Semiconductor heterostructure lasers are key enabling devices in our Internet society governed by the science of light: photonics. Outstanding properties of such lasers are the high efficiency of the conversion of electrical energy to photons, the small footprint, the modulation bandwidth, the excellent reliability, and the low cost. Important applications include optical communication, optical storage, sensing, printing, pumping for other lasers, and micro-invasive surgery. The vertical-cavity surface-emitting laser (VCSEL) arose from the ideas and contributions of several research groups. The first semiconductor laser emitting perpendicular to the surface of a semiconductor wafer traces back to the work of Melngailis in Ref. ^[1]. A surface-emitting laser diode on an InP substrate with a 90 μm long cavity and planar metal mirrors that operated at 77 K was demonstrated by Soda et al., which is often recognized as the inspiration for what is now known as a VCSEL ^[2]. Since this early publication on a surface-emitting laser diode, which is separate from the idea of using a grating to produce lasing perpendicular to the wafer surface for a conventional horizontal Fabry–Perot cavity, numerous research groups have contributed to the evolution and improvement of the device we now call a VCSEL, including the seminal idea of using semiconductor distributed Bragg reflectors (DBRs) as mirrors ^[3,4], the use of a DBR to form a surface-emitting laser ^[5–8], the use of quantum wells (QWs) placed at optical field intensity antinodes in the optical cavity ^[9–13], and the use of oxide apertures ^[14–17]. Many other significant and pioneering contributions on VCSELs, VCSEL arrays, and VCSEL applications were made by many research groups from the 1980s to the present ^[18–29]. These contributions are far too numerous to list or adequately describe in detail here. We nonetheless pay tribute to those distinguished researchers who led the epic journey of the VCSEL from the 1960s to the present day. Significant advances in epitaxial growth and device processing schemes circa the 1980s to the present have also contributed significantly to the advancement of VCSELs.

A modern VCSEL is constructed by two DBRs aligned in the vertical direction to form a vertical laser cavity, and the light output is along the same vertical wafer epitaxial growth direction normal to the wafer. Such a unique device structure is very different from an edge-emitting laser, which endows a VCSEL with many notable advantages and performance attributes as compared to other types of laser diodes. The small cavity volume brings a low threshold current, high quantum efficiency, and high-speed modulation at a low current. The symmetric transverse structure and surface emission of a VCSEL lead to a circular and non-astigmatic beam, good for coupling to an optical fiber and other optics. Surface emission also enables a large-scale two-dimensional (2D) VCSEL array and wafer level fabrication and testing for an overall low cost of manufacturing.

VCSELs have experienced exponential growth in key photonics application areas, for example, in data communication, optical sensing, laser printing, light detection and ranging (LiDAR), and illumination, to mention a few ^[18–29]. In this paper, we will focus on GaAs-based VCSELs for data communication and sensing that emit at 850–1000 nm. Our overview in Section 2 is applicable to InP-based and GaN-based VCSELs as described in, for example, Refs. ^[30–33].

The modern VCSEL is constructed by combining a gain region within an optical cavity with two highly reflective DBRs to form a vertical resonator, as shown in Fig. 1. The gain region is predominantly a set of one or more QWs or sheets of quantum dots that are bunched together and positioned at the central antinode of the standing wave on resonance of the vertical resonator to provide efficient optical amplification. The DBRs are composed of alternating layers of high and low refractive index materials, where the optical thickness of each layer is a quarter-wavelength. The DBR mirror attains a power reflectance ($R$) of 0.99 or larger at the resonant design wavelength. The top and bottom metal contacts inject current into the central active gain region. Oxide apertures are formed by the selective thermal oxidation of Al-rich AlGaAs layers. These oxide layers, which are typically 20–30 nm in thickness, confine the current and, at the same time, guide the optical field intensity within the VCSEL. The use of oxide apertures can effectively reduce the optical losses, reduce the threshold current, and enhance the VCSEL’s power conversion efficiency. The output light can be emitted by a top coupling DBR when the DBR’s $R$ is made to be below 0.999, while the $R$ of the bottom DBR mirror is made to be near 1.0. In general, the coupling DBR mirror has an $R$ at the emission wavelength between about 0.98 and 0.999. Alternatively, the $R$ of the top DBR may be set close to 1.0 by increasing the number of DBR periods, while the $R$ of the bottom DBR is set to below 0.999 so the VCSEL emission may be directed downward through the substrate and out into free space (usually into the air). In this scenario, ideally it is desirable that the substrate is transparent to the lasing wavelength, and thus the power absorptance of the bottom DBR is zero.

Figure 2 illustrates a small-signal model of a VCSEL together with the high-frequency driving source, consisting of a voltage source $v_s$ and a characteristic impedance $Z_0$. The dynamical behavior of semiconductor lasers can be described using a rate equation model ^[34]. The modulation rate of a VCSEL is limited by the intrinsic damping, self-heating effects, and electrical parasitics. The intrinsic modulation response of a semiconductor laser can be expressed by the transfer function ^[34] $$H_{\text{in}}(f)=A\times \frac{f_r^2}{f_r^2-f^2+i\frac{f}{2\pi }\gamma },$$(1)where $A$ is a constant, $f_r$ is the relaxation resonance frequency, $i$ is the imaginary unit, and $\gamma$ is the damping factor.

The relaxation resonance frequency is an oscillation frequency between the carriers and photons that interact via stimulated emission in the laser cavity. The relaxation resonance frequency increases with the square root of the bias current and is formulated as $$f_r=D\sqrt{I-I_{\mathrm{th}}},$$(2)where the $D$-factor is $$D=\frac{1}{2\pi }\sqrt{\frac{{\eta }_i\mathit{\Gamma }{v}_g}{q{V}_a}\times \frac{\partial g/\partial n}{\chi }},$$(3)and where $I_{\mathrm{th}}$ is the threshold current, $\mathit{\Gamma }$ is the optical confinement factor, ${\eta }_i$ is the internal quantum efficiency, $v_g$ is the photon group velocity, $V_a$ is the volume of the active region, $\partial g/\partial n$ is the differential gain, and $\chi$ is the transport factor.

The damping factor $\gamma$ limits the laser diode’s achievable bandwidth jointly with the relaxation resonance frequency, leading to a flatter frequency response as γ increases. The frequency response is a generally desired advantage of VCSELs for data transmission since this may improve the optical eye opening and thus the achievable error-free bit rate. The damping factor increases with an increase of the relaxation resonance frequency as given in the following equation: $$\gamma =K{f}_r^2+{\gamma }_0,$$(4)where the $K$-factor is $$K=4{\pi }^2({\tau }_p+\frac{ϵ\chi }{v_g\partial g/\partial n}),$$(5)${\tau }_p$ is the photon lifetime, and $ϵ$ is the gain compression factor. The damping offset ${\gamma }_0$ is inversely proportional to the differential carrier lifetime.

To achieve a high modulation bandwidth, a large $D$-factor and a reasonably low $K$-factor are preferred, i.e., we seek a high differential gain using strained QWs ^[35–37], a low photon lifetime by tuning the phase of the top DBR ^[38,39], and a high confinement factor by employing a short cavity and a small oxide aperture ^[40–42]. In reality, the relaxation resonance frequency and damping factor do not linearly increase with $\sqrt{I-I_{\mathrm{th}}}$ and $f_r^2$, respectively, because of thermal effects and additional loss mechanisms that occur at high currents ^[20]. The thermal effects in high-speed VCSELs can be relieved by reducing the series resistance of the VCSEL ^[43], using DBRs with a high thermal conductivity ^[44,45], employing copper-plated heat sinks ^[46], and bonding VCSEL chips to efficient heat sinks ^[47].

Electrical parasitics including bonding pads eventually limit the modulation bandwidth of a VCSEL. The transfer function introduced by electrical parasitics can be expressed by a single-pole low-pass filter transfer function as $$H_p(f)=B\times \frac{1}{1+i(f/f_p)},$$(6)where $B$ is a constant, and $f_p$ is the parasitic cutoff frequency. The parasitic resistances are mainly contributed by the series resistance of the DBRs, the junction resistance, and a resistance due to the oxide apertures. The parasitic capacitances arise primarily from the intrinsic diode junction, one or more oxide layers, the area of the mesa containing the aperture, and the metal contact pads. To reduce parasitic resistances, modifications of conduction and valence band interfaces in DBRs and modulated doping profiles are adopted while keeping any absorption loss to a minimum ^[48]. Thick insulating materials like polyimide and benzocyclobutene (BCB) with a low dielectric constant underneath the signal pad are used to planarize the mesa and reduce the pad capacitance ^[49,50]. Multiple deep oxidation layers and proton implantation are also employed to reduce the mesa capacitance ^[51–53].

The overall electrical small-signal modulation response of a VCSEL is given by multiplying the intrinsic transfer function with the transfer function introduced by the electrical parasitics, and thus $$H(f)=H_{\mathrm{in}}(f)\times H_p(f)=C\times \frac{f_r^2}{f_r^2-f^2+i\frac{f}{2\pi }\gamma }\times \frac{1}{1+i\left(\frac{f}{f_p}\right)},$$(7)where $C$ is a constant ^[34].

VCSELs are the dominant optical sources for multimode fiber (MMF)-based optical links in data centers and high-performance computers (HPCs). Since the network traffic exponentially increases year by year, wide modulation bandwidth is an essential requirement to meet the thirst for ever higher transmission data rates, and more and more power in data centers and HPCs is consumed. The heat dissipated leads to an operating environment reaching temperatures of 85°C, even with advanced cooling technologies. The latest communication system designs seek to develop VCSELs that are more energy efficient and that are capable of maintaining their room temperature modulation bit rates at high temperatures without adjusting the operating parameters.

Table 1 lists some typically reported modulation bandwidths and error-free bit rates of VCSELs via the standard on–off keying modulation format in a back-to-back measurement configuration from the past decade. The error-free bit rates are determined from transmission tests where the reported bit error ratio (BER) is $<{10}^{-12}$. The 850 nm wavelength is presently the standard for data communication based on systems using the standard OM3, OM4, or the new OM5 MMFs. Initially, conventional 850 nm VCSELs employed unstrained GaAs QWs with AlGaAs as the barrier layers. The highest data rate of 30 Gbps at 25°C was reported for a 6 μm oxide aperture diameter VCSEL with GaAs QWs in 2008 ^[55]. InGaAs QWs have a larger differential gain compared to unstrained GaAs QWs and are adopted for contemporary high-speed VCSELs for emission at or near 850 nm ^[36,37]. In 2009, error-free data transmission of 32 Gbps at 25°C was achieved with InGaAs QWs and double oxide layers by Chalmers University of Technology (CUT) ^[56]. An error-free transmission rate of 39 Gbps and open-eye operation at 40 Gbps at 25°C were reported in the same year by Technische Universität (TU) Berlin ^[62]. In 2012, CUT demonstrated a high-speed VCSEL with 44 Gbps error-free transmission and a modulation bandwidth of 28 GHz at 25°C ^[58]. This result was achieved by using a half-wavelength cavity for improved transport and a larger longitudinal optical confinement, adding four ${\mathrm{Al}}_{0.96}{\mathrm{Ga}}_{0.04}\mathrm{As}$ layers to reduce parasitics and tuning the photon lifetime by shallow surface etching. Later, error-free transmission of 57 Gbps in a VCSEL with a 3 dB bandwidth of 24 GHz at 25°C was reported ^[59]. This error-free transmission rate was pushed to 71 Gbps with two-tap feed-forward equalization in a collaboration between CUT and International Business Machines (IBM) Corporation ^[73]. A modulation bandwidth of 30 GHz was reached for a 3.5 μm oxide aperture diameter VCSEL by using a half-wavelength cavity, four deep oxide layers, tuning photon lifetime, and one oxide aperture at both the n- and p-DBR sides of the optical cavity ^[60,74].

Other groups also reported many exciting results on high-speed VCSELs in the 850 nm range. IBM and Finisar jointly reported a 55 Gbps directly modulated 850 nm VCSEL-based optical link in 2012 and a 56.1 Gbps link in 2013 ^[75,76]. Shi et al. reported a VCSEL with an open-eye operation at 41 Gbps at 25°C in 2015 ^[66]. This group used an oxide relief structure to reduce parasitics and Zn diffusion to define the optical aperture to a narrow spectral width and to reduce the differential resistance. In 2016, a University of Illinois Urbana-Champaign (UIUC) group reported a high-speed 850 nm VCSEL with error-free transmission at 50 Gbps and a 3 dB modulation bandwidth of 28.2 GHz ^[77].

VCSELs emitting at or in the range 980 to 1100 nm are now becoming important for data communication, most especially for wavelength division multiplexing and free-space optical communication ^[78]. Compared with the 850 nm wavelength range, in particular the 980 and 1100 nm wavelength ranges benefit from a larger temperature insensitivity, lower chromatic dispersion and lower transmission loss in the standard OM3 and OM4 MMF, a higher sensitivity of the photodetectors, and increased eye safety. InGaAs QWs emitting at 980 or 1100 nm exhibit a larger differential gain and lower transparency carrier density, promising a higher conversion efficiency, lower threshold current, higher modulation bandwidth, and higher reliability.

In 2007, error-free operation at 35 Gbps was demonstrated with a 980 nm VCSEL by the Coldren group at the University of California Santa Barbara (UCSB). This VCSEL used a tapered oxide aperture to reduce optical loss, multiple deep oxide layers to reduce the parasitic capacitance, and an optimized p-doping profile in the top DBRs to lower the loss and the resistance, showing a 3 dB modulation bandwidth larger than 20 GHz ^[50]. In 2011, TU Berlin demonstrated a 980 nm VCSEL realizing error-free transmission up to 44 Gbps at 25°C by shortening the cavity length to half-wavelength, using modulation doping and a binary DBR ^[40,79]. 980 nm VCSELs with 50 Gbps error-free transmission and 3 dB modulation bandwidth of about 27 GHz were demonstrated at 25°C by using a half-wavelength-thick optical cavity, tuning the VCSEL’s photon lifetime, and including current spreading layers ^[39,68,80]. In 2017, TU Berlin reported the simplicity VCSEL that employs a simplified epitaxial layer design with a 3 dB small-signal modulation bandwidth of 31–34 GHz ^[81,82]. This 1.5–2.5 μm oxide aperture VCSEL has a half-wavelength cavity and one oxide aperture at both the n- and p-DBR sides of the optical cavity for strong optical field and carrier confinements but without any deep oxide layers. Such a simple VCSEL structure is expected to improve the reliability and might ease manufacturing of large-diameter VCSEL wafers of 101.6 mm diameter or larger. In 2018 the same TU Berlin group reported a new record bandwidth of 35.5 GHz with their 980 nm VCSELs ^[69].

For VCSELs emitting at or near 1100 nm, Nippon Electric Company Ltd. (NEC) demonstrated error-free 25 Gbps operation at 25°C and a 3 dB modulation bandwidth up to 20 GHz in 2006 ^[83]. A 3 dB modulation bandwidth of 24 GHz was achieved with error-free operations at 30 Gbps and later 40 Gbps using buried type-II tunnel junctions ^[71,72].

Currently, over 90% of the MMF links are less than 100 m ^[84]. As the data centers become permanently larger, longer reach optical links up to $\sim 2\mathrm{km}$ at 20–25 Gbps are desirable ^[85]. Due to modal dispersion, it is challenging to realize error-free transmission at 20–25 Gbps across 2 km with a multi-mode VCSEL. To take care of the modal dispersion, VCSELs with a high side mode suppression ratio (a reduced spectral width) are sought for the long-distance transmission. The direct approach for VCSELs with a reduced spectral width is to reduce the oxide aperture. In 2013, TU Berlin reported 25 Gbps transmission across 1 km of OM4 MMF with a 3 μm oxide aperture VCSEL at 850 nm ^[86]. CUT used the surface relief method to control the modes and reported 20 Gbps transmission across 2 km of OM4 MMF with a 6 μm oxide aperture together with a 3 μm optical aperture VCSEL at 850 nm in 2014 ^[87]. Surface relief VCSELs show a lower current density than VCSELs with a reduced oxide aperture, demanding a more complicated fabrication and an increased threshold current.

Efforts have also been carried out to explore unconventional VCSELs with transverse coupled cavities for additional photon–photon resonances to extend the modulation bandwidth ^[88,89]. Such VCSELs demonstrated a 3 dB modulation bandwidth up to 37 GHz. The 3 dB modulation bandwidth of these VCSELs is very sensitive to the coupling strength between the cavities and the phase of the feedback light. The out-of-phase feedback is reportedly beneficial to extend the modulation bandwidth.

The ambient temperature can reach 85°C in data centers and HPCs, even with advanced cooling technology. Such a temperature might deteriorate the performance (data rate, energy consumption, reliability). VCSELs can maintain the room temperature performance without adjusting driving conditions and adding extra cooling systems by detuning the laser cavity resonance from the gain peak ^[90]. The gain peak of QWs shifts faster to larger wavelengths than the cavity resonance as temperature increases due, for example, to increasing bias current via joule heating. Therefore, as shown in Fig. 3, the cavity resonance is typically designed to be at room temperature at the longer wavelength side of the gain peak. When increasing the temperature, the gain peak and the cavity resonance wavelength will align, and the threshold current might decrease instead of increasing. The performance of various VCSEL designs at high temperatures in an on–off keying modulation format and measured in a back-to-back configuration is summarized in Table 2. Finisar reported 850 nm devices operating at 14 Gbps at 95°C in 2012 ^[91]. In 2013 Sumitomo described 850 nm VCSELs operating at 28 Gbps at 85°C ^[92]. Also, in 2013 CUT presented an 850 nm VCSEL operating at 40 Gbps at 85°C ^[93] and together with IBM in 2015 reached 50 Gbps at 90°C ^[94]. In 2016, UIUC reported an 850 nm VCSEL at 85°C ^[64] with an error-free data rate of 50 Gbps. VIS reported in 2018 a data rate of 25 Gbps at 150°C for their 850 nm VCSEL with a gain-to-cavity detuning of $\sim 15\mathrm{nm}$ ^[95]. For 980 nm VCSELs, TU Berlin reported error-free transmission at 30 Gbps at 120°C in 2011 ^[40] and 46 Gbps at 85°C in 2014 ^[68]. These results were exceeded in 2016 with 50 Gbps at 85°C for a 980 nm VCSEL ^[39]. For 1060 nm VCSELs, Chalmers reported a data rate of 40 Gbps at 85°C ^[70].

Power consumption becomes more and more important for both data centers and HPCs. Large power consumption increases the economic and ecological cost. Large power consumption might also deteriorate the long-term stability of VCSELs. The energy-to-data rate ratio $\mathrm{EDR}=(I·V)/\mathrm{BR}$ (in fJ/bit or mW/Tbps) is used to compare the energy consumption of different VCSELs, where $V$ and $I$ are the bias voltage (V) and current (mA), and BR (Gbps) is the bit rate at error-free operation ^[96]. Energy efficiency is commonly also quantified by the dissipated heat-to-bit rate ratio $(\mathrm{HBR}),\mathrm{HBR}=(I·V-P_{\mathrm{opt}})/\mathrm{BR}$, where $P_{\mathrm{opt}}$ (milliwatt, mW) is the optical output power of the VCSEL ^[96]. According to the International Technology Roadmap for Semiconductors projections for 2022, less than 100 fJ/bit is required for off-chip optical interconnects, and 10 fJ/bit is required for on-chip optical interconnects ^[100].

Large energy efficiency was reported for small oxide aperture diameter VCSELs ^[101–103]. The main reason is the larger $D$-factor originating from the smaller mode volume of small aperture diameter VCSELs, while the damping ($K$-factor) is more or less independent of the aperture size. Thus, a larger $-3\mathrm{dB}$ bandwidth at a smaller bias is realized. Table 3 summarizes the state of the art for energy-efficient high-speed VCSELs in an on–off keying modulation format for back-to-back data transmission at 850, 980, and 1060 nm. For 850 nm VCSELs, a record energy efficiency of 56 fJ/bit was achieved at 25 Gbps at 25°C for a 3.5 μm oxide aperture diameter VCSEL by TU Berlin and VIS in 2012 ^[101]. 145 fJ/bit at 35 Gbps at 25°C and 139 fJ/bit at 35 Gbps at 85°C were reported for a 980 nm 3 μm oxide aperture VCSEL by TU Berlin ^[98,106]. In 2014, Furukawa reported 76 fJ/bit at 25 Gbps at 25°C ^[108].

Today, short-reach optical interconnects are dominated by VCSELs combined with MMF and direct detection, offering low cost and small form factor. Existing IEEE standards (802.3bm) support 10 Gbps and 25 Gbps core rates in 4-fiber and 10-fiber arrangements to support up to 100 Gbps of traffic. However, a 400 GbE (Gigabit Ethernet supporting 400 Gbps) solution as envisioned by the IEEE Task Force (P802.3bs) based on 25 Gbps will require 16 fibers in each direction. Alternative solutions that lead to reduced fiber numbers could include higher bit rates combined with higher modulation formats to end up at 50+ Gbps per line. Four-level pulse amplitude modulation (PAM4) ^[109,110] is the presently most popular approach for higher-order modulation formats. Forward error correction and signal processing might help to surmount the 50 Gbps barrier reached by simple on–off keying.

Combining PAM4 and digital signal processing (DSP) techniques with coarse wavelength division multiplexing (CWDM, O-band) or short wavelength division multiplexing (SWDM) is enabling the next generation of high-speed optical VCSEL-based interconnects. Compared to the on–off keying modulation format, PAM is a way to transmit more bits into the same amount of time on a serial channel by using different signaling levels, as shown in Fig. 4. PAM4 has four distinct levels to encode two bits of data per symbol, essentially doubling the data rate of a connection. Generating or decoding more than two logical levels is technically more difficult, requiring more complex hardware and software and is probably energy inefficient. Already for PAM4, the VCSEL linearity, modulation response characteristics, and random and induced noise have become significant factors ^[111].

Recently, up to 94 Gbps PAM4 was demonstrated over single wavelengths using 850 nm VCSELs with a 20 GHz 3 dB bandwidth using energy expensive pre-emphasis and receiver equalization ^[112]. 180 Gbps PAM4 was demonstrated using four VCSELs and short wavelength division multiplexing ^[113]. 38 Gbps PAM4 data transmission was successfully realized using 1.3 μm wafer-fusion VCSELs, a wavelength that is important for silicon photonics ^[111]. To continue using the present OM3 and OM4 MMF infrastructure SWDM is one promising approach to reduce fiber counts and cabling density and save substantial capital and operational cost ^[114].

VCSELs are ideal sources for photonic integrated circuits (PICs). However, a VCSEL emits perpendicular to the wafer plane, while a PIC lies in the wafer plane. There are at least four approaches for coupling light from VCSELs to an in-plane PIC, as shown in Fig. 5. The most straightforward and direct approach is to use end-to-end coupling by placing a VCSEL at the end of an in-plane waveguide ^[115]. A spot-size converter is constructed to reduce the coupling loss between the VCSEL and an in-plane waveguide. The second approach is to use a 45° micro-reflector to direct the normally incident light from a VCSEL sitting on the top surface of a waveguide to an in-plane waveguide. A 45° micro-reflector can be fabricated mechanically or by etching, but the coupling efficiency is a very critical issue ^[116–119]. The third approach is based on a grating coupler. A buried diffraction grating is inserted in the top DBR of a GaAs-based VCSEL to couple the light into a horizontal waveguide ^[120]. This monolithic technology needs epitaxial regrowth. A more straightforward approach is to etch a surface grating into the top layer of the top DBR of a VCSEL needing no epitaxial regrowth. Such surface gratings achieved a coupling efficiency of 40% ^[121]. Recently, silicon grating couplers couple the light from VCSELs into the waveguide, showing 20 Gbps non-return to zero (NRZ) transmission at 1550 nm for packaged PICs ^[122–125]. The fourth approach is to use photonic wire bonding technology to connect a VCSEL to an in-plane waveguide ^[126]. The emerging photonic wire bonding technology has been used for chip-to-chip interconnects and chip-to-fiber interconnects. Very recently, hybrid integration of silicon photonics circuits and InP-based horizontal-cavity surface-emitting lasers by photonic wire bonding was demonstrated ^[127], which shows that the photonic wire bonding technology is a promising approach to realize the integration between a VCSEL and a PIC.

VCSELs show superior beam quality, low energy consumption, low cost, high reliability, and can be easily integrated to form 2D arrays, all requirements for sensing applications like mice or smartphones.

VCSELs used in optical mice first appeared in 2000 ^[22]. The tracking system in an optical mouse is based on the principle of laser self-mixing interference (SMI) ^[128–130]. As shown in Fig. 6, when the laser light emits to a scattering object, a small part of the scattered light is coupled back into the laser cavity and mixes with the strong laser field. When the movement of the object has a component along the direction of the laser beam, the phase of the reflected light continuously shifts with respect to the original laser light, resulting in a periodic variation of the feedback into the laser cavity at a frequency equal to the Doppler frequency ^[130]. The VCSEL is here used not only as a coherent light source but also an active filter, an amplifier, and an optical mixer. VCSELs with a small oxide aperture operate in a single mode, showing a high side mode suppression ratio (SMSR) at a very low threshold current and low power consumption, especially important for battery-powered optical mice. Large SMSRs in SMI sensors avoid signal distortion by mode beatings. The low output power of single-mode devices is acceptable for consumer electronics where the eye safety limit is 0.5 mW ^[21]. However, small oxide apertures of 3 μm are challenging for fabrication with high yield ^[131]. Next-generation tracking systems require single-mode and single-polarization VCSELs to avoid unwanted movement of the pointer due to polarization flips ^[132]. Many approaches have been proposed to break the radial symmetry of a typical vertical cavity laser. These approaches include the use of one-dimensional (1D) gratings ^[133,134], photonic crystals ^[135,136], and elliptical surface etching ^[137]. Other approaches are based on the epitaxial growth on higher-order substrates ^[138,139], highly strained QWs ^[140], elliptically shaped mesa geometries ^[141,142], and the use of external stress ^[143].

VCSEL-based sensors based on SMI have many industrial applications in, e.g., cable manufacturing; extrusion; steel, paper processing and other production equipment; and in measuring length and movements of objects ^[144]. These sensors measure the true object speed and displacement on virtually all surfaces and have a high speed and displacement accuracy outperforming currently available sensors. VCSEL-based sensors with the SMI effect also have a vast potential to measure the speed over the ground of cars with high accuracy on all surfaces and under all automotive conditions (e.g., rain, fog, snow) instead of measuring the wheel speed ^[144]. Thus, they are also useful in braking distance reduction, lip control, side-wind correction, parking assistance, and tire monitoring.

VCSELs have considerable potential in three-dimensional (3D) sensing, such as in gesture recognition, face recognition, eye movement tracking, as well as in 3D photography and autofocusing ^[26,145]. The incorporation of VCSEL chips in smartphones is nowadays a reality ^[23].

For 3D sensing applications, VCSELs have huge advantages as compared to competing approaches. VCSELs emit with a narrow spectral width of less than 1 nm, which is much lower than the emission width of LEDs, and have a temperature dependence of typically 0.06 nm/K ^[146]. This allows the use of narrowband filters to eliminate unwanted background signals and enhance the signal-to-noise ratio, which is a prerequisite for many outdoor sensing applications. The circular output beam with a low divergence angle and the configurability of large-scale arrays result in scalable output power and an efficient and easy beam combination with simple optics making modules very compact. VCSELs are surface mountable, and have a wide variety of packaging options, including a chip on board, surface mount, and plastic encapsulation. They can be easily integrated with other electronic components on a printed circuit board (PCB). This packaging approach can reduce the cost and improve the VCSEL-based module performance by minimizing parasitics. VCSELs have high power conversion efficiencies of typically $>40\%$ across a wide temperature range from 0 to 70°C with superior performance ^[26].

There are primarily two types of methods for 3D sensing. The first one is the time-of-flight (ToF) method ^[147,148]. The ToF VCSEL chip of modern smartphones is shown in Fig. 7(a). A pulsed ToF system (also called a LiDAR system) sends a laser pulse to the object and measures the time that the light pulse travels from the laser to the object and back to the detector, yielding the 3D position (or depth information). The peak power of VCSELs (or a VCSEL array) depends on the pulse duration, the duty cycle, and also the thermal operating condition. These three factors also contribute to the power conversion efficiency. Self-heating in a VCSEL leading to thermal roll-over limits the peak optical output power. Short pulses, shorter than the thermal time constant (typically about 1 μs), with a small enough duty cycle can eliminate this limitation. In 3D sensing, human eye safety is another critical issue that must be carefully addressed ^[149].

The first ToF technology for 3D sensing is called a scanning ToF sensor, where single pixels are scanned one at a time with a VCSEL via short pulses. In a scanning ToF sensor, the pulse duration is always 1–10 ns, and the duty cycle is less than 10% ^{[26,145,149,150]}. VCSELs for such outdoor applications require a large optical output power for covering larger distances and still achieving a large signal-to-noise ratio ^[144]. The second technology is based on a ToF camera and is called focal plane scanning ^{[148,151,152]}. In such a sensing system, as shown in Fig. 8(a), VCSELs and detector arrays are used. Focal plane scanning uses a beam from the laser array to illuminate the complete field of view on a surface. Each detector covers its own fraction of the field of view, which is illuminated by the VCSEL array. Detector signals are analyzed in the time domain, and the distance to particular points of the target is determined on the basis of time-interval measurements. The VCSEL array should show low coherence and high optical power but lower brightness as compared to the scanning ToF sensor. VCSELs have the intrinsic advantage in that they may easily be produced in the form of 2D electrically parallel arrays ^[152].

An alternative method for 3D sensing uses structured light ^[26,153]. As shown in Fig. 8(b), here a coded 2D pattern generated by a special projector or a light source modulated by a spatial light modulator illuminates a surface. An imaging sensor is used to acquire a 2D image. A VCSEL array for generating the pattern is shown in Fig. 7(b). If the surface is nonplanar, the image of the surface from the camera is distorted compared to the coded 2D pattern. The depth information can be extracted from the distortion of the known coded 2D pattern. The coded 2D pattern often consists of randomly distributed dots, which can be realized by the arrangement of the elements of the array.

Extensive research on surface metastructures or nanostructures has been carried out in the past decade, and many applications have been suggested to use their unique optical properties. High-index-contrast 1D wavelength gratings, also called high-contrast gratings (HCGs), and the 2D variation of such structures, generalized high-contrast metastructures (HCMs), have been and continue to be the focus of intense research. Here we focus on 1D HCGs and VCSELs with 1D HCGs. 2D HCMs have been comprehensively reviewed recently elsewhere ^[154,155].

A HCG reflector, also called a photonic crystal mirror or a guided-mode resonant reflector ^[156–158], is shown in Fig. 9(a). The grating bars (also called stripes) composed of the high-index material in the HCG are fully immersed in a low refractive index medium, e.g., the air, resulting in a high index contrast. The grating period ($\mathit{\Lambda }$) is a value in units of length that is between the wavelength in the high refractive index material and the wavelength in the low refractive index material. To simplify the modeling, the numbers of bars and periods and the length of the bars may all be taken to be infinite-size. Figure 9(b) shows that the first two waveguide array modes in infinite HCGs with real propagation constants have a $\pi$-phase difference in the output plane and cancel each other, causing a nearly 100% reflection ^[159]. When two such 100% reflectivity points are located closely in the spectrum, a high-reflectivity broadband is obtained. There are alternative interpretations for the observed high-reflectivity broadband of HCGs based on Fano resonance or guided-mode resonance ^[155,158]. A broadband and high-reflectivity HCG can serve as a reflector and replace the top DBR partially or fully ^[160–163]. Experimentally, HCG-VCSELs show an excellent mode selectivity and polarization control, even for large oxide apertures.

For the HCG design, the structure parameters can be optimized by rigorous coupled wave analysis (RCWA) ^[164] and analytical methods ^[159]. These two methods consider that HCGs are of an infinite size and the incident wave is an infinite plane wave. The infinite-size HCG model is useful to rapidly search the parameters of HCGs. However, in real devices the reflectivity may be reduced, caused by the finite size ^[165] and by imperfect (nonrectangular) stripes, which increases the threshold current, causing a reduced energy efficiency of the VCSELs. The finite-size incident wave can excite guide modes in the HCG by higher-order angular components. The excited guided modes reduce the reflectivity and increase the transmission ^[166]. These excited guided modes can redirect the incident wave to the in-plane direction, providing the potential for an integrated optical sensor and for integration with planar optical circuits ^[167,168]. Therefore, it is indispensable to calculate the finite-size HCG properties by a finite difference time domain method after choosing the HCG parameters using RCWA or analytical methods.

HCGs also show a strong ability to confine the field. The energy penetration depth of HCGs is much smaller than that of DBRs ^[169]. The small energy penetration depth is beneficial to construct a short cavity leading to a large confinement factor. The small mode volume is beneficial to enhance the Purcell factor, reducing the carrier lifetime to achieve a larger modulation bandwidth and energy efficiency ^[102,103]. In HCGs, the phase penetration depth, which partly controls the photon lifetime, can be tuned by the physical structure parameters without degrading the reflectivity, unlike the shallow surface etching method ^[38,170]. The increased confinement factor and the optimum photon lifetime combined with a high reflectivity are expected to increase the relaxation resonance frequency, lower the damping factor, and lower the threshold current, which are all factors that are decisive for high-speed modulation and energy-efficient operation.

In fabricating HCG-VCSELs, a sacrificial layer below the HCG layer is always removed to create a suspended HCG surrounded by a low refractive index material (for example, air). Alternatively, epitaxial AlGaAs that is selectively oxidized to form the low refractive index material AlGaO may be used as the underlying low refractive index layer. Therefore, a lattice matched material system is essential from the perspective of both device performance and fabrication feasibility. Recently zero-contrast gratings (ZCGs) have been proposed to replace completely or partly the top DBR ^[171–174]. The sacrificial layer removal step in the HCG fabrication is not required for the ZCG fabrication. The whole device fabrication is thus highly simplified, and the mechanical stability is improved. ZCGs also can achieve a high reflectivity beyond 99.5% due to the destructive interference at the output plane but with a narrower high reflectance bandwidth compared to HCGs, and they can show a better mode selectivity for ZCG-VCSELs. ZCGs are less complex in the overall VCSEL fabrication but they have rather tight geometrical tolerances. For a $\pm 10\mathrm{nm}$ variation of the etching depth of 304 nm, the wavelength of the maximum reflectivity shifts by $\pm 7\mathrm{nm}$, and the reflection phase change is $0.1\pi$ for a ZCG designed for use in the 980 nm range ^[175].

The first electrical HCG-VCSEL reported in 2007 was realized at 850 nm in the GaAs material system as shown in Fig. 10(a) ^[160]. It used HCGs to replace part of the top DBR and kept four DBR pairs to provide current spreading and active region protection. These devices showed a sub-milliampere threshold current and single-mode and polarization-selective operation. Later 1060 nm and 1550 nm HCG-VCSELs were realized ^[162,176]. The reflectivity of an HCG strongly depends on the HCG size, the HCG parameters, the incident angle, the polarization, and the sizes of the incident beam ^{[163,166,177–179]}. Thus, it is easy for an HCG-VCSEL to realize single-mode operation even with a large oxide aperture ^[165]. The output beam pattern can be shaped by the HCG, which was predicted in 2014 ^[166]. In 2018, HCG-VCSELs realized single-lobe, double-lobe, triple-lobe, “bow-tie,” “sugar cone,” and “doughnut” beam patterns without any external optics ^[177], as shown in Fig. 10(b). Figure 10(c) is the schematic of a nanoelectromechanical tunable HCG-VCSEL developed for a wide tuning range and fast tuning because of the compact HCG mirrors compared to existing DBR-based electrostatic-actuated microelectromechanical tunable VCSELs ^{[162,180,181]}. Such tunable HCG-VCSELs have been realized in the 850, 1060, and 1550 nm wavelength ranges.

HCG reflectors are resonant structures, very different from DBRs, and the reflection phase can be tuned while keeping a high reflectivity ^[170]. By adjusting the HCG parameters like period and duty cycle, different wavelengths of HCG-VCSELs on a single wafer can be simultaneously achieved ^[182]. Therefore, HCG-VCSEL arrays [as shown in Fig. 10(d)] with multiple uniformly spaced wavelengths can be constructed on a single HCG-VCSEL wafer for wavelength-division multiplexing (WDM), providing a promising way to increase the aggregate bandwidth of a single fiber. Optically pumped HCG-VCSEL arrays were reported with double silicon-based HCGs as reflectors for dense WDM at 1.55 μm ^[183]. A GaAs-based HCG filter array with different resonance wavelengths was demonstrated, targeting electrically pumped HCG-VCSEL arrays ^[184,185]. A first monolithic four-channel HCG-VCSEL array covering a wavelength span of 15 nm with 5 nm channel spacing was demonstrated at 980 nm for WDM applications ^[186].

Vertical coupling was also proposed to extract light from a Fabry–Perot microcavity composed of double HCG mirrors to a Si waveguide ^[187]. This is a feasible approach to realize the integration of a vertical cavity laser with in-plane PICs, as discussed in Section 3.E. Very recently, HCGs have been proposed to replace the DBR as reflectors and at the same time to route the emitting light into the in-plane waveguides ^[188,189], as shown in Figs. 11(a) and 11(b). This approach reduces the thermal resistance of the devices. Optically pumped hybrid vertical-cavity lasers with lateral emission into a silicon waveguide were reported using a silicon HCG at 1.5 μm ^[190]. Experimentally, a 3 dB frequency of 27 GHz was obtained for an optically pumped laser ^[191]. For the short wavelength range, SiN-based HCGs were proposed to serve as reflectors and extract the light into the in-plane SiN waveguide because SiN is transparent from 600 to 1100 nm, where silicon is not suitable for waveguide circuits. Figure 11(c) shows a half-VCSEL placed onto a SiN-based HCG surrounded by ${\mathrm{SiO}}_2$, thus forming a hybrid vertical-cavity silicon-integrated laser at 850 nm. The laser emission is in-plane directly into a SiN waveguide ^[192]. Continuous-wave electrically pumped operation was achieved for such a hybridly integrated laser, and a single-sided waveguide-coupled optical output power of 73 μW with a side-mode suppression ratio of 29 dB was reported with an oxide aperture diameter of 5 μm in the top half-VCSEL. These integrated laser sources in densely integrated circuits have the potential for applications in medical point-of-care devices, body implants for monitoring of glucose levels, and sensing devices integrated into smartphones. Another variation of the HCG-VCSEL uses a composite monolithic high-contrast grating and DBR as the top coupling mirror to relieve the fabrication complexity. The first electrically injected VCSEL that includes this type of a monolithic HCG was reported in 2018 ^[193].

VCSELs have unique advantages compared to edge-emitting lasers. Impressive progress has been achieved enabling energy-efficient and fast data communication in the past decade. Novel device designs based on the deeply developed insights to the device physics and to advanced modulation formats have been used to boost the performance of VCSELs. Data rates beyond 50 Gbps in the on–off keying modulation format have been demonstrated. An energy efficiency of close to 50 fJ/bit was reported at 25°C for a small oxide aperture VCSEL. An error-free transmission rate of 50 Gbps at 90°C was also demonstrated. To increase the capacity of MMF-based links, the PAM4 modulation format is employed, yielding a 100 Gbps data rate in a single lane. 400 Gbps via SWDM was also achieved.

VCSELs will continue to be the dominant laser sources in data centers and HPCs and might penetrate chip-to-chip and even on-chip optical interconnects. Single-mode VCSELs with a reduced spectral width will attract significant interest for extended distance transmission over MMF for data centers and HPCs ^{[66,86,87,194–198]}. Advanced modulation formats like carrierless amplitude phase (CAP) or discrete multi-tone (DMT) for high-speed VCSEL transmission are promising to increase the transmission data rates beyond 100 Gbps with a high spectral efficiency ^[199,200]. Low-loss and straightforward coupling between VCSELs and in-plane PICs is still challenging. The integration of VCSELs with PICs via metastructures makes the chips compact and power efficient. Such VCSEL-based integrated chips have important applications in optical interconnects, optical sensors, and consumer electronics. VCSELs with metastructures can provide different orbital angular modes ^[201] and laser modes ^[177], opening a new way for increasing the capacity of links.

VCSELs have unique advantages over silicon photonics technology, which is presently a hot topic ^[84]. VCSELs have a lower cost, lower power consumption (1.8 pJ/bit), and a smaller footprint ($\sim 8\mathrm{\mu m}$ emitting area in diameter) enabling a large bandwidth density, while a silicon photonic transmitter consisting of one continuous-wave semiconductor laser and a Mach–Zehnder modulator requires about 10 pJ/bit at 25 Gbps and takes up a footprint of about 300 μm. VCSELs will continue to be the lowest cost and highest energy efficiency light source for data links available for short-reach optical interconnects.

VCSEL-based sensors have wide applications in consumer electronics, car markets, Internet of Things, atomic sensing ^[202], and much more. VCSELs featuring low energy consumption, small footprint, and an array configuration will undoubtedly be the dominant laser sources for these and many other applications. As VCSELs will become essential for various sensor applications in our daily life, they will receive increasing attention from industry and academic communities.