Over the past few decades, the mainstream of integrated circuit (IC) industry has been mainly powered by Moore’s Law, which is targeted at achieving faster operation speed, less power dissipation, and lower cost^[1]. The key driving force behind Moore's Law is the ongoing MOSFET scaling down to nanoscales^{[2, 3]}. As shown in Fig. 1(a), the number of transistors that can be purchased per dollar is increasing, but since 2012 it almost encounters a bottleneck. Correspondingly, the feature size of the transistor is significantly reduced at first, but slowly decreasing to the order of ~10 nm in 2017. However, this trend is beginning to bump up against the fundamental physical limits on their size, which means the approaching end of the Moore’s Law. As shown in Fig. 1(b), the final International Semiconductor Technology Roadmap (ITRS) report predicts transistor scaling will end in 2021 and clearly states that it will no longer follow the path of decreasing process nodes after 10 nm^{[4, 5]}.

In fact, Moore’s Law is a techno-economic model that has enabled the information technology industry to double the performance and functionality of digital electronics roughly every two years within a fixed cost, power, and area^[6]. Fig. 2(a) shows that the main technological development trends will follow two paths in the More-than-Moore era. The first path we call “More than Moore”, is from the function, to achieve multi-functional expansion of the circuit. Another path we call “More Moore” is to continue to increase the density of integrated circuits through the miniaturization of devices. However, because traditional materials are difficult to break through their own limits, we may further promote the development of integrated circuits through the research and development of new materials, new structures, and new principle devices, and eventually move toward three-dimensional (3D) chip stacks integration^{[7, 8]}. Different and revolutionary strategies and approaches are highly and urgently needed to meet the requirements of faster, cheaper, and more energy efficient in the More-than-Moore era. So far, silicon photonics technology exhibits distinguished potential and has become one of the leading technological solutions for integrated photonics that target applications such as data centers, telecommunications, and high-performance computing (HPC)^[9]. As shown in Fig. 2(b), its total market size will increase dramatically in these fields. Approaching the opportunity of large scale marketization, silicon photonics have drawn intensive attention, driven by its advantages, market demand, and national strategy.

Silicon photonics have been widely investigated in recent years, which benefits from academic research efforts and available commercial complementary metal–oxide–semiconductor (CMOS) process for potential mass-production applications^[10-12]. Versatile passive- and active-silicon-based nanophotonic devices have been proposed for applications including optical interconnects, optical routers, remote telecommunications, modulation, and sensing. However, there are some challenging tasks need to be solved, including the intrinsic properties limits of pure silicon and the CMOS fabrication compatibility for on-chip silicon sources^[10]. Recently researchers have reported the integrated stimulated Raman scattering silicon lasers^[13], germanium-on-silicon lasers^[14], and hybrid III−V on silicon lasers^[15], but there are some problems, such as undesirably high optical external pumping, insufficient lasing emission, and complicated fabrication process. Hybridizing silicon waveguides and resonators/cavities with active materials have been considered as an alternative solution for lasing emission, optical modulation, and detection. This hybrid integration on silicon has shown potential advantages because of its low cost, easy processing, and various active materials^[16]. Many investigations about the hybrid integration on silicon have been demonstrated, including semiconductor nanotubes^[17], superconducting nanowires^[18], polymers^[19], single III–V semiconductor nanowires^[20], transition-metal dichalcogenides (TMDs)^[21] or graphene^[22]. In terms of the view, the natural host photonic structures of active materials deposited or grown on silicon are microcavities, because of their high quality factors (Q) and low mode volumes (V)^{[23, 24]}; therefore, large Q/V is helpful for laser threshold reduction, ultra-low voltage and energy-efficient optical modulation and ultra-sensitive label-free sensing.

Particularly, photonic crystal nanobeam cavity (PCNC) is considered as an ideal platform for on-chip integration, due to the advantages of an ultracompact footprint, enhanced light–matter interactions, high integrability with optical waveguides/circuits, and compatibility with CMOS processes^{[25, 26]}. To date, various optical devices based on PCNC have been demonstrated, such as optical lasers, optical modulators, optical switches/filters, and label-free sensors. By incorporating these photonic devices, versatile and reconfigurable photonic networks can be realized. Thus, PCNC-based devices in near infrared wavelengths could be potentially significant for future on-chip integrated silicon photonics. In this review, we will focus on photonics devices based on PCNCs.

Photonic crystal lasers, with large Q/V, have enhanced photon emission below threshold through the Purcell effect, and can operate at a higher modulation speed^[27]. Compared with two-dimensional (2D) photonic crystal slabs, the PCNCs with exceptional Q/V in a much smaller footprint have attracted particular interest. So far, PCNC lasers based on various materials have been demonstrated using different nanobeam cavities^[27-40], as summarized in Fig. 3. Insets show the device structures, materials, and threshold power, respectively. It can be found that III−V semiconductor compounds (gallium arsenide, GaAs and indium phosphide, InP) with high electron mobility have proved extremely successful for the realization of lasers. However, the major problems of III−V semiconductor compounds are high cost and poor CMOS compatibility, which limits its further development for on-chip integrated photonics application. Compared with III−V semiconductors compounds, CMOS-compatible SOI-based photonics devices are a promising platform for realizing low cost and high density on-chip integrated photonics. However, silicon cannot produce light directly due to its natural indirect band gap. Hence, recently silicon-based hybrid laser platforms, such as using high-performance III−V and 2D materials attached to silicon, have attracted much attention^{[37, 38]}.

For example, Lee et al. demonstrated an ultracompact nanobeam laser by effectively integrating a wavelength-scale unidirectional III−V materials onto a SOI waveguide^[37], as shown in Fig. 4(a). The light from the III−V laser was initially coupled to the III−V waveguide, and was connected to one end of the nanolaser. The light was then vertically coupled from the upper III−V waveguide to the lower SOI waveguide via a directional coupler in the overlapping region of the III−V waveguides and the SOI waveguides. At last, the light from the nanolaser propagated along the low-loss silicon waveguide^[37]. The SEM of the proposed III−V/Si nanobeam laser was shown in Fig. 4(b). It is worth mentioning that the coupling efficiency between the InGaAsP nanolaser and conventional SOI waveguide reached ~83%. The lasing started at a threshold pump power of ~0.2 mW, and the wavelength was 1556 nm in Fig. 4(c). It has demonstrated an efficient hybrid integration of a wavelength-scale photonic crystal nanolaser and a SOI waveguide. This nanoscale hybrid III−V/Si laser is considered as a promising platform for future compact, faster, and efficient silicon nanophotonics^[37].

Monolayer transition-metal dichalcogenides (TMDs) exhibit great potential to be the smallest and efficient optical gain media for low energy-consumption nanolasers due to its strong excitonic emission^{[38, 41]}. For example, Ning et al. firstly demonstrated the use of a silicon PCNCs and a monolayer TMD to generate a room-temperature laser operation in the infrared region^[38], as shown in Fig. 5(a). This was mainly due to the unique combination of a TMD monolayer with a silicon-transparent wavelength emission, and a high-Q silicon PCNC. Fig. 5(b) described the lasing emission spectrum under different pump power. As the pump power increased, it showed clearly the appearance of strong lasing peaks at 1052 nm of the first mode and 1132 nm of the second mode. To represent the lasing characteristics more accurately in Fig. 5(c), PL spectrum was measured at increasing pump levels, showing a clear lasing emission peak at 1132 nm and a very low lasing threshold ~ 97 μW. Moreover, the lasing Q-factor could be extracted from the data in Fig. 5(c) and this was the highest Q-factor of any 2D TMD-based laser reported so far. The SOI-based PCNCs would be considered as very attractive for integrated silicon nanophotonics at the wavelength transparent to silicon^[38]. Adding only a monolayer of non-silicon material may be the closest approach to a silicon laser as ever possible, which opens opportunities for CMOS process integration. The unique combination structure of MoTe₂–nanobeam cavity can possibly be used for 2D TMD-based electrically driven lasers via electrical injection^[33], and thus is a promising platform for optical communications.

Silicon photonics technology is poised to resolve short reach interconnects, and optical modulators are essential for such an interconnect. In addition, Pockels effect^[42], Kerr effect^[43] and the Franz–Keldysh effect^[44] are the main electric-optic (EO) effects that cause electric absorption or electric refraction. In order to realize the optical modulator, one method is based on changing the optical properties of the waveguide medium (i.e. refractive index or optical absorption) through linear EO effect^{[45, 46]} or free-carrier dispersion^{[47, 48]}. Another method is to control the properties of the waveguide medium by adopting an optical pumping, thereby actuating the nonlinear optical phenomena in the waveguide^[49].

Several parameters have been used to characterize the performance of EO modulator: footprint, modulation voltage, modulation speed, extinction ratio, and energy consumption. So far, various silicon hybrid EO modulators have been achieved. But for waveguide-based modulators, they have large footprints and high-power consumptions due to the interaction lengths about several tens of micrometers^{[22, 50]}. The micro-ring resonator (MRR)-based modulators increase bending loss and decrease Q factor, resulting in high power consumption and low modulation efficiency^{[51, 52]}. Recently, PCNC has been extensively used as EO modulators due to its excellent properties namely ultrasmall footprints and convenient integration with bus-waveguides. However, due to the absorbed distortion caused by carrier dispersion and the difficulty of direct PN doping in silicon nanobeam structures. Therefore, it is necessary to hybrid integrate with emerging materials, such as 2D materials, lithium niobate, EO polymers etc. In particular, the 2D materials and lithium niobate integrated on silicon platform are becoming fully CMOS-compatible. Existing PCNC-based EO modulators have been proposed including those based on a Si-graphene hybrid^{[53, 54]}, a Si-organic hybrid^{[55, 56]}, and a Si–lithium niobate hybrid^{[57, 58]}. Here, several works are listed in Table 1 in chronological order. This predicts that tremendous efforts have been made towards realizing smaller size and high-performance optical modulators by hybrid integrating with new materials.

We can find that many works based on the combination of PCNC and new materials have been reported^{[46, 53, 60]}. Especially, graphene as an active medium is attracting interest because of its high carrier mobility^[61] and gate-controllable broadband absorption^[62]. All-optical modulator has been theoretically proposed by using the strong Kerr effect of graphene with PCNC^[49]. However, the feasibility of these works is demonstrated through simulation, which makes it a promising candidate for achieving modulators experimentally.

Optical switch is widely used in optical communication^[63], optical computing^[64], optical interconnect and an optical information processing system^[65]. Due to the large capacity of the optical switch array and the high integration of silicon-based chips, the optical switch should have small size and low power consumption. Several structures have been proposed to achieve optical switch, including silicon microring resonators (MRRs)^[66] and Mach–Zehnder interferometers (MZIs)^[67]. However, the radius of MRR is generally in the order of tens of microns, and the arm length of MZI is also in the order of hundreds of microns or even millimeters. Therefore, the key technology in realizing the optical switch is to reduce the device size and energy loss.

PCNCs are increasingly gaining interest in optical switch due to the advantages of high Q/V. Therefore, it could be an effective method in realizing low power-consumption optical switches. For comparison, Table 2 summarizes the performances of some PCNC-based optical switches. For thermo-optic (TO) switch, Su et al. proposed and experimentally demonstrated a 2 × 2 TO crossbar switch based on dual-PCNCs^[69]. The cavity resonance wavelength could be red-shifted via thermally perturbing the refractive index of silicon, and then TO switch can be realized. Meanwhile, Su et al. proposed a compact 2 × 2 EO switch based on dual PCNCs with identical PN junctions respectively using the same schematic diagram^[73]. To realize the EO switch, an external bias voltage was applied on the PN junction. This changed the width of the depletion region, thereby tuning the refractive index of silicon PCNC. Miniature all-optical switches with high-speed and low power attract much attention in communication networks because of its versatility, such as optical logic operation^[77], wavelength conversion^[78], optical demultiplexing^[79], etc. So far, all-optical switch based PCNC has been achieved by several methods, such as using Fano resonance^[80], multi-channel switch^[75] or silicon-polymer hybrid structure^[81]. These works show that all-optical switches have great potential in improving optical information processing capacity and reducing the power consumption of on-chip all-optical signal processing^[80].

With the advancement of photonic integration technology, the filters have drawn much attention due to highly energy-efficient tunability^[82]. The tunability can be achieved through TO effects^[83], electromechanical effects, and opto-electro-mechanical effects^{[84, 85]}. Generally, direct reconfiguration based on TO effect is realized by integrating a microheater on a silicon waveguide^{[86, 87]}. Note that the TO effect is overwhelmingly preferred for high tuning efficiency, a large tuning range and those requiring simple fabrication based on SOI, due to the high TO coefficient of silicon (1.86 × 10⁻⁴ K⁻¹), corresponding the temperature sensitivity of ~80 pm/K^[88]. Zhang et al. demonstrated a TO tunable filter based on a suspended PCNC shown in Figs. 6(a) and 6(b). We can see that the TO tunability filters possess record high tuning efficiency of 21 nm/mW and the widest tuning range of ~43.9 nm^[89], as shown in Fig. 6(c). This device has been considered as an ideal platform for integrated photonics circuits, such as cross-bar switches and Bragg grating filters^[89], due to the advantages of ultra-high tuning efficiency and rapid response.

Ultra-sensitive and label-free detection of the analyte plays an important part in the field of homeland security, environment protection and medical diagnostics^[90-93]. Optical microcavities such as whispering gallery mode (WGM) cavities, Fabry–Pérot (F–P) cavities and photonic crystal (PhC) cavities are considered as promising candidates for label-free sensing^[94-97]. Particularly, PCNCs have extensively attracted attentions due to the advantages of ultrahigh Q/V, ultra-small footprint, and excellent CMOS compatibility properties^[98].

Hence, most research has focused on the optimization of PCNCs design to improve sensitivity, as shown in Table 3. With the rapid development of technology, micro-nano devices are moving towards high miniaturization and integration. Much research has been proposed for label-free sensing by integrating microfluidics with PCNCs. For instance, Yang et al. presented a nanoslotted parallel quadrabeam photonic crystal cavity sensor, with high sensitivity of 451 nm/RIU and high-Q of 7015 in aqueous environments at wavelength of 1550 nm^[104], as shown in Fig. 7(a). They also monitored streptavidin-biotin binding affinity in phosphate buffered saline (PBS) solution and the detection limit is down to 10 ag/mL, as shown in Figs. 7(b) and 7(c). In this configuration, the PDMS microfluidic channel is integrated with PCNC sensors, which provides a promising platform for multiplexing on chip and point-of-care medical diagnostics. In addition, they also show the potential application in single-molecule detection^[107]. On the other hand, works integrating multi-channel microfluidic have been proposed for multiplexing sensing. Mandal et al. demonstrated a nanoscale optofluidic sensor arrays with multiple channels based on PCNCs, as shown in Fig. 7(d)^[108]. To verify sensing ability of the nanoscale optofluidic sensor array, fluidic architecture is embedded in sensor array and achieving the RI resolution of 7 × 10⁻⁵, corresponding to the mass limit of around 35 ag in the measurement of water and CaCl₂ solution, as shown in Figs. 7(e) and 7(f)^[108]. This research opens the door for the detection sensitivity at the tens of attograms level in the field of label-free sensing and shows the multiplexing capabilities of this architecture.

In conclusion, with the rapid development of silicon photonics devices, higher integration, and miniaturization are required. Among these, PCNCs are considered as candidates for on-chip label-free sensing and multiple channel sensing, due to the advantages of an ultra-small footprint, ultrahigh Q/V, and excellent CMOS compatibility properties^[109].

To be implemented in practice, technical challenges are existed in manufacturing. The silicon photonics chip can be fabricated cost-effectively with CMOS-compatible technology. However, the fabrication tolerance limits the practical applications of PCNCs, which makes them impractical for high-yield production. Fabrication tolerance in the position and size of the PhC structures may result in fluctuations of resonance wavelength and Q factor^[110], and it also limits the parameter space for sweeping device design^[111]. So far, various strategies to improve the fabrication tolerance have been demonstrated, such as using UV-lithography in fabrication^[112], allowing precision wavelength trimming of devices^[113], controlling the Q by optimizing configurations^[114], and using a subwavelength grating (SWG) structure^[115]. An ultra-compact PCNC with large fabrication tolerance is desirable for silicon-based photonic integrated circuits and there is a need for further studies and industry efforts to realize outstanding fabrication tolerance.

In this paper, we review recent advances on photonics devices for lasers, modulators, switches/filters, and sensors based on PCNCs. It has been shown that PCNCs with ultrahigh Q/V, ultrasmall footprint are an idea platform for the monolithic integration and extending the capability of these optical devices, in which the key is that the PCNCs can greatly improve light-matter interaction. The optical devices show good characteristics and high-volume production, which are expected to benefit large-scale photonic-integrated circuits on silicon in the near future. Furthermore, photonic integration should not be required to surpass electronic integration, but its unique advantages should be used as a supplement to electronic integration to solve problems that electronic integration cannot solve in the More-than-Moore era.

This work was supported by the National Key R&D Program of China (Grant No. 2016YFA0301302 and No. 2018YFB2200401), the National Natural Science Foundation of China (Grant Nos. 11974058, 11825402, 11654003, 61435001), Beijing Academy of Quantum Information Sciences (Grant No. Y18G20), Key R&D Program of Guangdong Province (Grant No. 2018B030329001), Beijing Nova Program (Grant No. Z201100006820125) from Beijing Municipal Science and Technology Commission, Fundamental Research Funds for the Central Universities (Grant No. 2018XKJC05) and the High Performance Computing Platform of Peking University.