Heterogeneously integrated light emitting diodes and photodetectors in the metal-insulator-metal waveguide platform

2.1 Integration of light emitting diodes within the MIM waveguide

For heterogeneous integration of active semiconductor-based optoelectronic devices in a conventional vertically-stacked metal-optic platform, a p-i-n semiconductor heterostructure with direct top and bottom metal contact electrodes is placed into the MIM waveguide as shown in Figure 1(a). The MIM waveguide section consists of the vertical stack of Ag (200 nm)/BCB (benzocyclobutene, d = 500 nm)/Ag (>>200 nm) layers, and the MSM region contains Ag (200 nm)/a middle InGaAs active layer with InAlGaAs cladding layers (540 nm)/Ag (>>200 nm). Since the subwavelength-scale MSM/MIM multilayer thin film stack is bonded on the silicon substrate using thermocompression metallic bonding, the thickness of the bottom metal layer is much thicker than 200 nm. Thin titanium layers were used as an adhesion layer for the Ag metal thin films. The detailed fabrication process can be found in our previous works [22, 23] and Supplementary Figure S1.

In order to improve the radiative recombination rate as well as carrier confinement, the p-i-n InAlGaAs/InGaAs/InAlGaAs heterostructure was utilized [24] with the upper and lower InAlGaAs cladding layers highly doped at the volume concentrations of 2 × 10¹⁹ cm⁻³ and 5 × 10¹⁹ cm⁻³ for the p- and n-type doping, respectively. Table S1 describes the information of p-i-n InAlGaAs/InGaAs/InAlGaAs epitaxial layers. The inset of Figure 1(a) shows the cross-section of a fabricated sample at the interface between the active MSM region and the MIM waveguide on the silicon substrate. Red arrows in Figure 1(a) schematically represent the light propagation directions. To observe the guided light emission from the semiconductor’s active region through the MIM waveguide, rectangular light emission slits were milled by a focused ion beam (FIB) at the 200 nm-thick top metal layer. A single out-coupling slit is schematically described in Figure 1(a). The same FIB milling process can be used to define the electrical isolation between the top electrodes when multiple active MSM devices are integrated on the common bottom metal layer.

The electrical characteristics of the fabricated MSM structures were consistent with the typical behavior of pn junction diodes as shown in Supplementary Figure S2. Under the forward bias conditions, the current flow exponentially increases with the applied bias voltage, whereas the limited reverse current flow is observed due to the rectifying characteristics of typical pn junctions. As shown in Figure 1(b), we observe electroluminescence spectra under the forward bias conditions, and the light output power grows in proportional to the input current density as in typical LEDs. In order to measure the emission spectra, we used a microscope-based electroluminescence setup. While the current–voltage characteristics were recorded using a source measure unit (Keithley 236), vertically scattered light emission was collected through a microscope objective with a numerical aperture of 0.5. The output light spectra were analyzed by an infrared spectrometer with a linear photodetector array (iDus InGaAs 1.7, Andor).

To investigate the propagation modes for the one-dimensional vertical MIM stack, we used a two-dimensional eigenmode expansion solver (MODE, Lumerical) with an assumption that the metal cladding layers are sufficiently thick. For the material parameters in the calculations, we used the data from Palik [25]. More specifically, the refractive indices of BCB and Ag are 1.5 and 0.469 + 9.32i at 1550 nm, respectively. Without loss of generality, the wave propagation occurs along the horizontal axis (x-axis). In Figure 2(a), we show the propagation constants of several low order modes as a function of the insulator or dielectric layer (BCB material in our implementation) thickness, d, at a representative wavelength of 1550 nm for this planar MIM waveguide geometry. The propagation mode with the largest propagation constant is the antisymmetric transverse magnetic (TM) mode with an antisymmetric E _x field profile across the vertical y-axis as shown in the upper right panel of Figure 2(a). The electric field profiles for other modes (symmetric TM mode and the transverse electric (TE) mode) are also shown in Figure 2(a). When the thickness of the dielectric BCB layer is below 550 nm (d < 550 nm), the TE and symmetric TM modes are nearly cut off, but the antisymmetric TM mode is maintained. This means that the output light from the MSM region can be coupled to only the antisymmetric TM mode within the emission bandwidth of InGaAs LEDs near 1550 nm when d = 500 nm.

Figure 2:

MIM waveguide’s propagation constant and loss. (a) Propagation constant as a function of the BCB thickness, d, at a free-space wavelength of 1550 nm. Mode profiles of each mode supported in the MIM waveguide: Antisymmetric TM, symmetric TM, and TE mode. When d < 550 nm, the antisymmetric TM mode is guided while symmetric TM mode and TE mode are neatly cut-off. (b) Intensity of scattered light at out-coupling slit as a function of distance from the active MSM region. The measured exponential decaying behavior well corresponds to the two-dimensional FDTD simulation results.

In order to investigate the light propagation distance and attenuation along the fabricated MIM waveguide (d = 500 nm), out-coupling slits were patterned at various distances from the MSM active region, and the light output intensities were quantitatively measured by a two-dimensional infrared photodetector array through a microscope. An example of an out-coupling slit with a thin rectangular aperture perpendicular to the light propagation direction is schematically described in Figure 1(a). Owing to the ohmic losses in the metal cladding layers of the MIM waveguide, the output intensities decay exponentially with respect to the propagation distance as observed in Figure 2(b). A good agreement between the observed and calculated decay lengths was obtained when the effects of the titanium adhesion layer was considered. The simulations were performed using two-dimensional FDTD methods. More simulation results on the propagation losses can be found in Figure S4. Since the propagation constant of the antisymmetric TM mode for the free-space wavelength of 1550 nm is approximately 6.5 μm⁻¹, its effective wavelength along the MIM waveguide is less than 1 μm. The two-dimensional simulation results are valid in the regime where the beam width (along the z axis) is sufficiently larger than the effective wavelength. For the experiments, slit lengths of at least 10 μm were used.

Although the propagation distance of the MIM waveguide is shorter than that of typical dielectric waveguides, it can still be useful in on-chip optical transmission applications where compactness is a priority and there is sufficient power budget. The propagation distance of approximately 15 μm for the MIM waveguide shown in Figure 2(b) is specific to the implementation used in this study and may vary depending on the specific MIM waveguide geometry and materials used. The propagation distance can be improved by employing metal layers with lower optical losses [26]. For example, instead of the titanium adhesion layer, a germanium adhesion layer can be employed for reduced optical losses. We also observe that the output light from the out-coupling slit is strongly polarized perpendicular to the slit’s long axis (or parallel to the light propagation direction through the MIM waveguide (x axis)) as expected from the TM mode profile with non-zero E _x field components shown in Figure 2(a). This also confirms that the light output from the active MSM region is not coupled to the TE mode with negligible electric field components along the light propagation direction.

2.2 Integrated photodetectors

Under the reverse bias conditions, the MSM semiconductor-based light source shown in Figure 1(a) functions as a PD directly integrated with the MIM waveguide. Two-dimensional FDTD simulations and experimental results support the polarization-dependent operation of the integrated PD as described in Figure 3. For simulations, a finite Gaussian beam at a free-space wavelength of 1550 nm with a TM (electric field parallel to the x-axis) or TE (electric field parallel to the z-axis) polarization state was illuminated from the surface normal direction (propagating along the negative y-axis) toward the out-coupling slit whose long axis is along the z-axis, and the normalized electric field intensity profile was depicted. As expected, the incident TE-polarized wave is not coupled to the MIM waveguide. In case of the TM polarization, however, the illuminated wave first excites a gap plasmonic mode at the rectangular aperture, and then gets coupled to the propagating antisymmetric TM mode of the MIM waveguide. The coupled and propagated wave through the MIM waveguide is finally absorbed in the semiconductor block of the MSM section because its high refractive index leads to strong modal confinement and absorption between two metal planes. For experimental verification, linearly polarized laser light with a wavelength of 1520 nm was illuminated, and the polarization-dependent photocurrent generation was observed.

Figure 3:

Simulation and measurement results for integrated PDs. (a) The calculated electric field intensities for a MIM-MSM device are shown in both TE-polarized and TM-polarized input cases. The normally incident light efficiently couples to the MIM waveguide for the TM polarization case only. The input Gaussian beam size is 10 μm, and the BCB thickness is d = 500 nm. (b) and (c) High-speed integrated photodetector. (b) An example impulse response is shown for a PD at the bias of −0.3 V. The inset shows the SEM of the PD device used in the experiments. The scale bar represents 10 μm. (c) The frequency response obtained from the Fourier transform of the impulse response in (b) indicates the 3 dB bandwidth of ∼6 GHz.

The operation bandwidth of the integrated PD was investigated from its impulse response. A picosecond mode-locked laser with a pulse width of ∼0.5 ps (PriTel, Inc.) and a wavelength of ∼1550 nm was used as an input light source. The polarized pulsed light input was converted to the photocurrent through the embedded PD in the MSM section as described above, and the output signal was observed on an oscilloscope. Figure 3(b) presents an example of representative impulse responses measured at the reverse bias of −0.3 V for an MSM active region area of 15 μm × 15 μm. The rising and falling time are observed as 21 ps and 220 ps, respectively. Although the calculated 3 dB bandwidth from the Fourier transform of this impulse response was ∼6 GHz as shown in Figure 3(c), it could have been as large as 17.5 GHz when we consider the rising time only (∼0.35/τ _r for a rising time-limited bandwidth).

The overall bandwidth of typical p-i-n PDs is limited by both the carrier transit time and the RC time constant. When considering an MSM active area of 15 μm × 15 μm, the PD bandwidth from the RC delay time is estimated to be >100 GHz and therefore it is not a dominant factor that limits the overall PD bandwidth. For experimental verifications, various square-shaped MSM structures with side dimensions of 20 µm, 6 µm and 2 µm were also fabricated and compared. Dimension-independent PD bandwidths according to the normalized impulse responses in Figure S3 confirms that the PD bandwidth is indeed not constrained by the RC time constant but by the carrier transit time, especially because of the carrier diffusion tail and trapping observed in the slow falling edge of the impulse response in Figure 3(b). Due to the antisymmetric TM mode profiles, strong light intensities near the highly-doped semiconductor cladding layers result in non-negligible light absorption in the p-i-n PD’s diffusion region, which causes the long minority carrier diffusion tail [27]. We believe that the overall PD bandwidth can be further improved by proper epitaxial layer designs.

2.3 Chip-scale integration of light sources and photodetectors

To experimentally demonstrate intimate chip-scale integration of both light sources and photodetectors based on the heterogeneously integrated MSM stacks in the common MIM geometries, we fabricated a simple, short one-dimensional back-to-back link between an LED and a PD with the passive MIM waveguide as shown in Figure 4(a). The distance between the LED and the PD was ∼20 μm. For simultaneous operation of both devices, top electrodes of two MSM regions were oppositely biased, while the common bottom metal layer functions as the electrical ground. As described in the previous sub-sections, the MSM region for the LED is forward-biased for light emission, and the other MSM section is reverse-biased for photodetection operation. The spontaneously emitted light from the forward-biased MSM section is directly coupled to the one-dimensional MIM waveguide, and the propagated light is converted to the photocurrent at the other MSM section. Although vertical out-coupling is not necessary to establish this on-chip photonic link, the top metal cladding layer must have at least one discontinuity due to the electrical isolation between the oppositely biased MSM sections. Although some lights are scattered at this electrical isolation slit in the middle of the link, most of the propagated wave through the MIM waveguide is transmitted toward the PD (Figure S5).

Figure 4:

Electrically-driven light source and photodetector integrated in the MIM waveguide platform. (a) Simultaneous operation of multiple MSM sections connected with a one-dimensional MIM waveguide. The demonstrated electrically-driven optical link consists of an LED, MIM waveguide, and a PD. An SEM of the demonstrated optical link is shown in the inset. A slot in the middle of the link was ion-milled by FIB for electrical isolation. The scale bar represents 5 μm. (b) Photocurrent and system efficiency as a function of LED drive current. The gradual decline of overall link efficiency attributes to the reduced quantum efficiency of the LED section.

Due to the on-chip electrical-to-optical and optical-to-electrical conversion processes, the input and output signals for the on-chip optical link are the drive current to the LED (forward-biased MSM section) and the photocurrent from the PD (reverse-biased MSM section), respectively. Figure 4(b) shows the output photocurrent with respect to the LED input current as well as their ratio. Although the PD photocurrent monotonically increases with the LED drive current, the overall conversion efficiency decreases with the LED current. Although the exact reason is not experimentally verified, the gradual reduction in the link efficiency at high LED currents may be due to the LED’s reduced quantum efficiency, possibly caused by Auger recombination and thermal stress [28–30]. At a current of 200 μA, the corresponding current density reaches ∼800 A/cm² for the MSM area of 5 μm × 5 μm. At such high current density levels, nonradiative Auger recombination, which increases with the cube of the carrier density, can become more dominant and reduces the LED’s overall quantum efficiency. Furthermore, high current densities increase the LED temperature and cause thermal stress in the device, which can further reduce its performance.

The overall link efficiency is ∼0.005 A/A. One reason for the relatively low link efficiency is the low coupling efficiency between the MSM section and the MIM waveguide. The coupling efficiency can be affected by many factors, such as the thickness, position, and refractive indices of the semiconductor light emission layer within the MIM waveguide, the wavelength and polarization of light. The MSM LED emits light to multiple modes, while the MIM waveguide only supports several modes. As shown in the subpanel of Figure 2(a), the mode profile for the antisymmetric TM mode, which has the lowest propagation loss, has higher intensity distribution near the semiconductor(insulator)-metal interfaces. However, the active semiconductor region is located in the middle of the waveguide to provide the Ohmic contacts through the metal cladding layers, resulting in low spatial overlap with the antisymmetric TM mode. As a result, only a small fraction of the light emitted from the LED will couple to the waveguide propagation mode. This inefficiency in coupling also applies to the photodetector parts. The link efficiency can be further improved by using more advanced designs for mode coupling structures between the MSM sections and the MIM waveguides. The epitaxial layers can also be modified to better match the propagation mode, and enhance the Purcell factor. Although the Purcell enhancement has not been experimentally proven in our proof-of-concept MIM/MSM integration scheme, we believe that an improved adhesion layer and epitaxial layer design providing optical gains close to the metal-semiconductor interface may significantly improve the overall efficiency of light generation and detection in the MSM section.