From 2D to 3D: Strain- and elongation-free topological transformations of optoelectronic circuits

Significance

We demonstrate a general method to transform planar electronic and optoelectronic devices fabricated by conventional photolithography into a strain-free but topologically different geometry. The method is used to demonstrate hemispherical, retina-like imagers whose pixel spacings are unaffected by the topological transformation. Our approach overcomes the critical limitation of distortion encountered in transforming a planar circuit to a nondevelopable, 3D shape. The process opens up additional possibilities for making a variety of electronic circuits that conform to randomly shaped surfaces, including high resolution, and large field of view imagers that have the shape and form factor of the human eye.

Abstract

Optoelectronic circuits in 3D shapes with large deformations can offer additional functionalities inaccessible to conventional planar electronics based on 2D geometries constrained by conventional photolithographic patterning processes. A light-sensing focal plane array (FPA) used in imagers is one example of a system that can benefit from fabrication on curved surfaces. By mimicking the hemispherical shape of the retina in the human eye, a hemispherical FPA provides a low-aberration image with a wide field of view. Due to the inherently high value of such applications, intensive efforts have been devoted to solving the problem of transforming a circuit fabricated on a flat wafer surface to an arbitrary shape without loss of performance or distorting the linear layouts that are the natural product of this fabrication paradigm. Here we report a general approach for fabricating electronic circuits and optoelectronic devices on nondevelopable surfaces by introducing shear slip of thin-film circuit components relative to the distorting substrate. In particular, we demonstrate retina-like imagers that allow for a topological transformation from a plane to a hemisphere without changing the relative positions of the pixels from that initially laid out on a planar surface. As a result, the resolution of the imager, particularly in the foveal region, is not compromised by stretching or creasing that inevitably results in transforming a 2D plane into a 3D geometry. The demonstration provides a general strategy for realizing high-density integrated circuits on randomly shaped, nondevelopable surfaces.

Optoelectronic circuits shaped onto 3D surfaces can offer additional functionalities inaccessible to conventional planar electronics based on 2D geometries that are routinely fabricated by conventional photolithographic patterning processes. A light-sensing focal plane array (FPA) used in imagers is one example of a system that can benefit from fabrication on curved surfaces. By mimicking the hemispherical shape of the retina in the human eye, a hemispherical FPA provides a lightweight, compact imaging system with a low-aberration image, a low f number, and a wide field of view (1–8). Other examples include head-up and virtual reality displays that are often needed on curved or folded surfaces (9), wearable sensing devices (10–12), and conformal light absorption modules (13–16). Due to the inherently high value of these applications, intensive efforts have been devoted to solving the problem of transforming a circuit fabricated on a flat wafer surface into one of arbitrary shape without loss of performance or distorting the linear layouts that are the natural product of this fabrication paradigm.

Fig. 1A illustrates the deformation of a developable circuit from a flat to a cylindrical surface, representing a transformation between topologically equivalent surfaces. This transformation maintains a constant relative distance between two arbitrary fixed points on the surface (e.g., points 1 and 2) (17). Optoelectronic components fabricated on flexible substrates can be trivially bent to become functional devices on these developable surfaces (4, 18–21). The topological transformation from a developable to a nondevelopable surface, however, is a more general type of distortion that can morph a plane into a random 3D shape. Fig. 1B illustrates circuits fabricated on a stretchable plane (e.g., a deflated balloon) transformed into a nondevelopable spherical surface (an inflated balloon). This topological transformation results in a change in the relative distance between two arbitrary fixed points on the surface (points 1 and 2 in Fig. 1B) (17). Electronic components fabricated on brittle semiconductors attached to the deformed surface suffer from strain that may eventually lead to structural damage (10, 22). Furthermore, the increased distances between points can lead to loss of resolution in pixel arrays located on the initially flat surface.

Fig. 1.

Schematic illustration of a developable deformation vs. nondevelopable deformation process. (A) Circuits fabricated on a flexible plane and deformed into a developable semicylindrical shape that does not entail a topological transformation. The distance between points 1 and 2 along the surface remains the same after the deformation. (B) Circuits fabricated on a stretchable plane (a deflated balloon) and deformed into a nondevelopable, topologically distinct spherical shape (an inflated balloon). The distance between points 1 and 2 along the surface is dramatically increased after the deformation.

One extensively studied system that benefits from being shaped into a nondevelopable hemispherical architecture is the image-sensing FPA. It offers significant benefits if a retinal shape can be achieved without changing the interpixel spacing that results in loss of image resolution or image distortion. The retina is the nearly hemispherical light-sensitive layer on the back of the human eye on which an image is focused by the lens (23). In contrast to the shape and size of the retina, high-performance photodetector FPAs employed in modern cameras are flat due to limitations of conventional photolithographic fabrication. The imperfect match between planar FPAs and image planes using only a single-element convex lens such as that in the human eye results in a degraded image with a limited range of focus, a narrow field of view (FOV), and off-axis optical aberrations (2, 3, 24, 25). Consequently, additional optical elements are required to correct these aberrations that increase the complexity, weight, and cost, while often decreasing the functionality of the imaging system. Many efforts, therefore, have been made to shape the FPA into a hemisphere (1, 4, 5, 8, 19, 22, 26–29). Fabricating arrays on retina-like hemispherical surfaces (5, 22, 30, 31), however, introduces significant challenges. For example, thinning and deforming commercial complementary metal-oxide-semiconductor (CMOS) imagers (1, 31) (with integrated addressing circuits) provide a high pixel count, although the curvature must remain small to avoid the significant mechanical strain or distortions such as creasing or folding. Changes in pixel separation that must be corrected to avoid image artifacts and resolution loss associated with strain are also unavoidable. Larger deformations from a plane to a hemisphere have been achieved by placing bendable and stretchable metal interconnection “bridges” between pixels that relieve strain to create both concave (5) and convex (6) imagers. However, the gaps between pixels reserved for the bridges result in a loss of resolution, particularly near the central “fovea” at the point of maximum strain. Recently, origami-inspired hemispherical FPAs were reported (8, 27, 32) with high deformability and pixel counts that were achieved by cutting, folding, and mating sections to form an approximately hemispherical shape. This process does not result in a perfect conformation to a hemisphere, leading to undesirable optical aberrations and image stitching errors.

In this work, we overcome these deficiencies by employing well-established optoelectronics processing techniques to form a thin film, GaAs FPA on planar, flexible plastic foils. The hemispherical FPA (HFPA) is then achieved by transferring to an elastomeric handle and then allowing the circuits to shear and slip along the elastomeric surface during distortion, a method first introduced in making organic thin-film detector arrays (22). Specifically, a 15 × 15 thin-film GaAs photodiode FPA was fabricated on a flexible Kapton foil via cold-weld bonding (33) and subsequently nondestructively epitaxially lifted off (ND-ELO) (34, 35) from its parent (growth) substrate. The flexible FPA, attached to an elastomeric transfer handle with rows of detectors separated by plasma etching, is then deformed into a hemispherical shape that allows for shear slippage between the elastomer and the array surface and then transferred to a mating concave hemispherical substrate to achieve the HFPA. The HFPA shows nearly perfect fabrication yield (∼99%) and an external quantum efficiency (EQE) >80% between wavelengths of 650 nm and 900 nm. Moreover, the noise performance and detectivity are both comparable to those in commercially available charge-coupled detector (CCD) imagers (36). Note that the fabrication strategy is independent of the semiconductor materials choice and can achieve the same high pixel density on almost any arbitrarily shaped surface with a continuous first and second derivative, as on a planar surface.

Results and Discussion

Fig. 2 illustrates key steps in fabricating the HFPA. In Fig. 2A, a passive matrix, GaAs p-n junction photodiode array is fabricated on a flexible, 25-μm-thick E-type Kapton substrate. Details of the array fabrication process are described in ref. 4 and in Methods. Photodiode mesas on the array are connected only in rows, whereas the column connections are not patterned at this point (see SI Appendix, Fig. S1 for details). An aluminum etch mask with a pattern matching the rows of detectors on the mesa surface is formed on the back side of the Kapton substrate. Separately, a 100-μm-thick poly(dimethylsiloxane) (PDMS) membrane is spun onto a Si handle pretreated with a release agent. The Kapton substrate with the detectors facing the membrane is then attached to the PDMS (37, 38). Next, the Kapton is etched through to the PDMS surface using O₂ plasma, and the Al mask is removed using Cl₂ plasma as shown in Fig. 2B. This step removes the Kapton substrate between the rows of detectors, i.e., separates a 2D array plane into individual 1D lines of detectors.

Fig. 2.

Schematic illustration of the key steps of fabricating a hemispherical photodiode array. (A) GaAs p-n junction photodiode array connected in rows fabricated on flexible Kapton substrate (brown) with an Al etch mask (light gray) patterned on the backside, is laid flat onto a poly(dimethylsiloxane) (PDMS) membrane (purple). (B) The Kapton substrate is etched through to the PDMS surface using O₂ plasma. The Al etch mask is removed using Cl₂ plasma. (C) The PDMS membrane that supports the array is fixed on its edges and deformed by a centered PDMS hemispherical punch. The array is transferred to a matching hemispherical concave glass lens coated with UV curable adhesive. (C, Inset) Cross-section views from the xz plane and the xy plane during the deformation process. Kapton substrate (brown) supports Au connection lines (yellow) and photodiode mesas (gray) when the PDMS membrane (blue) is stretched. Rows of pixels are free to move in the x direction and have shear motion with the PDMS membrane in the y direction. (D) Array (connected in rows) transferred to the concave glass lens.

The PDMS membrane that supports the array is released from its Si handle, fixed on its edges, and deformed by a centered, PDMS hemispherical punch, as shown in Fig. 2C. The PDMS membrane thus undergoes a topological stretching into a nondevelopable surface (17) despite significant strain (∼7% in the center and ∼20% toward the edge; SI Appendix, Fig. S2). The pixels, however, do not change their spacing during stretching. Fig. 2C, Inset shows cross-sectional views of the array and PDMS membrane in xz and yz planes. In the xz plane, detectors (gray) together with in-row connections (yellow) and the etched Kapton (brown) move freely along the x direction without longitudinal strain when the PDMS membrane (blue) is stretched. In the yz direction, however, the detectors and connections are constrained by the Kapton film, and hence they shear along the PDMS stretched in the y direction. The shear along the membrane surface is allowed without strain due to the weak adhesion at the detector/PDMS interface (SI Appendix, Fig. S3) (22).

Shear-slip motion on PDMS has previously been observed and characterized in both organic (22) and inorganic (39) semiconductor systems. The governing factor that enables the slip is that the strain energy release rate must exceed the interface bonding energy between the surfaces. For typical inorganic semiconductor/PDMS interfaces, the slip can occur for shear strains >7% (39). In addition, due to the high Young’s modulus of the 25-μm-thick Kapton film (∼10³ times higher than PDMS), the stress along the detector rows induced by PDMS stretching is well below the yield strength, and the strain in the thin film can thus be ignored. Generally, shear-slip motion and nondevelopable deformation are applicable to any circuit structure as long as the shear-induced energy release rate exceeds the interface binding between the circuit and the substrate transfer stamp, and the stress induced by PDMS stretching does not exceed the material yield strength of the circuit materials. It is worth mentioning that the relative positions of the top (light-absorbing) surfaces of detectors on a row do suffer minor shrinkage due to the bending of the Kapton film. More controllable geometries can be achieved by employing predistortion offsets of the pixel spacings during the fabrication on the planar surface to achieve the target pixel spacings after transfer.

Next, the deformed array is brought into intimate contact with a hemispherical concave glass substrate coated with a thin layer of UV-light curable adhesive. The radius of curvature of the glass indentation matches that of the PDMS punch. The adhesive is cured, the punch is withdrawn, and the PDMS membrane is peeled off to complete the transfer (Fig. 2D). The approach described in Fig. 2 transforms the 2D tensile strain introduced during deformation to a simple separation and 1D bending process. It maintains the pixel spacing before and after deformation in the y direction. In the x direction, a second layer of detector rows can be applied in the same manner to fill in the gaps that arise during application of the first layer during stretching. Finally, using these same fabrication steps, an array of metalized Kapton pads is patterned and transferred to the concave substrate to connect rows of detectors and simultaneously form the column connections (see SI Appendix, Fig. S1 for details). The approach described in Fig. 2 is compatible with batch fabrication of imagers (SI Appendix, Fig. S6) with many high-performance materials including, but not limited to, Si, GaAs, InGaAs, etc.

Fig. 3A shows a GaAs p-n photodiode HFPA fabricated on a truncated concave hemispherical glass substrate with a radius of curvature = 9.2 mm, depth = 2.5 mm, and opening diameter = 12.7 mm. The 15 × 15 pixel array is centered within the substrate depression, providing high-resolution foveal imaging capability. A secondary, 4 × 2 pixel array is located along the lip of the depression that is transferred at the same time as the central array. It provides peripheral, but low-resolution vision similar to that sensed by the human eye. Furthermore, its application demonstrates the ability to transfer devices at angles >43° to provide a very large FOV (40).

Fig. 3.

(A) Photograph of a 15 × 15-pixel GaAs p-n junction photodiode array fabricated on a concave hemispherical surface. Additional 4 × 2 peripheral pixels that allow for motion detection at wide angles of view are also shown. (B) Scanning electron microscopic image of a portion of the photodiode array. (C) Schematic of a single pixel in the array. (D) External quantum efficiency (EQE) spectra of the photodiode in the wavelength range from 400 nm to 900 nm. (D, Inset) Current-voltage (I-V) characteristics of the photodiode in the dark (blue line) and under 64-nW, 530-nm light-emitting diode (LED) illumination (orange line). (E) Histogram of dark current of photodiodes on the 15 × 15 FPA. (E, Inset) Normalized dark current maps of the 15 × 15 GaAs FPA on the hemispherical surface. (F) Photocurrent vs. input optical power of a single photodetector in the 15 × 15 FPA. Red line shows a linear fit to the photocurrent at low-input optical power. The minimum detectable power is about 10⁻⁴ W/cm², and the 1-dB compression point is at 0.1 W/cm², giving a 30-dB dynamic range and a 10-bit grayscale resolution.

The scanning electron microscopic image in Fig. 3B provides a detailed view of the pixels shown in Fig. 3A. No metal or semiconductor cracks are observed as typically encountered for free-standing metal films subjected to similarly substantial strain (22, 41, 42). Metalized Kapton pads between the pixels form top electrical connections that enable the column readout of the passive matrix HFPA. Lateral misalignment between rows is due to the asymmetric shear slippage during deformation and transfer. This issue can be resolved by designing the array with a compensating offset between rows during fabrication before deformation. The one-to-one mapping of the plane onto a hemisphere makes this compensation both predictable and accurate.

Fig. 3C is a schematic illustration of the photodiode pixel in the HFPA (Methods). The current-voltage (I-V) characteristics of a photodiode under dark and 64-nW illumination at a wavelength of λ = 530 nm are shown in Fig. 3D, Inset. The dark current is 1.3 ± 0.4 nA (corresponding to 7.4 ± 2.1 μA/cm²) at −1 V for the individual detector. The current under illumination is 18.5 nA at 0 V and 23.3 nA at −1 V. Fig. 3D presents the EQE spectrum of a photodiode. We observe EQE > 80% at λ > 650 nm, which to our knowledge is above those reported for other hemispherical imagers (5, 8, 22, 27, 32). The photodetector noise equivalent power (NEP) is $\text{NEP}=\sqrt{2q{I}_D}/R\left(\lambda \right)$ under shot-noise–limited detection at −1 V, where $q$ is the electron charge, $I_D$ is the dark current, and $R\left(\lambda \right)$ is the responsivity at a given wavelength $\lambda .$ With EQE = 67.7% at $\lambda$ = 530 nm, then $R\left(\lambda \right)=0.29\hspace{0.17em}\text{A}/\text{W},$ and $\text{NEP}=7.03\times {10}^{-14}\hspace{0.17em}\text{W}/{\text{Hz}}^{1/2}.$ The specific detectivity of the detector is $D*=\sqrt{A\mathrm{\Delta }f}/\text{NEP},$ where $A$ is its area, and $\mathrm{\Delta }f$ is the bandwidth, giving $D*=1.89\times {10}^{11}\hspace{0.17em}\text{cm}\cdot {\text{Hz}}^{1/2}\cdot {\text{W}}^{-1}$ in a 1-Hz bandwidth. The $\text{NEP}$ and $D*$ are at the same order of magnitude as that of commercially available CCD imagers (36).

The normalized dark current map in Fig. 3E, Inset and histogram in Fig. 3E indicate the yield of the 15 × 15 photodiode array is >99% (223/225 photodiodes have a leakage current <40 nA at −1 V). The dark current of the detectors on the array is 9.1 ± 7.9 nA at −1 V, which is approximately seven times greater than for individual detectors due to sneak reverse currents from adjacent detectors. This can be eliminated by using a passive pixel sensor address transistor (43) at each pixel that can be transferred simultaneously with the detectors without change or complication of the existing process.

As shown in Fig. 3F, the detector dynamic range is determined from the detector photocurrent (black square) at $\lambda$ = 850 nm vs. incident optical power. A photocurrent compression of 1 dB from linear response (red line) sets the maximum intensity, $P_1,$ whereas $P_0$ is the lowest detectable optical power (root-mean-square noise power). The dynamic range (DR) is $\text{DR}=10\hspace{0.17em}\text{log}\left(P_1/P_0\right).$ At 0 V, $P_0$ = 10⁻⁴ W/cm², $P_1$ = 10⁻¹ W/cm², giving DR = 30 dB, corresponding to a 10-bit grayscale resolution. This, too, can be improved using passive pixel sensors at each detector in the array.

A conventional imaging system based on a planar FPA has a mismatch with the image plane of a single-element lens. Producing a high-resolution image thus necessitates additional optical elements that increase the complexity, weight, and cost of the system, while restricting the FOV. Using an HFPA, however, provides the possibility of using a single plano-convex lens, whose optical field curvature is matched with that of the curvature of the FPA to produce high-quality images (2, 3, 25). As shown in Fig. 4A, multiple rays illuminated from five point sources (3 cm wide) positioned at the origin can be focused onto the curved plane of the HFPA centered at 3.0 cm from the lens at a distance of 10 cm. This image plane has a radius of curvature of R = 9.2 mm in the center and gradually increases to 10.1 mm toward the edge as shown by the blue dashed line in Fig. 4B. An HFPA (black contour in Fig. 4B) with R = 9.2 mm is positioned coaxially with the lens. The simulated results (SI Appendix, Fig. S4) show a spot size of 13.4 μm and 38.3 μm for the images of point sources in the center and on the edge, respectively, corresponding to a 1.8× edge defocusing. In comparison, when a planar FPA is located at the same position as the HFPA, the simulated spot sizes are 13.4 μm and 73.9 μm, corresponding to a 4.5× edge defocusing.

Fig. 4.

(A) Ray-tracing simulation result of an object (3 cm wide) located 10 cm from a plano-convex lens (black contour). Rays from the object are focused by the lens onto the FPA surface (orange curve, 3.0 cm from the lens). (B) Magnified view around the hemispherical imager (black contour). The simulated lens focal surface (blue dashed line) has good overlap with the concave FPA surface (front curve of the black contour). (C) Photograph of the hemispherical FPA mounted on a 3D-printed substrate holder integrated with a 3D-printed lens holder. Also presented is a 48-channel probe card used to read currents generated by all pixels on the hemispherical FPA simultaneously. (D) Side view of the experimental setup for imaging acquisition. (E) Normalized photocurrent map on the 15 × 15 FPA showing images of letters “O,” “C,” and “M.” A leakage current threshold of 15.8 nA is applied to minimize obscuration of the images by the background sneak currents.

A single-lens imaging system using the fabricated HFPA is shown in Fig. 4C. The HFPA was mounted on a 3D-printed substrate holder. Rows and column electrical contacts are extended to the edge of the substrate holder and connected to a 48-channel probe card that is interfaced to the readout electronics. The plano-convex lens (diameter = 6 mm, focal length = 24 mm) is mounted on a 3D-printed lens holder and plugged into the substrate holder. The resulting system is mounted on a six-axis optical stage to capture images as shown in Fig. 4D. The diffuse emission from a λ = 525-nm LED illuminates an image formed by a glass slide patterned with 1-cm-wide “O,” “C,” and “M” apertures. Applying a leakage (sneak) current threshold of 15.8 nA, the images of these letters are acquired as shown in Fig. 4E. The lens provides the HFPA with a calculated array angular coverage of ∼15°, and a FOV of ∼112°. This is demonstrated by focusing the LED source (3 mm diameter) to ∼60° from the optical axis of the lens. The edge detectors on the HFPA generate a photocurrent two orders of magnitude larger than in the absence of the light source with a power of 23.2 nW. This demonstrates the object detection ability of the HFPA at a large viewing angle.

The resolution of the imaging system is currently limited to tens of micrometers by the need to manually align the lens with only a 10-μm alignment tolerance. A smaller pixel spacing of <5 μm is achievable by use of more precise optics. The array size is limited to 15 × 15 to minimize sneak currents. Arrays with pixel density and counts similar to those in commercial CMOS imagers are possible if the detectors are integrated with access transistors in each cell. Including more circuit elements does not change the process sequence, since our process uses conventional planar semiconductor fabrication methods until the transfer to the shaped surface occurs.

Conclusion

We demonstrate a general strategy to achieve topological transformations of optoelectronic devices from a 2D plane into a 3D surface by exploiting slippage of the circuits during deformation. We use this process to demonstrate retina-like hemispherical imagers by starting on a planar substrate and then transferring the array onto a hemispherical surface without loss of array resolution. This process results in defect-free metal interconnections and a fixed pixel spacing. The HFPA has an individual detector performance comparable to that found in conventional planar CCD imagers. The hemispherical shape enables simplified optical designs with reduced aberrations along with a large FOV. The combination of features and fabrication strategies demonstrated in this work introduce processing techniques and performance advantages that may lead to additional capabilities of next-generation conformable and foldable optoelectronic devices.

Methods

Epitaxial Growth.

The photodiode array employs a 200-nm undoped GaAs buffer layer, a 25-nm undoped AlAs sacrificial layer, a 25-nm Si-doped (5 × 10¹⁸ cm⁻³) GaAs contact layer, a 25-nm Si-doped (1 × 10¹⁸ cm⁻³) In_0.49Ga_0.51P window layer, a 150-nm Si-doped (1 × 10¹⁸ cm⁻³) GaAs emitter layer, a 2.5-μm Zn-doped (2 × 10¹⁷ cm⁻³) GaAs base layer, a 100-nm Zn-doped (6 × 10¹⁷ cm⁻³) Al_0.26Ga_0.74As back surface field layer, and a 200-nm C-doped (5 × 10¹⁸ cm⁻³) GaAs contact layer that are consecutively grown on an undoped (100) GaAs substrate using molecular beam epitaxy.

Array Fabrication.

Following growth, the surface native oxide is removed in buffered hydrofluoric acid (HF) for 90 s and rinsed in deionized (DI) water for 10 s. A 200-nm Au layer is deposited using e-beam evaporation on the epitaxial surface, and 5-nm Ir and 200-nm Au layers are sputtered onto a 25-μm E-type Kapton foil. The GaAs sample with epitaxial layer is bonded to the Kapton foil by applying heat (200 °C) and pressure (2 MPa) for 5 min under vacuum (10⁻⁴ mTorr) using an EVG 510 wafer bonder (EV Group Inc.). The bonded sample is then immersed in 17% HF solution maintained at 60 °C with 400 rpm (Brewer Science and Cost Effective Equipment, Model CEE100CB) agitation for 3 h to remove the AlAs sacrificial layer, thereby separating the epitaxial layers from the parent GaAs wafer using nondestructive epitaxial lift-off (ND-ELO) (35).

The Kapton substrate is fixed to a rigid Si handle to eliminate curling. All layers are photolithographically patterned using LOR 3A (MicroChem Corp.) and SPR 220 3.0 (MicroChem Corp.) bilayer photoresist. Photodiode mesas (150 μm diameter, 300 μm pixel pitch) are patterned using inductively coupled plasma (ICP) reactive-ion etching (RIE) (Cl₂:Ar₂:BCl₃ = 2:5:10 standard cubic centimeters per minute (sccm), 5 mTorr pressure, 500 W ICP power, 100 W forward power, 0 °C stage temperature for 7 min). The back contact lines (50 μm wide) are wet etched using TFA Au etchant (Transene Company Inc.) to pattern photodiode rows. A 1.2-μm-thick polyimide (PI2610; HD Microsystem) insulation layer is spin cast and cured at 250 °C for 5 h. The polyimide layer is patterned to expose the light detection area and back contact pads using O₂ plasma (O₂ = 80 sccm, 800 W ICP power, 300 mTorr pressure, 150 °C stage temperature for 10 min). Next, the Ti (10 nm)/Au (500 nm) top contact ring is deposited onto the photodiode mesas. A TiO₂ (49 nm)/MgF₂ (81 nm) antireflection coating is then patterned on the light detection area. A Ti (10 nm)/Al (200 nm) etch mask is deposited onto the reverse side of the Kapton substrate with a pattern that matches the photodiode rows and contact lines on the front substrate surface.

A 100-μm PDMS (Sylgard 184, base to curing agent weight ratio = 10: 1) membrane is spun (800 rpm) (Brewer Science and Cost Effective Equipment, Model CEE100CB) on a Si handle pretreated with a release agent (tridecafluoro-1,1,2,2-tetrahydrooctyl trichlorosilane) and cured at 100 °C for 3 h. The Kapton substrate is placed detector side down on the PDMS membrane. The Kapton area not covered by the Al mask is removed to separate photodiode rows, using O₂ plasma (O₂ = 20 sccm, 6 mTorr pressure, 500 W ICP power, 100 W forward power, 0 °C stage temperature for 25 min). The Al mask is then removed using Cl₂ plasma (H₂:Cl₂:Ar = 12:9:5 sccm, 10 mTorr pressure, 500 W ICP power, 100 W forward power, 0 °C stage temperature for 2 min). A thin layer of NOA 84 optical adhesive (4,000 rpm; Norland Products) (Brewer Science and Cost Effective Equipment, Model CEE100CB) is spin coated and precured using UV light. The PDMS membrane is peeled from the Si handle and attached to the bottom of a 3D-printed holder (0.5-mm-thick, 4-cm × 4-cm square shape with a 2-cm diameter clear aperture in the center for device transfer). The same uncured PDMS is also poured into a plano-concave lens (LC4942; 12.7 mm diameter, 9.2 mm surface curvature, 4.4 mm edge thickness, 2.0 mm center thickness; Thorlabs), and cured at 100 °C for 3 h to form a hemispherical transfer punch. The plano-concave lens (LC4942; Thorlabs) is deformed to match the curvature of the PDMS membrane and cured (0.15 W/cm², 1 cm from the sample surface, 5 min), after which the lens and PDMS membrane are separated. Photodiode rows are transferred from the membrane to the lens. The residual adhesive is removed from the concave lens surface using O₂ plasma (O₂ = 80 sccm, 800 W ICP power, 300 mTorr pressure, 150 °C stage temperature for 40 min). Sputter a layer of Ti (5 nm)/Al (100 nm). The Kapton column connection pads are transferred to the array to connect rows of detectors using the same techniques as described above. Residual adhesive is removed using O₂ plasma (O₂ = 80 sccm, 800 W ICP power, 300 mTorr pressure, 150 °C stage temperature for 40 min) and Al area that is not covered by Kapton pads is removed using Cl₂ plasma (H₂:Cl₂:Ar = 12:9:5 sccm, 10 mTorr pressure, 500 W ICP power, 20 W forward power, 0 °C stage temperature for 8 min) to finish the fabrication of the HFPA.

Pixel Dimension.

Each 150-μm diameter photodiode is connected in rows with adjacent pixels (300-μm center-to-center spacing) through the 50-μm-wide bottom contact lines supported by the 60-μm-wide Kapton foil strips. Top contact rings are extended out of the photodetection area with 150-μm × 20-μm contact pads and connected to adjacent units through a separately transferred layer of 80-μm × 60-μm column connection pads. An antireflection coating (ARC) is deposited on the top to enhance the optical absorption in the visible spectrum.

Characterization.

The current-voltage characteristics under dark and 64 nW illumination at λ = 530 nm are measured using a Keithley 2400 Source Measuring Unit (SMU). External quantum efficiency is measured using monochromatic illumination chopped at 200 Hz and coupled into an FG050LGA optical fiber oriented normal to the photodiode using a Lightwave Probe (Cascade Microtech). The output signal is collected by an SR830 lock-in amplifier. The light illumination power is calibrated using a reference 818-UV/DB Si detector (Newport). Dark current mapping and object imaging are measured using a 48-channel probe card (AccuProbe) interfaced with a Keithley 2400 SMU and a Keithley 2700 + Keithley 7705 switching unit. A customized LabView graphic user interface is programmed to collect output signals. A schematic of the signal collection mechanism is provided in SI Appendix, Fig. S5.