The downscaling of photonic components, such as light sources, waveguides, modulators, and photodetectors, is crucial for achieving large-scale photonic integration. Despite the rapid development of nanophotonic devices, fast and efficient light sources with small footprints have long been the missing piece. Over the last two decades, numerous micro- and nanoscale light sources have emerged, ranging from vertical cavity surface emitting lasers (VCSEL), (1,2) microdisk lasers, (3−5) photonic crystal lasers, (6−8) metal-dielectric nanocavity lasers, (9−11) spasers, (12,13) and nanoscale light emitting diodes (nanoLEDs). (14−16) As an important type of nanoscale light source, nanoLEDs have facilitated a wide array of research, including lensless microscopy, (17) self-emissive display, (18) optical interconnects, (14,15) and neuromorphic computing. (19) These research works span a wide range of applications, including lighting, display, communication, sensing, and computation.

In optical communication and computing systems, conventional LEDs are not widely used, primarily due to their slow modulation speed limited by the intrinsically slow spontaneous emission rate. However, with a prominent Purcell effect in nanocavities, the spontaneous emission rate can be significantly increased, and the spontaneous emission factor (β) can approach unity. Leveraging the Purcell effect, thresholdless lasing in nanolasers has been achieved. (20) In a parallel development, nanoLEDs that also harness the Purcell effect become an attractive alternative to nanolasers as fast on-chip light sources. Furthermore, compared to lasers, LEDs can work at lower input power densities and have higher power conversion efficiency at the low input power density regime before hitting the “efficiency droop” limit. (21) Owing to these merits, several studies have suggested that nanoLEDs can be a strong light source candidate for on-chip optical interconnects. (22−24) However, the experimental demonstration of a fast nanoLED is still lacking.

Here, we experimentally demonstrate a nanoscale light source with a high Purcell factor of 74, corresponding to a modulation bandwidth of 6.65 GHz. This demonstration serves as a proof-of-concept, optically pumped version of nanoLEDs for future high-speed optical communication and computation systems.

The Purcell factor is expressed as $F_{\mathrm{p}}=\frac{3}{4{\pi }^2}{\left(\frac{{\lambda }_{\mathrm{c}}}{n_{\mathrm{a}}}\right)}^3\left(\frac{Q}{V_{\mathrm{eff}}}\right)$, where λ_c is the cavity resonant wavelength and n_a is the refractive index of the active material. It reveals two strategies to increase the Purcell factor, and thus, the spontaneous emission rate in nanoLEDs: (1) minimizing the effective mode volume V_eff and (2) employing high quality (Q) factor cavities. (25,26) Minimizing V_eff or more broadly, minimizing the cavity size to the subwavelength scale in all three dimensions brings about two challenges. The first challenge is to simultaneously decrease the cavity size and mode volume below the diffraction limit. The second challenge is to employ a fully scalable cavity, namely, one that supports self-sustained electromagnetic field regardless of the cavity size. Dielectric cavities like defect-mode photonic crystals have extremely small mode volume; (6) however, at least one dimension of the cavity is larger than the diffraction limit. To achieve both goals at the same time, metallic cavities have emerged as a potential solution due to their exceptional mode confinement properties. (9,27) To address both challenges at the same time, we employ a metallic coaxial cavity, (20) in which the subwavelength coaxial cavity supports the cutoff free transverse electromagnetic (TEM)-like mode with a mode volume below 0.1 × (λ₀/n_a)³. It is worth noting that although the use of a high Q cavity increases the spontaneous emission rate, the emitted photons are also trapped in the high cavity due to strong optical feedback. Thus, increasing the cavity Q factor indefinitely is not a viable solution, and an optimized Q factor exists. (22)

The structure and material composition of the asymmetrical shifted-core coaxial nanocavity is depicted in Figure 1a. The cavity utilizes III–V semiconductor gain material composed of nine quantum wells of In_x=0.564Ga_1–xAs_y=0.933P_1–y (10 nm thick)/In_x=0.737Ga_1–xAs_y=0.569P_1–y (20 nm thick), of which the optical gain spectrum spans from 1.26 to 1.59 μm at room temperature (295 K). In the unshifted symmetric case, the InGaAsP ring has a uniform width (half of the cladding and core diameter difference) of Δ = 80 nm and a height of 300 nm. The 10 nm InP layer on top of InGaAsP quantum wells functions as a protection layer, resulting in a total III–V material height of 310 nm. In the vertical direction, the metal-dielectric interface at the top and the high dielectric contrast between InGaAsP and air at the bottom form the two mirrors in a Fabry–Perot cavity configuration, ensuring strong three-dimensional optical mode confinement. The nominal device footprint is about 320 × 320 × 310 nm³.

Figure 1b shows the Scanning Electron Microscope (SEM) image after the III–V ring formation, in which the sidewall tilting angle is controlled to be below 3°, thus mitigating the negative impact of the nonstraight sidewall on cavity Q. The postmetal deposition SEM image in Figure 1c shows a uniform silver (Ag) encapsulation around the entire cavity and properly fills the center of the InGaAsP ring. Furthermore, the top view SEM in Figure 1d provides a clear visual demonstration of samples with different amounts of core displacement. The details of the fabrication procedure are provided in the Supporting Information.

The modulation bandwidth of a spontaneous emission dominated light source is $f_{3\mathrm{d}\mathrm{B},\max }\approx \frac{1}{2\pi }\frac{1}{\sqrt{{\tau }_{\mathrm{p}}^2+{\tau }_{\mathrm{s}\mathrm{p}}^2}}$, where τ_p and τ_sp represent photon and spontaneous emission lifetimes, respectively. (22) Notably, τ_sp = τ_sp0/F_pβ, thus, a large F_p and unity β are desired to shorten the long intrinsic spontaneous emission lifetime τ_sp0. Since β describes the fraction of spontaneous emission coupled to the primary optical mode, it is essential to reduce the number of optical modes and to overlap the high F_p mode with the material gain spectrum. Employing the finite element method (FEM), we can identify all optical modes supported by the coaxial cavity, including photonic modes, hybrid photonic-plasmonic whispering gallery modes (WGMs), TEM-like modes, and gap plasmon modes. Among all modes, the TEM-like mode is self-sustained regardless of the cavity size (cutoff free) and nondegenerate, which allows us to design a single-mode cavity with exceptional scalability.

To simulate the optical properties of the coaxial cavity, we place a randomly orientated electric dipole within the optical gain region to excite all the optical modes supported in the cavity. As illustrated in Figure 2a, all the optical modes redshift when the core-shifting distance d increases, and the TEM-like mode becomes the dominant optical mode with d > 40 nm after the WGMs eventually move out of the material gain spectrum. This mode redshift can be observed in the experimental photoluminescence (PL) in Figure 2b, which shows a series of emission spectra from three devices with similar cladding and core radius difference (Δ ≅ 80 nm) but with different core-shifting distances d under a 250 μW average pump power.

Furthermore, we perform power-dependent PL measurements to investigate the pump power dependency of the nanocavity light source. In Figure 3a, comparative analysis with the unpatterned InGaAsP MQW’s PL peak at 1547.6 nm reveals that the emission peaks of the nanocavity light source qualitatively match the simulated resonances of TEM-like modes in Figure 2a. The inset of Figure 3a illustrates the log–log scale light–light curve of the single-mode light source, demonstrating a slope value greater than one. This suggests that the device operates in the amplified spontaneous emission (ASE) regime or that surface-recombination is dominant. (29,30) However, given that β ≈ 1 in this single-mode light source, the deviation from unity in the slope value should be attributed to surface recombination. Our observed slope value of 1.73 dec/dec closely aligns with previous findings of 1.62 dec/dec from an unpassivated InGaAsP ridge in a metal cavity nanoLED. (30)

In Figure 3b, we observe a blueshift in the center emission wavelength as the pump power increases, which is governed by two competing effects. First, the free-carrier plasma dispersion effect comes into play: with increasing optical pump power, the concentration of free carriers rises. This leads to an increase in the imaginary part of the refractive index while the real part decreases. Consequently, there is a reduction in the refractive index contrast vertically, resulting in a blueshift in the resonance wavelength. (31) Second, at higher pump intensities, self-heating occurs, which typically induces a redshift in the resonance wavelength. However, this effect is expected to be minimal due to the efficient heat dissipation provided by the Ag core and shell. Thus, our light source exhibits a slight blueshift in Figure 3b. Additionally, Figure 3b reveals a slight increase in the active Q factor with increasing pump power, consistent with the narrowing of emission linewidth in Figure 3a.

Figure 4 presents the measurement of carrier lifetime as well as the subsequent extraction of spontaneous emission lifetime and the corresponding Purcell factor. To ensure the measurement in the spontaneous emission dominant PL regime, Figure 4a shows our determination of PL and ASE regime via pump power-dependent time-resolved photoluminescence (TRPL) carrier lifetime τ_c measurement. Contrary to the steady-state PL measurements of Figure 3a, in which only the ASE regime can be accessed due to the low-sensitivity spectrometer, the high-sensitivity avalanche photodiode (APD) in our TRPL setup allows access to the PL regime. Within the spontaneous emission-dominated PL regime (pump power <10 μW), τ_c is relatively constant, indicating minimal pump power influence. Conversely, τ_c in the ASE regime exhibits a pronounced pump power dependency attributed to the transition from spontaneous emission to stimulated emission. Given that the Purcell effect is specific to the modification of the spontaneous emission rate, the Purcell factor measurement via TRPL should be performed in the PL regime (pump power <10 μW), where τ_sp predominantly relies on the cavity dimensions. Due to weak emission from some samples, the largest pump power (8 μW) within the spontaneous emission regime was selected to ensure enough photons were collected for precise TRPL measurements. The low optical excitation condition is further confirmed with a measurement of the InGaAsP MQW lifetime (τ_QW), see Supporting Information. The typical value of τ_QW of InGaAsP/InP MQW is 2–3 ns and can be shortened to <1 ns in nanopatterned structures due to increased surface recombination at the etched sidewalls. (32) The τ_QW from the bare InGaAsP MQW is measured as 1.78 ns and reduced to 0.77 ns after etching.

The carrier lifetime can be expressed as $\frac{1}{{\tau }_{\mathrm{c}}}=\frac{1}{{\tau }_{\mathrm{sp}}}+\frac{1}{{\tau }_{\mathrm{nr}}}$. (33) In our nanocavity, the nonradiative surface recombination is dominant (large metal loss and surface-to-volume ratio). A clear understanding of the carrier recombination dynamics in the gain material and a precise assessment of the surface recombination rate are essential to correctly extract the spontaneous emission lifetime in the nanocavity. The major carrier combination channels in InGaAsP MQWs are Shockley-Read-Hall (SRH) recombination, surface recombination, bimolecular radiative recombination, and Auger recombination, denoted by coefficients A′, B, and C, respectively. (30,34) In the low carrier injection regime, the Auger recombination term CN³ can be neglected. A′ is the combined nonradiative recombination coefficient of SRH and surface recombination. $A^{\prime }=\frac{1}{{\tau }_{\mathrm{nr}}}=\frac{1}{{\tau }_{\mathrm{sr}}}+\frac{1}{{\tau }_{\mathrm{SRH},\mathrm{bulk}}}$ with ${\tau }_{\mathrm{sr}}=\frac{V}{A_{\mathrm{S}}}·\frac{1}{v_{\mathrm{s}}}$, where τ_SRH:bulk is the bulk SRH recombination lifetime, τ_SR is the surface recombination lifetime, and v_s is the surface recombination velocity. For nanoscale devices, τ_SRH:bulk ? τ_SR generally, allowing the contribution from bulk to be neglected, thus facilitating a direct relationship between the measured lifetime and surface recombination velocity. One of two formulas can be used to determine the total carrier lifetime, depending on whether carriers are coupled with the cavity mode. (14) They are $\frac{1}{{\tau }_{\mathrm{cav}}}=\frac{F_{\mathrm{p}}}{{\tau }_{\mathrm{QW}}}+\frac{1}{{\tau }_{\mathrm{nr}}}\mathrm{an}\mathrm{d}\frac{1}{{\tau }_{\mathrm{un}}}=\frac{1}{{\tau }_{\mathrm{QW}}}+\frac{1}{{\tau }_{\mathrm{nr}}}$, where τ_cav is the lifetime of QW emission coupled to the cavity, τ_un is the lifetime of QW emission not coupled to the cavity, τ_QW is the intrinsic spontaneous emission lifetime of QW, and τ_nr accounts for carrier decay rates including all nonradiative recombination. According to a previous study, the bare InGaAsP has a surface recombination velocity on the order of ∼10⁴ cm/s. (35) By measuring the carrier lifetime of QW emission not coupled to the cavity (τ_un), we can calculate τ_nr and extract the surface recombination rate to be 1.8 × 10⁴ cm/s. Combining this information with the device carrier lifetime (τ_cav) and τ_QW, we can extract the devices’ Purcell-enhanced spontaneous emission lifetime and Purcell factor accordingly. Details of the device carrier lifetime measurements and spontaneous emission lifetime analysis can be found in the Supporting Information.

Figure 4b illustrates the correlation between spontaneous emission lifetime, the Purcell factor, and the gap distance between the Ag core and shell. We choose the gap distance, rather than the metal core/shell diameter or the InGaAsP ring width, to be the geometric parameter to vary, because the gap distance is the determining parameter of the Purcell factor. Purcell factors ranging from 18 to 74 are observed in our shifted-core coaxial cavity light source. It is noteworthy that surface recombination is primarily influenced by the surface-to-volume ratio rather than the symmetry breaking of the cavity, thus, shifting the metal core shall not affect the surface recombination rate.