughs. To figure out whether it is the crosstalk that leads to significant ER differences between two types of polarimetric FPAs，specialized MPA layouts have been designed to quantitatively characterize the crosstalk rates from adjacent pixels.

For instance，as shown in Fig. 15（a），the 0°-oriented pixels were all covered by Au layer. For these pixels，the signals generated by external radiations have been eliminated，and the measured ones are mostly contributed by the crosstalks from adjacent pixels. For pixels with other orientations，the crosstalk signals from their adjacent pixels can also be extracted using the similar layout designs.

The signal response of the pixel oriented in a certain polarization angle can be expressed as^［27］：

where P denotes the degree of polarization for the polarizer，V_{pi（i=1，…，4） represents the response signal of the corresponding pixel before MPA fabrication，qi and ri （i=1，…，4） are referred to as TM and TE transmittances，respectively，α is the polarization angle of the incident light. In this equation，121-Pqi+riVpi represents the response signal generated by the unpolarized portion of the incident lights transmitted through the polarizer，and Pqisin2θ-α+ricos2θ-αVpi,(θ=0,π4,π2,3π4) represents the response signal generated by the polarized portion of the incident lights transmitted through the polarizer.}

Taking the 0°-oriented pixels as an example，when these pixels were covered，the measured crosstalk signals from the adjacent pixels can be written as：

where χ denotes the crosstalk rate. In Eq. （10），the former and latter terms represent crosstalk signals contributed by the left and right 45°-oriented pixels，and upper and lower 135°-oriented pixels，respectively. The weak crosstalks from diagonal pixels were neglected. In ideal conditions，V_p2 and V_p4，q₂ and q₄，r₂ and r₄ are equivalent，therefore，Eq. （10） can be simplified as：

In this case，the sum of crosstalk signals from adjacent pixels is a constant independent of the polarization angle.

However，the measured signals of the covered pixels appeared as sinusoidal curves，as shown in Fig. 16. The discrepancies can be ascribed to the response nonuniformities existed in the FPA as well as the polarization nonuniformities induced by the MPA. Such nonuniformities led to different V_pi，q_i，and r_i values. In this case，according to Eq. （10），the crosstalk signals sinusoidally varied with the polarization angle. The sum of the response signals from adjacent pixels is also plotted in Fig. 16（a）. After fitting these two curves，it could be observed that their phases and periods were consistent. By calculating the ratio of these two groups of data，the crosstalk rate of 3.48% can be determined. Since both 0°- and 90°-oriented pixels are surrounded by two pairs of perpendicularly oriented pixels （45° and 135°），the depicted calculation process for 0°-oriented pixels is equivalent for 90°-oriented ones. As for 45°/135°-oriented pixels，the MPA layout as shown in Fig. 15（b） is utilized，and the measured and fitted results are shown in Fig. 16（b），by which the crosstalk rate is determined as 4.54%.

For the B-type polarimetric FPA，the same methods were adopted. Based on the measured data and fitted curves as shown in Fig. 17，the crosstalk rates from adjacent pixels of 0°/90°- and 45°/135°-oriented pixels were calculated to be 1.70% and 2.06%，respectively.

For the A-type polarimetric FPA，when the imposed crosstalk signals were detracted，as shown in Fig. 18，the ERs of 0°- and 45°-oriented pixels recovered to 16.37 and 17.94，respectively. For the B-type polarimetric FPA，using the same method，the ERs of 0°- and 45°-oriented pixels recovered to 17.22 and 25.97，respectively. Obviously，for both two types of polarimetric FPAs，the re-extracted ERs became closer，strongly inferring that it was their distinct crosstalk rates that led to their significant ER differences. As previously discussed，the lithography misalignment of ±1 μm can be covered by the opaque regions. Besides，with MPA directly fabricated on the FPA surface，the diffraction induced crosstalks have been suppressed. Therefore，the optical crosstalks have been essentially minimized. It is reasonable to believe that the electrical crosstalk dominates ER. Furthermore，it should be noted that after re-extractions，for two types of polarimetric FPAs，there still are noticeable ER differences between their 45°-oriented pixels. This could be caused by process tolerances among different fabrication batches.

Above all，expanding opaque regions up to 4 μm can effectively suppress the crosstalk rate. Even so，simultaneously setting all the opaque region widths as 4 μm could still be a preliminary design. For pixels with various orientations，it would be a more flexible method to accordingly expand their opaque regions to different extents，which is expected to implement more balanced ERs.

In summary，this work developed a monolithic integrated MW HgCdTe polarimetric FPA with remarkable radiometric performance and polarization selectivity. A complete flow including design，fabrication，and test was comprehensively described，which would be beneficial for facilitating the practical applications of HgCdTe polarimetric FPAs.

At design stage，systematical optimization for MPA layout has been carried out. Not only the structural parameters of line gratings，but also the crosstalks have been considered.

The feasibility and compatibility of proposed monolithic integration process have been testified，which achieved favorable ERs ranging from 17.32 to 27.41. More efforts in improving non-uniformity will be taken.

The crosstalk mechanism has been deeply discussed，which provided a helpful guidance on MPA designs. It was verified that properly expanding opaque regions implemented substantial ER enhancements at the cost of very limited radiometric performance scarification. With more flexible opaque region settings，the overall performance of the proposed polarimetric FPA is expected to be preferable.