Precise photoelectrochemical tuning of semiconductor microdisk lasers

Abstract

Micro- and nano-disk lasers have emerged as promising optical sources and probes for on-chip and free-space applications. However, the randomness in disk diameter introduced by standard nanofabrication makes it challenging to obtain deterministic wavelengths. To address this, we developed a photoelectrochemical (PEC) etching-based technique that enables us to precisely tune the lasing wavelength with sub-nanometer accuracy. We examined the PEC mechanism and compound semiconductor etching rate in diluted sulfuric acid solution. Using this technique, we produced microlasers on a chip and isolated particles with distinct lasing wavelengths. Our results demonstrate that this scalable technique can be used to produce groups of lasers with precise emission wavelengths for various nanophotonic and biomedical applications.

1. Introduction

Micro- and nano-scale laser batches – with each batch lasing at a deterministic wavelength having ultra-narrowband precision, and different batches lasing at uniquely distinguishable wavelengths – are critical for various nanophotonic applications. This includes on-chip photonic communications using dense wavelength division multiplexing¹, quantum photonic information processing², on-chip bioimaging using optical coherence tomography³, and biochemical sensing^{4, 5}. Unless elaborate and space-consuming tuning circuits are associated on-chip with such lasers, scalable fabrication of precise wavelength micro- and nano-laser batches is a major challenge. The specific lasing wavelength is determined by the physical dimensions of the laser cavity and the material’s band gap energy. The precision of fabrication scales with the wavelength of the beam used in lithography. While electron-beam lithography offers high resolution, it is too expensive and slow for large-scale production. In contrast, standard optical lithography, even using deep UV, offers lower precision.

Free-standing versions of such matched-wavelength micro-laser batches would also allow for novel applications such as optical barcoding of heterogeneous biological samples, where lasers with the same wavelength can be used to tag a specific cellular population⁶ or target specific molecules in multiplexed assays. Conventionally, cell-type specific tagging has been performed with organic fluorophores⁷, quantum dots^8–10, and fluorescent beads^11–13. Owing to the broad emission linewidth of these conventional biomarkers, only a limited number of non-overlapping spectral signatures can be distinctly identified and thus, only very few distinct cell types can be simultaneously tagged.¹⁴ Micro-lasers offer an advantage over conventional biomarkers due to their ultra-narrowband laser light emission, with linewidth on the sub-nanometer scale. Thus, a much larger number of cell types can be simultaneously identified using precisely wavelength tuned micro-lasers as biomarkers.

Recently, micro- or nano-disk lasers based on compound semiconductors have successfully been used as physically associated optical barcodes for living cells that are also sufficiently small to cause no observable perturbation of normal cell behavior.^15–17 In previous demonstrations, the inherent uncertainty in disk diameter introduced during standard lithography and dry-etching processes resulted in microdisks with virtually random wavelengths. These microdisk lasers, called Laser Particles (LPs), were utilized for stochastic¹⁸ combinatorial barcoding of hundreds of thousands of cells for imaging^{15, 19, 20} and flow cytometry.²¹ However, precise control over the wavelength of LPs is imperative to allow for more controlled association between wavelength and sample, instead of being stochastic.

To achieve precise wavelength control, in this study, we have investigated light-induced etching^22–24 of semiconductors as a technique to fine-tune the wavelength of microdisk lasers. We describe the mechanism and kinetics of photoelectrochemical (PEC) etching for indium gallium arsenide phosphide (InGaAsP) microdisks on indium phosphide (InP) substrate, and present a recipe for wavelength tuning that enables the production of a batch of microdisk lasers at a desired wavelength.

2. PEC etching of InGaAsP / InP microdisks

2.1. Fabrication of microdisk-on-pillar

We obtained a custom-ordered semiconductor wafer comprising a 200 nm-thick In_0.26Ga_0.74As_0.57P_0.43 layer and an InP buffer layer (300 nm thick), both epitaxially grown on an InP substrate using metal organic chemical vapor deposition (MOCVD). The unintentional (n-type) doping concentration of the epitaxial InGaAsP and InP layers is <10¹⁶ cm⁻³, while the substrate is n-type doped to 1×10¹⁸ cm⁻³. To safeguard the top active layer, we deposited a 250 nm-thick layer of SiO₂ using inductively coupled plasma chemical vapor deposition (Oxford ICP-CVD). To improve resist adhesion, we baked the SiO₂ coated wafer at 180°C on a hotplate for 5 minutes, followed by O₂ plasma priming (Anatech SP100). Next, we deposited a 2 μm-thick photoresist (SU-8 2002, Kayaku) using spin coating. The photoresist was exposed with a quartz photomask having circular patterns using a UV i-line contact aligner (Karl Suss MA6) and then developed to yield resist pillars on the substrate. The photoresist was hard-baked at 190 °C to enhance chemical resistance. The microdisk pattern on the photomask was then transferred onto the SiO₂ layer using a C₄F₈-SF₆-H₂ chemistry. Next, we removed the photoresist using O₂ plasma and a sulfuric acid dip. The SiO₂ mesas were then employed as etch masks to etch non-selectively into the semiconductor using a Cl₂-Ar chemistry (Oxford Instruments ICP-RIE). The etch depth was such that it passed through the InGaAsP layer and penetrated a few hundred nanometers into the underlying InP. We then used a 1:1 dilute hydrochloric acid (HCl) to partially etch the InP (by controlling etching time) underneath InGaAsP disks, producing a SiO₂-capped InGaAsP-disk-on-InP-pillar structure. Figure 1a shows a representative microdisk on a pillar as seen through scanning electron microscopy (SEM).

Figure 1.

(a) SEM image of a microdisk-on-pillar structure. Temporal variation of microdisk laser spectrum showing (b) no peak shift with continuous pump laser illumination in water, (c) no peak shift in sulfuric acid under no illumination, and (d) blue-shift and mode change of laser peak with continuous illumination in sulfuric acid. (e) SEM images of microdisks before (initial), after 60 s and after 100 s of PEC etching.

2.2. Real-time monitoring of PEC spectral shifts

To excite charge carriers, we employed a pulsed laser with a wavelength of 1064 nm, 10 ns pulses, and 2 MHz repetition, which was focused onto a single microdisk with an intensity significantly above the lasing threshold of typical pristine microdisks. The illumination area was large enough to cover the entire disk area. The excitation intensity was adjusted by an acousto-optic modulator and neutral-density filters. The output luminescence of the microdisk was analyzed with a grating-based spectrometer employing a high-speed InGaAs linescan camera (Sensor Unlimited 2048L).

We immersed microdisks in various solutions to compare the etching speed and light sensitivity of the microdisks. To precisely quantify the degree of etching, we monitored the temporal evolution of the laser emission spectrum from a single microdisk. We found that microdisks are relatively stable in water, both with and without excitation by the pump laser (Fig. 1b). The best results were from a dilute sulfuric acid solution (H₂SO₄:H₂O = 1:10000). Without illumination, the microdisk emission wavelength remained nearly unchanged over a duration of 6 minutes (Fig. 1c). However, under the continuous illumination of pump light with an intensity of 1 W/cm², we observed a relatively rapid shift of the lasing wavelength at a rate of around −0.6 nm/sec, as shown in Fig. 1d. Over time, the microlaser underwent mode hopping and eventually ceased lasing in about 6 minutes (Fig. 1d and Supplementary Movie 1). Both the blueshift and the mode transition to a lower mode number (indicated by the jump to a longer wavelength) indicate that the diameter of the microdisk continuously reduced during the excitation. SEM images confirmed the reduction of the InGaAsP disk diameter (Fig. 1e).

We separately immersed the pristine microdisks in a dilute sulfuric acid bath heated to 200°C and observed no etching of InGaAsP microdisks, indicating that heating alone is not sufficient to induce etching. This result supports the conclusion that the etching mechanism is photoelectrochemical, rather than photothermal.

2.3. Mechanism of photochemical etching

The PEC etching process in (n-type) III-V semiconductors in an acidic medium typically involves the oxidation of the material, followed by etching of the oxidized material by acid. Under sustained illumination, the anodic reaction results in a reduction in the dimensions of the microdisk, leading to a blue shift of the lasing wavelength. In the case of our study, the InGaAsP material with bandgap of 0.97 eV absorbs the pump photons at 1064 nm (1.16 eV), generating electrons and holes, while the InP (bandgap 1.34 eV) pillar and substrate do not absorb pump photons. When in contact with the dilute sulfuric acid solution, the InGaAsP layer forms an upward band-bending that leads to charge separation, and holes accumulate near the InGaAsP-solution interface. The holes initiate the oxidation reaction of InGaAsP, producing oxides that are dissolved in the acidic medium.^{25, 26} There are several anodic reactions possible at the surface of InGaAsP, one of which can be written as follows:^{27, 28} $InGaAsP+12h^{+}+6H_{2}O\to {In}^{3+}+{Ga}^{3+}+As(OH{)}_3+P(OH{)}_3+6H^{+}$.

On the other hand, photo-induced electrons drift downward in potential from InGaAsP to the undoped InP and then to the n-type InP substrate, from which the electrons are injected into the electrolyte solution, driving the hydrogen evolution reaction. Minimal cathodic dissociation is expected at the surface of n-InP, and the typical water reduction reaction, $2H^{+}+2e^{-}\to H_2$, prevails. Figure 2a illustrates the charge carrier diffusion path (green-red arrow) and representative PEC reactions.

Figure 2.

(a) Schematic diagram of a microdisk-on-substrate structure and the flow of electrons and holes enabling redox PEC reactions. (b) Band diagram depicting the energy levels along the charge carrier path between InGaAsP through InP pedestal and n⁺ InP substrate, with both InGaAsP and n⁺ InP substrate bounded by the electrolyte solution.

We refer to the band diagram presented in Fig. 2b, which shows the charge carrier path with the InGaAsP and InP substrate, both bounded by diluted H₂SO₄ (0.0019 M). The details of the band diagram derivation, including assumptions, are provided in Supplementary Note 1. Previous studies have demonstrated^29–32 that for PEC etching to occur, a local photovoltaic cell action is required, where one part of the semiconductor is at a more positive potential compared to another region. A uniformly illuminated surface does not create an open-circuit potential difference, which means that no etching occurs. This necessary condition is met in our current structure, where charge carriers are generated in InGaAsP but not generated in the electrically connected InP. This conclusion was further confirmed by releasing microdisks from the InP pillar and dispensing them on an insulating silicon dioxide surface. In this case, PEC etching was nearly absent, as the lasing peak from InGaAsP microdisks stayed at virtually the same wavelength, or blue shifted at orders of magnitude lower rate (Supplementary Movie 2). On the other hand, rapid PEC etching was observed from microdisks released on conductive gold surface (Supplementary Movie 3).

2.4. The kinetics of PEC etching

The rate equation for photoexcited electron or hole density $x$ in intrinsic $\left(\right[n]=[p\left]\right)$ or modestly doped (<10¹⁷ cm⁻³) semiconductor can be written as

$$\frac{dx}{dt}=-Ax-B{x}^2-C{x}^3-D(x)+q.$$ (1)

Here, $A$ is the Shockley-Read-Hall (SRH) recombination coefficient that corresponds to relaxation to traps, including surface defects, $B$ is the radiative recombination coefficient, and $C$ is the Auger recombination coefficient. The absorption rate of excitation photons is $q$, and $D\left(x\right)$ represents the carrier loss due to PEC dissociation reactions. The magnitude of this term is substantially smaller than the other terms. We note that $q=\alpha I/hv$ where $\alpha$ is the absorption coefficient of excitation light, and $I$ is the illumination intensity ($\alpha ={10}^4$ cm⁻¹ and $hv=1.16\mathrm{e}\mathrm{V}$ for 1064 nm).

For InGaAsP disks, we estimate $A\approx 2v_S/R\approx 6.7\times {10}^7{\mathrm{s}}^{-1}$, where $v_s=5000\mathrm{c}\mathrm{m}/\mathrm{s}$ is the surface recombination velocity,³³ and $R=1.5\mathrm{\mu }\mathrm{m}$ is the disk radius. We found literature values of $B=4\times {10}^{-10}{\mathrm{c}\mathrm{m}}^3/\mathrm{s}$ and $C=8\times {10}^{-29}{\mathrm{c}\mathrm{m}}^6/\mathrm{s}$.³³ The steady-state solution for $x$ in Eq. (1) is

$$Ax+B{x}^2+C{x}^3=q.$$ (2)

This equation is plotted in Fig. 3a as a function of $I$. At low intensity, $x$ is linearly proportional to $I$, but it increases sub-linearly with $I$ when $x$ exceeds about 10¹⁶ cm³.

Figure 3.

(a) Theoretical steady-state carrier density as a function of excitation laser intensity based on Eq. (2). (b) Experimental data showing the rate of wavelength changes. (c) Best fit curve based on Eq. (4).

We conducted experimental measurements of the rate of lasing wavelength shift at different pump laser intensities (time-average of the MHz pulse train) spanning six orders of magnitude. The results are presented in Fig. 3b. To measure wavelength shifts at laser intensities below lasing threshold, we used a brief pump above threshold while simultaneously collecting the lasing peak position. Then, the pump intensity is lowered to the interrogation value. Finally, the laser was briefly pumped above threshold again while collecting the laser peak position. We obtained the peak wavelength shift by subtracting the starting peak position in the second measurement above threshold from the end peak position in the first measurement above threshold. We observed a sub-linear trend in the peak wavelength shift across all incident power levels.

It is well-established that a Piranha solution, which contains H₂O₂ as an oxidation agent, can be used to etch InAs, GaAs, and InGaAs, but is less effective for etching InP. Based on this observation, it is reasonable to assume that the dissociation of phosphorus is the rate limiting step in the dissolution process of InGaAsP in the PEC etching. Since this reaction involves holes (as shown in the equation in Fig. 2a), the overall etching rate is proportional to the concentration of photoexcited holes ($x$). However, at higher values of $x$, holes become less efficiently driven to the surface, which leads to a dampening of the etch rate despite the increased generation of electron-hole pairs.³² This effect has been studied in the literature using models that accounts for charge transport mechanisms within the semiconductor, at the semiconductor-solution interface, and within the solution itself.^{32, 34, 35}. We have modeled the saturation effects with the following equation:

$$D(x)=\frac{k_{D}x}{\sqrt{1+x/x_s}},$$ (3)

where $k_D$ is the reaction rate coefficient and $x_S$ represent the saturation carrier concentration. The change of radius $R$ is equal to the change of lattice constant $L$ (equivalent to the length decreased by each dissociation reaction) per reaction times the reaction rate, given by

$$\frac{dR}{dt}=L\mathrm{*}D(x)L^3=\frac{k_D{L}^{4}x}{\sqrt{1+x/x_s}}$$ (4)

We fit the experimental data using Eq. (4), as depicted in Fig. 3c, and obtained $k_D{L}^4=1.1\times {10}^{-26}{\mathrm{c}\mathrm{m}}^4/\mathrm{s}$ and $x_S=1.6\times {10}^{17}{\mathrm{c}\mathrm{m}}^{-3}$. Using $L=0.58\mathrm{n}\mathrm{m}$ for InGaAsP, we obtain $k_D=970{\mathrm{s}}^{-1}$. For example, at $={10}^{18}{\mathrm{c}\mathrm{m}}^{-3}$, the reaction rate per unit cell is $k_D{L}^{3}x=0.19{\mathrm{s}}^{-1}$, which means one reaction occurs every 5.3 seconds per unit cell.

2.5. Layer-selective etching by excitation wavelength

We have previously developed wafers containing multiple InGaAsP layers with different alloy compositions, which were used to create random-wavelength LPs.^{15, 16, 19, 20} By exploiting the different bandgap energies of the individual layers, we can utilize the variation in spectral absorption [$q$ in Eq. (1)] to perform PEC etching on specific layers. To demonstrate this concept, we employed a bilayer wafer composed of In_0.74Ga_0.26As_0.57P_0.43 (bandgap at 1150 nm) and In_0.81Ga_0.19As_0.41P_0.59 (bandgap at 1275 nm) layers separated by a 300 nm InP layer. In Fig. 4, we show the effect of bandgap-dependent PEC etching on an array of pillars. Figure 4a shows the pillar structure after immersing the chip in dilute sulfuric acid in the dark, where no etching occurs. In Fig. 4b, we see a chip after exposing it to acidic medium and supercontinuum laser illumination through a long-pass filter that only allows wavelengths greater than 990 nm. The etched pillars indicate that both InGaAsP layers were etched, as evidenced by their reduced diameters between the intact InP layers. It is believed that electrons generated in the upper layer are transported all the way to the substrate. Figure 4c shows the same structure under illumination through a long-pass filter at 1250 nm, where the upper InGaAsP layer remained intact while the lower layer is etched.

Figure 4.

SEM images of a bilayer InGaAsP-in-InP pillar structure after immersion in dilute sulfuric acid under different illumination conditions: (a) Without illumination, (b) under illumination with wavelengths >990 nm, and (c) under illumination with wavelengths >1250 nm. Yellow arrows indicate etched InGaAsP layers. Scale bar 1 μm.

2.6. Role of SiO₂ capping layer

To ensure reliable PEC tuning, it is critical to protect the flat surface of InGaAsP disks from etching, which is achieved by the SiO₂ capping layer. Figures 5a (and Supplementary Fig. S1) present SEM images of microdisks after removal of the silica cap following PEC etching, revealing a normal morphology of the top surface and reasonably smooth side walls. When we removed the microdisks from the pillar using HCI etching, we found the bottom surface to be acceptably smooth. While the outer part of the bottom surface is in contact with the etchant solution, the light intensity reaching there is smaller than that on top due to optical absorption throughout the disk thickness, resulting in a less vigorous etch reaction.

Figure 5.

(a) SEM images of silica-capped microdisks with PEC etching for 30, 60, and 100 s, respectively, with the silica cap removed to show the top and side morphology. Also shown is the bottom surface of a post-PEC microdisk released from its pillar. (b) SEM image of uncapped microdisks after 30 and 60 s of PEC etching, revealing severe top surface roughening. (c-d) Temporal evolution of lasing peak intensity (mean and standard deviation of 5 disks) under PEC etch for (c) silica-capped and (d) uncapped microdisks.

In contrast, Figure 5b shows the top surface of microdisks that were etched without silica caps, which visibly affected their morphology with significant surface roughening in 30 s of PEC etch, progressively worsening with longer etch times (Supplementary Fig. S2). The porous surface indicates the production of solid components.

The significance of the silica cap is further emphasized by the temporal evolution of the laser peak intensity over an extended period. As depicted in Figures 5c and 5d, the peak intensity initially increases for both silica-capped and non-capped microdisks. In the case of silica-capped disks, this may be attributed to an increase in the Q factor, as observed in previous studies.^{23, 24} Conversely, for non-silica-capped disks, the observed roughening of the surface suggests that the Q factor is unlikely to increase. The apparent increase in lasing intensity is instead likely due to an increased scattering of laser light circulating in the disk, as reported in previous studies that employed different engineering methods.³⁶ However, the intensity trend is rather brief for non-capped disks, with the intensity rapidly decreasing after approximately 30 seconds of etching. Silica-capped disks demonstrate higher stability, with peak intensity being retained above the starting intensity for up to 120 seconds of etching on average.

3. Precision wavelength control of microdisk lasers

3.1. Sub-nanometer-scale wavelength control on disk-on-chip lasers

To achieve microlasers with a consistent wavelength, we created a 6×6 array of microdisks, each with a nominal diameter of 3 μm arranged in a square lattice of 30 μm spacing. As shown in the graph labeled (i) in Fig. 6, the initial lasing wavelength distribution of the disks produced by standard nanofabrication spans a range 9 nm around 1323 nm, which indicates a variation in disk diameter of approximately 18 nm.¹⁵ To address this, we used live-monitoring PEC etching to all the disks to a wavelength 1314.7 nm, and subsequently to 1309.6 nm, 1304.8 nm, and 1299.6 nm (Fig. 6 and Supplementary Fig. S3), each time with a standard deviation of 0.5 or 0.6 nm. The median lasing peak intensity in each of these narrowband distribution disks remained close to the starting median peak intensity (see Supplementary Fig. S4).

Figure 6.

Wavelength tuning by PEC etching. (a) Lasing wavelength histogram of an array of 36 disk-on-pillar samples (i) as fabricated by UV lithography and RIE, and the same sample tuned sequentially to give narrowband wavelength distributions of (ii) 1314.7 ± 0.5 nm, (iii) 1309.6 ± 0.5 nm, (iv) 1304.8 ± 0.5 nm, and (v) 1299.6 ± 0.6 nm.

3.2. Nanometer-scale wavelength control of isolated laser particles

Additionally, we achieved the fabrication of free-standing LPs by transferring wavelength tuned microdisk arrays onto a polydimethylsiloxane (PDMS) substrate (Fig. 7a). The wavelength distribution of two separate batches of microdisk LPs released on the PDMS substrate is presented in Fig. 7b, where the interquartile range is about 1.5 nm. The distribution is compared to the corresponding distribution observed after precision tuning while still on pillars. In both cases, the median lasing wavelength is slightly blue-shifted by less than 2 nm. This shift is partially attributed to the change in the surrounding medium’s refractive index in the two different situations — the on-pillar measurements were taken with the microdisks surrounded by the etchant solution, while the on-PDMS measurements were taken with one side of the microdisks in contact with the PDMS substrate and the other side by water. The broadening of the distribution for isolated particles is presumably caused by differences in pillar sizes, which affect lasing modes. We believe that the broadening can be reduced through improved engineering.

Figure 7.

Isolated microdisk LPs. (a) Optical micrograph showing LPs on a PDMS substrate. (b) Histogram of lasing wavelength of LPs in two different batches after on-chip PEC tuning.

4. Discussion

We have successfully utilized active PEC etching to precisely tune the lasing wavelength of InGaAsP microdisks across a 70 nm gain bandwidth. By applying this technique, we have created precision microdisks that emit lasing within an ultra-narrow bandwidth of <0.6 nm for on-pillar lasers and <1.5 nm for isolated particles. Our model accurately predicted the hole-induced oxidative etching rates based on the illumination intensity. Through automating the wavelength tuning process with a robotic setup, we can potentially achieve single-mode lasers that are precisely tuned deterministically to ultra-narrow wavelength bands with higher throughput, that can find applications in on-chip photonic communications, information processing, and sensing. Additionally, we have observed that free-standing particles isolated after releasing the disks from the pillars retained a narrow distribution with comparable median lasing wavelength values. These high-precision single-mode microdisk LPs could prove useful in barcoding for specific cell populations and assay molecules.