In-plane photonic integrated circuits (PICs) have been attracting more and more attention [1–5]. The in-plane PIC can integrate numerous active and passive photonic devices in a monolithic integration manner or a hybrid integration manner on a common substrate. The in-plane PIC exhibits diverse functions such as wavelength tuning, filtering, switching, routing, amplifying, modulation, and detection. One feature of the in-plane photonic integration scheme is the easy coupling of the edge-emitting laser by butt-joint coupling or evanescent coupling to the waveguide connecting the modulator, photodiode (PD), and other passive devices [6–8]. The in-plane photonic integration can be carried out in the process stage or the packaging stage in the wafer scale with a reduced cost, size, weight, and power (CSWaP). In many PICs, the wavelength-tunable laser and the power monitor PD are basic blocks [9–11]. Nowadays, the in-plane PICs are used in artificial intelligence (AI), sensing, data centers, light detection and ranging, and so on [12–17].

Compared to the edge-emitting laser used for the in-plane PIC, the vertical-cavity surface-emitting laser (VCSEL) has many advantages, such as high modulation speed, low cost, low power, small footprint, two-dimensional (2D) array, on-wafer testing, and circular beam [18–21]. Therefore, VCSELs have been proposed to serve as the laser sources of the in-plane PICs [22–27]. Since the emitted light is normal to the substrate for the VCSELs, VCSELs are the ideal laser sources for the photonic systems that have been used in three-dimensional (3D) sensing, optical computing, and deep learning [28–30]. In the future, the 3D vertically integrated photonic system based on VCSELs will be promising in 3D sensing and AI applications [28,29], compared with the bulky photonic system packaged with discrete photonic devices and optical elements.

VCSELs integrated with other active and passive photonic devices will be the basic blocks in the future 3D vertically integrated photonic system [28–30]. There have been many reported integrations of VCSELs with other photonic devices. Paraskevopoulos et al. and Germann et al. ever monolithically integrated an electro-optic modulator (EOM) section into the top distributed Bragg reflector (DBR) of a VCSEL [31,32]. The EOM-VCSEL has two p-i-n structures cascaded in an n-i-p-i-n or p-i-n-i-p manner. The EOM-VCSEL is expected to outperform the conventional VCSEL and has a small signal bandwidth ($-3\mathrm{dB}$) of larger than 60 GHz.

The VCSEL integrated with a PD has also been demonstrated. The p-i-n PD can be monolithically integrated below the $n$-type DBR, in the $n$-type DBR, or above the $p$-type DBR of the VCSEL for bidirectional data communication and sensing [33–35]. This monolithic integration scheme avoids the disadvantages of external power monitoring schemes with bulky, costly, and complex optics for VCSELs.

The integration of a VCSEL with passive optical elements provides the ability to manipulate the output beam from the VCSEL. The VCSELs integrated with a microlens, metasurface, and phase plate can realize collimated beams, deflected beams, focused beams, orbital angular momentum beams, and so on [36–39]. However, these optical elements are placed out of the optical cavities. As a neat photonic element, the high-contrast grating (HCG) with broadband high reflectivity has been intensively studied in the past years. The HCG as a highly reflective mirror is integrated into the VCSEL as one part of the optical cavity. The HCG-VCSEL shows single-mode operation, single-polarization output, on-chip beam shaping, high modulation bandwidth, and fast wavelength tuning [40–47]. Compared to the tunable micro-electro-mechanical system (MEMS) VCSEL with a movable DBR, the tunable HCG-VCSEL has a higher tuning speed with a mechanical frequency response bandwidth ($-3\mathrm{dB}$) of 27 MHz [48].

Wavelength-tunable HCG-VCSELs with n-i-p-i-n structures have been well studied [49–51], since tunable MEMS-VCSELs are important laser sources in optical coherence tomography (OCT), frequency modulated continuous wave (FMCW) light detection and ranging (LiDAR), optical interconnects, gas sensing, and so on [52–62]. The $n$-type suspended HCG and the bottom DBR form the vertical cavity. The $n$-type suspended HCG moves toward the active region by the electrostatic force under the reverse-bias voltage. The length of the optical cavity is shortened, and the wavelength is blue-shifted. To avoid optical absorption, the sacrificial layer beneath the HCG layer is intentionally chosen to be transparent at the operation wavelength and also be lattice-matched with the wafer substrate.

In future 3D vertically integrated photonic architectures, the VCSEL array [i.e., space division multiplexing (SDM)] with fast wavelength tuning ability [i.e., time division multiplexing (TDM)] in a wavelength division multiplexing (WDM) approach will enhance the throughput capacity. Parallel high-dimensional information transmission and processing can be achieved. Each wavelength can be considered as an independent channel. By increasing the quantities of available wavelengths, the throughput capacity of the system can be scaled, leading to a very high computational throughput [16,17,63–69]. In such a 3D vertically integrated photonic architecture, the power monitor PD is essential as in the in-plane PICs. Also, the VCSEL and its arrays with integrated power monitor PDs will be beneficial to the optical interconnects, sensing, and imaging applications.

Recently, VCSELs at 940-nm wavelength range have been attracting intensive attention, because they are ideal laser sources for sensing applications [70,71]. Compared with 850-nm wavelength range, at 940-nm wavelength range the background from the sun is lower, the red glow becomes less visible, eye-safety margins are higher, and the Si-based CMOS image sensor is still sufficiently sensitive for detection. Also, 940-nm VCSELs are important in optical interconnects [72–74].

In this paper, we demonstrate a detector-integrated VCSEL with a movable HCG mirror in an n-i-p-i-n manner like the tunable HCG-VCSEL. The movable HCG mirror for wavelength tuning is realized by removing the ${\mathrm{Ga}}_{0.51}{\mathrm{In}}_{0.49}\mathrm{P}$ sacrificial layer with a band gap larger than the photon energy of the 940-nm range. The device demonstrates functions such as a tunable HCG-VCSEL, monitor PD, and resonant-cavity-enhanced PD. The detector-integrated VCSEL with a movable HCG mirror has promising applications in the future 3D vertically integrated photonic systems for sensing, imaging, AI application, and so on.

The structure of the detector-integrated VCSEL with a movable HCG mirror is like the tunable HCG-VCSEL that is developed from our reported HCG-VCSEL in Ref. [43]. Figures 1(a) and 1(b) show the schematic and doping profile of the 940-nm detector-integrated VCSEL with a movable HCG mirror, respectively. The device has a bottom $n$-type 38.5-pair ${\mathrm{Al}}_{0.12}{\mathrm{Ga}}_{0.88}\mathrm{As}/{\mathrm{Al}}_{0.90}{\mathrm{Ga}}_{0.10}\mathrm{As}$ DBR on a GaAs substrate, a $\lambda$-cavity, $p$-type current spreading layers, a highly $p$-doped GaAs layer, an undoped ${\mathrm{Ga}}_{0.51}{\mathrm{In}}_{0.49}\mathrm{P}$ sacrificial layer, and an $n$-type top suspended HCG with four arms. The $p$-type current spreading layers are composed of three pairs of $p$-type ${\mathrm{Al}}_{0.12}{\mathrm{Ga}}_{0.88}\mathrm{As}/{\mathrm{Al}}_{0.90}{\mathrm{Ga}}_{0.10}\mathrm{As}$ DBRs. The cavity consists of three InGaAs quantum wells surrounded by AlGaAs barriers. The suspended HCG is formed by removing the ${\mathrm{Ga}}_{0.51}{\mathrm{In}}_{0.49}\mathrm{P}$ sacrificial layer, which is lattice-matched to the GaAs substrate and has a thickness of about 1 μm. The size of HCG is $12\mathrm{\mu m}\times 12\mathrm{\mu m}$. The length and width of the four arms supporting the HCG are 5 μm and 2 μm, respectively. Figure 1(c) shows the scanning electron microscope (SEM) image of the fabricated device. Figure 1(d) shows the SEM image of the suspended HCG. Theoretically, the ${\mathrm{Ga}}_{0.51}{\mathrm{In}}_{0.49}\mathrm{P}$ sacrificial layer has a band gap of 1.85 eV and the GaAs layer has a band gap of 1.42 eV [75]. Both materials are transparent at 940-nm wavelength range.

The vertical cavity of the HCG-VCSEL is formed by the bottom DBR, active region, and top HCG-based mirror (including the current spreading layers, the air layer, and the movable HCG). The PD part of the device is composed of the $n$-type HCG, the undoped sacrificial layer, and the $p$-type current spreading layers. The PD $n$-contact and the $p$-contact are reversely-biased to move the suspended HCG toward the active region by electrostatic force. As a result, the physical length of the optical cavity is reduced to tune the resonant wavelength. The current is injected into the quantum wells to provide a sufficient gain through the $p$-contact and the $n$-contact at the backside of the substrate. When the PD $n$-contact and the $p$-contact are applied with reverse-bias voltage, the $n$-type HCG, intrinsic ${\mathrm{Ga}}_{0.51}{\mathrm{In}}_{0.49}\mathrm{P}$ layer, and $p$-type GaAs form an n-i-p PD to monitor the power of the HCG-VCSEL. An oxide aperture is formed by the oxidation of the ${\mathrm{Al}}_{0.98}{\mathrm{Ga}}_{0.02}\mathrm{As}$ layer among the $p$-type current spreading layers to provide the current and optical confinements.

In our design, the field intensity in the semiconductor cavity is larger than that in the air layer. The HCG-VCSEL is a semiconductor-cavity dominant (SCD) design [50]. Theoretically, the wavelength tuning range with mode hopping is about 30 nm with the air gap changing from 1 to 0.7 μm for this SCD design, as shown in Fig. 2. As we know, the semiconductor–air coupling (SAC) region between the air cavity and the semiconductor cavity can determine the cavity structure of the tunable MEMS-VCSEL [50]. In the SCD design, the SAC region comprises a window layer (WL) on the top of the quarter-$\lambda$ high-index layer. When the SAC region comprises a WL on the top of the quarter-$\lambda$ low-index layer, an air-cavity dominant (ACD) design occurs. The tuning range and tuning efficiency of the ACD design are higher than those of the SCD design. A larger wavelength tuning range without mode hopping is possible with an ACD design, which is our future work.

The designed wafer was grown by the metal–organic chemical vapor deposition on a GaAs substrate. The fabrication process flow of the device is shown in Appendix A. The wafer was first processed for the AuGeNiAu metal for the PD $n$-contact by the lift-off process. The $n$-type top GaAs was selectively etched with the photoresist as the mask. The ${\mathrm{Ga}}_{0.51}{\mathrm{In}}_{0.49}\mathrm{P}$ layer was selectively etched by the HCl solution to expose the $p$-contact layer for the current injection into the active region. After the first mesa etching was finished, the TiAu metal was deposited for the $p$-contact by the lift-off process. The second mesa was formed by the inductively coupled plasma etching to expose the ${\mathrm{Al}}_{0.98}{\mathrm{Ga}}_{0.02}\mathrm{As}$ layer. The ${\mathrm{Al}}_{0.98}{\mathrm{Ga}}_{0.02}\mathrm{As}$ layer was oxidized under 420°C in the ${\mathrm{H}}_2\mathrm{O}$ atmosphere to form an oxide aperture for the current and optical confinements. The AuGeNiAu metal was deposited for the $n$-contact at the backside of the substrate.

To fabricate the HCG, the ZEP 520A electron beam resist was coated on the top GaAs layer. Electron beam lithography was used to define the grating pattern. Then the ICP etching was performed to transfer the grating pattern into the top GaAs layer with ZEP 520A as the mask to expose the ${\mathrm{Ga}}_{0.51}{\mathrm{In}}_{0.49}\mathrm{P}$ layer. The ${\mathrm{Ga}}_{0.51}{\mathrm{In}}_{0.49}\mathrm{P}$ layer under the grating pattern was selectively etched by the HCl solution. Finally, the suspended HCG was released with a critical pointing dryer. The calculated reflectivity spectra of the HCG under different structure parameters shown in Appendix B were obtained by the rigorous coupled-wave analysis method. The transverse electric (TE) HCG has a high reflectivity ($>99.5\%$) at 940-nm range with a period of 648 nm, a bar width of about 220 nm, and a thickness of about 140 nm.

As shown in Fig. 1, the detector-integrated VCSEL with a movable HCG has an n-i-p-i-n structure with three terminals. This device shows different functions when the terminals are connected in different manners.

The HCG-VCSEL can work with the injection current between the $p$-contact and the $n$-contact. Figure 3 shows the power-current ($P\text{-}I$) curve of a device without reverse-bias voltage between the PD $n$-contact and the $p$-contact. The threshold current is 0.6 mA under continuous-wave (CW) operation at room temperature (25°C). The left inset shows the spot image below the threshold current. A speckle pattern in the lasing spot image is shown in the right inset when the current is larger than the threshold current.

By applying a reverse-bias voltage between the PD $n$-contact and the $p$-contact with the setup shown in Appendix C, the lasing wavelength of the HCG-VCSEL can be tuned. Figure 4(a) shows that as the reverse-bias voltage is increased from 0 to 9 V, a continuous wavelength tuning range of 11.72 nm is first obtained. Above 10 V, the device hops to another longitudinal mode around 942 nm. Another continuous wavelength tuning range of 15.48 nm is obtained from 10 to 15 V. A total wavelength tuning range of 27 nm through mechanical MEMS actuation at 25°C is obtained with the reverse-bias voltage ranging from 0 to 15 V. Figure 4(b) shows the relationship between the wavelength of the fundamental mode and the tuning voltage, deduced from Fig. 4(a). The laser does not operate in a single transverse mode, possibly due to the imperfect fabrication of the HCG. The tuning range of the MEMS-VCSEL is determined by the smallest of the following three factors: the wavelength shift range resulting from the maximum movement range of the MEMS, the free spectral range, and the reflectivity bandwidth of the reflector or the gain bandwidth [50]. In future work, by operating on the steeper part of the wavelength tuning curves and/or reducing the number of DBRs in the $p$-type current spreading layers, a larger wavelength tuning range will be possibly obtained [41,76,77]. With the increase of the reverse-bias voltage, the HCG undergoes irregular deformation due to the electrostatic force. With the finite element method, the deformation of the HCG under the reverse-bias voltage of 8 V is calculated, as shown in Appendix D. The irregular deformation can lead to the poor performance of HCG, increasing the threshold current and limiting the mechanical movement distance of HCG. These impacts can be relieved by increasing the arm length and decreasing the arm width.

Wavelength tuning of the device can be achieved by applying a reverse-bias voltage between the PD $n$-contact and the $p$-contact. By measuring the photocurrent between the PD $n$-contact and the $p$-contact, the real-time power monitoring of the tunable HCG-VCSEL can be realized. The upper part of the tunable HCG-VCSEL is an integrated n-i-p PD. Even though the energy gap of ${\mathrm{Ga}}_{0.51}{\mathrm{In}}_{0.49}\mathrm{P}$ is 1.85 eV and GaAs is 1.42 eV [75], this integrated n-i-p PD can work at 940-nm wavelength range by the defect absorption to generate the photocurrent. The integrated n-i-p PD detects the scattered light inside the optical cavity. Thus, the laser output power can be measured without changing the laser cavity.

Figure 5(a) shows the $P\text{-}I$ curve of the HCG-VCSEL and the photocurrent curve of the integrated n-i-p PD under zero bias voltage with the setup described in Appendix C. The experiment was conducted in a dark room, and the dark current of the integrated n-i-p PD under zero bias voltage is in the order of 1 pA, which can be ignored. As the injection current increases, the output power of the device increases, and the photocurrent of the integrated n-i-p PD also increases. The photocurrent curve and power curve in Fig. 5(a) have a kink at the threshold current. This indicates that the photocurrent of the integrated n-i-p PD is directly related to the output power of the VCSEL. Figure 5(b) shows the photocurrent-power curve of the detector-integrated VCSEL with a movable HCG, which is deduced from Fig. 5(a). There is a kink at about 0.08 mW in the curve. Below the kink point, the slope of the curve is higher than that above the kink point. The possible reason is that when the current is below the threshold current, the device is in broadband spontaneous emission. Thus, the integrated n-i-p PD can absorb the broadband light and have a higher effective responsivity [33]. It should be noted that the integrated n-i-p PD operates by absorbing the scattered light from the cavity, not the light transmitting through the HCG. The real optical power absorbed by the integrated n-i-p PD is less than the measured optical power by an external detector. Thus, the measured photocurrent is in the order of μA.

To investigate the mechanism of photocurrent generation in the integrated n-i-p PD, a device without a patterned HCG was investigated in the experiments. The oxide layer in the device has been fully oxidized, and the oxidation aperture is closed (Appendix E). Therefore, the integrated n-i-p PD is electrically isolated from the lower part below the oxide layer to avoid the crosstalk from the lower part. The central wavelengths of external semiconductor lasers are 960 nm and 1062 nm, respectively. Figures 6(a) and 6(b) show the photocurrent-power curves with kinks, similar to Fig. 5(b). The kinks are related to the kinks at the power-current curves of the 960-nm and 1062-nm semiconductor lasers (Appendix E). The kinks in Figs. 5(b), 6(a), and 6(b) indicate that the photocurrent is generated by the integrated n-i-p PD. Deduced from Figs. 6(a) and 6(b), as shown in Fig. 6(c), the integrated n-i-p PD has a linear response within the optical power range from 40 to 90 mW. Figure 6(d) shows a linear response of the integrated n-i-p PD from 20 to 40 mW. However, the relationship between the photocurrent and the optical power is nonlinear across the whole incident power range. There is a kink in the photocurrent-power curve at the threshold current. Below the threshold current, spontaneous emission occurs, and stimulated emission occurs above the threshold current. Thus, the responsivity curve calculated by the ratio of the photocurrent to the power is nonlinear. According to the quantum efficiency [78] $${\eta }_e=R\frac{h\nu }{e},$$(1)and the responsivity $$R=\frac{I_{\mathrm{ph}}}{P},$$(2)where $h\nu$ is the photon energy, $e$ is the electronic charge, $I_{\mathrm{ph}}$ is the photocurrent, and $P$ is the incident power, the quantum efficiency of the integrated n-i-p PD with the light from the 960-nm semiconductor laser is about $5.4\times {10}^{-6}$ within the incident power ranging from 40 to 90 mW. This linear relationship deviates slightly at the higher powers ($>90\mathrm{mW}$), possibly caused by the electron trapping saturation in the integrated n-i-p PD [79]. Similarly, with the light from a 1062-nm semiconductor laser, the quantum efficiency is about $8.2\times {10}^{-8}$ within the incident power ranging from 20 to 40 mW. However, the integrated n-i-p PD has no response under the light from a 1550-nm laser (Santec TSL-510). The experiments conducted with the three laser sources indicate that the photocurrent of the integrated n-i-p PD decreases as the wavelength increases, and the cutoff occurs between 1062 and 1550 nm.

Since the integrated n-i-p PD has a linear photocurrent response with the optical power, the photocurrent responses in Figs. 6(a) and 6(b) are related to the deep energy levels in the band gap [80]. There are two possible mechanisms for the linear sub-bandgap absorption process, photo-assisted tunneling (PAT) and the photo-assisted Shockley–Read–Hall (PASRH) mechanism [78]. The PAT mechanism excites electrons in the valence band to deep energy levels by the photon which has an energy less than the band gap. Because the external electric field can change the spatial gradient of the energy band, the tunneling width becomes narrower, which helps the electrons at the deep energy levels tunnel to the conduction band. The absorption probability of sub-bandgap photons caused by these tunneling processes depends on the applied external electric field intensity and the volume density of defects [78,81]. Therefore, the generated photocurrent based on the light-assisted tunneling mechanism increases significantly as the external reverse-bias voltage increases. However, our experimental results do not show the strong dependence of the dark current and the photocurrent on the reverse-bias voltage, as shown in Fig. 7. The photocurrent almost remains the same, as the reverse-bias voltage increases under different incident powers. Hence, we think the PAT is not the root of the photocurrent in the integrated n-i-p PD [78].

Therefore, the linear relationship between the photocurrent and the power in Fig. 5(b) is likely due to the PASRH mechanism. The PASRH mechanism is that photons smaller than the band gap can ionize electrons at deep energy levels to the conduction band, and the empty quantum state at the deep energy levels can enhance the SRH process (i.e., PASRH) [82]. The experimental results in Figs. 6(a) and 6(b) can be explained by the expression [82] $$J_{\mathrm{ph}}=J_d+J_d{\sigma }_{\mathrm{op}}\mathrm{\Phi }(\frac{1}{e_n}+\frac{1}{e_p}-\frac{2}{e_n+e_p}),$$(3)where $J_{\mathrm{ph}}$ is the photocurrent density, $J_d$ is the dark current density, ${\sigma }_{\mathrm{op}}$ is the cross section of the optical transition, $\mathrm{\Phi }$ is the photon flux, and $e_n$ and $e_p$ are the thermal emission rates of electrons and holes from the impurity level to the conduction and valence bands, respectively. It can be found that the photocurrent is proportional to the magnitude of the optical power, which is consistent with the experimental results in Figs. 6(a) and 6(b).

The detector-integrated VCSEL with a movable HCG can also serve as a resonant-cavity-enhanced (RCE) PD with or without a reverse-bias voltage. The active region of the device serves as the absorption region and generates photocurrent in response to the external incident light. The photocurrent versus voltage ($I\text{-}V$) curves with and without 940-nm incident light are shown in Fig. 8(a), measured with a micro–nano spectral response measurement system (Zolix DSR300). This is a typical $I\text{-}V$ curve of an RCE PD, and the dark current is in the order of 1 nA. Since the top HCG-based mirror and the bottom DBR together with the absorption region form a resonant cavity, the resonant cavity can recycle the light within the cavity to effectively enhance the light absorption in the active region [83]. The relative responsivity of a device under the zero bias voltage is shown in Fig. 8(b). The response peak wavelength is around 926 nm. The device has an obvious wavelength selectivity as an RCE PD, and the quantum efficiency at the resonant wavelength is selectively enhanced.

It can be found that the resonant wavelength ($\sim 926\mathrm{nm}$) of the RCE PD is smaller than the lasing wavelength ($\sim 945\mathrm{nm}$) of the HCG-VCSEL shown in the inset of Fig. 8(b). The possible reason is that, as an HCG-VCSEL, the carrier density and junction temperature under current injection can affect the refractive index distribution of the device, and its lasing wavelength is longer than the resonant wavelength of the RCE PD [84].

To summarize, we have realized a 940-nm detector-integrated VCSEL with a movable HCG mirror in an n-i-p-i-n manner. The device can achieve three functions, tunable HCG-VCSEL, monitor PD, and RCE PD. As a tunable VCSEL, a total wavelength tuning range covering 27 nm is obtained. Because of the deep energy level absorption of the integrated n-i-p PD, the integrated n-i-p PD can measure the output power as a monitor PD without changing the laser cavity. The defect absorption mechanism of the n-i-p PD is possibly the PASRH mechanism. The detector-integrated VCSEL with a movable HCG can also serve as a PD by reversing the bias voltage or at zero bias voltage. We demonstrate the RCE PD function, which has a response peak wavelength of around 926 nm. This device configuration can be extended to other wavelength ranges.

A larger wavelength tuning range of the device and higher quantum efficiency will be achieved by optimizing the HCG, the offset between the cavity resonance and the gain peak, and cavity design in future work [50]. With the beam-shaping ability of the HCG, the detector-integrated VCSEL with a movable HCG can integrate the phase mask function and manipulate the phase [28,29,85,86]. The defect absorption mechanism and two-photon absorption mechanism used in the monitor PD can extend the operation wavelength and provide another freedom to choose the sacrificial layer in the device [87]. This detector-integrated VCSEL with a movable metastructure including 1D and 2D HCGs is promising in applications such as OCT, FMCW LiDAR, and sensing.