1. Introduction

Three-dimensional (3D) imaging and neural mapping techniques using fluorescence in miniaturized microscopes (miniscopes) have been rapidly developing in the past decade [1]. Miniscopes utilize miniaturized microscope optics and sensors packaged in a small (typically < 5 grams with dimensions on the order of 1 to 3 cm) format suitable for mounting directly to the heads of rodents with a surgically implanted cranial window [2,3]. Head-mounted miniscopes are advantageous for in-vivo neural imaging as they allow some form of mobility for the test subject enabling behavioral experiments to be performed simultaneously with image acquisition [4–8].

Typical miniscope light sources are light-emitting diodes (LEDs) with the use of excitation and emission filters and miniature CMOS cameras capable of taking widefield images [9,10]. LEDs are an attractive light source in miniscopes due to their high output efficiency, extreme brightness, small form factor, and availability in many emission wavelengths [11]. However, three-dimensional optical sectioning imaging techniques beyond widefield microscopy are often employed when imaging biological tissue to achieve higher resolution or improved contrast [12–14]. In particular, microscopes using patterned light sources used in structured illumination microscopy (SIM) can surpass the optical diffraction limit, and image 3D volumes [15–17]. Typically, these light sources use spatial light modulators (SLM), laser interference patterns, or digital micromirror devices (DMDs) to modulate pixel illumination to produce multiple structured patterns or phases. The optically sectioned image is then digitally reconstructed from the various phases recorded by the camera module [18]. Implementing an onboard DMD within a miniscope is prohibitively complex as the optical design becomes too cumbersome to implement without increasing the size or weight of the miniscope to sizes that are no longer feasible for freely roaming animals. One solution uses a benchtop optical setup to produce patterned images and delivers them to the miniscope optics via an optical fiber bundle [5,17,19]. While this simplifies the optical design within the miniscope, it adds a second tether from the miniscope in addition to the electrical connection. It prevents progress towards a fully wireless miniscope [9,20].

An integrated matrix-addressable microdisplay as an illumination source enables a simple optical design, patterned illumination, and does so without a fiber tether. The ubiquity of commercial liquid crystal displays (LCD) and organic light-emitting diode (OLED) displays are an attractive choice, but neither is capable of the high brightness needed to penetrate deep into tissue or would require long exposure times on the camera sensor; either of which does not advance miniscopes towards high-frame rate imaging with large 3D volumes [21,22]. Given these limitations, micro light-emitting diode (microLED) microdisplays pose as an ideal OS-SIM light source for miniscopes given their extraordinary brightness, and efficiency [23,24]. Additionally, ultra-high density microLED displays have been demonstrated with individual emitters on the scale of 1 µm [25].

MicroLED microdisplays as an OS-SIM imaging source technology bring a unique set of challenges to overcome. For simplification, rather than building a fully addressed display with individual emitters, microLED pixels patterned as lines, or stripes, have been demonstrated as an OS-SIM light source [26,27]. While these manuscripts show feasible results, the form factor of the driving electronics is too large, and the beam output is too wide to effectively use micron-scale imaging in miniscopes. To overcome the former, demonstrations have been shown that place the microLED display on a flexible interposer to allow for miniaturization and a flexible tether for a miniscope [28,29].

This study demonstrates a striped microLED microdisplay with direct addressing on a long (70 cm) flexible cable for an OS-SIM light source in a benchtop setup. The overall design of the display is described in detail with accompanying performance metrics. The OS-SIM implementation on a benchtop setup using a microdisplay as a light source is presented with a discussion on future work to miniaturize the system.

2. Experimental methods

2.1 MicroLED light source fabrication

LED wafers (4" diameter) of epitaxially grown GaN/InGaN on patterned sapphire substrates with a 455 nm emission center wavelength were used for this study. Additionally, the foundry deposited 30 nm of annealed Indium-Tin-Oxide (ITO) to the top p-type GaN layer to form a transparent and low-resistance ohmic contact. Shipley S1818 positive photoresist was spin-coated and laser patterned with a Heidelberg DWL 66+ lithography system to act as a dry-etch mask. The ITO was plasma etched in an Oxford PlasmaPro 100 using Methane and Hydrogen gases. The photoresist was stripped, and a bi-layer resist stack up of Kayaku LOR 5A, and Shipley S1805 was coated and patterned for lift-off with e-beam evaporated Ni/Au (20/50 nm) current spreading contacts. Next, Shipley SPR220 4.5 was coated and patterned as a dry-etch mask for microLED pixel formation. The GaN/InGaN was mesa-etched using Cl/BCl$_3$ plasma in an Oxford PlasmaPro 100. Aluminum contacts (80 nm) to the now exposed n-type GaN layer were thermally deposited and lifted off with a bi-layer resist stack of LOR 30B and SPR220 4.5. SiO$_2$ (500 nm) was deposited via plasma-enhanced chemical vapor deposition (PECVD) in an Oxford PlasmaPro 80. Via holes were etched into the SiO$_2$ with an Ar/CHF$_3$ plasma in an Oxford PlasmaPro 100 with SPR220 4.5 used as a dry-etch mask. Aluminum interconnects (500nm) were plasma sputter deposited in an AJA sputter system, patterned with SPR220 4.5, and dry-etched in a Cl/BCl$_3$ plasma. An Additional SiO$_2$ passivation layer (500 nm) was deposited via PECVD, and via holes were etched as described before to form wirebond pad sites. Bond pads of Cr/Au (20/600 nm) were e-beam deposited and lifted off using an LOR 30B and SPR 220 4.5 resist stack. The wafers were then stealth diced into individual dies (9.5 x 9.5 mm$^2$) by DISCO. The individual die were then wire-bonded using a semi-automatic wirebonder to rigid-flex printed circuit boards (PCB). Figure 1 shows a microscopic image of the fully fabricated and diced microdisplay.

 

Fig. 1. Microscope images of a fully fabricated and diced microdisplay. a) Microscope images stitched together of an entire die. Shadow artifacts are from non-uniform brightness distributions in image stitching. b) Microscope image centered on the microdisplay active area.

Download Full Size | PDF

2.2 System design and implementation

The microLED display design consists of three distinct subsystems: the microLED display itself, a rigid PCB that interfaces with a computer over USB and acts as the display driver for the microdisplay, and a flat flexible cable (FFC) interposer that addresses the microLED display. The rigid PCB houses a Teensy 3.6 microcontroller for programmable control of the PCB electronics. The PCB acts as a passively addressed display driver allowing for a common cathode to be selected with a multiplexer and 100 S/R latches (flip-flop switches) to supply current to the microLED stripes. A power supply section controls the supply voltage to the S/R latches (which dictates the supply current) to the microLEDs, effectively acting as a brightness controller. The PCB has connector housings to interface with the FFC interposer, the cable for the camera sensor, and available I/O pins for additional miniscope peripheral component control.

The decision to place the driving electronics on the rigid PCB rather than on the microLED die simplifies system integration but introduces the challenge of needing a flexible cable with many individual wires. For a target display with up to 100 rows and 100 columns, 200 individual wires are required for a passive matrix scheme. The FFC interposer was designed to contain 200 wires in a 4-layer PCB design. The FFCs were sourced through Epec Engineered Technologies due to their ability to create a long flexible PCB (70 cm maximum length) with tight trace and space tolerance (4 mil / 100 $\mathrm{\mu}$m), and the ability to laser drill small plated vias (4 mil / 100 $\mathrm{\mu}$m). This enabled us to design a 10 x 10 mm square space with 50 bond sites per side. Each bond site is 100 $\mathrm{\mu}$m wide with a 200 $\mathrm{\mu}$m pitch, which also correlates to the approximate minimum tolerance for most semi-automatic wirebonding systems. The four vias at the corners of the microdisplay bond sites were designed for M1 mounting screws and are separated by 12 and 14 mm, horizontally and vertically, respectively. These dimensions are on the order of the length scale of miniscopes to allow for simple integration of the microLED display in small form-factors.

2.3 Fixed tissue preparation

The sample is fixed mouse cortical tissue (4% paraformaldehyde) expressing green fluorescent protein (GFP) driven by proteolipid protein promoter (PLP-eGFP). The GFP labels oligodendrocyte cells in the cortex. For fixed tissue imaging, a 500 µm thick coronal slice of the brain tissue was mounted on a microscope slide in VECTASHIELD Plus Antifade Mounting Media (H-1900)[30].

3. Results

3.1 Microdisplay demonstration

Figure 2(a) shows an image of a fully illuminated microstripe display. Each microstripe is 15 $\mathrm{\mu}$m wide and 2 mm long. The microstripes are separated from each other by 5 $\mathrm{\mu}$m, making the center-to-center distance between each microstripe to be 20 $\mathrm{\mu}$m. The overall dimension of the microstripe display is 2 mm x 2 mm. Figure 2(b) demonstrates an alternating pattern with two rows turned on and two turned off. The display shows excellent fidelity, with only one microstripes failing to show illumination. This is due to either failed wirebonds or electrical connection failure between the FFC and the rigid PCB, as these stripes do emit light when operated through micromanipulator probes. Figure 2(c) shows the complete system with a working micro-display attached to the FFC interposer, which is connected to a rigid PCB to control and drive the microdisplay.

 

Fig. 2. a) The brightfield microstripe display containing 100 lines showing different illumination patterns for the SIM light source. b) Darkfield image of the display. The lines are 2000 $\mathrm{\mu}$m x 15 $\mathrm{\mu}$m in dimension. The total size of the display is 2 mm x 2 mm. c) Photograph of the micro-display attached to the FFC interposer with the rigid control PCB.

Download Full Size | PDF

Figure 3 details luminance, current, voltage, and spectral information for a single microstripe LED. Current density increases with the increasing applied voltage, as seen in Fig. 3(a). The light output power of the microstripe increases with increasing current density as more injected electrons are converted into photons (Fig. 3(b)). Figure 3(c) shows the external quantum efficiency (EQE) as a function of current density. The EQE initially increases with increasing current and reaches a peak before it starts decreasing again. The peak EQE is observed around 10 A/cm$^2$. A higher Shockley-Read-Hall (SRH) non-radiative recombination rate at higher current densities causes the observed efficiency droop [23]. This efficiency droop gets worse with smaller active area devices having a higher ratio of the perimeter to total area of the mesa [31]. Figure 3(d) shows the emission spectra of the blue light with a peak emission wavelength at approximately 455 nm. The emission spectrum is measured using an Ocean Optics USB4000 spectrometer. The spectrum shows a full-width-half-maximum (FWHM) of 23.76 nm when fitted with a Gaussian function. An additional plot fitted with a bigaussian distribution is shown in the Supplement 1. Variation of the peak emission wavelength for individual microstripes across a die and the impact of self-heating on the peak emission wavelength are also shown in the Supplement 1.

 

Fig. 3. Electrical and luminance performance of a single microstripe. a) Current density as a function of applied voltage. b) The light output power vs. current density plot. Power increases as the current density increases. c) External quantum efficiency (EQE) as a function of current density. EQE reaches its peak around 10 A/cm$^2$ and then decreases. d) The light emission spectrum with a peak wavelength at 455.5 nm and FWHM of 23.76 nm extracted from Gaussian fit. e) Angular distribution of microstripe display with two contiguous rows turned on. Normalized output power vs. the angle at which the optical power is measured. The angular distribution has an FWHM of 81 degrees f) The same angular distribution plot as (e) in polar coordinates.

Download Full Size | PDF

Figure 3(e) and Fig. 3(f) shows the angular distribution of the microstripe display. The angular distribution was measured using a goniophotometer setup built using a stage rotation system (Thorlabs ELL18K) and a photodiode power sensor (Thorlabs S120VC). The microstripe display and photodiode were placed 3 cm apart. The light output power is measured as the stage with the mounted microstripe display is rotated. A linear and polar plot of the angular distribution of light emission from two consecutive rows (49 and 50) is shown. The angular distribution has an FWHM of approximately 80 degrees, but also it is more directed near the center. It differs from standard Lambertian distribution with an expected FWHM of 120 degrees. The microstripes have a current spreading contact that acts as a partial aperture effectively narrowing the emission profile.

3.2 OS-SIM imaging

A standard epifluorescence microscope configuration was used for SIM experimentation. The microLED array was imaged onto the sample using a 4f relay consisting of a tube lens Olympus SWTLU-C) and an objective lens (Olympus Plan N 20x 0.4 NA). The emitted fluorescence from the sample was collected by the objective lens and reflected by a short-pass dichroic mirror (Thorlabs DMSP505R). The reflected fluorescence passed through a long-pass emission filter (Chroma 525/20m) and was focused onto the image sensor (FLIR BFLY-U3-23S6C-C, CMOS) using a 180 mm tube lens (Thorlabs TTL180-A). The axial position was controlled using a piezo objective scanner (Thorlabs PFM450E). To determine the lateral resolution of the system, 1.1 µm beads (Fluoro-max G0100) were imaged in the focal plane of the microscope. The intensity profile across the center five beads was measured and fit to a Gaussian function. The FWHM of the measured bead profiles was found to be 1.18 $\pm$ 0.3 µm. See Fig. S8 in the supplemental information for an image of the beads. The pixel to micron calibration was found using an airforce target (see Fig. S9).

Five micron beads (Fluoro-max G0500) were imaged to demonstrate the capabilities of the microLED OS-SIM microscope to reject out of focus fluorescence in the absence of tissue. The beads were imaged with a fluorescent slide underneath to create out of focus fluorescence. The p-WF and OS-SIM images are shown in the supplementary Fig. S10. The contrast improvement from p-WF to OS-SIM is more than 3x. Additionally, due to the strong emission response from the 5 µm beads, images were acquired with a framerate of 40 frames per second (FPS), demonstrating that in an optimized, ideal system, high frame rates are achievable with this light source.

To demonstrate the optical sectioning capabilities in brain tissue, the microLED array was programmed to generate three phase-shifted sinusoidal fringe illumination patterns at 0, 90, and 180 degrees. The microLED patterns were comprised of 8 stripes on, and 8 stripes off to achieve a spatial frequency of .003125 µm$^{-1}$. The resultant spatial frequency at the sample was found to be 0.06 µm$^{-1}$. A step size of 2.5 µm was selected for a complete axial scan range of 47.5 $\mathrm{\mu}$m. Data were recorded at 3 FPS, and 9 frames were taken for each illumination pattern. Frames were examined for motion artifacts before being averaged to reduce noise. OS-SIM reconstruction was performed on the resulting images following the method outlined in [14]. A brain slice from PLP-EGFP mice expressing EGFP in oligodendrocytes was used as the imaging sample [30].

To evaluate the optical sectioning capabilities of the microLED array, the OS-SIM reconstruction was compared to a pseudo-widefield (p-WF) reconstruction. A 75 x 123 µm$^{2}$ field of view was selected within the center of the microLED due to the illumination inhomogeneity across the microLED. Homogeneity was determined by imaging the microLEDs on a thin fluorescent sample prior to recording in tissue (see supplementary Fig. S6). The p-WF reconstruction was created by averaging the images for the three-phase illumination patterns. p-WF and OS-SIM reconstructed images for various slices in the z-stack taken with 2.5 µm steps are shown in Fig. 4(a). Four cells at the center of the scan range were selected for analysis. ROIs were drawn in ImageJ for each cell, and mean intensities were extracted at each focal plane for both p-WF and OS-SIM reconstructions. An average offset was calculated from a portion of the camera’s FOV that was not illuminated by the microLED source. This offset was removed from each trace. The data was normalized, and Gaussian approximations were made for each trace using the MATLAB Curve Fitting Toolbox. Figure 4(b) shows the resulting traces for the p-WF (blue) and OS-SIM (red) across the center of the axial scan range with the plane for which the cell was in-focus at the 0$^{th}$ position. Calculated FWHM values for the two cells averaged traces were 54.12 µm and 16.85 µm for the p-WF and OS-SIM traces, respectively.

 

Fig. 4. MicroLED OS-SIM results in EGFP-PLP brain slice a) Comparison between pseudo-widefield (top) and OS-SIM (bottom) for z-stack taken with 2.5 µm size steps. The scale bar is 25 µm. b) Normalized intensity vs. axial position profiles for OS-SIM reconstruction and p-WF, and expected response (Theor.) by a convolution with a simulated cell of average cell diameter. c) Zoomed-in image of a cell at one focal depth showing (i) pseudo-widefield (p-WF) (ii) OS-SIM reconstruction. The scale bar is 15 µm. d) Normalized intensity line profile for the cells shown in (c). e) Percent contrast calculated from line profile. Contrast values were 44.31% and 86.92% for p-WF and microLED OS-SIM reconstruction (SIM), respectively.

Download Full Size | PDF

The theoretical optical sectioning strength was calculated using the Stokseth approximation to be 14.26 µm based on the spatial frequency of the illumination pattern, 0.0571 $\mathrm{\mu}$m$^{-1}$ measured at the sample plane [12]. The expected axial profile was calculated by a convolution with a simulated cell of average cell diameter (shown in yellow dashed line, Fig. 4(b)).

An in-focus cell (Fig. 4(c)) at z = -10 µm was selected to illustrate the background fluorescence reduction capabilities of microLED-SIM. Line intensity profiles were taken across the selected cell in the p-WF and OS-SIM reconstructions. A moving average filter was used to smooth the noise, and the traces were normalized for illustration (Fig. 4(d)). Percent contrast was calculated from the smoothed intensity profiles prior to normalization (Fig. 4(e)). The microLED-SIM data showed a contrast of 86.92% for OS-SIM reconstruction compared to 44.31% for p-WF for this cell.

4. Discussion

The luminance-current-voltage (LIV) characterization of the microstripes indicates the potential of the display to be used as a light source. The light output power of 3.5 mW at 100 A/cm$^2$ is high enough to cause fluorescence in the genetically modified brain slice samples. Operating the microstripe display at voltages more than 3 V gives enough light output power to be used in a benchtop setup. A lower voltage operation is possible for the miniscopes. The light coupling in the benchtop system is weaker as the microstripe display is located further away compared to the miniscopes. In miniscopes, the light source is placed much closer to the lens; hence a lower light output power is needed. The divergence angle of the light emission can be decreased using microlenses or apertures to create a more directed beam for improving the light coupling in the future. A narrower light beam would also ensure a better separation of the individual microstripes. The peak emission wavelength of 455.56 nm elicits green fluorescence emission in PLP-EGFP brain slice samples. The microLEDs with a different epi-GaN structure can be fabricated to have a peak emission centered at 470 nm to increase the sensitivity even further.

The experimental results outlined above illustrate the capabilities of the microLED array for performing OS-SIM. Although the calculated optical sectioning strength, represented by FWHM of the theoretical axial response to an approximated cell in the FOV, was slightly better than the experimentally obtained OS-SIM optical sectioning strength representation (FWHM of the OS-SIM Gaussian fit data), the axial response of the OS-SIM implementation still had a significantly smaller FWHM than the response of the pseudo-widefield data. The contrast increase and background fluorescence reduction, calculated with the line intensity traces in Fig. 4(c), further illustrate the microLED array’s OS-SIM capabilities compared to widefield illumination.

Previous work has demonstrated OS-SIM capabilities in a miniature microscope using a coherent fiber bundle to relay structured illumination patterns generated with a LED and digital micromirror device (DMD) [17]. This demonstration of microLED-SIM expands upon Supekar et al. by providing means for patterned illumination to be implemented on board of the miniature microscope. This development removes the theoretical field of view limitation imposed by the size of the fiber bundle’s active area. Additionally, there is no sacrifice in the illumination source’s optical power due to fiber coupling. A miniature microscope utilizing this microLED array would require a much less complicated optical setup than one utilizing a DMD. The microstripe display will be mounted onto a miniscope design and connected to the driver board using the flexible connector. The onboard implementation of the microstripe display removes the fiber bundle. However, the flexible connector adds the extra tether in this system, thus creating similar limitations for rodents’ movement. Additionally, while FFCs are usually good for bending, the non-circular-symmetric shape limits the torsional movement. This will further impact the experiments requiring rotational couplings with rodents turning. Implementing the driver PCB as an integrated circuit (IC) will pave the way for the microstripe display to be wireless and fully onboard while removing the FFC tether. The small form factor of the microLED-based light source will enable easier integration with the existing miniscope design. The integration with the miniscope will also increase the total light output power at the sample, allowing a smaller camera exposure time and faster imaging speeds. The increased output power can also be traded-off for a more directed light beam by integrating microlenses and apertures on top of the light emitting area. Additionally, the miniscope with onboard microstripe display can be used for in-vivo brain recordings with optical sectioning in freely moving mice.